The Chemical Physics of Solid Surfaces, Vol. 10: Surface alloys and alloy surfaces

THE CHEMICAL PHYSICS OF SOLID SURFACES T H E C H E M I C A L P H Y S I C S OF S O L I D S U R F A C E S Volume 1...

Author: D. P. Woodruff

44 downloads 592 Views 29MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

THE

CHEMICAL

PHYSICS

OF SOLID

SURFACES

T H E C H E M I C A L P H Y S I C S OF S O L I D S U R F A C E S

Volume 1 CLEAN SOLID SURFACES Volume 2 A D S O R P T I O N AT S O L I D S U R F A C E S Volume 3 CHEMISORPTION SYSTEMS Volume 4 F U N D A M E N T A L S T U D I E S OF H E T E R O G E N E O U S CATALYSIS Volume 5 S U R F A C E P R O P E R T I E S OF E L E C T R O N I C M A T E R I A L S Volume 6 COADSORPTION, PROMOTERS AND POISONS Volume 7 PHASE TRANSITIONS AND ADSORBATE R E S T R U C T U R I N G AT M E T A L S U R F A C E S Volume 8 G R O W T H A N D P R O P E R T I E S OF U L T R A T H I N E P I T A X I A L LAYERS Volume 9 OXIDE S U R F A C E S Volume 10 SURFACE ALLOYS AND ALLOY SURFACES

TH E CH EMICAL PHYSICS OF SOL! D SU RFACES

EDITED D.P.

BY

W O O D R U F F

B.Sc. (Bristol), Ph.D., D.Sc. (Warwick)

Professor of Physics, University of Warwick

VOLUME

I0

SU RFACE ALLOYS AN D ALLOY SU RFACES

2002

ELSEVIER AMSTERDAMSAN DIEGO

BOSTON - SAN

- LONDONFRANCISCO

NEW - SINGAPORE

YORK-

OXFORD - SYDNEY-

- PARIS TOKYO

ELSEVIER SCIENCE B.V. Sara Burgerhartstraat 25 P.O. Box 211,1000 AE Amsterdam, The Netherlands 92002 Elsevier Science B.V. All rights reserved. This work is protected under copyright by Elsevier Science, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier Science via their homepage (http://www.elsevier.com) by selecting 'Customer support' and then 'Permissions'. Alternatively you can send an e-mail to: [email protected], or fax to: (+44) 1865 853333. In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (+1) (978) 7508400, fax: (+1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London Wl P 0LP, UK; phone: (+44) 207 631 5555; fax: (+44) 207 631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of Elsevier Science is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission ofthe Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier Science Global Rights Department, at the mail, fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made.

First edition 2002 Library of Congress Cataloging in Publication Data A catalog record from the Library of Congress has been applied for.

British Libary Cataloguing in Publication Data A cataloque record from the British Library has been applied for.

ISBN 0-444-51152-0 (Vol. 10) ISBN 0-444-41971-3 (Series) O The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.

Contributors to Volume I0

D.A. ADAMS

Institute of Physics and Astronomy, University of Aarhus, DK-8000 Aarhus C, Denmark

J.N. ANDERSEN

Department of Synchrotron Radiation Research, Institute of Physics, Lund University, S-223 62 Lund, Sweden

C.J. BADDELEY

School of Chemistry, University of St Andrews, St Andrews, Fife KY 16 9ST, UK

U. BARDI

Dipartimento di Chimica, Universith di Firenze, Via G. Caponi 9, 50014 Firenze, Italy

C.J. BARNES

School of Chemical Sciences, Dublin City University, Dublin 9, Republic of Ireland

J.C. BERTOLINI

Insitut de Recherches sur la Catalyse- CNRS, 2, avenue Albert Einstein, F-69626 Villeurbanne Cedex, France

G. BOZZOLO

Ohio Aerospace Institute, 22800 Cedar Point Rd., Cleveland, OH 44142, USA and NASA Glenn Research Center, Cleveland, OH 44135, USA

J.E. GARCES

Ohio Aerospace Institute, 22800 Cedar Point Rd., Cleveland, OH 44142, USA and Centro Atomica Bariloche, 8400 Bariloche, Argentina

J. HRBEK

Department of Chemistry, Brookhaven National Laboratory, Upton, New York 11973, USA

Y. JUGNET

Insitut de Recherches sur la Catalyse- CNRS, 2, avenue Albert Einstein, F-69626 Villeurbanne Cedex, France

G.L. KELLOGG

Sandia National Laboratories, Albuquerque, NM 87185-1415, USA

M. POLAK

Department of Chemistry, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel

H. NIEHUS

Humboldt-Universit~t zu Berlin, Institut fur Physik, Oberfl~chenphysik und Atomsto6prozesse, InvalidenstralSe 110, D- 10115 Berlin, Germany

J.K. NORSKOV

Centre for Atomic-scale Materials Physics and Department of Physics, Technical University of Denmark, DK-2800 Lyngby, Denmark

J.A. RODRIGUEZ

Department of Chemistry, Brookhaven National Laboratory, Upton, New York 11973, USA

vii A.V. RUBAN

Centre for Atomic-scale Materials Physics and Department of Physics, Technical University of Denmark, DK-2800 Lyngby, Denmark

L. RUBINOVICH

Department of Chemistry, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel

M. SCHMID

Institut fur Allgemeine Physik, Technische Universit~it Wien, A- 1040 Wien, Austria

H.L. SKRIVER

Centre for Atomic-scale Materials Physics and Department of Physics, Technical University of Denmark, DK-2800 Lyngby, Denmark

S. SPELLER

Research Institute for Materials, University of Nijmegen, Toernooiveld 1, 6525 ED Nijmegen, The Netherlands

P. VARGA

Institut fur Allgemeine Physik, Technische Universit~it Wien, A- 1040 Wien, Austria

E. VLIEG

NSRIM Department of Solid State Chemistry, University of Nijmegen, Toernooiveld 1, 6525 ED Nijmegen, The Netherlands

D.P. WOODRUFF

Physics Department, University of Warwick, Coventry CV4 7AL, UK

...

Vlll

Preface During the late 1960s and 1970s the commercial availability of ultra-high vacuum (UHV) systems allowed the development of a plethora of new techniques which were devised to probe materials in a surface-specific fashion, and this in turn led to the creation of modem surface science; the study of the structural, electronic and chemical properties of extremely well-characterised surfaces on an atomic scale. When David King and I first conceived this series of volumes in the later 1970s our objective was to recognise the growing maturity of this new scientific discipline which was already starting to apply these techniques in a combined fashion to understand surface processes. In the 20 years since the first volume was published, this perception has certainly proved to be well-founded, and while new techniques have continued to appear, they have rapidly been assimilated into the general armoury of methods (the increasing pervasiveness of scanning probe microscopies is evident again in the current volume), and it is the combination of methods which has proved most effective. The topic of the present volume, Sufluce Alloys and Alloy Sufluces, provides new insights into a mixture of old and new problems. The surfaces of bulk alloys have long been known to be of practical interest for their chemical properties, be it novel activity or selectivity to certain reactions in a way which differs from the constituent elements in isolation or novel passiveness to corrosion. It has also long been known that the surface composition of such alloys commonly differs from that of the underlying bulk, and some of the basic thermodynamics of this segregation is far from new. Nevertheless, our understanding of these chemical and physical phenomena is far from complete, and the application of surface science methods to investigate these phenomena is a manifestation of a general trend to the study of surfaces of increasing complexity. A phenomenon which has been fully recognised far more recently is that of surface alloy formation - the intermixing of substrate atoms and adatoms in the outermost atomic layer, or few atomic layers of a solid, to form a stable ultra-thin alloy phase which may be in equilibrium with an essentially elementally pure substrate and may even involve the intermixing of elements which are immiscible in the bulk. There are now many examples of these surface alloys, and quite complex alloying and de-alloying behaviour may be observed as a function of surface stoichiometry. It is this combination of surface alloys and alloy surfaces which is addressed in the chapters of this volume. The first three chapters, by Ruban, Skriver and N~rskov,by Bozzolo and Garces, and by Polak and Rubinovich, are concerned with different theoretical descriptions of some of these phenomena from which one gains physical

1X

insight and predictive powers into the mixing, segregation and ordering phenomena. There follows a series of chapters based on experimental studies of surface composition, ordering and structure based on a variety of different materials and techniques. Schmid and Varga show, in particular, the remarkable power of scanning tunnelling microscopy, when chemical discrimination of the elemental components of an alloy surface is possible, to gain an atomic-scale understanding of some of the effects of segregation and ordering. Kellogg describes phenomena particularly in the Cu/Pb system based on information from many techniques but including the application of low energy electron microscopy to follow the processes of alloying and de-alloying. Speller and Bardi and Adams and Andersen describe the results of extensive structural studies of Pt-Sn alloys and Al-alkali surface alloy phases respectively. The latter systems, involving mixtures of superficially 'simple' metals, show a remarkably rich range of ordering and intermixing phenomena. Woodruff and Vlieg describe some detailed quantitative structural aspects of some metallic surface alloys including the systematics of surface alloy layer atomic rumpling and associated effective atomic radii, while Barnes surveys the structural aspects of surface alloys on Cu(100). Finally, in this group of chapters, Niehus discusses some results on the surface order of bulk alloys, especially Cu3Au and related systems, and ordered overlayers on these surfaces. The final group of chapters by Bertolini and Jugnet, Rogriguez, Hrbek and Baddeley address issues more directly related to the chemical properties of these surfaces, the first three of these chapters being concerned directly with the relationship of the nature of the surface alloys (and alloy surfaces) and their reactivity, while Baddeley turns the problem round in addressing the issue of adsorbate-induced modification of surface segregation; not only does the alloy surface modify the reactivity, but also the reaction modifies the surface alloy. March 2002

D.P.Woodruff

Contents Preface

viii

Chapter 1 (A.V. Ruban, H.L. Skriver and J. Norskov)

Local equilibrium properties of metallic surface alloys 1. Introduction 2. Surface energy 2.1 Monoatomic solids 2.2 Alloys 3. Stable surface alloy configurations 4. Generic classes of surface alloying 4.1 Mixing energy 4.2 Segregation energy 5. General trends for the surface mixing energies in transition metal alloys 6. General trends for the surface segregation energies in transition metal alloys 7. Island formation: multilayer versus monolayer growth 8. Bulk-type ordered surface alloys 9. Alternative ordered structures on the surface Acknowledgement References

1 2 2 5 7 8 9 10 11 13 15 19 23 27 28

Chapter 2 (G. Bozzolo and J.E. Garces)

Atomistic modelling of surface alloys 1. Introduction 2. The BFS method 2.1 Calculation of the BFS strain energy 2.2 Calculation of the BFS chemical energy 2.3 The BFS reference state in surface alloys vs. epitaxial growth 3. BFS modelling of surface alloys 3.1 Calculational procedure 3.2 Au/Ni(110) 3.3 Pd/Ni(110) 3.4 Pd/Cu(100) 3.5 Pd/Cu(110) 3.6 Cu/Pd(110) 3.7 Pt/Cu(100) 3.8 Au/Cu(100) and Au/Cu(110) 3.9 Cu/Ni(110) 3.10 (Cu, Au)/Ni(110) 4. Conclusions Acknowledgements References

30 36 39 44 45 47 48 51 60 62 68 68 68 72 78 79 82 83 83

Chapter 3 (M. Polak and L. Rubinovich) Alloy surface segregation and ordering phenomena: recent progress 1. Overview 2. Segregation in multi-element alloys 3. Surface segregation in ordered alloys 3.1 Prediction of order/segregation interplay by means of a simple model 3.1.1 Equiatomic binary alloys 3.1.2 Non-equiatomic binary alloys 3.2 Case studies 3.2.1 Compositional variations in Cu3Au(100) and CuaPd(100) 3.2.2 Surface order in PtsSn(111) and Co3Pt(111) 3.2.3 Segregation characteristics of aluminide surfaces 4. Segregation in a bi-phase binary alloy 5. Summary References

86 90 96 97 97 99 101 101 104 105 109 113 115

Chapter 4 (M. Schmid and P. Varga) Segregation and surface chemical ordering- an experimental view on the atomic scale 1. Introduction 2. Chemical discrimination on bimetallic surfaces with atomic resolution by STM 2.1 True topographic effect 2.2 Difference in local electronic density of states 2.3 Tip-surface interaction 3. Segregation on alloys- surface and subsurface composition 3.1 Segregation 3.2 Preferential sputtering and segregation in the altered layer 4. Chemical ordering of alloy surfaces 4.1 Bulk chemical order 4.2 Fundamentals of surface chemical order 4.3 Chemical order of close-packed alloy surfaces 4.4 fcc(100) surfaces 4.5 Site-specific segregation 5. Implications for adsorption on alloys 5.1 Chemical affinity 5.2 The ensemble effect 5.3 The ligand effect 6. Conclusions Acknowledgement References

118 120 121 123 125 127 127 128 130 130 131 134 140 141 144 144 145 147 148 149 149

xii

Chapter 5 (G.L. Kellogg) Surface alloying and de-alloying of Pb on single-crystal Cu surfaces 1. Introduction 2. Experimental and theoretical techniques 2.1 Experimental 2.2 Theoretical 3. Atomic structure, surface alloying and de-alloying 3.1 Pb on Cu(111) 3.2 Pb on Cu(100) 3.3 Pb on Cu(110) 3.4 Pb on stepped surfaces of Cu 3.5 Summary of Pb surface alloy and overlayer structures on single-crystal surfaces of Cu 4. Concluding remarks Acknowledgements References

152 154 154 157 158 158 165 172 175 178 178 180 180

Chapter 6 (S. Speller and U. Bardi) Surface alloys and alloy surfaces: the platinum-tin system 1. Introduction 2. Methods 3. The platinum-tin system 3.1 Low index surfaces of the Pt3Sn alloy 3.1.1 Pt3Sn(111) 3.1.2 Pt3 Sn(001) 3.1.3 Pt3Sn(110) 3.2 Surface alloys obtained depositing tin on platinum surfaces 3.2.1 Sn-Pt(111) 3.2.2 Sn-Pt(100) 4. Discussion 4.1 Surface atomic structure of the bulk Pt3Sn alloys 4.2 Defects and disorder on Pt3Sn alloy surfaces 4.3 Multilayer and single layer surface alloys 5. Conclusion Appendix: Notes on nomenclature References

184 185 190 191 191 197 202 207 207 209 210 212 215 217 219 220 221

Chapter 7 (D.L. Adams and J.N. Andersen) Alkali-aluminum surface alloys 1. Introduction 1.1 Background 1.2 Present work 2. Experimental methods 2.1 LEED measurements 2.2 LEED analysis 2.3 The surface structures of clean AI(111), (100) and (110)

225 225 226 228 228 228 229

xiii 2.4 Core-level measurements 3. Adsorption on AI(111) 3.1 AI(111)-(2x2)-Rb and Cs phases formed at 100 K 3.2 AI(111)-(~/3x~/3)R30~ Rb and Cs phases formed at 100 K 3.3 AI(111)-(4x4)-Na phase formed at 100 K 3.4 AI(111)-(~/3x~/3)R30~ Na, K and Rb phases formed at 300 K 3.5 AI(111)-(2~/3x2~/3)R30~ phase formed at 300 K 3.6 AI(111)-(2x2) -Na phase formed at 300 K 3.7 Ternary surface alloys formed by coadsorption on Na and K, Rb or Cs on AI(111) at 300K 4. Adsorption on AI(100) 4.1 AI(100)-(2x2) -Na phase formed at 100 K 4.2 Al(100)-(~/5x~5)R26.6~ phase formed at 240 K 4.3 AI(100)-c(2x2)-Li and Na phases formed at 300 K 4.4 Al(100)-c(2x2) -2Li phase formed at 400 K 5. Adsorption on AI(110) 5.1 AI(110)-c(2x2)-Li and Na phases formed at 300 K 5.2 AI(110)-(4xl)-3Na phase formed at 300 K 6. Phase transitions 6.1 AI(111)-(~/3xx/3)R30~ and Rb 6.2 AI(100)-c(2x2)-Na 6.3 AI(100)-(q5xq5)R26.6~ 7. The role of DFT calculations 7.1 AI( 111)-(q3xq3)R30~ and K 7.2 Al(111)-(2x2)-Na 7.3 Al(100)-(~/5x~/5)R26.6~ 7.4 AI(100)-c(2x2)-Na 7.5 Al(100)-c(2x2)-Li 7.6 AI(100)-c(2x2)-2Li 8. Summary and conclusions Acknowledgements References

229 233 234 235 235 237 240 241 243 245 246 246 247 248 253 253 254 257 258 261 262 264 264 266 267 268 268 269 270 273 273

Chapter 8 (D.P. Woodruff and E. Vlieg) The structure of surface alloy phases on metallic substrates 1. Introduction 2. Case studies 2.1 Cu(111)/Sb and Ag(111)/Sb: interracial stacking faults 2.2 Ni(111)/Pb: a case of strongly suppressed surface alloy rumpling 2.3 Mn and non-magnetic metals on Cu(100), Ni(100) and Pd(100): effect of local magnetic moments 2.4 Surface alloys formed by Sn on Cu, Ni, Pt and Rh surfaces: effect of changing substrate lattice parameter and surface orientation on rumpling amplitude 3. Interatomic distances in surface alloys 4. More complex systems 5. Conclusions References

277 278 278 286 288 291 293 298 301 302

xiv

Chapter 9 (C.J. Barnes) Surface alloy formation on Cu{100} 1. Introduction 2. Cu{ 100 }-c(2x2)-X (X=Au,Pd,Mn) surface alloys 2.1 Geometric and electronic structure 2.2 Growth mechanism of Cu{ 100}-c(2x2) surface alloys 3. Surface alloy formation upon Co, Fe and Ni adsorption 4. Surface alloy formation upon alkali and alkaline earth metal adsorption 4.1 The Cu{ 100}/Li surface alloy: the coverage dependent (2xl)---)(3x3)---~(4x4) transition 4.2 The Cu { 100 }-c(2x2)-Mg surface alloy 5. De-alloying transitions: adsorption of group IIIA, IVA and VA metals 5.1 The Cu { 100 }/Pb system 5.2 De-alloying in the Cu{ 100}/Bi system 5.3 Surface alloy formation in the Cu { 100 }/In and Sn systems 5.4 De-alloying transitions for transition metal adsorbates 6. Underlayer 2D alloys and overlayer to underlayer transitions 6.1 The Cu{ 100 }-c(2x2)-Pd overlayer to underlayer transition 6.2 Cu{ 100 }/Pt: the Cu{ 100 }-c(2x2)-Pt underlayer alloy 6.3 Cu{ 100 }/Ir: the unusual case of p(2xl) underlayer formation 7. Formation of ordered multilayer alloys 7.1 The Cu { 100 }-p(2x2)- 1 ML Pd surface alloy 7.2 The Cu{ 100 }-c(2x2)-Pt multilayer alloy 7.3 The Cu{ 100}-(4x2)-pgg-Mn structure 8. Conclusions Acknowledgements Reference

305 308 308 315 322 326 326 331 333 334 339 341 343 345 345 347 349 351 351 355 356 358 359 359

Chapter 10 (H. Niehus) Surface and sub-surface alloy formation connected with ordered superstructures

1. Introduction 2. Surfaces of ordered bulk alloys 2.1 Preparation dependent surface composition: NiAI 2.2 Surface properties of alloys with identical surface composition 2.2.1 Cu3Au(110) 2.2.2 Cu3Au(100) 3. Surface alloys of bulk immiscible constituents 3.1 Sub-surface alloy formation: iridium on Cu(100) 3.2 Intermixing versus phase separation: copper on Ir(100)-(5xl) 4. Alloy surfaces as substrates for ordered superstructures 4.1 Vanadium on Cu3Au(100) 4.2 Vanadium oxide on Cu3Au(100)-O 5. Summary Acknowledgement References

364 366 366 372 373 375 378 378 389 393 394 396 399 400 400

XV

Chapter 11 (J.C. Bertolini and Y. Jugnet) Surface structure and catalytic activity of palladium overlayers with 1,3butadiene hydrogenation 1. Introduction 2. Experimental approach 3. The 1,3-butadiene hydrogenation reaction 4. Surface and reactivity of Pd based alloy surfaces 4.1 General points 4.2 Surface composition and reactivity of Pd5Ni95 and Pd5Pt95 polycrystals 4.3 Influence of the surface orientation on reactivity 4.3.1 A solid solution in the whole range of composition: PdsNi92(111 ) and (110) 4.3.2 A system with a tendency to ordering: Pd50Cu50(111) and (110) 5. Surface and reactivity of Pd deposits 5.1 Pd in compression on Ni and Cu 5.1.1 Case thermodynamically favouring A on B: Pd on Ni Pd on Ni(111) Pd on Ni(110) 5.1.2 Case of A on B unfavourable: Pd on Cu(110) 5.2 Pd in tension on Au(110) 6. Summary and conclusion Acknowledgements References

404 407 409 413 413 414 418 418 421 423 423 423 423 424 428 430 433 434 435

Chapter 12 (J.A. Rodriguez) Electronic and chemical properties of palladium in bimetallic systems: how much do we know about heteronuclear metal-metal bonding? 1. Introduction 2. Photoemission studies 3. Thermal desorption studies 4. CO chemisorption studies 5. Models for bimetallic bonding 6. Theoretical studies 6.1 Charge redistribution in bimetallic bonding 6.2 Core-level and valence-band shifts 6.3 CO chemisorption 7. Conclusion Acknowledgement References

43 8 439 445 448 454 455 455 458 460 462 462 462

XV1

Chapter 13 (J.A. Rodriguez and J. Hrbek) Interaction of sulphur with bimetallic surfaces: effects of structural, electronic and chemical properties 1. Introduction 2. Repulsive interactions between gold and sulphur on transition metal surfaces 3. Interaction of sulphur with Ag/Ru(0001) and Cu/Ru(0001) 4. Admetal promoted sulphidation of Pt(111) and Mo(110) 5. Bimetallic bonding and the prevention of sulphur poisoning 6. Conclusion Acknowledgement References

466 467 475 482 488 492 492 492

Chapter 14 (C.J. Baddeley) Adsorbate induced segregation at bimetallic surfaces 1. Introduction 1.1 Bimetallic surface chemistry - traditional ideas 1.1.1 Ensemble effects 1.1.2 Electronic effects 2. Adsorbate induced segregation 2.1 Thermodynamic considerations 3. Techniques for characterising adsorbate induced segregation 3.1 Photoelectron spectroscopies 3.1.1 X-ray photoelectron spectrscopy (XPS) and Auger electron spectroscopy (AES) 3.1.2 Photoelectron microscopy (PEEM, SPEM) 3.2 Ion scattering spectroscopies 3.2.1 Low energy ion scattering (LEIS) 3.2.2 Medium energy ion scattering (MEIS) 3.3 X-ray absorption spectroscopies 3.3.1 Extended X-ray absorption fine structure (EXAFS) 3.4 Vibrational spectroscopies 3.4.1 Infra-red spectroscopy 3.5 Other techniques 3.5.1 Scanning tunnelling microscopy (STM) 3.5.2 Low energy electron diffraction (LEED) 3.5.3 Nuclear magnetic resonance (NMR) 4. Conclusions References

508 508 509 510 515 515 516 516 517 517 521 522 522 523

Index

527

495 495 497 499 500 500 505 505 505

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces D.P. Woodruff, (Editor)

Chapter 1

Local equilibrium properties of metallic surface alloys A. V. Ruban, H. L. Skriver, and J. K. NOrskov Center for Atomic-scale Materials Physics and Department of Physics, Technical University of Denmark, DK-2800 Lyngby, Denmark INTRODUCTION A great variety of structures are formed after deposition of one (or several) metals on the surface of another [1]. The deposited metals may form alloys with each other or they may form islands with some microstructure [7,8] with the substrate in the first or deeper layers [ 1-6]. Alloy formation at the surface may be observed even in those cases where there is phase separation in the bulk [9-11 ]. If the size mismatch between the deposited and substrate atoms is large, misfit dislocation structures may be formed [ 12-14]. A detailed theoretical prediction of such structures from very general considerations based on first-principles total energy calculations, is very demanding, since it includes the kinetics of the deposition, growth, and diffusion in the system under the relevant external conditions. Nevertheless, there are some surface alloys, the structures of which, although being metastable, mostly reflect the thermodynamics of the ground state of the system. This is so, since at ordinary temperatures the entropy driven diffusion of the deposited material into the bulk is very slow, and, hence, at time scales which are long in terms of surface kinetics, but short in terms of bulk diffusion, a local equilibrium may be established in the surface region [ 15,16] In such cases a local-equilibrium structure may be obtained theoretically by minimization of the free energy of the system under the constraint of a fixed alloy composition in the surface region [8,17-24]. Although this approach is very similar to the one used for bulk systems, it should be modified due to the specific features introduced by the surface. First of all, since the structure of the underlying bulk system is fixed, it acts as the source of an external field for the surface alloy, creating, for instance, epitaxial strain. Secondly, since the surface is an open system, it allows the formation of a great variety of different structures, which may not have any connection at all to the crystal structure of the substrate. Finally, the surface is a spatially inhomogeneous system, and thus different alloy components have their own

specific preference for different parts of the surface region, which will cause the segregation of alloy components to the various layers. The surface segregation phenomena play a major role in establishing the composition of the surface alloy in each layer, and therefore any thermodynamic-like consideration of the surface alloy formation should start by investigating the segregation behavior of the alloy components. This behavior is in fact naturally incorporated into the theory if, instead of the free energy of the surface region, the surface free energy is considered [15]. In general the surface energy is a complex function of the structure, composition, and alloy configuration in each layer of the surface region, and thus the optimization to find the equilibrium state should be made in the phase space of all these parameters. Moreover, for different amounts of the deposition element there may exist different equilibrium phases or mixtures of them. However, to categorize in a simple way the behavior of surface alloys the surface energy of a monolayer of a pseudomorphic random alloy of the deposited and substrate elements can be used. The main advantage of the surface energy curve of a random surface alloy is the fact that its general features can be described by only two physically well-determined parameters" the solution and segregation energies of the deposited element in the surface layer of the substrate. Four different combinations of theses two parameters lead to four generic cases of surface alloy behavior. Another advantage of this kind of theoretical consideration is the fact that such a surface energy curve, or the corresponding solution and segregation energies, may easily be obtained in first-principles or ab initio calculations based on density-functional theory [15,24,25] using only the atomic numbers of the alloy components and the crystal structure of the bulk as input parameters, which gives a reliable first insight into possible surface alloy behavior. In this chapter we discuss the general trends for the formation of transition metal surface alloys obtained by first-principles calculations [15,24,25]. We also present some examples where the behavior of the surface alloy appears to be more complicated than expected from the simple considerations based on the knowledge of the surface energy curve of a monolayer of a random alloy and the bulk phase diagram. In all cases considered here we assume that there is no exchange of atoms through the vacuum region due to either evaporation or condensation, since such processes do not affect mass transfer towards the surface region in most metallic systems at ordinary temperatures, where a surface alloy may exist for a sufficiently long time.

2. S U R F A C E E N E R G Y 2.1.Monoatomic solids The surface energy is the energy required to create one unit of surface area. Such a process is schematically illustrated in Fig.l, and its energy is the difference between the total energies of system 2, in which the additional surface area A has been created, and system 1, which is the initial state. Thus the surface energy, u , is 1

2

Y--"~(Eto,-

El

tot)

9

(1)

Here, the surface energy is determined per surface area, and the total energies, i Eto ~ , correspond to the complete systems (extensive quantities). In theoretical calculations another but equivalent definition of the surface energy is usually used, i.e., 1 (E~ury Et, Ulk) Y=-, ,o, - ,o,, 9

(2)

ns

Here, E ~ur tot is the total energy of the surface region, which usually consists of several layers the crystal and electronic structures of which are different from ~ is the total energy of a bulk region equivalent in their bulk counterparts, E tot size to the surface region, and n, the number of atoms at the surface. Thus, the surface energy in (2) is normalized per number of atoms at the surface.

Fig. 1. The surface energy of a monoatomic solid. A is a new surface created due to a change of the form of the crystal under the conservation of the number of atoms. Light grey color schematically indicates the surface region.

Although it is very difficult to measure surface energies, they may today relatively easily and reliably be calculated from first-principles [26,27], even in the cases of quite open surfaces [28]. In Fig. 2 we show the surface energies of metals in the 4d-series of the Periodic Table obtained by first-principles calculations [27]. The energies in the figure exhibit a parabolic-like behavior as a function of the atomic number. Such a behavior is explained in terms of the occupation of the valence d-band by the Friedel model [29,30] in which the surface energies follow the same trend as the corresponding cohesive energies and can be estimated from

11 1

WNd(IO-N a)

u = 2"-0- 1 -

(3)

,

where W is the width of the d-band, N d the number of valence d-electrons, and Zsand zb are the coordination numbers of the atoms at the surface and in the bulk, respectively. It follows from (3) that the transition metals with a half-occupied d-band have the highest surface energies, the magnitude of which increases down the Periodic Table from the 3d to the 5d metals due to a corresponding increase in the d-band width [31 ]. Formula (3) also shows the dependence of the surface

l::: 1.5

~ v

I

>, 1.0 (D (--

(D

o 0.5 '1:: r 0.0

hcp bcc bcc hc:p hcp fcc

Rb Sr

Y

fcc

fcc hcp

Zr Nb Mo Tc Ru Rh Pd A cl Cd

Fig. 2. The energies of the most closed-packed surfaces of the metals in the 4d transition series obtained from first-principles [27].

energy of a transition metal on the surface coordination number z~. With decreasing z,, or with increasing number of "broken" bonds, the surface energy increases, and thus the surface energy of open surfaces may be quite large. The later is a consequence of the localized bonding provided by valence d-states. In contrast, for the "simple" metals the free-electron like contribution to the bonding dominates, making their surface energies much less sensitive to the surface orientation. 2.2. Alloys. In the case of alloys the composition of the surface region may differ from that of the bulk and therefore (2) should be modified to take into account the energy of an exchange of atoms between the bulk and the surface region as sketched in Fig. 3. As usual, we assume that the bulk is infinitely large compared to the surface region and therefore such an exchange does not influence the composition of the bulk. Let us consider a binary A~_cBc alloy. The energy of removing a B atom is minus the chemical potential of the B-component, -/~ B 9At T = 0 K, when there is no contribution from the entropy term, E to t

-UB = - ' - - - - ~

,

(4)

ON B

where NB is the number of B atoms in the bulk. Thus, the surface energy of an alloy is

Fig. 3. Schematic exchange of A and B atoms between the surface region and the bulk.

1

=

--( ns

surl Etot

bulk -Etot

-

Z i = A B, I.IiA N i )

9

(5)

As in the case of a monoatomic solid E `ury is the total energy of the surface tot region having a given composition and configuration, A N i the number of A and B atoms which have been exchanged between the surface region and the bulk. For a binary A,_~Br alloy A N A = - A N B if no vacancies are formed in the surface region, and thus using the concentration variable, c =c B = N B I N (N = NA + NB), (5) can be rewritten in the form: 1

surl

}" = --- ( E tot t/s

bulk

- -

E,o , ) - n l u A c

9

Here, n I is the number of layers in the surface region, A c = c , - c

(6)

the

difference between the concentration c, in the surface region and the concentration in the bulk, and /~=/2 B--/./A the effective chemical potential of the bulk alloy, which may be determined by ,,-.,(O)-bulk

U =

0 lgL-tot Oc

Ebulk ---

1

0 __tot

,

~

(7)

NOc

where the first energy is per atom. At non-zero temperatures one should instead of the total energy of the system consider its free energy by adding the corresponding entropy contribution - T S . In general, it is a quite complicated problem to obtain the chemical potential since the concentration derivative should be taken along the minimal path in the phase space of short and long-range order and other parameters which define the equilibrium alloy configuration and structure at each concentration. However, this problem is greatly simplified in the case of a dilute alloy, where all the configurational effects become negligible, because to lowest order they are proportional to c 2. In this case, which in fact corresponds exactly to the deposition of one element (B) on the surface of another (A), the effective chemical potential is defined as u =

cgE(t~l-bulk(Al_cnc ) Oc

,

(8)

where E(~ l B ) is the total energy (per atom) of a random A~_cB~ alloy and the derivative is taken at c = O. tot

-

c

3. S T A B L E S U R F A C E A L L O Y C O N F I G U R A T I O N S A small amount of material deposited on a pure host crystal will always be metastable at non-zero temperatures, since the gain in entropy by dissolving into the bulk, which is roughly A S = k l n ( N b l N s ) , where Nb and N~are the number of sites in the bulk and at the surface, respectively, will drive the deposited material away from the surface. However, as has already been mentioned, near room temperature bulk diffusion in a metal is extremely slow, and a local equilibrium is usually established in the surface region. The local equilibrium surface alloy configuration and structure may be found by minimization of the surface free energy, or if several different phases may exist, by finding a convex hull of the lowest free energies of different phases at different alloy compositions (at T=0), or more generally by a common-tangent construction which is completely analogous to the usual treatment of the bulk systems. The procedure is illustrated in Fig. 4. Given the surface energy curve in Fig. 4, the surface alloy with an overall concentration Co of atoms deposited at the surface will, instead of forming a homogeneous solution, H, separate into two distinct phases, say S and P, with concentration Cs and Cp, respectively, if Cs < C o < C e . The relative fraction of the S and P phases is determined by the lever rule as c e - c 0 to C o - C s , which also implies that the energy of the phase equilibrium of S and P will be a straight line between the points S and P in the surface energy diagram.

$

H

p

S 't=

I I I I I I !

0

Cs

Co

c

Cp

1

Fig. 4. Sketch of a common-tangent construction for the surface energy of an alloy, c is the coverage of the deposited material.

Such a phase diagram has, for instance, recently been calculated for a Mn/Cu(111) surface alloy [20]. Although these calculations include only the simplest alloy configurations in the limit of an infinitely large pseudomorphic surface it gives a better understanding of the initial stages of surface alloy formation during deposition growth of Mn on Cu(111), and, in particular, the formation of islands of a ~ x ~ Cu2Mn ordered alloy. 4. G E N E R I C CLASSES O F S U R F A C E A L L O Y I N G To categorize in a simple way the behavior of surface alloys we will use the so-called surface energy curve which is the surface energy of a pseudomorphic monolayer of a random AcB~_c alloy on the surface of B Although such a surface alloy is almost never realized in practice, it is quite useful in theoretical considerations. First of all, the surface energy of such an alloy may easily, and quite accurately, be determined by first-principles calculations [15,24]. Secondly, it allows one to categorize the deposition behavior in a simple way, and to predict some general features of real surface alloys. In Fig. 5 we show the surface energy (per substrate atom) for four different 0.8

AgcPt,_JPt(111)

Ag~Cul_JCu(100) !

i

i

i

I

!

I

I

I

0.9 0.7

0.7

0.6

0.5

E

o

0.5 0.00

0.25 0.50

0.75

1.00

0.3

0.00

0.25 0.50 0.75 1.00

>, t._

(i) t(D o 't:: :D r

1.4

RucAul_JAu(111 )

Pt~Cu,JCu(111 ) i

i

i

0.9

1.2

0.8

1.0

0.7

0.8 0.6

0.( 10 0.25

0.50

0.75

1.00

0.6

0.00 0.25 0.50 0.75 1.00

C

Fig. 5. Surface energy curves for a monolayer of a random alloy on surfaces of pure metals.

systems obtained by first-principles calculations [15]. These surface energy curves naturally fall into four distinct generic classes which may clearly be recognized by their curvatures and slopes. For instance, the surface energy curve of Agr has a positive curvature and negative slope, while the surface energy of Ru~Au~_c/Au(lll) exhibits negative curvature and positive slope. In this section we show that these two features of the surface energy curve in fact correspond to the mixing and segregation energies of the deposited element in the surface of a substrate. These energies may easily be obtained by first-principles calculations and thus the general trend of the surface alloying can be established.

4.1. Mixing energy The surface alloy mixing energy is determined similarly to case of bulk alloys as A BI

Cmi x "- y

~

/B

-~

AIB

--C y

A B~ IB

where

y~

-~

--(l--c)

y

B

,

(9)

is the surface energy of a monolayer of a random AcB~-c

alloy on a B substrate,

yA/B the surface energy of an infinite pseudomorphic

monolayer of A on B, and yB the surface energy of B. Thus, the straight line which connects yA/B and yB in Fig. 5 represents the energy of the standard state, which is a mixture of infinitely large islands of B and A on B, given by the last two terms in (9). It is obvious from the consideration in the A B~ IB previous section that if the surface energy curve y ~ -~ goes above the standard line there should be a phase separation of the surface alloy into islands of pure B and A elements in the surface layer. In contrast, if the surface energy curve goes below the standard state line, then alloying will occur on the surface. Since, the surface energy curve is obtained for a pseudomorphic alloy on a fixed lattice of the substrate, its behavior can be related directly to the type of so-called effective interactions which are responsible for the ordering of A and B atoms on the surface. That is, if the multisite interactions are small in the system, which is usually the case for metallic alloys on a fixed lattice, the mixing energy can be written in terms of pair potentials between alloy components, v~AA ,viAB ' and v i BB for each coordination shell i at the surface as

1

Emix

=---c(1-c) 2

where

Zi

E

1

i

z (v~A+v88--2vAB)=----C(1--C) i i 2

E i

ziV

i

'

(lo)

is the coordination number of the i-th coordination shell at the surface

10

and Vi the so-called effective interactions. Since the nearest-neighbor interactions are usually the strongest, the mixing energy is roughly proportional to minus the effective interaction at the first coordination shell. Thus, if the mixing energy is negative, i.e. the surface random alloy is stable against separation into islands of pure A and B elements, the effective interaction at the first coordination shell is positive, which means that a surface alloy has a tendency towards ordering. Such an ordering usually takes place at low temperatures. This is indeed the case for the two systems, presented in Fig. 5: AgcCUl_c on Cu(100) [12,15] and Pt~CUl~on Cu(111) [ 15], while the deposition of Ag on Pt(111) and Ru on Au(111) should lead to the formation of islands of the deposited element and the substrate. Although there appears to be no experimental data for the Ru/Au(111) system, the surface alloy structures of the Ag/Pt(111) have been thoroughly investigated experimentally [32-35], and island formation is wellestablished. In fact Ag islands have a finite size and they may form different, droplet- or stripe-like, structures, exhibiting a quite fascinating behavior with temperature, which unfortunately is beyond the scope of the present considerations. 4.2. S e g r e g a t i o n energy

Another distinctive feature of the surface energy curve is its slope. In fact the slope of yAB,_/B is simply the segregation energy of the deposited element to the surface layer at a given concentration: ~r,,ABj_ /B e segr = ~ C

,

(11)

which is the energy of transfering an atom of the deposited element from the bulk to the surface. On the other hand, esegr'--l.ls--12 i.e., the segregation energy is equal to the difference of the effective chemical potentials in the surface layer and in the bulk, where the chemical potentials in the surface layer are defined by (at T=0): ,

Os=

0 Eto,urr t ( A c B l _ C/B)

Oc

.

(12)

If the segregation energy is negative, as in the case of Ag on Pt(111), the deposited element stays at the surface. If the segregation energy is positive, as in the case of Pt on Cu(111) and Ru on Au(111), the deposited element should go into the deeper layers of the surface region (if the transfer of deposited element into the bulk is kinetically hindered). Usually, the deposited element

11

appears to be capped by a monolayer of the substrate, which is a process that may be observed in deposition experiments due to the quite fast diffusion of atoms between the surface and subsurface layers. The experimental data, a discussion of which may be found in [15], confirm the above mentioned general features of the formation of surface alloys. In general the surface segregation energy is different for different alloy concentrations of the deposited element (as one may see it even changes sign in the case of Ag on Cu(100)). Such a change in the surface segregation energy is in fact related to the alloying behavior, presented by the mixing energy, and therefore, the surface segregation can be characterized by a single parameter, which is the initial slope of the surface energy curve of the segregation energy of a single impurity of the element deposited at the surface of a substrate. 5. GENERAL TRENDS FOR THE SURFACE MIXING ENERGIES IN TRANSITION METAL ALLOYS In Table 1 we present the sign of the curvature of the surface energy curve calculated from first-principles [ 15] for the closed-packed surfaces of the 4d and 5d metals (fcc(111): Rh, Pd, Ag, Ir, Pt, Au; bcc(110): Nb, Mo, Ta, W; hcp(0001): Tc, Ru, Re, Os). Since the sign of the mixing energy is opposite to that of the curvature, a "+" in the table means a negative mixing energy or alloy formation, and a "-" means that alloying of the deposited element in the surface layer of the substrate is energetically unfavorable against island formation. It is clearly seen that 4d-4d, 4d-5d, 5d-4d and 5d-5d combinations exhibit similar patterns of "+" and "-" signs. This is because the bonding in t~ansition metals as well as in transition metal alloys is mainly determined by the valence d-electrons [29,31 ], which form quite localized bonds in contrast to the free-electron like bonding found in the simple metals. As a result the d band occupation is the main parameter for the characterization of the bonding in this case. In general, alloying in the surface alloy cases follows the trends observed in the corresponding bulk systems [31,36]. However, there are exceptions due to several factors. One of these is the crystal structure of the host (or substrate), which may play crucial role in the alloying [37], especially when the substrate is an earlier transition metal. This is, for instance, reflected in the asymmetry of the alloying behavior of A-B and B - A systems (see, for instance, W - M e and M e - W or Ta-Me and Me-Ta). Another factor which may change the alloying at the surface is the epitaxial strain of the surface alloy due to its pseudomorphic attachment to the substrate. This concerns especially systems with elements that differ considerably in size

12

where

the

resulting

epitaxial

surface more favorable of Au on Ni(110) reconstruction

[9]. H o w e v e r ,

energy

usually

makes

alloying

the pseudomorphic

at t h e

if t h e e p i t a x i a l s t r a i n is r e l i e v e d b y a s t r u c t u r a l

of the surface layer, the alloying may disappear,

the case of Ag growth hexagonal

strain

[ 3 8 ] . T h i s is t h e c a s e , f o r i n s t a n c e , i n t h e i n i t i a l g r o w s

on Cu(111)

surface

as observed

[ 12,15] for a higher coverage

alloy separates

into islands

of Cu

in

of Ag. Here, and

a c(2xl)

Ag phase.

Table 1 T h e sign of the curvature of the surface energy curve: "+" corresponds to surface alloy formation, "-" to island formation, and . .=. . to zero curvature. C o l u m n s are labelled by the deposited element and rows by the substrate.

Zr Zr Nb

+

Nb

Mo

Tc

Ru

Rh

Pd

Ag

Hf

Ta

W

Re

Os

Ir

Pt

An

+

+

+

+

+

+

+

+

+

+

+

+

+

+

4-

+

+

+

+

-

-

+

+

+

+

+

+

+

+

+

=

.

+

+

+

-

+

+

+

+

+

+

-

-

+

+

-

+

+

+

=

-

-

+

-

+

+

+

+

-

-

-

4-

+

+ +

Mo

+

+

Tc

+

+

+

.

Ru

+

+

+

.

Rh

+

+

+

+

-

Pd

+

+

+

+

-

Ag

+

.

Hf

=

+

+

+

+

Wa

-

-

+

+

.

.

.

.

=

W

+

+

+

+

.

.

.

.

+

+

Re

+

+

+

.

+

+

+

.

.

.

.

.

.

.

.

.

.

.

. .

-

.

.

.

+

+

+

-

.

-

.

.

+

.

+

+

+

+

+

+

+

+

+

+

+

+

+

-

.

.

. +

+

+

+

+

.

+

+

+

+

+

+

+

+

=

-

-

+

+

+

+

=

Pt

+

+

+

+

+

+

+

-

+

+

+

+

-

Au

+

.

+

+

+

+

.

.

The surface orientation also plays a very important formation,

since

corresponding

cases,

is

coordination

cases, especially periodic

alloying when

determined

numbers,

the substrate

belongs

surface

energy

curve

will

not

.

.

+

interactions

are surface

the may

simple

in Fig. 4 and

described

above. To have even a qualitative understanding

and

the

specific. In some

to the IVb-VIIIb

have

.

-

play an important

behavior

+

role in the surface alloy

presented

need further input.

the alloying

may

.

effective

z~, w h i c h

table, the multisite interactions

the

by

.

.

+

.

.

.

Os

.

.

.

.

Ir

.

.

.

.

group

parabolic

be more

in the

role. In those complex

shape than

one may therefore

13

6. GENERAL TRENDS FOR SURFACE SEGREGATION ENERGIES IN TRANSITION M E T A L ALLOYS In most cases the experimental techniques used to study surface phenomena do not seem to yield consistent values for the surface segregation energies. One important exception is the special case of an atom of atomic number Z+I in a host of atoms of atomic number Z, where the surface segregation energy may in fact be extracted with a high degree of accuracy from X-ray photoemission spectroscopy (XPS) measurements of surface core-level shifts (SCLS) [39]. In contrast they may be calculated quite accurately by modern first-principles methods [ 18,25,40]. In Fig. 6 we show the results of such calculations for transition metal alloys using grey scales for presenting absolute values of the surface segregation energies. To differentiate their sign we use filled circles for the negative segregation energies which correspond to segregation of the impurity (solute or deposited element) to the surface of the host (substrate). If the segregation energy is positive, the impurity prefers to be in the bulk or to be covered by the host. The actual values of the surface segregation energies can be found in Ref. 25. Similar to the case of the mixing energies in Table 1 one observes a pattern of segregation behavior which repeats itself for each combination of transition series. Again, the main feature of this pattern, an hourglass shape formed by the elements of the matrix which correspond to the negative surface segregation energy, is a consequence of the bonding along each transition metal series. The surface segregation energy is roughly proportional to the difference of the surface energies of the alloy components (under the condition, that they are determined for the given surface, structure, and lattice spacing of the alloy or of the host, in the case of an impurity). The main contribution to the surface energy in transition metal alloys is due to the bonds broken by the surface, and the energy involved is a parabolic function of the number of d-electrons as given by the Friedel model [25]. Thus, the surface segregation energy of a d metal impurity in another d-metal host may be estimated from: ._

E segr A~B

Here,

WB

O[WBNaB(IO--NB)--WA_~BNA(IO--NaA)]. d

d

(13)

and WA~ B are the d-band widths of the host (B) and the impurity

(A) in the host, respectively,

N di

the number of d-electrons in the host and

the impurity, and 0 = 0 . 0 5 [ 1-~/z/zb] , where zs and zb are the coordination numbers at the surface and in the bulk, respectively. The dependence on the surface coordination number means that the segregation energy in transition

14 metal alloys may increase dramatically for more open surfaces.

Fig. 6. Surface segregation energies of transition metal impurities (solute) for the closedpacked surfaces of transition metal hosts.

15

The deviations from the hourglass behavior predicted by the Friedel model (13) are due to the crystal structure effects, which originates from the local character of the interatomic bonding and its dependence on the number of valence d-electrons. The later determines the sequence of crystal structures which is the same along each transition metal series, except for the 3d transition metals where magnetic effects occur. Although the structural energy difference (bcc-fcc or bcc-hcp) in the pure transition metals is of the order of 0.2-0.4 eV, the structural energy difference in the segregation energies in some cases reach 1 eV [25], which makes it a very important parameter in the general analysis. 7. ISLAND F O R M A T I O N : GROWTH

MULTILAYER

VERSUS MONOLAYER

In this and the next section we will consider several examples which illustrate the application of the stability analysis based on the surface energy curve. We will start with the deposition of an element which does not form alloys (at low temperatures) with the substrate in the bulk and on the surface. Hence, there should be a formation of islands of the pure, deposited element incorporated in the surface of the substrate. Such structures, for instance, are usually formed during epitaxial growth of Co, Fe, and Cr on different surfaces of Cu: Systems which are well studied experimentally (see, for instance, [22,41-45], and references therein). Since the behavior of all the above mentioned systems is similar, we will consider here the growth of Co on Cu(111) during a submonolayer deposition. This case has been investigated thoroughly and it is found that the deposition of Co at low temperature (150K) leads to the growth of three-layer islands of Co with one subsurface layer, which at room temperature transform into twolayer islands of Co capped by one layer of Cu [22]. Note, that in contrast to the case considered here, most experimental investigations, e.g., Fe on Cu, have been carried out because of the interest in magnetic multilayers and therefore the amount of deposited element has usually been quite large. However, our main interest is the equilibrium structures formed during the initial epitaxial growth with up to one monolayer of the deposited element. Let us first mention, that the formation of islands of pure Co on Cu is an obvious consequence of the bulk phase diagram [36]: Co and Cu do not form alloys up to the melting temperature. Further, the size mismatch of Co and Cu is very small, and thus the alloying behavior will not be altered at the surface. The capping of Co islands by Cu is explained on the basis of surface segregation arguments knowing the fact that the surface energy of Cu is less than that of Co. Therefore it is no surprise that the surface energy curve for a monolayer of a random Co~Cu~_r alloy, obtained in first-principles calculations

16

[22] has the form shown in Fig. 7. The initial slope of the surface energy curve (at small concentrations of Co) in the figure indicates that the surface segregation energy is positive (it is equal 0.33 eV, for Cu(111) surface [25]). Hence, the Co islands which will form during epitaxial growth will be capped by Cu atoms, if diffusion at the surface is sufficiently fast. It is important to notice that the fact that the surface energy decreases when the Co coverage exceeds about a quarter of monolayer, is a consequence of the phase separation of Co and Cu in the bulk and does not mean that Co islands with no Cu on top will be stable (locally) at the surface. Let us demonstrate how the growth mode can be understood and obtained from the surface energy curve. To do so, one needs the surface energy of n layer (pseudomorphic) Co-structures on Cu(111) as a function of n shown in Fig. 8 [22]. In the limit n ~ oo the surface energy of Co,/Cu(111) is

YC~ Here

=

u

]/Co(lll)+yiCtolCu(lll) n E co-~cu sol

and

u

(14)

"

are the surface energies of fcc(111) Co and Cu,

ColCu(lll)

respectively ' u the Co/Cu(111) interface energy, and b u l k solution energy of Co in Cu.

COxCU~_x/Cu(111) ......

"

I

'"

I

"

I

'

I

"

> o.7 E~ L_

E

e

9 0.6

I:= .,

.5

o.o

i

i

i

o.4

i

Fig. 7. Surface energy of a monolayer CocCu~-c on Cu(111).

i

o18

1.0

E sol c~

the

17 It follows from (14) that the generally negative slope of the surface energy curve is due to the positive solution energy of Co in Cu, and it simply reflects the fact that the formation of Co islands is an energetically favorable process, since it "removes" Co from the bulk. One may analyze the stability of an n layer island against separation into islands of different heights by a common tangent construction, or in this particular case simply by the convex hull of the lowest surface energy points. The procedure is shown in Fig. 8. First we draw a line from the point n=0 (the surface energy of Cu(111)) to the surface energy of Co2/Cu(111). This line is below the surface energy of Co~/Cu(lll), and thus monolayer islands of Co are unstable against separation into a pure Cu surface and two-layer islands of Co. If we neglect the effect of island boundaries, the system should gain 0.39 eV per surface atom as a result of such a separation. We can continue this process and find that two-layer islands are unstable against separation into a pure Cu surface and three-layer islands, and so on. However, the energy of the separation is reduced for every step: In the case of the separation of Co2/Cu(111) into pure Cu(111) and Co3/Cu(111) it is only 0.11 eV as shown in Fig. 8. For large n the gain in energy due to the separation into a pure Cu surface and n+ l-layer island is Co / Cu( ~ ~~)

AE,=y

1

------y n+l

1.0

Cu( ~~ ~)

n

--------y n+l

!

!

AE 1 = 0.39 eV

v

0.0

.

--~.~..

"-"Z,-'~ CoJCu(111) ""-~',~ AE 2 = 0.11 eV

e--

'1=

(15)

,

(111)

>

o

Co.+,/ Cu( ~ ~ ~)

-1.0 COxOU,_,/Cu (111 ) Con/Cu (111 )

00 -2.0

I

I

I

1

2

3

Number of layers (n)

4

Fig. 8. The surface energy of Co,/Cu(111) as a function of the number of Co layers. Broken lines correspond to the energy of a mixture of the those structures which they connect.

18

which using (14) can be rewritten as AE n m 1 AEo= 1 ( yCO(lll) _ yCu(ll])+ Y i n tCo/Cu(lll) ) er n+l n+l

(16)

"

Hence, if A E0>0 , the deposited material should constantly undergo "island" separation, during which low islands transform into higher islands and clean surface areas. This is exactly the Volmer-Weber epitaxial growth mode and since the condition is satisfied for the Co/Cu(111) system multilayer epitaxial growth is energetically favorable. One can also see from (16) that the energy gain due to an increasing height of the islands reduces quite fast when n is small. If one includes the effect of the step-edges and the additional microfacets created by the formation of multilayer islands the energy will quite fast become positive. A simple estimate of the neglected effects allows one to explain the stability three-layer islands of Co on Cu during the initial deposition at low temperatures [22]. Next, we consider the capping of Co islands and find the equilibrium height of the capping slab. In Fig. 9 we show the calculated surface energies of

,0

c

C ...... "

>

0.5 -

~

0.0

]

u(111) -.N

C~"%.

co

Cu/Co/

CuZCo/

CuZCo/

]

C o e ' ~ . ...... "\,Cu/Coe' Cue'CoJ . \ " .....- ~ ...................L~...................,'

a~o -0.5 t1:l

o .

o9

.

.

.ocolCu(!! ) .

.

/~ ......- A c u j c o j c u ( 1 1 1 )

_\_ I

..A..

0

1 Number

2

3

of

4

layers (n+m)

5

Fig. 9. The surface energy of different multilayer CUmCOn overlayers on Cu(lll) as a function of n + m . The circles mark the surface energy of Co,/Cu(lll) and the triangles correspond to an additional capping of Co layers by Cu. The dashed-doted line shows the energy of a mixture of a clean Cu surface and two Co layers capped by a Cu monolayer.

19 capped overlayers of Co on Cu(111). First, one can again observe that any, e.g., Cu/Co~/Cu(111), structure is unstable against separation into a pure Cu surface and Cu/Con§ islands. This is schematically shown for Cu/Co/Cu(111), the energy of which is above dot-dashed line, representing the mixture of pure Cu surface and Cu/Co2/islands, by about 0.18 eV. But again, this energy drops quite fast for small n, and for the next island separation of Cu/Co2/Cu(111) into pure Cu(111) and Cu/Co3/Cu(111) it is about 0.05 eV only. In fact, the gain in energy due to such an island separation _ Co/Cu(lll) for Cum/Con/islands is equal to 2/(n + 1 ) Y i n t e r for large n. The interface energy in the case of a phase separated system is usually positive (proportional to the mixing energy) and thus this result simply reflects the ordinary phase separation in the bulk. Now, following the change of the surface energy of Co islands of a fixed height, one finds that there is a substantial gain in energy when Co layers become capped by the a single layer of Cu. In fact, this energy independently of the height of the Co layers is about 0.3 eV, which is simply the value of the segregation energy. A further increase in the of height of the Cu cap does not lead to a corresponding gain in energy, and thus, the capping stops (when the height of the Co islands is greater than one, a one-layer capped configuration is in fact the most stable configuration, although the energy difference between one-layer and multilayer capped configurations is very small). In this section we have considered examples of systems where the alloy behavior on the surface remains the same as in the bulk. As has been mentioned this is basically due to the fact that the size of the alloy components is practically the same. On the other hand, it is now well-known that alloying behavior on surfaces may change due epitaxial strain of the surface alloy [38] when the alloy components have different sizes. Such an alloying in this case is simply a consequence of the release of the epitaxial strain energy, which is positive and reaches its maximal value for an overlayer of a pure deposited element on the substrate. 8. B U L K - T Y P E ORDERED SURFACE ALLOYS A very good initial guess at the structure of a surface alloy may actually be obtained from the bulk phase diagram for the deposited element-substrate system. This is so, simply because, if there are no specific surface effects, the observed structures would have to be those found in the bulk phase diagram. Since the concentration of the deposited element should be considered small (it is actually "almost" zero, but in the case of local equilibria only the substrate atoms close to the surface may participate in the alloy formation, and thus the "effective" concentration of the deposited element could be quite high), the surface alloy will usually have the structure of the first ordered phase in the

20 substrate-rich part of the phase diagram. This kind of surface ordered alloy would be trivial, if the surface did not add some specific features. The simplest surface specific feature of an ordered phase is the fact that there usually are different truncations of the bulk ordered alloy by the same surface orientation. In this case the problem is to find the stable truncation which, as we will show in this section, is usually directly related to the surface segregation energy of the deposited element to the corresponding surface of the substrate. Let us consider the deposition of A1 on a (110) surface of Ni. According to the bulk phase diagram, the addition of A1 to Ni in the limit T=0K must lead to the formation of Ni3A1 in pure Ni. Therefore, the surface alloy formed during such a deposition may have a structure which corresponds to Ni3AI(ll0). Ni3A1 has the L12 structure, and therefore two different truncations are possible for the (110) surface as shown in Fig. 10: The ordered phase can be truncated either by a layer of pure Ni or by an ordered p(2xl)-NiA1 layer, which alternate in the [ 110] direction of ordered Ni3A1.

Fig. 10. Two different truncations of the A3B-LI2(ll0) surface: A pure A layer or an equiatomic p(2xl)-AB layer.

21 The segregation energy of A1 into the first layer of a Ni(110) surface or the surface energy curve can be calculated using first-principles methods [24]. One finds that the energy of segregation to the first layer is approximately -0.3 eV while the energy of segregation to the second and deeper layers is almost zero [24]. This is clearly seen from the initial slope of the surface energy curve of random AlcNi~_c alloys in the first (surface) and in the second (subsurface) layers shown in Fig. 11. In this figure the two squares at c=0.5 correspond to the two different possible truncations of Ni3AI(ll0): a monolayer of p ( 2 x l ) - N i A 1 ordered layer on the surface and a monolayer of p ( 2 x l ) - N i A 1 ordered layer on the surface but capped by Ni atoms. From this result it is clear that the NiA1truncation of the surface alloy is the most stable, and the energy gained by forming this truncation with respect to the Ni-truncation is about 0.15 eV, which is approximately half the segregation energy. Another important result presented in this figure is the behavior of the surface energies of partially ordered p(2xl)-NiA1 alloys in the surface layer. Such partially ordered alloys have the same ordered p ( 2 x l ) structure, but the excess of Ni atoms form partial antisite defects on the A1 sublattice. One can

2.0

oE 1.9 ~ cr

~---~random alloy in the 1st layer O--Orandom alloy in the 2nd layer ~partially ordered alloy in the 1st layer

1.8

..-%

1.7'

0 ~ 1.6

1.5

0.00

Ni/p(2xl)NiAI p(2xl )NiAI/Ni

~

I

0.25

~

I

0.50 O

~

I

0,7'6

a

1.00

Fig. 11. The calculated surface energies of Ni(ll0) with random, partially ordered, and p(2xl) ordered NiA1 layer on the surface and in the subsurface layers (capped by a Ni layer). The dotted line indicates the energy of the two-phase system for a given c: The pure Ni(ll0) surface and the ordered NiA1 alloy in the first layer.

22 see in Fig. 11 that the surface energies of the partially ordered alloys go above the line which connects the surface energy of the pure Ni(110) surface (c=0) and the completely ordered p(2xl)-NiA1 alloys in the surface layer. This is a very general feature, which holds not only in the case of surface alloys, but also in the case of bulk systems. It is connected to the concentration dependence of the ordering energy and means that at low temperature the partially ordered alloys should undergo phase separation if the alloy composition is not stoichiometric. That is, if the A1 coverage is less than half a monolayer, the surface of Ni(110) will be covered by pure Ni and completely ordered p(2x 1)-NiA1 islands. A similar growth of the ordered Ni3A1 alloy is observed experimentally during deposition of A1 on the (100) surface of Ni [46]. Here the formation of a stable c-(2x2) ordered NiA1 alloy was found on the surface while the second layer was an almost entirely pure Ni layer and the third layer was enriched by A1. This type of structure corresponds to the NiA1 termination of the Ni3AI(100) surface, which also has an alternative truncation. The surface segregation energy of A1 on the (100) surface of Ni is only about-0.1 eV, and as has been shown [24], the NiA1 termination is more stable than Ni termination by approximately half of this value.

1,3

E

0 *"~

m

11st random )2nd random Plst ordered 12nd ordered

"

1.2 "i

>

1.1 -

r-

1.0-

0

0.9

'1:: 09

0.8 0.00

I

I

0.25

I

I

0.50 C

I

|

0.75

I

1.00

Fig. 12. The surface energies of random and p(2xl)-ordered Pd~Cul_c alloys in the first (surface) and second (subsurface) layers.

23

A system which exhibits a behavior somewhat different from A1-Ni is PdCu. The first ordered phase in the Cu-rich region of the Cu-Pd bulk phase diagram [47] is L12-Cu3Pd, and therefore it is not a surprise that the growth of Pd on Cu(110) leads to the formation of surface alloy with the corresponding bulk ordered structure [21]. However, in contrast to the growth of A1 on Ni(110), Pd does not segregate to the (110) surface of Cu. This can be seen in Fig. 12 where the first-principles results for different surface alloys of are presented [21 ]. In fact, although the segregation energy of Pd into the first layer is positive, but very small (less than 0.05 eV: It is the initial slope of the surface energy curve for the random alloy in the first layer) the main driving force behind the final surface alloy configuration is the segregation energy of Pd into the second layer, which is -0.23 eV. As a result the energy gain of having the Cu truncation at the surface is about 0.1 eV relative to the CuPd truncation. The reason, why the energy of segregation to the second layer is so large is the fact that the (110) surface is quite open: as one can see from Fig. 9, the second layer is in fact not covered by the surface atoms. In the case of NiPt random alloys, this even leads to a segregation reversal at the (110) surface. Like the case considered above this is directly related to the quite large energy of segregation to the second, subsurface layer [48,49], which is greater than the energy of surface segregation to the first layer. Therefore, in general one should be very careful in making predictions for more open surfaces: simple surface segregation arguments may not work at all. 9. ALTERNATIVE ORDERED STRUCTURES ON THE S U R F A C E

In Fig. 12 we have also shown that partially ordered (2x l) CuPd alloys in the subsurface layer (c < 0.5) are unstable against separation into islands of pure Cu(110) surface and ordered (2x l) CuPd islands capped by Cu atoms. This is indeed observed experimentally [21]. However, at a very low coverage of a few percent, ordered - C u - P d - one dimensional chains aligned along the closed-packed [ 110] direction are formed in the surface layer, see Fig. 13. Although, this may look as a change in the ordering behavior of the surface alloy, the effect is entirely consistent with the ordering behavior in the bulk and is in fact related to the specific features of the structure of the surface itself. Namely, the strongest effective interaction (see (10)) which is responsible for the ordering in CuPd is the effective interaction for the first coordination shell [50]. All the other interactions are rather small. This means that the main gain in the ordering energy is due to CuPd ordering in the closed-packed direction. Since the (110) surface is anisotropic, the Pd atoms first tend to form order in this specific direction, forming thereby ordered CuPd strings at very low Pd coverage. The reason, why such ordered chains

24

Fig.13. (a) STM image following deposition of small amounts of Pd on Cu(ll0). Linear chains are observed, which are aligned along the closed-packed direction (70x70 Ang.). (b) Atomically resolved image of an island of pure Cu at coverage 0.28 ML Pd.

are not covered by Cu atoms, is the fact that the energy gain by this process does not counterbalance the energy cost of creating the steps, which must appear during such a process. At higher Pd coverage when PdCu islands of ordered alloy start to form the perimeter-to-area ratio of the islands drops dramatically and hence the energy balance changes in favor of capping CuPd islands. One may also see in Fig.13 that such islands have a preferential alignment along the [ 110] direction. In the example considered above the ordering of the deposited element and the substrate leads to the formation of distinct long-range structures: chains and islands. However, it may not always exhibit itself as long-range order even below the order-disorder transition temperature. This kind of behavior is observed in the Cu-Pd system, but during a deposition of Pd on Cu(111) in the temperature range between - 8 0 - 250 C which is well below the orderdisorder transition temperature of Cu3Pd in the bulk (about 500 C[47]). In this case similar to the growth of Pd on Cu(110) considered above and on Cu(100) [51,52] one may expect a formation of ordered (2x2) Cu3Pd surface alloy consistent with the (111) surface of L12-Cu3Pd alloy. Nevertheless, a formation of bands of a quite stable random CuPd alloy along the steps at the surface has been observed [53]. The Pd concentration in this alloy depends on the subsequent heat treatment, and varies between 0.18 and 0.31 at.% of Pd. As a matter of fact although the alloy configuration seems to be completely random without any distinct long-range order features, the analysis of the STM image shows that almost all of the Pd atoms are surrounded by Cu atoms

25

0.9

E

0

m

0.8

O-~Orandom

PdcCU~_c alloy

0 C~

,- 0.7

0 tO 0 0

m 0.6

"1:::

............

:3

......- A

CuaPd

O9 0.5

,

0.00

I

0.25

Cu2Pd ,

I

0.50

0.75

Fig. 14. First-principles results for the surface energy of random and ordered surface alloys on Cu(111). CuzPd and Cu3Pd are ~ x qr~ and (2x2) ordered alloys correspondingly. The dotted line is the stability line which is the surface energy of a disordered alloy with the maximal possible value of the SRO parameter for a given concentration.

in the first coordination shell. The energy gain due to such a short-range order (SRO) in the (111) fcc layer can be expressed in terms of the effective pair interactions at the first coordination shell, V~, defined in (10), as [54]"

ESRO= -1- Z l 2

C (1-c)

V~ er

,

(17)

where Zi--" 6 is the coordination number of the first coordination shell for fcc(111), c the concentration of Pd, and oc~ the so-called Warren-Cowley SRO parameter for the first coordination shell. The value of the SRO parameter in the case where all Pd atoms are surrounded only by Cu atoms reaches its minimal value, which is - c / ( 1 - c ) [54], and therefore the energy of the SRO effects is - 3 c 2V 1 " As a result the total mixing energy of an alloy with the maximal SRO is the mixing energy of the random alloy, given by (10) plus the ordering energy (17), which yields - 3 c V 1 . This energy is a linear function of the

26 concentration and therefore the energies of (2x2)-Cu3Pd and x/3 x ~-3 CuzPd, as well as the energies of random alloys with maximal value of the SRO parameter lies practically on the same stability line. This means that all these structures are equally stable and may coexist on the surface. This is shown in Fig. 14 where the results of the first-principles calculations for the surface energies of random and ordered Pd~Cul_~ alloys [53] are presented together with the surface energy of the disordered alloys having the smallest possible value of the SRO parameter at the first coordination shell.

Fig. 15

~f3 x ~

- A2B (a) and (2x2)-A3B (b) structures on the triangle lattice.

27

Such an unusual behavior is in fact a consequence of the highly (infinitely, to be precise) degenerated ground state (at T=0K) of an alloy on a triangle (fcc(111), hcp(0001)) lattice with positive nearest neighbor interactions due to frustration effects [54]. There are, for instance, infinitely many random alloy configurations of A3B alloys the energy of which are equal to the energy of the 2x2-A3B alloy. Such a degenerate ground state for alloy compositions different from A2B (or AB2) leads to the so-called surface induced disorder in the case of the (111) surface of L12-A3B and L10-AB ordered alloys [55]. The only exception is the A2B alloy on the triangle lattice which has a Vr3 x ~ - A2B structure in the ground state. If the alloy composition exceeds 1/3, then again the ground state becomes infinitely degenerate. Nevertheless, its energy will be higher than the stability line connecting the surface energies of A and ~ x ~ - A2B, since 1/3 is the maximal concentration at which atoms of one alloy component can be surrounded exclusively by the atoms of the opposite type on the triangle lattice and at this composition there is only one way to arrange every triangle to be A2B. This makes the ~ x ~r~-A2B surface structure special in the d e p o s i t i o n experiments for the fcc(111) and hcp(0001) surfaces in the case of ordered alloys, and is the reason why it is so frequently observed in the deposition experiments [56-60]. It should be noticed, however, that the L12(lll)-A3B and ~ x ~ A2B ordered structures are equally stable (on the same stability line) only if the effective interactions for more distant coordination shells are zero. If this is not the case, then the relative stability of these structures will depend on the value of the other interactions. The first difference actually appears at the second coordination shell in the surface layer (which corresponds to the third coordination shell in the bulk). The corresponding contribution from V2 to the mixing energy of the L l z ( l l l ) - A 3 B is again - 3 c V 2 , while it is zero in the case of the ~ x ?r~-A2B alloy, that is, the ordering at the second coordination shell will favor the L12(111)-A3B ordered structure. Exactly the opposite situation occurs with the effective interactions at the third (fourth in the bulk) coordination shell, V3, which give zero and - 3 c V3 contributions to the mixing energy of the L12(111)-A3B and phases, respectively.

~r~ x ~

-A2B ordered

ACKNOWLEDGMENT The Center for Atomic-scale Materials Physics is sponsored by the Danish National Research Foundation.

28

REFERENCES [1] D.A. King, D.P. Woodruff (eds.), Growth and Properties of Ultrathin Epitaxial Layers, Elsvier, Amsterdam, 1998. [2] U. Bardi, Rep. Prog. Phys. 57 (1994) 939. [3] J. Wintterlin and R.J. Behm, in Scanning Tunneling Microscopy I, 2nd ed., edited by H.J. Guntheroldt and R. Wiesendanger, Springer-Verlag, Berlin, 1994. [4] R.Q. Hwang, C. Gunter, J. Schroder, S. Gunter, E. Kopatzki, and R.J. Behm, J. Vac. Sci. Technol. A 10 (1992). [5] D.D. Cahmbliss, R.J. Wilson, and S. Chiang, IBM J. Res. Dev. 39 (1995) 639. [6] S.C. Wu, S.H. Lu, Z,Q. Wang, C.K.C. Lok, J. Quinn, Y.S. Li, D. Tian, F. Jona, and P.M. Marcus, Phys. Rev. B 41 (1990) 3353. [7] K.-O. Ng and D. Vanderbilt, Phys. Rev. B 52 (1995) 2177. [8] G.E. Thayer, V. Ozolins, A.K. Schmid, N.C. Barlet, M. Asta, J.J. Hoyt, S. Chiang, and R.Q. Hwang, Phys. Rev. Lett. 86 (2001) 660. [9] L.P. Nielsen, F. Besenbacher, I. Stensgaard, E. Lagsgaard, C. Engdahl, P. Stolze, K.W. Jacobsen, and J.K. Norskov, Phys. Rev. Lett. 71 (1993) 754. [10] C. Nagl, E. Platzgummer, O. Haller, M. Schmid, and P. Varga, Surf. Sci. 331 (1995) 831. I1 l] J.L. Stevens and R.Q. Hwang, Phys. Rev. Lett. 74 (1995) 2078. [12] P.T. Sprunger, E. Lagsgaard, and F. Besenbacher, Phys. Rev. B 54 (1996) 8163. [13] H. Brune, H. Roder, C. Boragno, and K. Kern, Phys. Rev. B 49 (1994) 2997. [14] C. Gunter, J. Vrijmoeth, R.Q. Hwang, and R.J. Behm, Phys. Rev. Lett. 74, (1995) 754. [15] A. Christensen, A.V. Ruban, P. Stolze, K.W. Jacobsen, H.L. Skriver, J.K. Norskov, and F. Besenbacher, Phys. Rev. B 56 (1997) 5852. [16] A. Senhaji, G. Treglia, B. Legrand, N.T. Barret, C. Guillot, and B. Villete, Surf. Sci. 274 (1992) 297. [17] S. Blugel, Appl. Phys. A 63 (1996) 595. [18] T. Asada, S. Bulgel, Physica, B 237-238 (1997) 359. [19] Ch. Ross, S. Schirmer, M. Wuttig, Y. Gauthier, G. Bihlmayer, and S. Blugel, Phys. Rev. B 57 (1998) 2607. [20] G. Bihlmayer, Ph. Kurz, and S. Blugel, Phys. Rev. B 62 (2000) 4726. [21] P.W. Murray, S. Thorshaug, I. Stensgaard, F. Besenbacher, E. Lagsgaard, A.V. Ruban, K.W. Jacobsen, G. Kopidakis, and H.L. Skriver, Phys. Rev. B 55 (1997) 1380. [22] M.O. Pedersen, I.A. Bonicke, E. Lagsgaard, I. Stensgaard, A. Ruban, J.K. Norskov, and F. Besenbacher, Surf. Sci. 387 (1997) 86. [23] L.T. Wille, B. Nonas, P.H. Dederichs, and H. Dreysser, Phil. Mag. B 78 (1998) 643. [24] A.V. Ruban and H.L. Skriver, Comp. Mat. Sci. 15 (1999) 119. [25] A.V. Ruban, H.L. Skriver, and J.K, Norskov, Phys. Rev. B 59 (1999) 15990. [26] M. Methfessel, D. Henning, M. Schefler, Phys. Rev. B 46 (1992) 4816. [27] L. Vitos, A.V. Ruban, H.L. Skriver, and J. Kollar, Surf. Sci., 411 (1998) 186. [28] J. Kollar, L. Vitos, B. Johansson, and H.L. Skriver, Phys. Stat. Sol. (b) 217 (2000) 405. [29] J. Friedel, The physics of metals (ed. J.M. Ziman), p.494 Cambridge University Press, New York (1969). [30] M.C. Desjonqueres and D. Spanjaard, Concepts in Surface Physics, Springer, Berlin, 1996. [31] D. Pettifor, Bonding and structure of molecules and solids, Clarendon Press, Oxford, 1995. [32] A.F. Becker, G. Rosenfeld, B. Poelsema, and G. Comsa, Phys. Rev. Lett., 70 (1993), 477.

29 [33] H. Roder, R. Schuster, H. Brune, and K. Kern, Phys. Rev. Lett. 71 (1993) 2086. [34] U. Struber and K. Kuppers, Surf. Sci. Lett., 294 (1993) L924. [35] P. Zeppenfeld, M.A. Krzyzowski, Ch. Romainczyk, R. David, G. Comsa, H. Roder, K. Bromann, H. Brune, and K. Kern, Surf. Sci. Lett., 342 (1995) L1131. [36] F.R. de Boer, R. Boom, W.C.M. Mattens, A.R. Miedema, and A.K. Niessen, Cohesion in Metals: Transition Metal Alloys, North-Holland, Amsterdam, 1988. [37] A.V. Ruban, H.L. Skriver, and J.K. Norskov, Phys. Rev. Lett., 80 (1998) 1240. [38] J. Tersoff, Phys. Rev. Lett. 74 (1995) 434. [39] B. Johansson, N. Martenson, Phys. Rev. B 21 (1980) 4427. [40] B. Nanos, K. Wildberger, R. Zeller, and P.H. Dederichs, Phys. Rev. Lett. 80 (1998) 4574. [41] M.T. Kief and W.F. Egelhoff, Jr. Phys. Rev. B 47 (1993) 10785. [42] J. Jandelleit, Y. Gauthier, M. Wuttig, Surf. Sci. 319 (1994) 287. [43] J. Giergeil, J. Shen, J. Woltersdorf, A. Kirilyuk, and J. Kirschner, Phys. Rev. B 52 (1995) 8528. [44] J. Shen, J. Giergiel, A.K. Schmid, J. Kirschner, Surf. Sci. 328 (1995) 32. [45] C. Pflitsch, R. David, L.K. Verheij, R. Franchy, Surf. Sci. 468 (2000) 137. [46] D.J.O'Connor, M. Draeger, A.M. Molenbbroek, Y. Shen, Surf. Sci. 357/358 (1996) 202. [47] P.R. Subramanian and D.E. Laughlin, J. Phase Equilibria 12 (1991) 231. [48] I.A. Abrikosov, A.V. Ruban, H. L. Skriver, and B. Johansson, Phys. Rev. B 50 (1994) 2039. [49] L.V. Pourovskii, A.V. Ruban, I.A. Abrikosov, Yu. Kh. Vekilov, and B. Johansson, Phys. Rev. B 64 (2001) 35421. [50] Z.W. Lu, D.B. Laks, S.-H. Wei, and Z. Zunger, Phys. Rev. B 50 (1994) 6642. [51] P.W. Murray, I. Stensgaard, E. Lagsgaard, F. Besenbacher, Phys. Rev. B 52 (1995) R 14404. [52] P.W. Murray, I. Stensgaard, E. Lagsgaard, F. Besenbacher, Surf. Sci., 365 (1996) 591. [53] A.B. Aaen, E. Lagsgaard, A.V. Ruban. and I. Stensgaard, Surf. Sci., 408 (1998) 43. [54] F. Ducastelle, Order and Phase Stability in Alloys, North-Holland, Amsterdam, 1991. [55] J. Neugebauer, M. Scheffler, Phys. Rev. Lett. 71 (1993) 577. [56] W. Schweika, D.P. Landau, K. Binder, Phys. Rev. B 53 (1996) 8937. [57] S. Oppo, V. Fiorentini, and M. Scheffler, Phys. Rev. Lett. 71 (1993) 243. [58] P. Baily, T.C.Q. Noakes, D.P. Woodruff, Surf. Sci. 426 (1999) 358. [59] D. Tian, H. Li, S.C. Wu, F. Jona, and P.M. Marcus, Phys. Rev. B 45 (1992) 3749. [60] D. Tian, A.M. Begley, and F. Jona, Surf. Sci. Lett. 273 (1992) L393.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces 30

D.P. Woodruff, (Editor)

Chapter 2

Atomistic modeling of surface alloys Guillermo Bozzolo a'b and Jorge E. Garces a'c

aOhio Aerospace Institute, 22800 Cedar Point Rd., Cleveland, OH 44142, USA bNASA Glenn Research Center, Cleveland, OH 44135, USA CCentro Atomico Bariloche, 8400 Bariloche, Argentina 1. INTRODUCTION Different experimental techniques provide detailed information on the structure and composition of surface alloys and alloy surfaces, sometimes leaving little doubt regarding the often complex patterns that appear, for example, during the process of surface alloy formation [1-13]. There are cases, however, in which some level of modeling is necessary in order to reconcile the sometimes conflicting conclusions that can be drawn from different experiments. Whether it is the often unpredictable alloy surface composition and structure, due to segregation or surface defects, or the several active degrees of freedom during deposition of different types of atoms on an arbitrary substrate (a polycrystalline surface of a monatomic crystal, an alloy surface, etc.), atomistic modeling is essential in identifying isolated features, analyzing correlations, or simply allowing for the study of a wide range of possibilities not accessible via experiment as, for example, the study of metastable structures. Whether it is the analysis of different bulk or surface alloy phases, it is difficult to predict what can be expected during the corresponding process of formation. From the modeling standpoint, the process is extremely complex, and any attempt to develop a useful modeling tool would be almost hopelessly limited by the excessive number of variables that should be taken into account in order to provide a satisfactory description of the actual process. Increasing computer power, alone, is not necessarily the only answer, nor is the development of detailed theories that, regardless of the computer power available, are sometimes difficult to implement. However, the purpose of atomistic modeling is not to exactly reproduce every detail of the real process, but to be able to identify the main features and driving mechanisms of a certain specific behavior. The amount

31

of detail that can be considered satisfactory is clearly dependent on the problem at hand, but in spite of the particular characteristics of any given system, a few fundamental ingredients are necessary: 1) simplicity in the implementation of the physical theory and the ensuing calculations, 2) universality in the set of parameters or potentials used (i.e., complete transferability) and 3) versatility, in terms of a minimum number of restrictions on the type and number of elements and type of structures. Satisfying these minimum requirements is essential for the successful modeling of surface alloys which has been, so far, mostly limited to either a single-element substrate or the deposition of one single element at a time. To extend modeling to more complex systems would require complete freedom from the restrictions imposed by not fulfilling any or all of the above mentioned conditions. While a demanding challenge, the recent trend of combining first-principles methods with quantum approximate methods has resulted in steady progress in this area, allowing for increased understanding on the atomic processes that govern the phenomenon of surface alloy formation. Although different in their formulation, all quantum approximate methods rely on the simplicity introduced by a global description of the interaction between atoms, but at the same time, such generalization often translate into limitations thus failing to provide an ideal framework that would allow for a general and unrestricted application. In this work, we will concentrate on the description and application of one such method, particularly suitable for the study of surface phenomena. The Bozzolo-Ferrante-Smith (BFS) method for alloys [14] satisfies most of the requirements imposed on quantum approximate methods, in terms of simplicity, accuracy, generality and ease of implementation, with the added advantage that its novel interpretation of the alloy formation process is free of constraints that would limit its applicability to arbitrary systems. To a great extent, this lack of restrictions in the general formulation of the BFS method relies on the ability to properly define the parameters of each constituent element. In high symmetry situations (i.e., Cu on Cu(100), where both the adsorbate and the substrate atoms are of the same atomic species), the implementation of the method is generally straightforward. That is also the case when the different participating elements have the same bulk symmetry (i.e., Pd on Cu(100), where both elements form fcc bulk solids) [15]. It is not necessarily true, however, that the bulk symmetry of each element dictates the structure of the alloy, as it is most generally the case that phases of other symmetries can and do form [ 16]. This is even more so in the case of surface alloys, where not even a thorough knowledge of the bulk phases constitutes a sufficient basis for the determination of the structures that form on the surfaces. Depending on the characteristics of the surface, even immiscible metals in the bulk are known to form ordered surface structures [ 17-21].

32

From a theoretical standpoint, the traditional approach for the determination of an alloy structure implies, in principle, a search through any possible configuration until the most energetically favorable is found. While current first-principles methods, coupled with a substantial increase in computational power, have made this approach a standard practice for the calculation of phase diagrams of (mostly binary) bulk alloys, the complexity of surfaces makes quantum approximate methods a necessary tool to supplement the existing techniques and the growing body of experimental data. However, the study of surfaces and surface processes has been a severe test for quantum approximate methods, which usually rely on parameters or potentials determined from bulk properties, thus limiting their effectiveness in the low-symmetry environments represented by surfaces. One way to circumvent this obstacle is by formulating the method on the basis of a one-to-one mapping between any arbitrary bulk or surface environment onto an equivalent, ideal, bulk-like one. If such correspondence is uniquely established, then the parameterization becomes universal and equally applicable to bulk or surface problems. The BFS method for alloys satisfies this requirement by assigning to every atom i, regardless of its location and environment, a set of equivalent bulk crystals whose states of isotropic compression or expansion (and the corresponding difference in energy with their ground states) are taken as a measure of the defect in the real crystal where atom i is located. Three equivalent crystals for a given atom i are thus defined to completely describe the different aspects in the process of alloy formation. The first of these equivalent crystals describes purely structural effects. This is achieved by considering every neighboring atom as if it was of the same atomic species of atom i but retaining the actual positions that they have in the real system. The second equivalent crystal of atom i describes chemical effects, considering every neighboring atom by its true chemical identity, but forced to occupy lattice sites of an equilibrium, otherwise monatomic, crystal of species i. A third equivalent crystal is also defined in order to ensure a complete decoupling between the first two, eliminating any structural information in the calculation of the chemical effects. Each of these equivalent crystals shows some degree of departure from a certain equilibrium reference state R i. The energies associated with these departures represent, respectively, the strain (s), chemical Co (c), and chemical reference (~,) energies which, properly coupled, represent the contribution of atom i to the total energy of formation AH of the actual system. The choice of the reference state R i is, in most cases, a straightforward task. For example, the reference state of a Cu atom in a pure Cu crystal is, clearly, fcc. That is also the case for a Cu atom in a Cu3Pd L 12 ordered alloy. For a Cu atom in an ordered CuPd B2 alloy, however, the reference state is bcc. For general situations, a substantially useful degree of freedom in the methodology is therefore

33

introduced by allowing the reference state R i to have symmetries other than the one corresponding to the ground state of each constituent element. With the appropriate means for computing the parameters corresponding to arbitrary reference states, a complete characterization of R i is not less straightforward. For the sake of simplicity, it is convenient to illustrate this issue with an example. It is well known that A1 deposition on a Ni substrate leads to the formation of an fcc-like Ni3A1 film [22], followed by a transitional region leading to a bcc-like NiA1 pattern [22,23]. A1 atoms adopt the symmetry of the substrate (fcc) for low coverages, slowly transitioning to a different symmetry (bcc) as the A1 coverage increases. The layer-by-layer individual distortions from equilibrium lattice sites translate into what can be seen as a continuous transition from Al-fcc to Al-bcc. Other examples include the deposition of Cu on Ag(100) [24], where Cu layers transition from bct to bcc with increasing coverage. In general, varied situations ranging from grain boundaries, where each individual atom sees a different environment that could be best described by a particular intermediate state between the symmetries describing each grain, to liquids and amorphous materials, could be described with this approach. It is clear that with the novel way of partitioning the individual atomic contributions to the energy of formation, added to the appropriate determination of the reference state for each atom, the method provides a tool that is general enough to tackle equally general situations: a) The representation of arbitrary situations in terms of bulk equivalent crystals removes any distinction between bulk, surface or interface situations, all dealt with the same degree of accuracy, b) the calculation of the strain energy in terms of equivalent monatomic crystals lifts any restriction on the structural characteristics of the system at hand, c) the calculation of the chemical energy in terms of perfect crystals lifts restrictions in terms of the number of different atomic species that can be included and d) the atomby-atom determination of the reference state lifts any restriction on the number and type of phases that can be formed by any group of atoms. From a computational standpoint, the usefulness of the method relies on the simplicity of the calculations needed for the determination of the three equivalent crystals associated with each atom i. This is accomplished by building on the simple concepts of Equivalent Crystal Theory (ECT) [25,26], as will be discussed in detail below. The procedure involves the solution of one simple transcendental equation for the determination of the equilibrium Wigner-Seitz radius (rWSE) of each equivalent crystal. These equations are written in terms of a small number of parameters describing each element in its reference state, and a matrix of perturbative parameters Aji, which describe the changes in the electron density in the vicinity of atom i due to the presence of an atom j (of a different chemical species), in a neighboring site. The determination of parameters for each atom in

34

its reference state, whichever it is, is also a straightforward task, as it can be easily accomplished with first-principles methods when experimental input is not available. While we have restricted the examples shown in this work to systems for which experimental input exists, it is important to note that the possibility of expanding the input database by means of first-principles calculations allows the methodology to be applied to any arbitrary system. In addition, the ability to obtain every necessary parameter from the same source has the additional advantage of giving the BFS method much needed consistency in terms of the parameterization used. The primary set of parameters needed are the equilibrium values for the Wigner-Seitz radius, the cohesive energy, E c, and the bulk modulus, Bt~ Two additional single-element parameters are determined as a function of the parameters in the primary set: a screening length ~., that accounts for the screening of atoms beyond the nearest-neighbor layer, and a parameter tx, which represents a measure of the electron density in the overlap region between neighboring atoms. Moreover, the parameter tx is determined by requiting that the maximum strain energy that can be assigned to a given atom is given by the cohesive energy, thus allowing for a direct and simple calculation that also establishes the boundaries for the range of validity of the method. Thus, the primary set of parameters describing any arbitrary reference state for a given atom is then {rws E, E c, l, ~, o~}, where l, a scaling length, replaces B o in order to allow for a closer correspondence with the universal binding energy relationship (UBER) of Rose et al. [27], which is usually written in terms of I rather than B o. A detailed description of the operational equations and the role of each parameter will be presented in Sec. 2. A second set of parameters for element i, the BFS parameters Aji, account for the changes in the electron density in the vicinity of atom i due to the presence of an atom j. These parameters can be computed with first-principles methods by means of a straightforward calculation based on the energy of formation and equilibrium atomic volume of all the binary systems that can be formed with the participating elements. For more accurate results, the parameters Aji can be computed as a function of the concentration of element j in the vicinity of atom i. In some cases it is possible to fully parametedze a given system from experimental input, as will be done in every example presented in Sec. 3. However, the determination of complete primary parameter sets for every possible element in every possible reference state, as well as the associated secondary sets for the concentration-dependent binary cases, is not an easy task as, in most cases, it would require input that is not available from experiment. Once again, this issue can be properly and systematically dealt with by means of first-principles calculations. For example, reference states can differ from the ground state symmetries of the participating elements in the alloy. In those situations, it is strictly neces-

35

sary to rely on first-principles methods for the determination of the equilibrium properties of such crystals. To illustrate this point, we focus on one particular example, namely, the continuous transformation between a bcc and an fcc structure (Fig. 1). Several intermediate configurations can be singled out and the firstprinciples calculation of the primary set of parameters can be made for each one of these configurations. In doing so, each one of the relevant parameters can be written as a function of a single quantity, properly defined to identify each step in the transformation. For the particular case of the bcc <-> fcc transformation, illustrated in Fig. 1, this parameter is r - -~. This procedure can be generalized to include transformations between any number of elemental crystallographic structures. Once this parameterization procedure is completed, the primary set of single-element parameters is general enough so as to allow for the identification of the appropriate reference state for every atom in the system under consideration. In this work, recent progress on BFS modeling of surface alloys will be summarized, with the main purpose of exploring the possibilities that become available with the synergy between a computationally simple and physically sound methodology, increasing computing power, and a substantial practical foundation based on powerful techniques for experimental analysis. For the problem at hand, surface alloys, the general formulation of BFS is not just convenient, but necessary. While it is true that a large number of applications deal with rather simple systems (i.e., deposition of one element on a monatomic substrate), there is a growing body of experimental evidence dealing with more complex situations, thus imposing challenging demands on any modeling effort. It is therefore important to establish a modeling tool for experimentalists based on an operating procedure with a minimum number of constraints, thus allowing for the systematic analysis and interpretation of specific observed features. a

a

r

Fig. 1' Relation b e t w e e n the fcc (c _ 1 ) and bcc (a -" "T ) structures.

36 2. THE BFS M E T H O D

The BFS method has been applied to a variety of problems, ranging from the determination of bulk properties of solid solution fcc and bcc alloys and the defect structure in ordered bcc alloys [28] to more specific applications including detailed studies of the structure and composition of alloy surfaces [29], ternary [30] and quaternary alloy surfaces and bulk alloys [31,32], and even the determination of the phase structure of a 5-element alloy [33]. Previous applications have focused on fundamental features in monatomic [26] and alloy surfaces [29]: surface energies, reconstructions, surface structure and surface segregation in binary and higher order alloys [34,35] and multilayer relaxations [36,37]. While most of the work deals with metallic systems, the lack of restrictions on the type of system that can be studied translated into the extension of BFS to the study of semiconductors [38]. In what follows, we provide a brief description of the operational equations of BFS. The reader is encouraged to seek further details in Refs. 28-35, where a detailed presentation of the foundation of the method, its basis in perturbation theory, and a discussion of the approximations made are shown [14]. The BFS method provides a simple algorithm for the calculation of the energy of formation of an arbitrary alloy (the difference between the energy of the alloy and that of its individual constituents). In BFS, the energy of formation AH is written as the superposition of elemental contributions e~ of all the atoms in the alloy AH - E ( E ' i - E i ) i

- EEi

(1)

i

where E i' is the energy of atom i in the alloy and E i is the corresponding value in a pure equilibrium monatomic crystal. In principle, the calculation of AH would simply imply computing the energy of each atom in its equilibrium pure crystal and then its energy in the alloy. In BFS, beyond directly computing the difference E~for each atom in the alloy, a two-step approach is introduced for such a calculation in order to identify contributions to the energy due to structural and compositional effects. Therefore, E~ is broken up in three separate contributions" a strain energy (Es), a chemical energy (Ec), and a chemical reference energy (Ec~ While there is a certain level of arbitrariness in how this separation is implemented, it is only meaningful when a good representation of the initial and final states of the actual process is obtained by properly linking all contributions. This is achieved by recoupling the strain, chemical and chemical reference contributions by means of a coupling function, gi, properly defined to provide the correct asymp-

37 totic behavior of the chemical energy contribution. Each individual contribution ~t can therefore be written as

S gi(eC Co -E i ) Ei = E i +

(2)

The BFS strain energy contribution s is defined as the contribution to the energy of formation from an atom in an alloy computed as if all the surrounding atoms were of the same atomic species, while maintaining the original structure of the alloy. To visualize this concept, Fig. 2.a represents the atom in question (identified with an arrow) in an equilibrium position in its reference, ground state crystal (arbitrarily represented by a simple cubic lattice). Fig. 2.b shows the same atom in the alloy being studied (also arbitrarily represented by a different crystallographic symmetry). The reference crystal and the alloy differ in two basic aspects. First, atoms of other species may occupy neighboring sites in the crystal and, second, the crystal lattice may not be equivalent in size or structure to that of the ground state crystal of the reference atom. In Fig. 2.b, the different atomic species are denoted with different symbols from that used for the reference atom, and the differences in size and/or structure are denoted with a schematically different atomic distribution as compared to the ground state crystal shown in Fig. 2.a. The BFS strain energy accounts for the change in energy due only to the

(a)

(b)

(c)

k

d

h

al

Fig. 2: (a) A pure, equilibrium crystal (reference atom denoted by the arrow), (b) a reference atom (denoted by the arrow) in the alloy to be studied (atoms of other species denoted with other shading) and (c) the same reference atom in a monatomic crystal, with the identical structure of the alloy to be studied, but with all the atoms of the same atomic species as the reference atom, for the calculation of the strain energy term for the reference atom. The strain energy is the difference in energy of the reference atom between (c) and (a).

38 change in geometrical environment of the crystal lattice (fromFig. 2.a to 2.b), ignoring the additional degree of freedom introduced by the varying atomic species in the alloy. In this context, Fig. 2.c shows the environment 'seen' by the reference atom when computing its BFS strain energy contribution. The neighboring atoms conserve the sites in the actual alloy (Fig. 2.b), but their chemical identity has changed to that of the reference atom (Fig. 2.a) thus simplifying the calculation to that of a single-element crystal. The BFS strain energy term represents the change in energy of the reference atom in going from the configuration denoted in Fig. 2a to Fig. 2.c. In this sense, the BFS strain energy differs from the commonly defined strain energy in that the actual chemical environment is replaced by that of a monoatomic crystal. Its calculation is then straightforward, even amenable to first-principles techniques. The chemical environment of atom i is considered in the computation of ~c, the first term in the total BFS chemical energy contribution, where the surrounding atoms maintain their identity but are forced to occupy equilibrium lattice sites corresponding to the reference atom i. Following the convention introduced in Fig. 2, Fig. 3.a shows the reference atom in the actual alloy (similar to Fig. 2.b), while Fig. 3.b indicates the atomic distribution used in computing the BFS chemical energy Ec (note that the lattice used in Fig. 3.b corresponds to that of the reference crystal of the reference atom, as shown in Fig. 2.a). The total BFS chemical energy is then the difference between the energy of the reference atom 2.a). in Fig. B.2.b, ec, and its energy in its ground state crystal ~C~ The chemical reference energy ~Co (Fig. 2.a) is included in order to completely free the chemical energy from structural defects, taking into account the possibility that the reference atom is not in a full-coordination environment (as is the case (a)

(b)

Fig. 3: (a) The reference atom (denoted by an arrow) in the actual alloy environment and (b) the reference atom surrounded by a chemical environment equivalent to that in (a) but with the different neighboring atoms occupying equilibrium lattice sites corresponding to the reference, ground state of the reference atom.

39

E

m

~~~li:

9

=

..t_ gi

.

+

_

9

Fig. 4: Schematic representation of the BFS contributions to the total energy of formation. The left hand side represents the reference atom (denoted by an arrow) in an alloy. The different terms on the right hand side indicate the strain energy (atoms in their actual positions but of the same atomic species as the reference atom), the chemical energy term (atoms in ideal lattice sites) and the reference chemical energy (same as before, but with the atoms retaining the original identity of the reference atoms).

in or near a surface). This is accomplished by recomputing the contribution Ec defined before, but once again assuming that all atoms are of the same species as the reference atom. As mentioned above, the BFS strain and chemical energy contributions take into account different effects, i.e., geometry and composition, computing them as isolated effects. A coupling function, gi, restores the relationship between the two terms. This factor is defined in such a way as to properly consider the asymptotic behavior of the chemical energy, where chemical effects are negligible for large separations between dissimilar atoms. Within the framework of this discussion, the total BFS contribution Ezof each atom in the alloy can be graphically depicted by the combination of strain and chemical effects shown in Fig. 4. In what follows, we present the necessary equations and concepts needed for the computation of each energy term.

2.1. Calculation of the BFS strain energy The BFS strain energy can be computed by any method appropriate for the calculation of pure element crystals. Due to its consistency with the determination of the chemical energy contribution, we choose the ECT [25,26] for its computation. ECT is based on an exact relationship between the total energy and atomic locations, and applies to surfaces and defects in both simple and transition metals as in covalent solids [25]. Lattice defects and surface energies are determined via

40

perturbation theory on a fictitious, equivalent single crystal whose lattice parameter is chosen to minimize the perturbation. The energy of the equivalent crystal as a function of its lattice constant, is given by a UBER [27]. The method can be easily applied to calculate surface energies, surface reconstructions and bulk distortions of metals and semiconductors. ECT is based on the concept that there exists for each atom i, a certain perfect equivalent crystal with its lattice parameter fixed at a value so that the energy of atom i in the equivalent crystal is the BFS strain energy contribution Es. This equivalent crystal differs from the actual ground state crystal only in that its lattice constant may be different from the ground state value. We compute Es via perturbation theory, where the perturbation arises from the difference in the ion core electronic potentials of the actual defect solid and those of the effective bulk single crystal. For the sake of simplicity, the formal perturbation series in ECT is approximated by simple, analytic forms which contain a few parameters, which can be obtained from experimental results or first-principles quantum mechanical calculations. The simplified perturbation series for Es is of the form ~"iS --

* Ec' F[al(i)] + ~ F [ a 2 (*i , J)] + ~ F [ a 3 (*i , J, k)]

j

j,k

+

* p, ~_F[a4(i, p,q

q)] l

J

(3)

where F[x]

= 1 - (1 + x ) e - x

(4)

Four different contributions to the energy of atom i, which find their origin in four different perturbations, are singled out. The linear independence attributed between these four terms is consistent with the limit of small perturbations which is assumed in the formulation of ECT. Correspondingly, four different equivalent crystals have to be determined for each atom i [25]. The first term, e[a~(i)], contributes when average neighbor distances are altered via defect or surface formation (i.e., changes in coordination). It can be thought of as representing local atom density changes. In most cases this "volume" term is the leading contribution to E~ and in the case of isotropic volume deformations, it gives Es to the accuracy of the UBER [27], given by Eq. 4. The value of a~(/), the scaled lattice parameter of the first equivalent crystal associated with atom i, is chosen so that the perturbation (the difference in potentials between the solid containing the defect and its bulk, ground state equivalent crystal) vanishes. This requirement translates into the following condition from which a~(i) is determined:

41

gRPe_t~R 1 + MRPe-(a+ ~)R2 =

2

rf e-(a+ S(rj))rj

(5)

j(defect)

where the sum over the defect crystal or surface is over all neighbors within nextnearest-neighbor distance. 1) is the actual distance between the reference atom i and a neighbor atom j, N and M are the number of nearest-neighbor (NN) and next-nearest-neighbors (NNN), respectively, of the equivalent crystal (12 and 6 for fcc, 8 and 6 for bcc). The ECT parameters p, {~ and 9~for all the elements used in this work are listed and described in Table 1. S(r) is a screening function given by

i

S(r)=

(6)

1 - cos ~(r 2 _ rl)_]

for r I < r < r 2, S(r) = 0 for r < r I and S(r) = 1/~, for r > r 2, and R 1 and R 2 are the NN and NNN distances in the equivalent crystal of lattice parameter a~, which is obtained by solving Eq. 5. The equivalent lattice parameter a 1 is related to the scaled quantity al(i) via 9

- rWS E

al =

1

(7)

where rws E is the equilibrium Wigner-Seitz radius, l is a scaling length,

Table 1 Computed input parameters for Ni, Cu, Pd, Pt and Au Experimental results Lattice Parameter (/~)

Cohesive Energy (eV/atom)

Bulk Modulus (GPa)

Ni

3.524

4.435

187.48

6

Cu

3.615

3.50

142.12

Pd

3.89

3.94

Pt

3.92

Au

4.078

ECT parameters

(A-1)

(A)

(A -1)

3.015

0.270

0.759

6

2.935

0.272

0.765

195.83

8

3.612

0.237

0.666

5.85

288.54

10

4.535

0.237

0.666

3.78

180.74

10

4.339

0.236

0.663

42

l =

Ec 12~Borws E ,

(8)

cl is the ratio between the equilibrium lattice constant and rws E and where B o is the bulk modulus. The higher order terms are relevant for the case of anisotropic deformations [25]. The second term, F[a*2(i,j)], is a two-body term which accounts for the increase in energy when N bonds are compressed below their equilibrium value. This effect is also modeled with an equivalent crystal, whose lattice parameter is obtained by solving a perturbation equation given by

NRPle-o~R~- NR~e -~176+ A a R ~ (Rj - Ro) e-~(RJ- Ro)

= 0

(9)

J

where ~ - 4t~ for metals [25], R 1 is the NN distance of the equivalent crystal associated with the deviation of NN bond length Rj from R o, and R o is the bulk NN distance of a pure crystal of lattice parameter a e, at whatever pressure the solid is maintained. A 2 is a constant determined for each metal [25]. The scaled equivalent lattice parameter is then , a2 =

(R-~)-rWSF-,

l

(10)

The third term, F[a~(i,j,k)], accounts for the increase in energy that arises when bond angles deviate from their equilibrium values of the undistorted single crystal, and the fourth term, F[a:(i,p,q)], describes face diagonal anisotropies (see Ref. 25). For the topics of interest for this work, these two terms can be neglected, as typical contributions from these anisotropies are exceedingly small for fcc and bcc metals. When ECT is applied to the study of surfaces of monatomic crystals, all four terms should be included in the calculations. However, when considering rigid surfaces (i.e., no interlayer relaxation) all bond lengths and angles retain their bulk equilibrium values, thus Fta2J = Fta3J -- F [ a 4 ] - 0. The rigid surface energy is therefore obtained by solving for the "volume" term represented by Fta~l only. If we consider a rigid displacement of the surface layer towards the bulk, as is the case in most metallic surfaces, the higher-order terms become finite: some bonds are compressed, contributing to Fta2J, the bond angles near the surface are distorted as well as the difference in length between face diagonals in some cases, generating an increase in energy via Fta~l and Fta:J, respectively. For the cases studied in this work, those additional contributions to ~ are generally small, usu9

,

,

9

43

ally representing 1% to 2% of the total energy. In this approximation (i.e., ignoring the third and fourth term in Eq. 4), the method can be further simplified by avoiding the solution of Eq. 9 and determining the bond-length anisotropy term, , F[a2], with an alternative scheme [26]. In this approximation, which we will call ECT in the rest of this work, the corresponding energy contribution is directly computed using Ns

F~2

=

E ~

Mn v~mn

~_~

n= lm= 1

-~nmnF(amn)

(11)

where N s is the number of atoms in the solid, 0,, = 1 if a*mn
emn c1

9

amn -

_ rws E

1

(12)

where Rmn is the distance between atoms m and n. Following the guidelines described above for the implementation of ECT, we now apply this formalism to the calculation of the BFS strain energy contribution, s , of atom i. To do so, the ECT perturbation equation is written in terms of the distances rj between atom i and its nearest- and next-nearest-neighbors,

N RPl'e-a'"~+ M RPie

: E rj e J

(13)

where rj denotes the distance between the reference atom i and its neighbor j, and the sum runs over NN and NNN. This equation determines the lattice parameter of a perfect equivalent crystal where the reference atom i has the same energy as it has in the geometrical environment of the alloy under study. R 1 and R e denote the NN and NNN distances in this equivalent crystal. Once the lattice parameter of the (strain) equivalent crystal, s , is determined, the BFS strain energy contribution is computed using the UBER [27], which contains all the relevant information concerning a single-component system:

S

= E

1-(l+a

i ) e -as"

where the scaled lattice parameter a Si * is given by

(14)

44

s* ai

(a s - a~) = q

(15)

1

where q is the ratio between the equilibrium Wigner-Seitz radius and the equilibrium lattice parameter ae i. 2.2. Calculation of the B F S chemical energy

The BFS chemical energy is obtained by a similar procedure. As opposed to the strain energy term, the surrounding atoms retain their chemical identity, but are forced to be in equilibrium lattice sites of an equilibrium (otherwise monoatomic) crystal i. The BFS equation for the chemical energy is given by

NRfie-O~iRl.l. gRPie-( Oq+~i) - ~a. INijrfie-Oqjrl.i" gijrPie-I Oqj+~i)r2I

(16)

J

where N 0. and Mij are the number of NN and NNN of species j of atom i in the actual alloy. The chemical environment surrounding atom i is reflected in the parameters o~ij, given by

O~ii --- (Xi -i-Aji

(17)

where the BFS parameters Aji (a perturbation on the single-element ECT parameter ~i) describe the changes of the wave function in the overlap region between atoms i and j. Once Eq. 16 is solved for the equivalent chemical lattice parameter c , the BFS chemical energy is then

E

C

i(

= TiEc

C*

where Ti = 1 if is given by C* ai

c.)

(18)

1 - ( 1 + a i )e -ai

aiC*> 0 and Ti =

-1 if

aiC*< 0 and the scaled chemical lattice parameter

(a/C- a~) = q

li

.

(19)

It is worth noting that the BFS parameter A j i is, in a sense, as much a singleelement parameter that describes element i as any of the parameters previously

45 introduced, as it describes changes in the electron density in the vicinity of atom i due to the presence of atom j. It is, figuratively, a conditioned response of atom i to the presence of atom j, and it does not depend on the relative location of these two atoms, as would be the case in the context of a traditional potential approach, where there would be a distance-dependent interaction. Moreover, the definition of the chemical defect, represented by the r.h.s of Eq. 16, requires all atoms to be located at NN and NNN distances in the equilibrium reference state of atom i, so that each time the parameter Aji enters in the calculation of the chemical defect, it does so in the same way, without requiring any adjustment for the actual distance separating atoms i and j. As mentioned before, it is necessary to introduce structural information in the chemical energy in order to compensate for the fact that chemical interactions are accounted only in equilibrium situations. The coupling between strain and chemical effects is achieved by introducing a coupling function gi, given by

gi = exp(-a/s*)

(20)

where the scaled lattice parameter a S* i is defined in Eq. 7. The pure element parameters a e, E c, l, o~, ~, and the BFS parameters used in this study are listed in Tables 1 and 2.

Aji and Aij

2.3 The BFS reference state in surface alloys vs. epitaxial growth The determination of the reference state is instrumental in properly describing the alloy formation process. All the examples to be discussed in the following section deal with situations where both the adatoms as well as the substrate atoms are referenced to a state with the symmetry of the substrate as it is the purpose of this work to deal with the many details leading to the formation of different surface alloy phases. This is not always the case in general situations (i.e., epitaxial growth with a symmetry different from that of the substrate, grain boundaries,

Table 2 BFS parameters Aij and

Aji Ilk -1]

i-j

Aij

Aji

i-j

Aij

Aji

Ni-Cu

0.0309

-0.01063

Cu-Pd

-0.0431

-0.0495

Ni-Pd

-0.0396

-0.0478

Cu-Pt

-0.0585

-0.0441

Ni-Au

0.0622

-0.0506

Cu-Au

-0.0604

-0.0513

46

etc.), and for the sake of completeness it is convenient to define the framework for a general determination of the reference state. The example of pseudomorphic Cu growth on Ag(100) [24] is simple enough to illustrate this point. Experimental results show that the first layer of Cu is such that the distance separating Cu atoms from the substrate (1.7 A) is comparable to Cu-fcc interplanar spacing (1.8 A) but the distance between coplanar Cu atoms, identical to that of Ag(100) where the NN distance is 2.9 A, corresponds to the Cu-bcc lattice parameter. The reference state for such Cu atom is, clearly, Cu-bct. Moreover, the step height between the fourth to eighth Cu layer is 1.4 A, similar to the interplanar spacing in Cu-bcc solid. Therefore, the reference state of Cu atoms in different layers is different, transitioning from a bct to a bcc phase. To model the deposition process step by step, the first layer of Cu atoms has to be properly parameterized (i.e., its reference state has to be properly defined) setting the correct framework for deposition of the second layer, and so on. To fix ideas, consider the first layer of Cu atoms, closest to the substrate. As mentioned earlier, the bcc <-> fcc transition can be modeled as a series of steps where, - ac (Fig 9 1) varies from -~- ~ to 1. When writing Eq. 13, it is necessary to consider that the reference state for the Cu atom is bct, thus requiting the inclusion of every atom that participates in that transformation. A total of eighteen atoms have to be considered in the immediate environment, a) eight of which remain nearestneighbors throughout the transformations as well as b) two atoms that remain next-nearest neighbors; c) four other atoms that transition from nearest-neighbors to next-nearest-neighbors and d) four atoms evolve from higher order neighbors

afO, O)=ra

a(O, e)=rc

d( 0 , @)=rb

,/(0, e)=rd

Fig. 5: Eighteen neighbors of the reference atom (large open circle) affected during the bcc to fcc transformation, shaded according their distance to the reference atom (see text and Eqs. 2124).

47 to next-nearest-neighbors, as shown in Fig. 5. The distances between the reference atom and each of these groups of neighbors can be written as a function of and c as follows

ra

=

rb=

1+

(21)

(~),~f2

(22)

rc~- c

(23)

ra = c

(24)

The screening function S ( r ) , defined in Eq. 6, is essential in properly dealing with those atoms that evolve from next-nearest-neighbors (bcc) to nearest-neighbors (fcc), becoming a function of ~:

S,
(~-~)~/1+122) a-cos (rc-(~)~]l+

(25)

for r a < r < r d. In general, the value of ~ can be determined by studying the dependence of the BFS strain defect Q (1.h.s. in Eq. 13) on the volume per atom, Q = 08 rae p-a(#)r, + 4rPe-([a(#)+S,]rD

+4~e -(a(#)+~)r~ + 2rPe-(a(#)+~)ra

(26)

Once ~ is determined, it is a straightforward procedure to identify the single element parameters in this reference state, {r w s E, E c, l, ~,, oc}~, via first-principles methods. 3. BFS MODELING OF SURFACE ALLOYS In this section, we introduce an operating procedure for applying the BFS method to the modeling of surface alloys, illustrated later with a number of selected examples chosen so as to focus on different aspects of the methodology as well as to investigate particular features that characterize each system. The examples are chosen from a limited number of elements (Cu, Pd, Ni, Au, Pt) for which experi-

48

mental results exist and for which the BFS parameters are readily obtained from experimental databases (see Tables 1 and 2). Each example highlights a different aspect that should be considered in the early stages of surface alloy formation, and as a whole, the set of examples addresses the role of modeling as a tool for understanding the observed results. The first example, Au/Ni(110), serves the double purpose of illustrating the operational procedure and highlighting the fact that surface alloy phases can greatly differ from their bulk counterparts: Au and Ni are immiscible metals but form a peculiar one-layer surface alloy for coverage below 0.5 ML Au [17,39,40]. The next example, Pd/Ni(110) [41-43], shows an interesting competition between alloying and stress release effects, leading to particular surface alloy structures different from those that could be expected from simply maximizing the number of favorable Ni-Pd bonds. This is followed by a thorough examination of the Cu-Pd system [44-65], where the difference between Pd/Cu(100) [4459] and Pd/Cu(110) [60] surface alloys highlight the role of the crystallographic characteristics of the substrate in the surface alloy that results, followed by Cu/ Pd(110) [61,62], where it is shown that the same set of parameters can be used for any application to a given system (i.e., whether Pd is deposited on Cu or viceversa). The Pd/Cu(100) system is particularly interesting when contrasted to Pt/ Cu(100) [66], the subject of the following example, as both lead to the same type of surface alloy but through different processes, which can be identified by the modeling algorithm even at low Pd or Pt coverages. Similarly, the study of Au/ Cu(100) [67] and Au/Cu(110) [68-71] underscore the importance of understanding the fundamental atomic interactions and their dependence on the symmetry of the substrate: nearly identical substitution mechanisms in either case lead to different high coverage behavior in spite of the fact that the same surface alloy phases form for low coverage. Finally, after a brief explanation of the main features observed in Cu/Ni(100) [72,73] and (110), we discuss the codeposition of Au and Cu on Ni(110) in order to understand the competing factors between the individual binary cases and the interactions between Au and Cu leading to a 3element surface alloy [74].

3.1 Calculational procedure The objective of this section is to introduce the BFS-based methodology for a detailed study of the most important features of surface alloy formation. The methodology assumes no a priori information on the system at hand. The only input necessary consists of the basic parameterization of the participating elements and lattice structures needed, as described in Sec. 2, and a catalogue of atomic distributions, where each configuration represents a state accessible by the system under study. Each entry in the catalogue is a computational cell popu-

49 lated by a collection of atoms with no predetermined chemical identity. Clearly, the construction of the catalogue is driven by the need to answer specific questions regarding the behavior of the system. It should be constructed with enough detail to address specific features of a given system and enough generality to be able to be applied to any other system. Each configuration consists of a slab (fcc for all the examples studied in this work), terminated in a surface ((100) or (110) in the examples) where a fixed number of non-interacting additional atoms occupy sites in the overlayer (O), constituting a 'reservoir' from which atoms will be drawn to study different coverage arrangements for the adatoms. This basic configuration (i.e., slab + adatoms in the overlayer) is schematically shown in Fig. 6. This cell is periodic in the two directions parallel to the surface (S) plane. A large portion of the cell is assigned for the assembling of the different types of atoms. In what follows, only a fraction of the active portion of the cell will be shown, depending on the area of the surface plane needed to describe a specific feature. Once this catalogue of configurations is defined, the energy of formation of each state is computed with BFS and the results are plotted in the form of an energy level spectrum. While it is true that the system will most likely reach those states of lowest energy, it is also true that other metastable states, close in energy to the lowest energy state, have a role in determining the behavior of the system (i.e., the closer these states are to the lowest energy state, the greater the likelihood that these states have for appearing in the actual system). In most cases, building configurations with a small number N of deposited atoms, besides being a simple task, yields information regarding the main driving mechanisms in the process of surface alloy formation. Once the energy spectrum of these configurations is determined, a good understanding of the low coverage 0

0

0

0

0

0

0

oA 0

o 0

0

0

0

9

OtlhO ,qp

o

o

9 0

0

0

0

o

. . . . . 9

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

oOto

0

0

Reservoir

0

0

9 O 0

9 9 0

0

9 O

0

0

0

o

o

o

0

0

0

9 0 9

9 OoO 9

9

9 0

Active computational cell

Fig. 6: Schematic top view of a fraction of the computational cell. Adatoms are denoted with large and small black spheres, depending on whether they occupy overlayer or surface sites. Substrate atoms are denoted with open circles, and ejected substrate atoms are denoted with large grey disks. The example also shows a 'reservoir' of substrate atoms (far left). A (110) cell is obtained from this generic cell by appropriately changing the spacing between rows of atoms.

50

regime can be achieved by linking those states based on their likelihood to appear during the deposition process. This procedure is schematically shown in Fig. 7, where different portions of the catalogue of configurations for fixed values of N (upper row) are assigned an energy level spectrum whose states are linked by 'decay chains' based on the possibility that a given state can be physically derived from a previous one. In doing so, it is possible to identify a sequence of configurations leading to a specific pattern, as it is equally possible to rule out other configurations that could be mistakenly accepted as likely valid states for the system. In what follows, each configuration will be identified by a simple notation describing the most salient features that defines it. All atoms will be identified by their chemical identity and the plane in which they are located: overlayer (O), surface (S), first plane below the surface (lb), etc. Exchanges between adatoms

N

M

N+I

i" ~ ~176 1;oi..... 'i: ^

^

^

^

o o o o o o o

^

I

O O n n n n n

o o o o o o_o_o_oo

o

ogogogo o o o o o

o

o o o o o

o

o o

o

o

o

o

I

~ i .......o

o

9

o

9

9 o

o~

ii

.

9 o

o_o_o_o

o

ogogogo o o o o

o

Io

o 9 9

9

oOo oeo o . . . . .

9

9 9

9

o

Fig. 7: Operational procedure for B FS calculations of static catalogues. For each coverage, a catalogue is built and the energy spectrum computed. 'Decay chains' link physically connected states with lower (or, eventually, higher) energy.

51

and substrate atoms will also include a reference to their relative location after the exchange. For example, B(S)A(O)I denotes a B adatom in a surface site (B(S)), displacing the substrate A atom to an overlayer site (A(O)) at NN distance (subindex '1'). Distances greater than NN distance are denoted with the subindex 'f'. While this nomenclature can be useful for a full description of the different features that define a given configuration, in some cases specific 'defects' can be succintly described with a shorthand notation particularly suited for the visualization of the ordered patterns that emerge. If 'a' and 'B' denote substrate atoms and adatoms, respectively, then 'A' and 'b' will denote those same atoms when they occupy overlayer and surface sites, respectively. Moreover, 'b A' will represent the atom 'A' in a nearest-neighbor site of the substitutional atom 'b'. It is then straightforward to establish a correspondence between the extended notation and this shorthand, graphic, way of denoting the defects: b A = B(S)A(O)I, b+A = B(S)A(O)f, bb = B(S)B(S)I, etc. For simplicity, we restrict our calculations to zero temperature and avoid, in most cases, a detailed treatment of individual or collective relaxations. For example, in terms of collective relaxations, previous work using ECT for the study of Cu (110) surfaces shows a 6% decrease in the interplanar spacing between the first two layers, and just 0.20% increase between the next two layers, in excellent agreement with experiment [75,76]. No reconstruction is observed. It is therefore unlikely that multiplanar relaxations would play a major role in the relative energy between the low-lying energy states. Local relaxations, however, potentially have a more pronounced effect. As will be seen below, the main feature responsible for the behavior observed in most surface alloys relates to the number and type of bonds formed, but it is important, particularly in situations of low symmetry or large lattice mismatch between the participating elements, to consider such relaxations. 3.2. Au/Ni(110) The deposition of Au on Ni(110) leads to the formation of a one-layer surface alloy [ 17], in spite of the fact that Au and Ni are immiscible metals. Starting with a single gold adatom deposited on a hollow site on the Ni(110) substrate, we consider five possible configurations, shown in Fig. 8. Table 3 displays results for the energy of formation per atom of each cell. These configurations include a Au atom located: (a) in the overlayer (Au(O)), (b) in a substituted site in the surface plane (Au(S)), with the substituted Ni atom in the overlayer nearest-neighbor site (Au(S)Ni(O)I), (r same, with the substituted Ni atom in the overlayer, far from the impurity (Au(S)Ni(O)f), (d) in the first plane below the surface (lb) with the displaced Ni atom somewhere in the overlayer, (Au(lb)+Ni(O)), and (e) two planes below the surface plane (Au(2b)+Ni(O)). The intermediate columns indi-

52

(a)

0

0

0

0

(b)

Q

0

0

0

0

9

0

0

0

0

0

0

9

0

0

0

0

0

0

0

0

0

0

0

0

0

(c)

0 0

0

0

0

0

0

0

0

0

0

0

9

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

(e)

0 0

Fig. 8: Configurations with one Au atom (black disk), (a) occupying an overlayer site, a surface site with the displaced Ni atom (large grey disk) in (b) a NN site, (c) far away, and the Au atom in subsurface layers: (d) in the first layer and (e) second plane below the surface ( the Au(2b) atom is shown as a black ring underneath a Ni(S) atom (open circle).

Table 3 BFS energies (eV/atom) eAu s

gAu

eCu - EAu Co

EAu

AH

(a)Au(O)

1.3670

0.2810

-0.3506

1.2683

-0.62912

(b)Au(S)Ni(O)]

0.2125

0.6840

-0.4574

-0.1003

-0.76245

(c)Au(S)Ni(O)f

0.5326

0.5186

-0.4469

0.3009

-0.79218

(d)Au(lb)+Ni(O)

0.6873

1.6610

-0.2449

0.2806

0.66555

(e)Au(2b)+Ni(O)

3.5988

2.6699

-0.9820

0.9581

1.47555

Configuration

One Au atom in different locations (see Fig. 8). The strain energy, glue and chemical energy contributions (in eV/atom) are listed in the second, third, and fourth column, respectively. The fifth column displays the total contribution to the energy of formation from the Au atom and the last column shows the total energy of formation per impurity atom of the cell.

53 Table 4 Energies of formation AH (eV/atom) Configuration

AH overlayer

AH Surface

AH 1-below

AH 2-below

AH

(a) Free surface

0.0

1.6097

0.13198

0.0

(b) Au(S)Ni(O)1

2.4516

1.4093

0.11835

-0.0002

-0.76245

(c) Au(S)Ni(O)f

2.4604

1.4080

0.11703

-0.0002

-0.79218

Contributions per layer to the energy of formation AH (in eV/atom) of (a) a free Ni surface, (b) a surface with a Au atom in a substitutional site in the surface plane, with the substituted Ni atom nearby, (c) same, when the Ni atom is somewhere else in the overlayer. cate the values of Ju ' g A u ' and ECau-EAu CO . The energy of formation of the entire computational cell, AH, is also listed. The small difference in AH between configurations (b) and (c) can be easily explained in terms of the larger number of surface Ni atoms in (c), affected by the impurity atom and the substituted Ni atom in the overlayer. This is illustrated in Table 4, where the contributions from each layer is listed for each case. As a reference, the corresponding contribution from a free Ni surface (i.e., no Au atoms present) is also included. Most configurations with two Au atoms are energetically favored with respect to those with a single Au atom: the results show that a Au dimer immersed in the surface layer, with the substituted Ni atoms forming a dimer somewhere else in the overlayer (i.e., not sharing next-nearest-neighbors) has the lowest energy (with both dimers oriented in the [110] 'close-packed' (cp) direction) with respect to the two random Au(S) atoms and two substituted Ni(O) atoms in other locations. Following the structure of Table 4, Table 5 displays the contributions per atom from different layers for the following configurations: (a) two isolated Au adatoms in the overlayer (2Au(O)), (b) a Au dimer in the overlayer (Au(O)Au(O)l), (c) a Au dimer in the surface plane with the Ni dimer in the overlayer in nearest-neighbor sites, ([Au(S)Au(S)I,Ni(O)Ni(O)I]I), (d) a Au dimer in the surface plane with a Ni dimer somewhere else in the overlayer ([Au(S)Au(S)I,Ni(O)Ni(O)I]f), (e) two Au atoms in surface sites with a Ni dimer somewhere else in the overlayer ([Au(S)Au(S)f, Ni(O)Ni(O)I]f), (f) a Au dimer one plane below the surface with the Ni dimer somewhere else in the overlayer (Au(lb)Au(lb)l+Ni(O)Ni(O) 1) and (g) a Au dimer two planes below the surface with the substituted Ni dimer in the overlayer (Au(2b)Au(2b)l+Ni(O)Ni(O)l). These last two cases are included to highlight the fact that Au dimers penetrate at most into the surface layer. The energy of formation of the cell per impurity atom is also listed, as well as the results for a free surface. The last entry in Table 3 and

54 Table 5 Energies of formation AH (eV/atom) AH (O)

AH (S)

AH (lb)

AH (2b)

0.0

1.6097

0.13198

0.0

(a)2Au(O)

1.26830

1.4141

0.11679

0.0

-0.62912

(b)Au(O)Au(O)l

0.80682

1.4151

0.11687

0.0

- 1.08063

(c)[Au(S)Au(S)I,Ni(O)Ni(O)I]I

2.01953

1.2915

0.11016

-0.0004

-1.04367

(d) [Au(S)Au(S)1,Ni(O)Ni(O)1 ]f

2.03769

1.2812

0.10741

-0.0004

-1.14302

(e)[Au(S)Au(S)f, Ni(O)Ni(O)l] f

2.03769

1.5096

0.12453

-0.0001

-1.19294

(f)Au(lb)Au(lb)l+ Ni(O)Ni(O) 1

2.03769

1.4176

0.14167

-0.0022

0.37305

(g)Au(2b)Au(2b)l+ Ni(O)Ni(O)I

2.03769

1.4229

0.10852

-0.1701

1.6533

Configuration Free surface

AH

Contribution per layer to the energy of formation AH (in eV/atom) for several configurations with two Au atoms (0.036 ML).

B+B b+b+A+A

bAB

.411-

0.0

-0.16

-o.41

BB -0.45

bb + AA ~ -0.51

b+b+AA -0.56

Fig. 9: Decay chain starting with two Au atoms in the overlayer (B+B), decaying to either a Au dimer in the overlayer (BB) or exchanging sites with Ni surface atoms. A and B denote Ni and Au atoms, respectively, in the overlayer while b indicates a Au atom in a surface site. b A indicates a Au atom in a surface site and a Ni atom in the overlayer in nearest-neighbor sites. The gain in energy (in eV/atom) with respect to the B+B state is also indicated.

55 (a)

0

0

0

0

0

0

0

0

(b)

0

0

0

0000

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

(d)

O0

000 0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

O0

(f)

000

(g)

0

O0

O0

(e)

0

0

0,0

(c)

0

0

0

O0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

000

9

9

(h)

011

011

OQ

Fig. 10: Schematic representation of the Ni(110) surface with four Au atoms in the overlayer. Open circles indicate Ni atoms in the surface layer, Au atoms in the overlayer are indicated by black disks. These configurations correspond to a Au coverage of 0.067 ML.

56

Table 6 Energy of formation AH (eV/atom) Config.

AH

Config.

AH

a

- 1.29000

e

-1.06248

b

-1.08063

f

-1.06191

c

- 1.07830

g

- 1.05311

d

-1.07193

h

-1.03737

Energy of formation per impurity atom AH for Au coverage of 0.067 ML, for the configurations indicated in Fig. 10.

the last entry in Table 5 show that, in spite of the ordering found in the surface plane (i.e., the tendency for Au atoms to form dimers and occupy substitutional sites in that plane), the Ni-Au system phase separates in the bulk. These results underscore the possibility that after deposition dimers tend to form on the overlayer along the cp direction, and later occupy substitutional sites only in the surface plane of the Ni substrate, a characteristic feature observed in the STM image [17]. The results in Tables 3 and 5 also indicate that two isolated Au adatoms realize a greater gain in energy by diffusing along the surface and forming a dimer (b) than if they exchange places with Ni surface atoms (c). With the diffusion mechanism becoming more important at higher temperatures, dimer formation would be expected in the adlayer prior to exchange to be favored. The final state, (d) ([Au(S)Au(S)I,Ni(O)Ni(O)I]f), that can be reached if the Au diffusion mechanism dominates, is however higher in energy than another alternative (e) [Au(S)Au(S)f, Ni(O)Ni(O)I] f, a state that can only be reached if the exchange mechanism of isolated adatoms dominates. Fig. 9 schematically shows the possible explanation for the dimer formation process, highlighting the fact that the lowest in energy not always is the one found experimentally. In a cell with 60 atoms in each plane, the case with NAu = 4 corresponds to a coverage of 0.067 ML. The cp chain (Fig. 10.a) has the lowest energy per impurity atom among all the possible configurations with all four Au atoms in the overlayer, as compared with other configurations shown. Not surprisingly, the next possible configuration corresponds to two cp dimers, far from each other (Fig. 10.b). This is followed by different island shapes, as illustrated in Figs. 10.c-h. The corresponding values of AH are listed in Table 6 in order of decreasing energy. However, there are several configurations with lower energy for the same coverage, characterized by Au atoms substituting for Ni atoms in the surface layer, with the displaced Ni atoms forming cp chains of 4 atoms in the overlayer. The difference between these configurations is in the relative position of

57

(a) O

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

9 0

0

0

0

0

(c) 0

0

0

0

0

9 0

(d) 0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

9

9 0

0

0

0

0

0

9

0

0

0

0

0

0

0

0

9 0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

9

(e) O

0

0

0

0

0

9 0

0

@00@

@0@|

@@@@

9

9

0

(b) O

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

9 0

0

0

0

0

0

0

9 0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

9

(f) 0 9

0 9 0

0 9

9 0

@@|

0000

@000

0

0 9

9

9

9 0

Fig. 11: Schematic representation of the Ni(ll0) surface. Open circles and large grey disks indicate Ni atoms in the surface layer and overlayer, respectively, Au atoms in the surface plane are indicated by black disks. These configurations correspond to a Au coverage of 0.067 ML. Table 7 Energy of formation AH (eV/atom) Config.

AH

Config.

AH

Config.

AH

a

- 1.39647

c

- 1.34664

e

- 1.34122

b

- 1.34872

d

- 1.34286

f

- 1.33580

Energy of formation per impurity atom, AH, for Au coverage of 0.067 ML, for the configurations indicated in Fig. 11.

58

(a) O

9

9 o

0

0

0

@

0

0

0

@

0

0

0

0

@

0

0 9

0

o

o

o

0

0

0

@

0

0

0

0

0

0

0

0

@

0

0

0

@

0

@

0

0

@

0

o

o

@

9

o

0

0 0

0

0

0

0

0

0

0

0

@

0

0

0

0

0

@

0

0

0

0

0

0

9 0

9 0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

@

0

0@00

0

0

0

9

0

0

0

9

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

o

0

0

@

o

0

o

0

o

0

o

0

o

0

o

0

o

0

o

0@@@

o

0

o

9

o

0

o

9

o

0

o

J(C~ 0 ~

o

0

o

@0@@|

o

o

(e)

@|

0

(b) O

o

0

9

0

(d)

9 0

@@@@|

9

(~

9

0

0

9

0

0

0

0

0

0

0

0

0

0

0

0

9

9

9

@

0

0

0

0

0

0

0

0

0

0

0

0

9

9

9

0

0

0

0

9

9

0

0

0

0

0

0

0

0

9 0

0

0

0

0

0

0

9

|

o

0

0

0

0

0

0

9

0

0 9

9

9 0

| 0

0

0

0

0

0

0

0

0

Fig. 12: Schematic representation of the Ni(110) surface for (a-c) 0.13 ML and (d-f) 0.17 ML Au coverage. Open circles and large grey disks indicate Ni atoms in the surface plane and overlayer, respectively. Black disks indicate Au atoms in the surface plane.

the Au atoms inserted in the surface layer. Fig. 11 indicates six of the lowest energy configurations and Table 7 lists the corresponding energies: all of them are lower than (a) in Table 6, indicating that arrangements that include long Ni chains surrounded by Au dimers are among the most likely configurations. The lowest energy configuration (Fig. 11.a) is characterized by the insertion of individual Au atoms and the formation of the Ni chain, while all the other configurations (except Fig. 11.d) display Au dimers in different locations relative to the Ni chain. The spread in energy between these states is quite small (0.06 eV/atom). The fact that Fig. 11.a has the lowest energy suggests that the insertion of isolated

59

Table 8 Energy of formation zl/4 (eV/atom) 0 = 0.13 M L A u

0 = 0.17 M L A u

Config.

z~

Config.

ZkH

a

-1.44847

d

-1.49735

b

-1.39317

e

-1.39389

c

-1.33580

f

-1.37870

Energy of formation per impurity atom, ZkH,for Au coverage of 0.13 ML (left column) and 0.17 ML (right column), for the configurations indicated in Fig. 12. Au atoms is preferred, but as discussed above, the mechanisms leading to this final state might be less favorable than those leading to alternative configurations, as shown in Fig. 9 (i.e., the highly, energetically favored formation of adlayer dimers over insertion in the surface plane at elevated temperatures). From these results one can already see, even at this very low coverage, some indication of the trends which, ultimately, would lead to the situation found experimentally. Distinctive features can be identified: the penetration of Au atoms in the surface layer, the formation of Ni chains in the overlayer along the cp direction, and the linkage of the Au atoms and the substituted Ni atoms by means of an intermediate surface Ni atom. As the coverage increases, configurations that appear in the ground state group for NAu = 8 include those where the Au adatoms form long chains along the cp direction in the overlayer, as shown in Fig. 12. This could be taken as an indication of growth of the Au film on the Ni substrate, in competition with the formation of the surface alloy. A first hint of this alternative can already be seen at a relative low coverage: Table 8 shows some results for 0.13 and 0.17 ML Au coverage, with the corresponding configurations represented in Fig. 12. A possible explanation for this change in growth patterns can be found in terms of the surface energy of Au being much lower than that of Ni. For low coverage, Au atoms benefit from locating themselves in the surface layer, with the Ni atoms forming islands in the overlayer. At one point, the increase in surface energy due to the existence of large Ni islands becomes larger than any gain generated by the intermixing of Au and Ni atoms in the surface plane. Therefore, configurations with Au islands on the Ni substrate become energetically favored, thus reverting to a growth mode where Au atoms tend to form a pure Au layer. The breaking point between these two regimes seems to be around 0.5 ML Au. For higher coverage, there is experimental indication that alternative 3D patterns exist [39].

60

3.3. Pd/Ni(110) In a Ni(ll0) surface with terraces and steps, Pd atoms deposited on a clean terrace seem to nucleate chains or elongated islands with preferential growth along the [110] direction of maximum diffusion, but also suggesting growth along the [001] direction [43]. Scanning tunneling microscopy results suggest that the total surface of the islands is nearly twice the quantity of Pd deposited, suggesting the presence of a Pd-Ni surface alloy. These islands are of monatomic height (1.5 A1.6 A), consistent with bulk Pd interlayer spacing (1.4 A), but not ruling out that these might be islands of mixed composition. The morphology of the substrate is influential in the shape and distribution of these islands: there is a difference on whether they form on a clean terrace or in the vicinity of a step, where Pd atoms show a tendency for attaching themselves to irregularities in the steps. In agreement with experiment, pseudopotential density functional theory (DFT) calculations [41] show that Pd-Ni alloying is favored, although there is a weak dependence of the energy on the actual number of Pd-Ni bonds, that is, for a given Pd coverage, there is no substantial difference between different ordering 0

9

|

9

0

0

0

9

0

0

0

0

0

0

0

0

0

0

0

0

0

9

0.0427 0

9 0 0

9 0

0.0189

0

9

0

0

0

0

0

0

0

0

9

9

0

0

0

9 0

0

0

0

0

0

9 0

0

0

9

9

9

0

0

0

0

0

0

0.01040

@

0 9 0

0.0209

9

|

0

|

0.0210 0

9

9

0

0

0

0

0

0

0

9 9 0

0.0

Fig. 13: A Ni(110) surface (open circles) with two Pd atoms (black disks) in different arrangements of surface sites and two displaced Ni atoms (large grey circles) in the overlayer. The energy difference (in eV/atom) between each configuration and the lowest energy state is shown.

61

patterns in the surface alloy. BFS modeling of a simple case with just two .Pd atoms, as shown in Fig. 13, suffices to show the small energy spread between states characterized by different ordering patterns. These results are comparable to DFT calculations of an 8xl cell with two Pd atoms, showing an energy difference of approximately 30 meV between configurations where the two Pd atoms are located at neighboring sites or away from each other by 10 A, suggesting that the number of Pd-Ni bonds does not significantly change the energy. However, in spite of these small differences, the lowest energy configuration corresponds to a configuration that maximizes the number of Pd-Ni bonds. All these results are consistent with the solid solution bulk behavior, also to be expected in a Pd-rich surface of a PdNi alloy [42]. The weak dependence of the energy on the number of Pd-Ni bonds means that there could be an alternative process competing with alloying effects, that can be identified as the stress release linked with the size mismatch between Pd and Ni. Evidence can be obtained from DFT calculations showing that a disordered surface alloy is favored due to the competition between alloying effects (Pd-Ni bond creation) and stress release effects (leading to 'segregation'). A simple way to model this competition consists of considering different ordering patterns, as A

w

A

~

w

w

(a) p(2xl) A

w

w

~

~

w

v

(c) unrelaxed p(4xl)

A

dpclecl

dNiNi

A

~.

w

I,

A

O---G

A

9

O---G

w

9

(d) relaxed p(4x 1) (b) p(2x2)

Fig. 14: Pd (black disks) and Ni (open circles) in different ordered cells: (a) p(2xl), with a NiPd-Ni-Pd chain, (b) p(2x2), with two displaced Ni-Pd-Ni-Pd chains, (c) rigid and (d) relaxed p(4xl), with alternating Pd-Pd and Ni-Ni dimers.

62

shown in Fig. 14: (a) a p(2xl) cell with Ni-Pd chains, (b) a p(2x2) cell with displaced Ni-Pd chains and a p(4xl) cell with Pd-Pd and Ni-Ni (c) rigid and (d) relaxed dimers. BFS results and DFT calculations agree on the fact that Pd-Pd enlarged pairs (2.674 A (BFS) vs. 2.66 A (DFT)) together with Ni-Ni compressed pairs (2.381 A (BFS) vs. 2.36 A (DFT)) (Fig. 14.d) are favored over the p(2xl) and p(2x2) cells, restricted by symmetry to remain in epitaxial sites. While configuration (d) does not maximize the number of Pd-Ni bonds, it is the only one that allows for stress release. The gain in energy due to the relaxation of the PdPd and Ni-Ni dimers is enough to make this configuration more favorable than any other ordered arrangement, as can be seen in the energy level spectrum corresponding to these configurations, shown in Fig. 15. As the Pd coverage increases, not only the number of Pd-Ni bonds decreases but there is also less opportunity for stress release. The stress release mechanism stabilizes the disordered surface alloy and as its likelihood diminishes, a dealloying process sets in.

3.4. Pd/Cu(100) The abundant experimental results for Pd/Cu(100) make this system an ideal framework for modeling the many recognized features during the early stages of surface alloy formation [44-59]. The first question is whether Pd atoms deposited on Cu(100) do or do not penetrate in the surface layer, and if they do, what hap-

unrelaxed PdPd-NiNi

0.075

p(4xl) 0.050

0.025 Ni-Pd-Ni-Pd II

p(2x2)

N~ relaxed PdPd-NiNi

Fig. 15" Energy difference AE (in eV/atom) between unrelaxed and relaxed configurations of a Pd/Ni(110) surface alloy for 0.5 ML Pd coverage.

63 (a) Pd(O)

0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0000000 (b)[Pd(S)+Cu(O)]l

@

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0000000 (c)[Pd(S)+Cu(O)]f

@

0 0 0 0 0 0 0 0 0000000 0 0 0 0 0 0 0 0 0000000

(d)[Pd(lb)+Cu(O)]2

@

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (e)[Pd(lb)+Cu(O)]f

@

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (O[Pd(2b)+Cu(O)]f

|

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Fig. 16: Side view of a Cu slab with a (100) surface, showing different stages of interdiffusion of a Pd adatom: (a) the Pd adatom (black disk) in the overlayer (Pd(O)), (b) occupying a surface site with the ejected Cu atom (large grey circle) at NN distance ([Pd(S)+Cu(O)]I), (c) same, but with the Cu(O) atom decoupled from the Pd(S) atom ([Pd(S)+Cu(O)]f), (d) occupying a site one layer below the surface layer and the Cu atom at NNN distance ([Pd(lb)+Cu(O)]2), (e) with the Cu atom decoupled from Pd(lb) ([Pd(lb)+Cu(O)]f) and (f) with Pd in the second layer below the surface and the Cu atom in the overlayer ([Pd(2b)+Cu(O)]f).

pens to the ejected Cu atoms. Fig. 16 shows the set of configurations designed to study the different options for the Pd and Cu atom and the ability of the Pd atom to interdiffuse in the Cu slab. The corresponding energy level spectrum, shown in Fig. 17, clearly indicates a strong preference for a [Pd(S)+Cu(O)] 1 state, as measured by the gap between this and any other state in the spectrum. At this level of coverage, the position of the Pd(lb) or the Pd(2b) state indicates that it is highly unlikely that Pd will penetrate the Cu slab beyond the surface plane. The tightly 'bound' ground state (lowest energy) hints to the possibility that this particular arrangement could constitute the building block of more complicated structures during the growth process. This can be seen by examining the case of two Pd atoms. The lowest energy configuration, shown in Fig. 18, corresponds to one where both Pd atoms locate themselves initiating a chain in surface sites along the [010] direction with the ejected Cu atoms closely attached to them, in agree-

64 0.70

-

(a) Pd(O) 0.60 -0.50 -0.40 0.30 0.20

-

0.10

-

.........4~ (f) [Pd(2b)+Cu(O)]f ~.........~(e) [Pd(lb)+Cu(O)]f " " " " ~ d ) [Pd(lb)+Cu(O)] 2 ~"'~(c)

[Pd(S)+Cu(O)]y

(b) [Pd(S)+Cu(O)]I

0 . 0 0 _

Fig. 17" Energy spectrum of the configurations shown in Fig. 16. AE is the difference between the energy of formation per adatom (in eV/atom) of a configuration and the lowest energy state. ment with experiment [44,45]. Moreover, the structure of the ground state can be seen as two coupled N = 1 states: [Pd(S)Cu(O) 1, Pd(S)Cu(O)l] 1. Continuing this coupling scheme at higher coverage results in the formation of the c(2x2) structure, in agreement with experiment. The ejected Cu atoms either migrate to nearby steps or nucleate on top of alloyed areas resulting in subsurface Pd, as shown in Fig. 19.a-c. The preference for the formation of a Pd c(2x2) structure is highlighted by the fact that alternative ordering patterns (even those that show a slight departure from the c(2x2) ordering) are much higher in

0

0

0

0

0

0

9

0

o0o

0~0

0

0

0 0

0 0

Fig. 18" Lowest energy configuration with two Pd atoms on a Cu(100) plane, corresponding to 0.033 ML Pd coverage. Pd atoms are denoted with black disks. Ejected Cu atoms are denoted with grey disks while surface Cu atoms are indicated with open circles.

65

(a)

0

0

0

o

0

0

0

0

0

(b)

0

0

9

9

0

o

0 0

o

0

0 0

0

0

9

0

0

0

0

0

0 0~0

0

0

o0o@o

0 0

0~0 0 0 0 0 0 0 9 0 0 0 0 0

9

0

0

0

0 0

0 0

0

0

0

0

0~0

0

o@o@o

0 0

01~0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

0

0

0

0 o

(c) 0

0 o

0 0

0

0

0

0

0 0

0

0 0 0

0 o

0

0 0

0

0 o

0 o

0

o@o@o

0

0~0

0 0

0

o 0~0

0 0

0 0

0

0.2 --~

(a) [Cu(i)+4Pd(S)[lOO]]f

(d) 0

(b) [Cu(i)+3Pd(2x2)n]l+Pd(lb) (c) [Cu(i)+4Pd(2x2)]1

Fig. 19: (a-b) Two low-lying states (in energy) with four Pd atoms (black disks) and four ejected Cu atoms (large grey disks) in the overlayer of a Cu (open circles) (100) substrate. (c) Lowest energy configuration showing four Pd atoms forming a c(2x2) structure. (d) The energy level diagram indicates the difference in formation energy (in eV/atom) with respect to the ground state. All three configurations include a four atom Cu island in the overlayer (Cu(i)), a number m of Pd atoms in surface sites following a c(2x2) pattern (mPd (2x2)) and, in configuration (b), a Pd atom in the first layer below the surface (Pd(lb)).

66

o@o@o@o @o@o@o@

o@o@o@o @o@o@o@

0

0

0

0

0

0 0

0

0 0

0 0

0

0

0

0

0

0 0~0

0 0

o|174

0

0

0

0 0

0

0~0

0

(a) 0.198 eV/atom

(b) 0.326 eV/atom

0 0

0 0

o@o@o

0~0

o@o@o 0 0 0

0

0 0

0

0

0~0

0

0

0

0

0

0~t0 0 0 0 0 0 0 0 0 0 0 0 9 9 0 0 0 0 0 9 9 0

0

0

0

0

0 0

0 0

0 0

0

0

0

0

0 0

0

0

0

0

o@o@o@o |

0 0 0 0 0

0

0~0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 9 9 0 0 0 9 9 O

(c) 0.0

Fig. 20: A Cu island on a Cu(100) surface (open circles) (a) alloyed with a c(2x2) Pd structure in the surface layer, (b) somewhere between the c(2x2) Pd structure and a terrace and (c) adjoining a terrace. Cu atoms in the overlayer are indicated with large grey disks, Pd atoms with black disks. The energies of (a) and (b) are referred to the lowest energy configuration (c).

energy and therefore unlikely to occur, as illustrated in Fig. 19.d. However, the low-lying energy states (all including c(2x2)-like patterns) also include states where some Pd atoms interdiffuse below the surface layer, a trend enhanced for higher Pd coverage (Fig. 19.b). While this feature hints to the possibility of much more complex alloying patterns for Pd coverage close to 0.5 ML, it also establishes a limit for the validity of the restrictions imposed in the current calculation, as subsurface Pd penetration would require a full treatment of collective atomic relaxations. In fact, experiment indicates a clock-rotated phase with the c(2x2)p4g symmetry [44], consistent with a substantial displacement of the subsurface atoms [44]. In spite of the fact that these are static calculations, the motion of Cu islands or Cu atoms in the adlayer can be modeled by analyzing configurations where Cu resides on top of surface Pd atoms, attached to a nearby step, or somewhere in between, as suggested in Fig. 20. The results show that Cu migrates to nearby steps, with decreased mobility for increasing Pd coverage (i.e., the difference in energy between alloyed spots and steps decreases).

67 Summarizing the results for Pd/Cu(100), the BFS calculations successfully reproduce all the features observed experimentally up to 0.5 ML Pd coverage [44,45], some of which are represented in Fig. 21: the alloying of Pd atoms in the surface in the [010] and [100] directions, the nucleation of Cu islands on top of alloyed areas, the decreased mobility of Cu islands with increasing Pd coverage, the formation of a c(2x2) phase as the chains converge, and the interplay between the c(2x2) phase and the initiation of second layer growth at increasing coverages before the c(2x2) is fully completed.

N=I

N=2

N=3

N=4

N=5

N=6

N=7

m

m

N=8

0"90 ~ m m m

0.80 -0.70 0.60 0.50 0.40 0.30 0.20 0.10

m

m

m

m

0.00 ooo o o

9

o~o

ooo

ooo o

o

e

o

o

oee

eeoee

~

o

o 9o

o

oe o

o ee 9 o

0

0

0

0

0

0

o

e

ooo

9 o

o

o

9 o

eeoee 0

0

0

oeo eOo@e_ _

o e o

oeeeo

o

0

e9 o @ e ee

9 o

9e o

o

e e eo

e

0

0

0

0

0

o e o

e o e

eO o 9 9 ee o 9 o

o 9e 9 o ee 9 o 9

0

9e eo 9 0

e 0

o e ee e| 9 0

o 9

Fig. 21: Summary of the energy level spectra for N Pd atoms (N = 1.....8). The lowest energy state is indicated with a thick line. The thick dashed line indicates the first configuration that includes a Pd atom below the surface layer (Pd(lb) or Pd(2b)). The dotted lines indicate the first configuration that includes Pd atoms in the overlayer (the fact that such states do not appear for N=I, 6 and 8 is because configurations with Pd(O) were not included in the catalogue of selected distributions (N = 6, 8) or because they exceed the highest energy shown (0.90 eV/atom)). The lowest energy configuration for each value of N is also shown. The unit cell is rotated by 45~ with respect to previous figures depicting a (100) surface.

68

3.5. Pd/Cu(110) Once again, it is seen that Pd atoms exchange sites with Cu surface atoms [60,77]. However, the particular features of an fcc (110) surface turn this Pd surface atom into a strong nucleation point for Cu atoms in the overlayer, which in turn stimulates the formation of Cu chains in the overlayer, as opposed to Cu islands attached to alloyed surface regions. An energy level diagram representing this process is shown in Fig. 22. Configurations with neighboring Pd atoms in surface sites are substantially higher in energy than those where Pd atoms interact only through Cu adatoms. Within the BFS framework, and in agreement with experiment, no additional features, like Pd penetration in subsurface layers, are observed. For increasing Pd coverages the growth pattern remains the same, resulting in the formation of Cu chains in the adlayer, coupled with alternating Cu-Pd-Cu surface atoms, as shown in Fig. 23. Not surprisingly, this experimentally observed pattern [60] corresponds to the [010] direction in a (110) termination of an L12 Cu3Pd alloy. The lack of alternate ordering patterns that compete energetically with the one observed, leads to the conclusion that for higher Pd coverages, the Cu-Pd chains observed will eventually coalesce into what amounts to a Cu3Pd (110) surface, with full coverage of Cu with a 2:1 Cu:Pd ratio, as observed experimentally [60]. 3.6. Cu/Pd(11 O) Experimental results for this system show that the observed growth pattern is substantially different from the systems where Pd is deposited on a Cu substrate [61,62]. The energy level diagram shown in Fig. 24 indicates that Cu does not substitute for Pd surface atoms. Instead, the state of minimum energy corresponds to the Cu atom attached to a Pd step edge in the [110] direction, with the step edge acting as a nucleation center, favoring the growth of 1D Cu chains. While this trend is also observed at higher coverages, the preference for separate anchoring points in the step edge is diminished at higher coverages, suggesting a 2D growth of the linear chains. Still, it is expected that a clear separation between 1D chains will dominate the low coverage regime. The energy levels in Fig. 24 correspond to selected configurations with increasing Cu coverage, clearly showing that Cu atoms are always trapped by a substrate edge or existing Cu islands. Consistent with the modeling results, the Cu 'fingers' in the [110] direction are observed experimentally for room temperature and low (<0.1 ML) Cu coverage, as well as the formation of 2D islands for higher Cu coverages [60]. 3.7. Pt/Cu(100) From the modeling standpoint, the deposition of Pt on Cu(100) is best described

69 N=0 0.00

-0.20

N =1

N=2

N=3

m

--

-0.40 - -

bAA

-0.60 - -

A

AbA

-0.80 - -

AA

bAAA

-1.00 - -1.20

o

-1.40

o

o

o

o

o

o

o

o

o

AbAA

@ o

9

bAAAA

@ 0 -1.60

o

o

o

AbAAA ~'

AAbAA

-1.80 Fig. 22: Energy level spectra for N = 0 through N = 3 states, starting with the seed state b A, a Pd atom in a Cu surface site (N denotes the number of extra Cu atoms). Subsequent configurations result from the addition of Cu atoms in the overlayer. AE (in eV/atom) indicates the difference in energy between any given configuration and the initial state in the chain (bA). The inset displays one of the configurations shown in the diagram, where a Pd atom is in a surface site and three Cu atoms occupy adjoining sites in the overlayer. In the Pd/Cu(ll0) case, a larger value of the ECT parameter X was used (X = 10 ,~) in order to account for the asymmetry of the surface.

0

|

0

o

0

0

@ @ 9

0

0

o

0

0

9 9

0

o

AAA A b b

0

Fig. 23: Lowest energy configuration for two Pd atoms in surface sites (black disks), linked by a Cu chain along the close-packed direction in the overlayer (large grey disks).

70

N=I

3.25

--

3.00

m

2.75

--

2.50

--

2.25

m

2.00

--

N=2

110

N=3

N=4

i

IIO

1.75

1.50

--

IIDO.. 1.25 13(X3

roB8

1.00 -

/

0.75

Fig. 24: 'Decay' chain for Cu deposition on Pd(110) for increasing Cu coverage (N denotes the number of Cu atoms). Open circles represent Cu atoms in an overlayer site, and solid squares represent Pd steps. Cu atoms attached to the right of the step represent Cu 'fingers' growing in the [ 110] direction. If two consecutive fingers are at more than nearest-neighbor distance, they are represented by slightly separated circles, otherwise, consecutive fingers in contact with each other represent 2D Cu islands attached to a Pd step edge. AE is the energy of formation of the computational cell. A solid circle represents a Cu atom in a surface site, with the displaced Pd atom (in the overlayer) represented by a solid square. within the framework of the previous results found for Pd/Cu(100), due to the fact that in both cases, the basic step in the beginning of the alloy formation process is the insertion of a Pt (or Pd) atom in a surface site, with the ejected Cu atom in a NN site in the overlayer. In spite of this similarity, the coupling between subsequent defects is different in each case, thus leading to different surface alloy patterns. Fig. 25 shows the two possible coupling schemes between two X(S)Cu(O) 1 (with X = Pd or Pt) defects and their respective energy differences with the lowest energy configuration. In the Pd/Cu(100) case, the lower

71

(a)

o

o o 0~0 0 0~0 0 0 0 0 0 0 0 0

0

o

Pd/Cu(100)"

0.0

Pt/Cu(100)"

0.0776

(b)

0 0 0

0 0

0

9

0

o0o

O

0

O

0~0 0 0

0

Pd/Cu(100)"

0.008

Pt/Cu(lO0):

0.0

Fig. 25: Coupling between two X(S)Cu(O) 1 'defects' on a Cu(100) surface (X = Pd, Pt). In each case, the energies listed (in eV/atom) are referenced to the lowest energy configuration. Cu(S), Cu(O) and X(S) atoms are denoted with open circles, large grey disks and black disks, respectively.

strain of the Pd atoms in the surface sites allows for c(2x2) growth, where two Pd atoms are located in NNN sites sharing one of the ejected Cu atoms (Fig. 25.a), thus maximizing coordination for the whole set of atoms. In the case of Pt/ Cu(100) the opposite is true. Pt atoms 'repel' each other, only linking through the Cu(O) atoms (Fig. 25.b). The effect of coupling Pt(S) atoms with the Cu(O) island can be factored out by considering two configurations where the only difference is the relative location of the two Pt(S) atoms, as shown in Fig. 26, which is consistent with the experimentally observed p ( l x l ) LEED pattern [66]. The consequence of the different coupling schemes and the relative location of the Cu island that forms with the ejected Cu atoms can be seen in the Net = 4 case. Three configurations should be considered, as shown in Fig. 27: (a) a p(2x2) pattern, (b) random Pt(S) atoms and (c) a c(2x2) pattern. The lowest energy configuration corresponds to a p(2x2) pattern, closely followed by a ran-

(a)

0

0 9

0

0

0

0 0

0

0

9

0 o

0.0

0

0

o0

0~0

(b)

0 0

0

0

0

0

9 9

0 0

0 0

0

o

0

o0

0~0

0.05

Fig. 26: 'Repulsion' between Pt atoms (black disks) inserted in Cu(100) surface sites (open circles). The configuration where both Pt(S) atoms are located in nearest-neighbor sites is 0.05 eV/atom higher in energy than the one where both atoms are randomly located in the surface layer. In both cases, the energy gain due to coupling to the ejected Cu(O) atoms (large grey disks) is ignored by locating the Cu atoms away from the Pt(S) atoms.

72

(a)

o

o

o

0

o

o@o

0

e@o@e o@o 0

0

9

o

9 0

o

(b)

o

o

o

9

o

0 0

0

o

9

o

o

0 0

0.0 (0.004)

9

0

0

o

0

o

o@o

0

0.118 (0.003)

0

o

o

o@o| o@o 0

(c)

o 9

0

0

o

0

o

eQe

0

o0o@o e@e 0

0

0

0

o

o

o 0

0

0.164 (0.0)

Fig. 27: Four Pt atoms (black disks) inserted in Cu(100) surface sites. The configuration with the lowest energy of formation corresponds to a p(2x2) pattern, where the Cu(O) island (large grey disks) is attached to the p(2x2) patch. The highest energy configuration corresponds to a c(2x2) surface alloy. The energy gain (in eV/atom) due to coupling to the ejected Cu(O) atoms can be ignored by locating the Cu atoms away from the Pt(S) atoms, as noted in the results in parenthesis.

dom distribution of Pt(S) atoms. The c(2x2) structure, while highly favored in the case of Pd/Cu(100), is now the highest energy configuration. It is important to note that ignoring the stabilizing effect of the Cu(O) island nucleating the surface alloy patch would lead to exactly the opposite results, where the c(2x2) pattern would be favored at any coverage over random or p(2x2) distributions. For higher Pt coverage, the p(2x2) pattern is therefore favored until a critical coverage of 0.25 ML Pt is reached, when the most favorable substitutions are those leading to a c(2x2) pattern, in agreement with experiment [66]. Therefore, the Pt deposition on Cu(100) can be summarized by the sequence of a p(lxl), p(2x2) and c(2x2) patterns, with increasing Pt coverage. It is also interesting to note that Pt interdiffusion in Cu is expected, like in the Pd/Cu(100) case. In spite of the higher strain induced by larger Pt atoms inside the Cu substrate, the alloying effect of the strong Pt-Cu bonds results in the net effect of Pt interdiffusion leading to the formation of a Cu3Pt phase in the nearsurface layers.

3.8. Au/Cu(100) and Au/Cu(ll0) These systems are ideal examples to study the importance of properly understanding the low coverage regime and the way it influences further growth with increasing coverage. Experimental results for Au deposition on Cu(100) are consistent with a p(lxl) pattern which evolves into a c(2x2) pattern for coverages around and above 0.5 ML Au [67]. In the case of Au/Cu(ll0) however, it is observed that a c(2x2) pattern, characteristic of low coverage, is followed by

73 (a)

0 0

o

0 0

0 0

0

0

(b) 0

0

0

0

o0o 0

0

0

0

o

0

0

0

0

0

0

0

9

(d)

0 0

(f) 0

0

0

0

0

0

0

0

0

0

0

9

0

0

0

0 0

0

0 9

0

0 O

0

0

O O

0

0

O

0

0

0

0

0

0

0

0

0

0

0

0

@ 9

0

0.33991

(g)

0 0

0.00422

0

0

0

0

0 0~0

0

0

0

e|

0

0.0

(e)

0

0.03638

0 0~0 O 0 0 0 0 0 O O

O

0

0

0.45808

(c)

0

0.0

0

0

(h) 0 0

0

9

0

0

0

@ 0

0

0

0.04742

@ 0

0

0

0

0

0.05169

Fig. 28: Configurations with one Au atom (black disk) deposited on a Cu(100) surface (open circles) located (a) in the overlayer, in a surface site with the displaced Cu atom (large grey disk) (b) in a neighboring overlayer site, (c) sharing a surface Cu atom as a nearest neighbor and (d) somewhere else in the overlayer. The difference in energy of formation (in eV/atom) of each cell with respect to the lowest energy configuration is indicated. (e-h) Same configurations for a Cu(110) substrate.

74

dealloying of Au from the Cu surface for higher coverages [68-71]. The similarities and differences between these two growth patterns can be derived from the low coverage regime, as shown in the following analysis. For the deposition of just one Au atom on a Cu(100) or Cu(110) slab (with 60 surface atoms), four basic configurations should be considered, as shown in Fig. 28: (a) the Au atom in an overlayer site (Au(O)) or occupying a surface site with (b) the displaced Cu atom at a nearest-neighbor site (Au(S)Cu(O)I), (c) sharing a common Cu(S) NN (Au(S)Cu(S)ICu(O)I) and (d) somewhere else in the overlayer (Au(S)Cu(O)f). Figs. 28.e-h show the same configurations for a Cu(ll0) substrate. For Cu(100), the lowest energy configuration corresponds to Au(S)Cu(S)ICu(O) ], where the displaced Cu atom shares one Cu(S) NN with Au(S) (Fig. 28.c). The substantial energy gain realized by an exchange with a Cu surface atom (0.45 eV/atom) favors individual substitutions that would lead to a p(lxl) pattern. For Cu(ll0) however, the Au(S) atom would rather keep the ejected Cu(O) atom as a nearest-neighbor, due to the lower coordination of atoms in a (110) surface site as compared to those in a (100) surface. Subsequent deposited Au atoms will adapt to this basic scheme, but clearly inducing the formation of different surface alloys. In the case of Cu(100), three different mechanisms compete: the insertion of Au atoms in surface sites following a specific pattern, the formation of Cu islands in the overlayer as a result of the surface substitutions Au(S)Cu(O), and the optimum location of Cu(O) atoms (a)

0

0

0

0

9

0

0 0

9

0

0

0

0

9 0

0

0

9 0

0

0

0.00340

0

0 0

0.00791 (c)

0 0

0

O~ 0~0

0

9

0 0

0 0

9 0

(b) 0

0

0

o| o0o

0

0

0.00784 0 0

0

0 0

O~ 01~0

(d)

0 0 0

0 0

0

0 9

0

0 9

o

0

0 0

oO

0~0

0.0

Fig. 29: Lowest energy configurations for two Au atoms (black disks) in a Cu(100) surface. Large grey disks and open circles denote Cu(O) and Cu(S) atoms, respectively.

75

(a)

0

0

0

0 0

0

0

0 0

0 9

0

0

0 o

(b)

0

00

9

9 9

0 0 9

9

0

0.00955

0

oO

o

0

0~0

o0o

0

0

Oe~O

o0o 9

9

0.00595 (d)

(c)

0 O~ 0 0 0~0 0 0~0 0 0 0 0 0

0

0

0

0

0.00369

0

9 9 0 0 0 0 0 9 0~0 0 0 0~0 0 0 0~0 0 0

0

0

0.00

Fig. 30: Lowest energy configurations for three Au atoms (black disks) in a Cu(100) surface. Large grey disks and open circles denote Cu(O) and Cu(S) atoms, respectively.

with respect to the Au(S) atoms. The fact that the lowest energy state includes Au(S) in specific surface sites (Fig. 29.d), indicates that the small energy gains listed in Fig. 29 make it highly probable for individual Au atoms to substitute in isolated surface sites (Fig. 29.b). This feature can be clearly seen in the NAu = 2 and NAu = 3 cases. After examining a large selection of possible distributions, it can be shown that the ground state band (low-lying energy states) is formed by a variety of configurations, some of which are shown in Fig. 29, each showing the effect of the above mentioned features in the surface alloying behavior. The ground state corresponds to the two Au(S) atoms following the beginning of a c(2x2) pattern, linked to the Cu(O) chain in a similar way as that shown in Fig. 28.c. The band of states close to the NAu = 3 ground state underscore the role of each feature in the growth pattern: states with Cu(O) islands completely detached from the Au(O) c(2x2) patch are low in energy, but even lower are those that retain the same type of links as those seen in Fig. 28.c and Fig. 29.d, thus favoring a particular orientation for the Cu(O) island relative to the c(2x2) patch. This can be seen in Fig. 30, where selected low energy NAu = 3 states are shown. As expected, the lowest energy state corresponds to that configuration where a) Au(S) atoms locate themselves forming a c(2x2) pattern, b) ejected Cu(O) atoms form a linear chain along

76 (a)

(b) 0

o 0

0

9

o

9

o@o o|174 o o0o 0

9

0

o o

9

9

0

0

o0o |174 o0o

o 0

0

0

O

O

O

O

O

O

O

O

O

O

9

O

0.00325

0.00666

(c)

(d) 0

o

o

0

0

9 0

9 0

oOo

0

9

oOooo ooo

0

9

o

o

o

o

o

o

0

0

(f)

o

0

0

0

oOo oOoOo o o0o 0.00099

0

9 0

9

oOo

9

0

o

o

o

0.00110

(e) 0

0

oOo oOoOo

0.00213

0

9

9 0

9

o

9 0

o

9

o

0

0

o0o oOoOo o0o 0

0

0

0

9 9

0

o 0

9 9

0

o 0

0.0

Fig. 31" Lowest energy configurations for four Au atoms (black disks) in a Cu(lO0) surface. Large grey disks and open circles denote Cu(O) and Cu(S) atoms, respectively. the [110] direction and c) the c(2x2) patch and the Cu(O) island are linked in a way that maximize the number of 'knots' seen in Fig. 28.c. The case NAu = 4 lends itself to a more complete analysis of the multiple arrangements that can be obtained regarding the shape of the Cu(O) island (a compact island or a linear chain), and the several symmetric distributions of the Cu(O) atoms that maximize the favorable linkage between Au(S) and Cu(O) atoms (Fig. 28.c). Fig. 31 shows a number of configurations highlighting this competition. It is clear, however, that during a steady deposition of Au on Cu(100), the highly favored insertion of

77

o

o

9

o

9

0

o0o0o0o 9 o0o| o@o@o@o 9 o0o|174

o

0

0

9 0

9 0

0

9 0

0

9

9 0

0 9

0

0

9 0

0

9 0

0 9

9 0

0

Fig. 32: High Au (black disks) coverage on a Cu(100) surface. Surface Cu atoms are denoted with open circles. Ejected Cu atoms forming an island in the overlayer are noted with large grey disks. Au atoms in the surface plane leads to a p ( l x l ) pattern for low coverage (Fig. 29.b). As the coverage increases, only one particular arrangement is favored, one where there is a balance between the formation of the c(2x2) surface alloy followed by the formation of a Cu island linked to the surface alloy in a particular way and orientation. This process can continue indefinitely for increasing Au coverage, which is in complete agreement between the experimentally observed dominance of a p ( l x l ) pattern followed by the increasing presence of a c(2x2) pattern of Au atoms inserted in Cu terraces [67]. Fig. 32 shows the projected pattern for higher coverages. The pattern that emerges is consistent with the 1:1 termination of the Cu3Au surface, which has been shown to be the one with the lowest surface energy [29]. The slight difference in linkage between the substitutional Au(S) atom and the ejected Cu(O) atom in the (100) and the (110) surfaces, is responsible for the different growth patterns observed in each case. The fact that Au(S) and Cu(O) remain at NN distance translates into a different island growth process. For 0

9

0

9

0

@

0

0 0 0

0.0

@

0

0

0

0

0

0

@

9 9 0

@

0

0

0

0

0

0

0.04624

Fig. 33: Two Au(S) atoms (black disks) and two ejected Cu(S) atoms (large grey disks) on a Cu(110) surface (open circles).

78 Table 9 NAu = 2 (0.03 ML)

NAu = 4 (0.067ML)

Nau = 8 (0.13ME)

NAu = 15 (0.25 ME)

NAu = 30 (0.5 ML)

0.23

0.17

0.14

0.05

-0.53

AE

Difference in energy of formation AE (eV/atom) between the energy configuration with all the Au atoms forming an island in the overlayer and the configuration where Au and Cu form a surface alloy similar to the Cu3Au 1:1 surface. example, for NAu = 2, it is seen that the lowest energy state is that with a Cu chain attached to the Au(O) dimer, as shown in Fig. 33. Increasing Au coverage leads to the formation of a pattern in the surface plane that is equivalent, like in the case of Au/Cu(100), to the ordered 1:1 Cu3Au(110) termination. However, as opposed to the Cu(100) case, the Cu(O) chain remains 'attached' to the surface alloy patch, as seen in Fig. 33, where the ground state for NAu = 4 is shown. This growth pattern can persist indefinitely. However, as the Au coverage increases, it can be seen that states characterized by the dealloying of the Au atoms and the formation of a Au island in the overlayer begin to compete with the Cu3Au 1:1 pattern seen at lower coverages. The difference in energy between configurations reflecting each type of behavior decreases as the Au coverage increases. Table 9 displays the difference in energy of formation between states that follow the 1:1 Cu3Au surface pattern (see Fig. 34) and those where the Au atoms form an island in the overlayer. The formation of the surface alloy remains favorable until approximately 0.5 ML Au coverage, where the dealloying process has definitely started.

3.9. Cu/Ni(llO) Experimental results have been reported for Cu/Ni(100), where epitaxial layers of Cu have been observed via photoelectron diffraction [72,73]. The BFS results

0

9

0

9

0

0

9

0

9

0

0

@ 00

0

0

0

0

Q

0

0

0

0

0

0

0

0

0

Fig. 34: Ground state for two Au atoms (black disks) deposited on a Cu(110) slab. The ejected Cu atoms are denoted with large grey disks. Open circles denote Cu(S) atoms.

79 Table 10 Energy of formation of N-atom chains (eV/atom) Ncu = 4

Ncu = 8

Ncu = 12

Ncu = 20

N-atom chain

-0.38663

-0.46833

-0.49557

-0.52654

Two N/2 chains

-0.22322

-0.38663

-0.44110

-0.49386

Energy of formation (in eV/atom) of a Ni(110) slab with (a) N Cu atoms forming a chain in the cp direction and (b) two separated chains of N/2 atoms. for Cu/Ni(110) and (100) indicate that Cu does not exchange sites with surface Ni atoms, forming chains along the cp direction rather than forming compact islands. However, as the Cu coverage increases (on a Ni(100) slab), the difference in energy between long Cu chains and compact Cu islands becomes smaller, ultimately resulting in the formation of a Cu film for 1 ML coverage. Table 10 shows the energy of a linear chain of N atoms in the cp direction and the energy of two separated chains with N/2 atoms in the same direction, referenced to a Ni (110) surface with 250 atoms. The relevance of this example, so far the only one in this review dealing with epitaxial growth, becomes apparent in the context of the following example (Cu,Au)/Ni(ll0), where of the competition between alloying and segregation effects dominates. It should be noted that if the modeling of Cu~i(110) was carfled beyond the low coverage regime, the formalism described Sec. 2.3 should be used, in order to allow for the possibility of distortions in the Cu film.

3.10. (Cu,Au)/Ni(ll0) After modeling Cu/Ni(ll0) and Au/Ni(ll0), and having described the interactions between Cu and Au, it is interesting to study the codeposition of Cu and Au on Ni(110) and discuss, within the framework of the binary cases, the behavior of such a complex system. In what follows, A, B and C denote Ni, Au and Cu atoms, respectively. We computed the energy of nearly 500 configurations including all the possible arrangements of two atoms of each species. We then studied the possible connections between these states based on the premise that the system will evolve by directly accessing states with lower energy or through small energy barriers disregarding states with much higher energy, as it was done in the Au/Ni(110) case. In doing so, we establish 'decay chains' where the initial state (i.e., all four atoms dispersed in the overlayer: B+B+C+C) decays into other states characterized by different ordering of these four atoms. Examining the low-lying metastable states (i.e., those with energy slightly higher than that of the lowest energy state), makes it possible to understand the

80 competition between individual trends and new ones that result from the interaction between the two different types of adatoms. Fig. 35 summarizes the results, considering only those states that have the possibility of evolving into an accessible configuration with lower energy. Two main decay chains are seen. The first one is characterized by the association of atoms in the overlayer and the insertion of Au dimers in surface sites along the cp direction. There is a substantial gain in energy whenever two Au atoms couple along the cp direction (B+B+C+C -> BB+C+C), later decaying into three types

B+B+C+C

b+A+B+C+C

bA+B+C+C ~.~

-0.195 bA+CC+B -0.236 b+b+AA+C+C -0.281 .

|

" ~ 0~

-0.217

bAB+c+c . -0.258

BB+C+C

bb+AA+C+C -0.256

BB+CC -0.388 ' -0.419 'bb+AA+CC bb+AAC+C -0.439

-0.408 BCB+C -0.421

-0.604 bACCA+b ~-0.606 C A b+b+CAAC~ ,

b+CBAC -0.644

bb+CAAC ~ -0.622

CBBC - --0.616 ,~~

~l bcb+AAAC ~ -0.626

Fig. 35" Main decay chains for two Au (B) and two Cu (C) atoms. Each level represents an accessible state, labeled with the difference in energy (in eV/atom) with respect to the initial state. A, B and C denote Ni, Au and Cu atoms in overlayer sites, while a, b and c denote the same elements in surface sites. Two or more consecutive elements (BB, BCB, etc.) denote chains along the cp direction.

81

of states, described by 1) the formation of cp chains (CBBC), 2) Cu-Au ordering in the surface plane (bcb) with the expelled Ni atoms somewhere else in the overlayer (bcb+AAAC) (no further gains are realized by formation of 4-atom chains, for example, bcbc) and 3) the insertion of a Au dimer in the surface with a CAAC chain in the overlayer. The second main decay chain involves the immediate insertion of at least one Au atom in the surface (B -> bA). This is less energetically favored than the first decay in the first chain (B+B -> BB) thus making the ensuing states in the chain less likely to be observed experimentally. Depending on the order in which 'events' take place, the chains evolve into states characterized by individual insertions (b A) associated with different chain structures in the overlayer (b+b+CAAC, b+CBAC, b+b+ACCA, etc.). From the decay scheme shown in Fig. 35, it is also clear that 1) every possible arrangement is favored as long as it forms in the cp direction, 2) insertion of Cu atoms in the surface is not favored, unless it is part of a Cu-Au trimer (bcb), 3) Cu atoms mostly remain in the overlayer anchoring cp chains (CAAC), 4) Au dimers tend to form in the surface plane, 5) no state where the simultaneous individual behavior of either Au or Cu atoms (bb+CC+AA) is favored by any chain, 6) individual Au atoms in surface sites are favored but probably due to the stabilizing effect of long, mixed chains in the overlayer and 7) the interplay between Cu and Au reflects in the fact that Cu atoms bracket Au atoms when forming chains in the overlayer, and Au atoms bracket Cu atoms in surface chains. While these results constitute an indication of trends that could be expected during the process of surface alloy formation, they are also valuable for understanding the interaction between the different components. As mentioned earlier, Ctt/Ni(110) is clearly characterized by the formation of a Cu overlayer, while Au/ Ni(110) is solely described by the surface insertion of Au dimers. Hints of both individual behaviors are seen in these results, but also a clear picture of the interaction between them emerges. The best example is the structure of the most likely configuration: bcb, where the trend for Au insertion is augmented by the addition of a Cu atom which is readily accepted as part of the chain due to the energetically favorable characteristics of the Cu-Au bond thus formed. In this sense, a Au/Ni(110) feature dominates the structure of the ternary system, albeit modified by the presence of Cu. On the other hand, a Cu/Ni(110) feature (Cu atoms remaining in the overlayer) is less dominant, as demonstrated by the energetically less favored formation of BCB or CBC chains. The formation of cp chains, both in the overlayer or surface, is a characteristic shared by both systems which also appears in any favored ternary configuration. In Fig. 36 we schematically represent the influence of each one of the individual processes in the observed behavior of the ternary system, including new features that arise from the interaction of Cu and Au atoms.

82

Au/Ni(110)

Cu/Ni(110)

bb+AA

CC

(Cu+Au)/Ni(110) bb

~

1~ bcb

.~~"

C...C

CBBC~)A~ bb+CAAC

Fig. 36: Dominance of Au/Ni(110) and Cu/Ni(110) features in the early stages of (Cu+Au)/ Ni(110) surface alloy formation. The Au/Ni trend to form Au dimers in the surface dominates. The mixed states most likely to appear in the ternary system include the formation of bcb chains due to the favorable Cu-Au bonds thus established. The tendency of Cu to remain in the overlayer is a secondary effect, leading to Cu-anchored cp chains or a Au-Cu-Au (BCB) chain in the overlayer with higher energy than its equivalent in the surface. 4. CONCLUSIONS The atomistic modeling of surface alloys is, regardless of the simplicity of the computational tool used, a complex and challenging problem. With the advent of powerful experimental techniques and the availability of equally powerful theoretical methods, there has been substantial progress, as demonstrated by the large number of systems for which reliable data and understanding is currently at our disposal. Still, by any measure, most of the work remains to be done. In terms of modeling, the field of surface alloys offers a unique opportunity to test and validate existing methods, as well as defining the role of modeling itself, often thought of as a direct, atom-by-atom representation of the actual system. Of course, the role of modeling is, and perhaps should only be, limited to providing an efficient tool for answering basic questions leading to a deeper understanding of the driving mechanisms of any given process. In this work, we summarized recent progress in the modeling of surface alloys by linking, via a number of selected examples, the different tools available for a comprehensive analysis of such systems: a powerful and versatile quantum approximate method for the energetics of the system, appropriate and efficient use of experimental input for its parameterization and, perhaps more importantly, an approach for dealing with the multitude of issues characteristic of such complex systems. Focusing on a simple and straightforward algorithm for a systematic analysis of the basic features present in the early stages of formation, it was possible to

83 identify the seed for different growth patterns and, in doing so, efficiently supplement the abundant experimental data. Having raised the necessary confidence on the validity of the technique used, the BFS method for alloys, and on the operational procedure defined for its implementation, it is possible now to further the understanding of ever more complex systems with the same ease of use and interpretation that characterize these basic examples. ACKNOWLEDGMENTS Fruitful discussions with N. Bozzolo are gratefully acknowledged. This work was supported by the HOTPC program at N A S A Glenn Research Center.

REFERENCES [1] U. Bardi, Rep. Prog. Phys. 57 (1994) 939. [2] B. V. Andryushechkin, K. N. Eltsov, V. M. Shevlyuga, C. Tarducci, B. Cortigiani, U. Bardi and A. Atrei, Surf. Sci. 421 (1999) 27. [3] E. McLoughlin, A. A. Cafolla, E. A1Shamaileh and C. J. Barnes, Surf. Sci. 482/485 (2001) 1431. [4] R. A. Bennett, S. Poulston, N. J. Price, J. P. Reilly, P. Stone, C. J. Barnes and M. Bowker, Surf. Sci. 433/435 (1999) 816. [5] S. Chaturvedi and J. A. Rodriguez, Surf. Sci. 401 (1998) 282. [6] I. Meunier, G. Treglia, B. Legrand, R. Tetot, B. Aufray and J.-M. Gay, Appl. Surf. Sci. 162/ 163 (2000) 219. [7] Ch. Konvicka, Y. Jeanvoine, E. Lundgren, G. Kresse, M. Schmidt, J. Hafner and P. Varga, Surf. Sci. 463 (2000) 199. [8] W. A. Hofer, J. Redinger and P. Varga, Solid State Commun. 113 (1999) 245. [9] E. Lundgren, B. Stanka, M. Schmidt and P. Varga, Phys. Rev. B 62 (2000) 2843. [ 10] D. P. Woodruff, D. Brown, P. D. Quinn, T. C. Q. Noakes and P. Bailey, Nuclear Instruments and Methods in Physical Research B 183 (2001) 128. [11] P. Bailey, T. C. Q. Noakes and D. P. Woodruff, Surf. Sci. 426 (1999) 358. [12] J. P. Reilly, D. O'Connell and C. J. Barnes, J. Phys.: Condens. Matter 11 (1999) 1. [13] E Besenbacher, L. Pleth Nielsen and P. T. Sprunger, The Chemical Physics of Solid Surfaces, ed. D. A. King and D. P. Woodruff (Elsevier, Amsterdam, 1977) Vol. 8. [14] G. Bozzolo, R. D Noebe, J. Ferrante and C. Amador, J. Comput.-Aided Mater. Design 6 (1999)1. [15] J. E. Garces, H. O. Mosca and G. Bozzolo, Surf. Sci. 459 (2000) 365. [16] G. Bozzolo, J. Ferrante, R. D. Noebe and C. Amador, Scripta Mater. 33 (1995) 1907. [ 17] L. Pleth Nielsen, E Besenbacher, I. Stensgaard, E. Laegsgaard, C. Engdahl, P. Stoltze, K. W. Jacobsen and J. K. Norskov, Phys. Rev. Lett. 71 (1993) 754. [18] E. I. Altman and R. J. Colton, Surf. Sci. Lett. 304 (1994) L400. [ 19] Y. -R. Tzeng, H.-T. Wu, K. -D. Shiang and T.T. Tsong, Phys. Rev. B48 (1993) 5549. [20] D. D. Chambliss, R. J. Wilson and S. Chiang, J. Vac. Sci. Technol. A10 (1993) 1992. [21] E. G. McRae and R. A. Malick, Surf. Sci. 177 (1986) 53. [22] D. J. O'Connor, M. Draeger, A. M. Molenbroek and Y. G. Shen. Surf. Sci. 357/358 (1996) 202.

84 [23] V. Shutthanandan, A. A. Saleh and R. J. Smith, Stiff. Sci. 450 (2000) 204, and references therein. [24] M. Dietterle, T. Will, A. M. Kolb, Surf. Sci. 396 (1998) 189. [25] J. R. Smith, T. Perry, A. Banerjea, J. Ferrante and G. Bozzolo, Phys. Rev. B44 (1991) 6444. [26] G. Bozzolo, J. Ferrante and A. M. Rodriguez, J. Comput.-Aided Mater. Design 1 (1993) 285. [27] J. H. Rose, J. R. Smith and J. Ferrante, Phys. Rev. B 28 (1983) 1835. [28] G. Bozzolo and J. Ferrante, Phys. Rev. B 45 (1992) 12191. [29] G. Bozzolo, J. Ferrante and R. Kobistek, J. Comput.-Aided Mater. Design 1 (1993) 305. [30] F. Honecy, G. Bozzolo and B. Good, Appl. Surf. Sci. 137 (1999) 157. [31] G. Bozzolo, R. D. Noebe and J. E. Garces, Scripta Mater. 42 (2000) 403-408 [32] G. Bozzolo, H. Mosca, A. Wilson, R. D. Noebe and J. E. Garces. MetaU. Trans. A (in press). [33] G. Bozzolo, R. D. Noebe, J. Ferrante and A. Garg, Structural Intermetallics 1997, M. V. Nathal, ed., The Minerals, Metals and Materials Society, Warrendale, PA 1997. [34] B. Good, P. Abel and G. Bozzolo, Surf. Sci. 456 (2000) 602-607 [35] G. Bozzolo, J. Ferrante and R. D. Noebe, Surf. Sci. 377/379 (1997) 1028. [36] P. Legare, Solid State Commun. 106 (1998) 87. [37] P. Legare, G. F. Cabeza and N. J. Castellani, Surf. Sci. 441 (1999) 461. [38] G. Bozzolo, J. E. Garces and P. Abel, Surf. Sci. (in press). [39] L. Pleth Nielsen, I. Stensgaard, E. Laegsgaard and F. Besenbacher, Surf. Sci. 307/309 (1994) 544. [40] D. O. Boerma, G. Dorenbos, G. H. Wheatly and T. M. Buck, Surf. Sci. 307/309 (1994) 674. [41] J. -S. Filhol, D. Simon and P. Sautet, Phys. Rev. B 64 (2001) 085412. [42] M. Abel, Y. Robach, J. -C. Bertolini and L. Porte, Surf. Sci. 454/456 (2000) 1. [43] L. Porte, M. Phaner-Goutorbe, J. M. Guigner and J.-C. Bertolini, Surf. Sci 424 (1999) 262. [44] P. W. Murray, I. Stengsgaard, E. Laegsgaard and F. Besenbacher, Phys. Rev. B 52 (1995) R 14404; [45] P W. Murray, I. Stensgaard, E. Laegsgaard and E Besenbacher, Surf. Sci. 365 (1996) 591; [46] G. W. Graham, P. J. Schmitz and P. A. Thiel, Phys. Rev. B 41 (1990) 3353; [47] T. D. Pope, K. Griffiths, P. R. Norton, Surf. Sci. 306 (1994) 294; [48] T. D. Pope, M. Vos, H. T. Tang, K. Griffiths, I. V. Mitchell, P. R. Norton, W. Liu, Y. S. Li, K. A. R. Mitchell, Z. -J. Tian and J. E. Black, Surf. Sci. 337 (1995) 79. [49] G. W. Graham, Surf. Sci. 171 (1986) L432; [50] Y. G. Shen, J. Yao, D. J. O'Connor, B. V. King and R. J. MacDonald, J. Phys.: Condens. Matter 8 (1996) 4903; [51] T. D. Pope, K. Griffiths, V. P. Zhdanov and P. R. Norton, Phys. Rev. B 50 (1994) 18553; [52] J. Yao, Y. G. Shen, D. G. O'Connor and B. V. King, J. Vac. Sci. Technol. A 13 (1995) 1443; [53] R. Tetot, J. Kudrnovsky, A. Pasturel, V. Drchal and P. Weinberger, Phys. Rev. B 51 (1995) 17910; [54] M. Valden, J. Aaltonen, M. Pessa, M. Gleeson and C. J. Barnes, Chem. Phys. Lett. 228 (1994) 519; [55] J. Yao, Y. G. Shen, D. J. O'Connor and B. V. King, Surf. Sci. 359 (1996) 65;

85 [56] G. W. Anderson, T. D. Pope, K. O. Jensen, K. Griffiths, P. R. Norton and P. J. Schultz, Phys. Rev. B 48 (1993) 15283; [57] T. D. Pope, G. W. Anderson, K. Griffiths, P. R. Norton and G. W. Graham, Phys. Rev. B 44 (1991) 11518; [58] J. Kudrnovsky, S. K. Bose and V. Drchal, Phys. Rev. Lett. 69 (1992) 308; [59] C. J. Barnes, E. A1Shamaileh, T. Pitkanen, P. Kaukasoina and M. Lindroos, Surf. Sci. 492 (2001)55. [60] P. W. Murray, S. Thorshaug, I. Steensgaard, F. Besenbacher, E. Laegsgaard, A. B. Ruban, K. W. Jacobsen, G. Kopidakis, H. L. Skriver, Phys. Rev. B 55 (1997) 1380. [61] E. Hahn, E. Kampshoff, A. Fricke, J.-P. Bucher, K. Kern, Surf. Sci. 319 (1994) 277. [62] J. P. Bucher, E. Hahn, P. Fernandez, C. Massobrio, K. Kern, Europhys. Lett. 27 (1994) 473. [63] S. C. Wu, S. H. Lu, Z. Q. Wang, C. K. C. Lok, J. Quinn, Y. S. Li, D. Tian, F. Jona, and P. M. Marcus, Phys. Rev. B 38 (1988) 5363. [64] C. Massobrio, P. Fernandez, J. Chem. Phys. 102 (1995) 605. [65] P. Fernandez, C. Massobrio, P. Blandin, J. Buttet, Surf. Sci. 307/309 (1994) 608. [66] R. Belkhou, J. Thiele and C. Guillot, Surf. Sci. 377/379 (1997) 948. [67] E Theilmarm, R. Matzdorf and A. Goldmann, Surf. Sci. 387 (1997) 127. [68] I. Fujinaga, Surf. Sci. 86 (1979) 581. [69] C. W. Graham, Surf. Sci. 184 (1987) 137. [70] J. C. Hansen, M. K. Wagner and J. G. Tobin, Solid State Commun. 72 (1989) 319. [71] B. J. Knapp, J. C. Hansen, J. A. Benson and J. G. Tobin, Surf. Sci. 188 (1987) L675. [72] W. E Egelhoff Jr., J. Vac. Sci. Technol. A 7 (1989) 2060. [73] W. E Egelhoff Jr. and D. A. Steigerwald, J. Vac. Sci. Technol. A 7 (1989) 2167. [74] H. O. Mosca, J. E. Garces and G. Bozzolo, Surf. Sci. 454/456 (2000) 707. [75] D. L. Adams, H. B. Nielsen, J. M. Andersen, Surf. Sci. 128 (1983) 294. [76] I. Stensgaard, R. Feidenhans'l and J. E. Sorensen, Surf. Sci. 128 (1983) 281. [77] J. E. Garces, G. Bozzolo, P. Abel and H. O. Mosca, Appl. Surf. Sci. 167 (2000) 18.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces 86

D.P. Woodruff, (Editor)

Chapter 3

Alloy surface segregation and ordering phenomena: recent progress M. Polak and L. Rubinovich

Department of Chemistry, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel 1. O V E R V I E W

Theoretical and experimental studies of surface segregation equilibrium phenomena in metallic alloys have been focused traditionally on substitutional solid solutions with elemental constituents (and non-metal impurities) assumed to be randomly distributed among the crystal lattice bulk and surface sites. Only in recent years more attention have been paid to the role of compositional order in surface segregation [ 1]. Ordering in alloys can be classified according to its range as short-range order (SRO) or long-range order (LRO), and is manifested in diverse structures. The sign and magnitude of the interatomic interactions are key factors in the compositional structural nature of equilibrium bulk phases in a multi-element (binary, ternary, etc.) system formed as function of temperature and overall composition (documented as phase-diagrams). In particular, in case of effectively attractive interactions between two elements, A and B, of a binary alloy,

v _ v AB = -1( u AA +u BB - 2 u AB ) >0 2

(1)

(uAA, uBB and u AB being the interaction energies between the corresponding atoms), the formation of ordered metallic phases is quite common ("chemical" or "intermetallic" compounds with fixed composition, or alloys with a certain range of concentrations). In such ordered alloys atoms of one element tend to be surrounded by atoms of the other element in periodic crystal sub-lattices (Fig.l). At finite temperatures LRO is never perfect, and in addition, there are local fluctuations in composition, known as short-range order (SRO, see Fig.2),

87

A

o w

a

b

Fig. 2. 2D schematics of a binary ordered alloy crystal with LRO alone (a), and with LRO accompanied by short-ranged compositional fluctuations (b). The probability of finding atom A (or B) at a lattice site is expressed by means of the circle size (and colour): higher probability corresponds to larger (and brighter) circles. The largest white circle in (b) represents 100% probability. which unlike LRO does not vanish at the order-disorder transition temperature. In case V < 0, bonding of like atoms is energetically preferred, leading in principle to separation of the alloy into a mixture of A and B rich solid-solution phases, each with nearly homoatomic SRO clusters, compared to short range AB mixed or heteroatomic clusters in the former case of V >0. In other words, the tendency to order (or phase-separate) is manifested to some, local degree also in most solid solutions, where the distribution of atoms in the crystal is not entirely random, and should be incorporated too in any theoretical quantitative evaluation of surface segregation phenomena. Moreover, many alloys of

88 practical importance are comprised at temperatures below solubility limits of two or more phases in a certain micro- (or nano-) structure with ordered regions or clusters (characterized by LRO and SRO) embedded in solid-solution matrixes (with SRO). Since not only the LRO and SRO contributions are temperature dependent, but also the solid solution bulk compositions and relative amounts of phases (each with its distinct surface segregation behaviour), the segregation characteristics of such a multi-phase alloy surface can be even more complex than the single-phase cases. Surface segregation of an alloy constituent, which is very common in substitutional (and interstitial) solid solutions, is expected to be manifested to a less extent in ordered alloys [ 1]. Thus, the process of segregation in A-B alloys, whereby atoms of one constituent element populate preferentially the surface layer, can be viewed as a sort of near-surface phase separation, which is typically incompatible with ordering tendencies*. Actually, segregation is expected to disrupt order and break energetically favourable A-B bonds, and hence as an endothermic process in strongly ordered systems, it may not occur at all, at least at relatively low temperatures. At higher temperatures this suppression is expected to diminish, as entropy-driven segregation with progressively higher levels prevails, until in case of a transition to a disordered phase, it becomes maximal usually around the range of the transition temperature (Fig.3). Then, in the solid-solution high temperature regime, the segregation level eventually decreases with temperature as an exothermic process. The resultant peaked segregation vs. temperature curve expected under certain conditions in strongly ordered systems has been predicted for alloys with LRO [2] and solid-solutions with strong SRO [3], and observed experimentally in several cases (e.g., Refs.3-5). The interplay of LRO and segregation can lead to other types of segregation curves, as described in section 3. Another complication, worthwhile mentioning in the context of developing insight into phenomena of surface segregation in the presence of ordering tendency, emerges when V is strongly composition dependent [6,7], or even changes sign, as in the case of the Fe-Cr system [8,9]. Describing of the equilibrium state of the macroscopic system by means of a statistical-mechanical approximation or Monte-Carlo (MC) simulations is one of the two main aspects of surface segregation theory, while the second aspect deals with the segregation energetics related to "microscopic" atomic interactions. Early experimental data on surface segregation phenomena in solid solutions were usually analyzed by means of the Langmuir-McLean theory [ 10]. This simplistic approach predicts monolayer segregation that decreases monotonously with temperature, and enabled to derive "segregation enthalpy" * Yet, as discussed in Sec.3, in certain bulk truncated terminations of ordered alloys the two tendencies can be compatible.

89

o

r o

~

,AS 0 < 0

Temperature Fig.3. Schematics of the evolution of equilibrium segregation with temperature in alloys with order-segregation competition: (a) dominant surface segregation tendency (LangmuirMcLean behaviour), (b) dominant ordering tendency. Signs of enthalpy and entropy of segregation are indicated. and "excess entropy" from experimental surface compositions vs. temperature (Fig. 3), but fails to account for the above mentioned complex segregation in alloys with interaction-induced strong ordering tendencies. Hence, together with the development of experimental techniques and the fast increase of relevant data, more elaborate theoretical approaches to surface segregation phenomena became necessary [1]. A better starting point for theoretical studies of LRO/segregation interrelations [11,12] became the Bragg-Williams (BW) statistical-mechanical approximation adapted for multilayer surface segregation while still assuming random distribution of atoms at identical layer and sublattice sites. It is based on Ising type rigid lattice model with constant bond energies, ignoring surrounding-dependent pair bonding and many-body interactions (an Ising type model that does consider composition dependent local interactions was introduced recently [9]). As further steps, basic SRO effects on surface segregation were treated by means of the statisticalmechanical cluster variation method (CVM) [13-18], and the free-energy expansion methods (FCEM, described in the next section) [1,3,9,19,20]. On the other hand, Monte-Carlo simulation methods [21-38] are capable of taking into account such contributions as atomic vibrations and surface atomic relaxation [29]. When combined with the embedded atom method (EAM)[23-25] as an improved energy model, or its modified version (MEAM) [37,39,40], MC simulations overcome several drawbacks of the above Ising type models. Yet,

90

the latter analytical approach can be helpful in predicting basic effects of atomic long-range and short-range order on surface segregation in alloys, including multi-component and dilute systems. This chapter is focused on the most recent theoretical and experimental efforts aiming at unravelling the diverse phenomena of segregation/ordering interplay. The issue was reviewed by us comprehensively about two years ago [ 1], and new topics are addressed here in three separate sections: (i) Theoretical formulation of multi-layer segregation in a multi-element solid-solution alloys (ternary alloys in particular) with emphasis on the role of short range order. It is followed by model calculations for NiA1-Cu solid solution. (ii) Evaluation of surface segregation trends for several classes of ordered alloy surface structures, including case studies, primarily in terms of segregation/ordering energetics. In view of the prominence of LRO effects, they constitute a central topic in this review. (iii) The complex segregation behaviour in a bi-phase system comprising of ordered clusters in a solid solution matrix. 2. S E G R E G A T I O N IN M U L T I - E L E M E N T ALLOYS Compared to numerous studies of surface segregation phenomena in binary alloys [1,41], quite fewer studies have been devoted to the theory of surface segregation in multi-component (in particular, ternary) metallic alloys. Characteristic phenomena as co-segregation and site competition were addressed originally by Guttmann [42] using a regular solution model. Later, Wynblatt and Hoffmann [43] used a monolayer segregation model with more accurately approximated total free energy, and this formalism was modified to include the prediction of possible compositional phase transitions [44]. However, as mentioned above, a more accurate description demands taking into account short-range order (SRO), as well as multilayer segregation. Free-energy approximate expressions that take into account SRO in the bulk of dilute binary [45,46], or multi-component [47] alloys were derived previously. However, their application to alloy surfaces is somewhat problematic, since upon segregation a solute can become a major constituent at the surface, thus violating the assumed low concentration. SRO correction for the binary alloy free energy, which is symmetric with respect to the alloy constituents, and thus overcomes this difficulty, was derived in the Ising model based "free-energy concentration expansion method" (FCEM) [1,3,19]. Being more accurate than the mean-field Bragg-Williams (BW) theory, and simpler to apply compared to the quasi-chemical and cluster variation methods, FCEM agreed quite well with MC simulations of segregation [19], while demanding much less computational efforts. Recently the FCEM approach was extended to

91

the case of alloys with any number of components [20]. An approximate SRO formula for multi-component alloy was constructed by adapting the corresponding binary alloy formula as a boundary case, and by making the multi-component alloy expression symmetric with respect to its components. The FCEM expressions for binary alloys were obtained using the Ising model Hamiltonian and an expansion of the partition function and free energy in terms of solute concentration [1,3,19]. The free energy of a binary alloy (A solute, S - solvent) reads,

A A F - k TE (c A In cA + cS In cS) + E Ahmcinm m 1 + - E VmASI2cA-1)(2c A - 11-

(2)

2 {mn}

- ~ kTc~m(1-cA)cA(1-cA)IexF(-2vAS/kT)+2vAS/kT-1 ) {mn} where

I cm

..l

is the concentration of a constituent I on a lattice site m, Ah~n

IJ

denotes a layer "field" (assuming that the lattice site m ~ p-layer), and Vmn is the effective pair interaction strength (see eq.1) between atoms of constituents I and J on lattice sites m and n. Rearranging the third term gives,

-il ~

AS mn( Acm,)( _

{mn}

cAa)_ _

(Vmn(CmC a .an + cAcB))+ 1 ZVmn {mn} 2 {mn}

The last, constant term can be omitted, and contributions related to the interaction Vm AS in eq.2 can be rewritten in a form symmetric with respect to the constituent concentrations.

F - k T E I cA

lnc A + c S lnc S

) +EAhmc A Am -

m

AS c A cS + c S Ac Vmn (mnmn)

m

+

-E A S A S (exF(_2Vmn AS /kT) + 2Vmn {mn}(+kTcmcmCnCn AS /tc/'-T-

(3) 1)

92

Generalization of this formula to multi-component alloys is straightforward [201, I I 2Ahmc mm,I :/:S

F-kT~2cllncI+ mI

)

CmCn CmCn . +

(4)

- {mn},{IJ}~~kTcmCmCnCn(exF(-2VIJn/kT)+2VIJn/kTJ I J Pair probability of finding atoms of types I and J on lattice site m and n is given by the formula,

plmJn - CmC I nJ + CmCnCmC I I J nJ(l_exp(_2VIJ/kT)) that coincides with the corresponding formula for a binary alloy [1,19]. In case of a ternary alloy (A,B - solutes, S - solvent) the free energy is given by, F - k T ~ ( c A In c A + c B l n c B + c S l n c S ) + m

A BB) +E (A Ahmcm + Ahmcm m r.,AB( A B B A'~ Vmn ~CmCn + CmCn J + VmAC( A S S A + n~CmCn+CmCn) l/maB( B S +CmCn S B 1+ n ~CmCn -

Z

{mn} kTcmcmC A B nA cnB( ex F(- 2 VAnB/kT 1 +2Vmn AB/kT - 1) + A S nA c nS( ex F(-2Vmn AS/kT 1 + 2Vmn AS/kT - 13 + kTcmcmC kTcmcmC n c n

ex

-

2Vmn

+ 2Vmn

(5)

93 Formulas within the BW mean field theory are obtained by omitting the

SROrelatedcontributionscontaining(ex~-2VIJ/kT)+2VIJ/kT-l]

from

eqs.3-5. The method was applied to the elucidation of effects of interatomic interactions and SRO on surface segregation in Ni-8%A1-4%Cu as a model ternary solid solution [20]. The results were then compared quantitatively to mean field calculations, and inspected in terms of the pertinent energetic parameters and effects of temperature. A primary consideration in choosing the Ni-A1-Cu system (Ni solvent) for model calculations was the relatively strong attractive Ni-A1 interactions (which lead to significant SRO effects on surface segregation in Ni-9at%A1 solid solution [ 1,3,19]). These effects are expected to be operative also in alloys containing a third constituent in low concentrations. Copper was chosen since Ni-Cu binary solid solutions (with quite weak repulsive interactions) had been extensively studied earlier and the corresponding energetic parameters are fairly known [48]. The energetics of the model was based on three nearest-neighbor (NN) interactions, V NiCu , V NiAI and V AlCu and two surface fields, Ah A1 and Ah Cu all listed in Table 1 In order to obtain the equilibrium layer compositions the free-energy (eq.5) was minimized numerically [20]. The alloy constituent concentrations calculated in the FCEM approximation for the first three atomic layers of the Ni-8at%A1-4at%Cu(111) surface are shown in Fig.4. A distinct surface phase transition characterized by a sharp jump in surface concentrations appears at 1075 K. Below this temperature the alloy surface is strongly A1 depleted and Cu rich, while at the transition A1 rises and Cu decreases, both reaching rather moderate segregation levels above it. The segregation behavior at all temperatures is indicative of site competition. ~

*

Table 1 Energetic parameters used in the model (in meV) v~C~ r~lc~ I~IN~ ahc~ AhAI -12.5" 31"* 136"* -120" -570*** *Ni-Cu energetic parameters were taken from Ref.48; v NiCu are enhanced at the surface by a factor of 1.5. **Estimation obtained from the heat of mixing [49] ***The surface field Ah Al for Ni-AI(100) has been determined as -680 meV [3], with -450 meV due to the difference in surface tensions [50]. Keeping the same elastic strain contribution, the estimation (-570 meV) takes into account the smaller nunaber of (111) broken bonds per atom (3 vs. 4).

94

Cu = 0.8

(p=O)

0.6

-""

-...

p=l

_

~'~ ~ ~

.~ .~

0.4 '"--..

p=2

"'""-..

i

i

~ 0.2

900

950

t .................................. ..... i. . . . . . . . . F .........

i

1000

1050 1100

r .......

1150 1200

""

1250

Temperature, K

0.16 ml

Surface (p=0)

.~ 0.12

e.~

p=2

0.08

...............................

./. 0.04

......

......-

.....................

/

~176176

....

900

"~176176176176176176

950

1000

A

1050

1 1 0 0 1150 1200 1250

Temperature, K

.; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....................... p=2

t

0.8 . -. - -. - - -.

. ." ~

~ .- ." -. ~

o 9

~

0.6

..--"r ~ 0.4 Z 0.2

900

Surface

950

1000

1050

1100

1150 1200

1250

Temperature, K

Fig.4. The near surface concentrations of fee Ni-8at O~AI-4at o~Cu(111) calculated in the FCEM approximation. (p=l and 2 correspond to the first and second under-layer, respectively.)

95

The predicted phenomena can be explained in terms of the different energetics and atomic coordination numbers involved. In particular, because of the reduced surface coordination, the Ni-A1 prominent mixing tendency is stronger in the bulk, leading to diminished surface concentrations of these constituents ("co-desegregation") throughout the temperature range below the transition. Consequently, in spite of its lower surface field (Table 1) the overall effective driving force for Cu segregation in the ternary alloy is significantly stronger than for A1, and even at temperatures above the transition the (diminished) surface enrichment by Cu is higher.

Cu 0.8 0.6 0.4

I FCEM

i

Ni-4at%Cu IlBw

0.2

-r .........

0 900

I

i

950

1000

I

p

i

I

i

1050

1100

1150

1200

1250

Temperature, K

0.2

Ni-$at%A1 FCEM

0.15

A1

I

......

i

r

BW

FCEM

0.1

0.05

0 900

____.~I 950

I

1000

i

~ 1050 ,

I

I

1100

1150

1200

1250

Temperature, K

Fig.5. Calculated solute surface concentrations for Ni-8at%A1-4at%Cu(111): thick solid lines - FCEM, thin solid lines - the BW-type approximation. Dashed-dotted lines - solute surface concentrations for the binary alloy Ni-Sat%Al(111) and Ni-4at%Cu(111) surfaces calculated in the FCEM approximation. Note the enhancement of Cu segregation induced by ternary alloying and short-range order effects.

96

The segregation of copper is further enhanced by short-range order that suppresses surface segregation of the solute (A1) interacting strongly with the solvent (Ni) [3,51]. Actually, SRO amplifies the interaction induced effects on segregation, without changing the general trend. Thus, as can be seen in Fig.5, the sharp transition in Cu surface concentration as predicted by the FCEM calculations occurs at a considerably higher temperature as compared to the results of the mean field (BW) theory that neglect interatomic correlations. Furthermore, it can be expected that the A1-Ni strong mixing tendency which diminishes surface concentration of both these constituents in the ternary Ni-8at%A1-4at%Cu alloy, would promote Cu surface segregation far beyond the driving forces operative in the corresponding binary alloy Ni-4at%Cu. Conversely, A1 segregation should be suppressed relative to its segregation levels in the binary Ni-8at%A1 alloy. The FCEM results for the binary alloys, shown in Fig.5, indeed exhibit below the transition temperature Cu surface concentration much lower (and A1 concentration much higher) than in the ternary alloy. To summarize this section, the multi-layer FCEM calculations predict strong segregation of Cu associated primarily with the Ni-A1 strong mixing tendency (attractive interactions) that effectively repels these constituents from the surface into the alloy bulk in an apparent site competition process. It appears to be operative also following a compositional phase transition, when the surface solute concentrations tend to be slightly below the respective binary alloy moderate segregation levels. Part of the former enhanced Cu surface segregation is associated with short range order effects that shift the transition to a higher temperature. These calculations can be further extended to other nominal compositions of this alloy, and the energetic parameters can be varied as to their general effects on site competition and surface phase transitions in ternary alloys. 3. SURFACE SEGREGATION IN ORDERED ALLOYS

Compared to SRO effects on surface segregation in solid solutions, the role of LRO should be naturally more prominent and common. Its elucidation requires calculations that take into account various factors contributing to the "net" segregation characteristics in ordered alloys including the temperature dependence: the crystal bulk structure and surface orientation, effective bulk and surface interatomic interactions (NN, non-NN) in relation to segregation driving forces, deviation from exact stoichiometry, possible surface relaxation and reconstruction, atomic vibrations, etc. This section attempts to quantify some of these factors and present several possible scenarios of segregation/order interplay.

97

Spatial ordering in the bulk of alloys and "classical" surface segregation in completely random solid solutions (without LRO or SRO) are both exothermic processes, which are enhanced at lower temperatures and accompanied by an entropy decrease. As discussed in our previous review [1] and mentioned in see.l, their interplay in ordered alloys can completely modify the segregation behaviour resulting either in endothermic or exothermic surface segregation, depending primarily on the energy balance of the respective tendencies. In the former case segregation is hampered, and an increase in its equilibrium level with temperature can be expected due to the enhancement of compositional disorder that disrupts the near-surface LRO, and is associated with increased configurational entropy. 3.1 Prediction of order/segregation interplay by means of a simple model As a first step, the interplay of surface segregation and long-range order in a binary alloy A~B~_~can be qualitatively evaluated by comparing the effective interaction strength (V) as a measure of ordering tendency with the "surface field" (Ah) reflecting the segregation basic driving force, similarly to the original approach of Moran-Lopes [2]. In this simple nearest-neighbour (NN)

pair interaction model, as the "segregation/order factor", r

1-71,

gets larger the

I - - I

balance tips more towards segregation. To obtain more quantitative estimation of the effects, r has been used as a parameter in FCEM calculations for two types of ordered structures with ideally equiatomic bulk truncated surface, assuming segregation limited to the three outmost atomic layers.

3.1.1 Equiatomic binary alloys Among possible equiatomic surfaces of equiatomic bulk alloys (e.g., B2(l10), B32(110), and L10(lll), see Fig.l) the calculations focused on bee B2(110). For low or moderate values of r (<-~10) a peaked segregation curve is predicted (Fig.6). Thus, starting with equiatomic bulk truncated composition at low temperatures, one of constituents (depending on sign of the non-zero surface field) segregates as disordering proceeds. At higher values of the ratio (r>-~10) full monolayer is formed at low temperatures, and the segregation decreases with temperature monotonously (Langmuir-McLean type behaviour). The role of bulk off-stoichiometry is exemplified in Fig.7. Even slight negative deviations diminish considerably segregation levels, while positive deviations lead to strong enhancement relative to the levels calculated for the exactly equiatomic bulk. These somewhat surprising findings can be understood in terms of the dominant bulk ordering tendency, by which excess atoms (>50%) are effectively pushed out from the bulk (due to its reduced coordination,

98 ordering tendency at the surface is weaker). This strong dependence of segregation on small deviations f r o m the bulk stoichiometry should be taken into account in any analysis of ordered alloy segregation data (see below).

1.0

"

"

".:. :. .:.:.:.;. ~. .~. .~. ". . :. '. .i . . . . . . . . . .

[

12

0.8 <

0.6

-

. ....

. ...... - ....' "

""-~ ~176176 !

2

.......

0.4 ~

0.2

~176176176176176176176176176 ~176

Ws

. . . . . . . . ~ 1 7~176176176 6

0.0 1.0

2.0

3.0

4.0

5.0

Temperature, kT/V

Fig.6. The B2(110) surface average (solid lines) and sub-lattice (dotted lines) concentrations of the segregant in AB model alloy as a function of reduced temperature calculated in the FCEM approximation for different segregation/order factors r (indicated near the plots). The difference in sub-lattice concentrations corresponds to the surface LRO parameter that vanishes at the surface transition temperature Ts that coincides with the bulk transition temperature Tb independently of r.

1.0 0.9 0.8 <

0.7 0.6 0.5 0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

Temperature, kT/V

Fig.7. The B2(110) alloy surface average concentration as a function of temperature calculated in the FCEM approximation for model AcBl_c alloys with stoichiometric (c=0.50) and near-stoichimetric (c=0.49,0.51) bulk concentrations (segregation/order factor r =8.9).

99

1.0

11 0.8 <

Ts

Tb

0.6 0.4

--1

3.5

............................

0.2

.3...5. ......... I

7

0.0

_ __1 . . . . . . .

0.4

0.6

0.8

............................. i

_1 . . . . . . . . . .

1.0

I. . . . . . . . . .

1.2

1.4

..... i

(i'iiill .....

I. . . . . . . . . .

1.6

i

1.8

Temperature, kT/V Fig.8. Variation with temperature of the average segregant concentration at the L12(100) surface (solid lines) and at the first underlayer (dotted lines) in AB3 model alloy calculated in the FCEM approximation for different segregation/order factors r (marked near the plots). Arrows indicate order-disorder transition temperatures (for r =3.5, Ts=Tb). 3.1.2 N o n - e q u i a t o m i c binary alloys Another class of ideally bulk-truncated equiatomic surfaces of alloys (e.g., L12(100) and DO3(100) with AB3 bulk stoichiometry), exhibits more diverse segregation/order interplay compared to the previous class ("non-segregated", Cs = Cb, equiatomic termination). Because Cs > Cb, the segregation vs.

temperature curve is not necessarily peaked. Equiatomic termination predicted for r <10 is followed by a peaked curve only for intermediate values of r ( r >4.5) (Fig.8), but for lower r values (<4.5) a monotonous decrease below the equiatomic level is anticipated. At high values of the segregation/order factor (>-~10), the behaviour resembles that of the previous class, namely, a full monolayer is formed at low temperatures, and then the segregation level monotonously decreases. Moreover, the diversity is manifested also by the Table 2 Relationship of surface/bulk transition temperatures calculated for L12(100) Segregation/order factor r<2 4> r >2 7.8>r>4 8>r>7.8

Surface induced order/disorder or neither Ts < Tb (SID) Ts = Tb Ts > Tb (SIO) Ts = Tb

r>8

Ts < Tb (SID)

100

disordering temperatures of the surface (Ts) vs. bulk (Tb ). Depending on r, they can coincide, Ts can exceed Tb (surface-induced order, SIO), or be lower than Tb (surface-induced disorder, SID). SIO is promoted by surface compositions close to equiatomic (Fig.9), which correspond to intermediate values of the segregation/order factor, while SID occurs for high and low values of r, as shown in Table 2.

1.0 0.8 <

0.6 0.4

p=2

0.2 0.0 0.4

0.6

0.8

1.0

1.2

I

I

1.4

1.6

1.8

Temperature, kT/V

Fig.9. Surface Induced Order at the L12(100) surface in AB3 model alloy calculated in the FCEM approximation for r=-7. p=l and 2 correspond to the first and second under-layer, respectively.

1.0 """-,,i"

|

* """

r': Z", -.

0.8 <

0.6 0.4 0.2

,

~

~

0.0

9

0.4

""

I

~

"

J. . . . . . .

0.8

.,.~

o.s-

i I"

I

1

1.2

1.6

2.0

2.4

Temperature, kT/V

Fig.10. Average (solid lines) and sub-lattice concentrations (dotted lines) of L12(100) and L12(lll) surfaces calculated in the BW and the FCEM approximations (thick lines) for r =3.5.

101 In addition to the segregation/order factor, and depending on its magnitude, the crystal structure and surface orientation can strongly affect the surface composition in ordered alloys. For example, unlike the case of the equiatomic bulk truncated composition of L12(100), LRO tends to maintain the L12(111) surface with nominal bulk concentration (0.25). Therefore, the two ordered surfaces are expected to exhibit quite different segregation characteristics for the same r value (Fig.10). Moreover, SRO causes pronounced changes of surface sublattice and average compositions associated with a considerable reduction of the order-disorder transition temperature (especially in fcc alloys). 3.2 Case studies It is instructive to review order/segregation interplay in specific alloys in view of the diverse predictions furnished by the model calculations of the previous section. We have chosen to focus on the well-studied Cu3Au(100) surface, and to assess comparatively the segregation/order interplay in a large number of equiatomic aluminides. Pertinent theoretical and/or experimental recent findings for Cu3Pd, Pt3Sn and C03Pt are addressed too.

3.2.1 Compositional variations in Cu3Au(lO0) and Cu3Pd(llO) The surfaces of Cu3Au alloy (bulk structure L 12) was studied thoroughly by various techniques and theoretical approaches, especially in relation to the order-disorder transition [52-63]. Recently, medium-energy ion-scattering spectroscopy (MEIS) measurements confirmed the stabilization of bulktruncated equiatomic termination for this surface at low temperatures. Starting at about 500 K, the Au atoms in the surface layer begin to move to the second 0.6

9

I

9

9

9

9

9

g

9

9

9

I

9

I

!

g

~

g

v

0.5 t-

.o I..

er C O

r

0.4 0.3 0.2

<

o.o |

250

t

300

#.

, |

i

350

"

9

400

"

9

450

"

500

" ". ,~

"

"

"

550

9

600

"

9

650

9

700

"

750

Temperature (K)

Fig. 11. Variations in the atomic concentration of Au atoms in the first ( 9 ) and second ( i ) layer of Cu3Au(100) measured by MEIS as function of temperature [64].

102 Cu3Au

[001]

surface

t.O

~t

::$: ::$: ~ 0.8

::$: ::$:: $

~:

~: :r

~: ~

1.0

Cu3Au Dt3GDOTZi':=0.9

surface

.

.

0

o

.

.

0r162

~: 0.6 _

-**_ .........

[001]

OOOOO T/T : 0 . 8

T/T.= 1. t ***** T/To= 1.2

0

0

0

,,~

~

0.6

12t

o

0

0.2

llx

0

o

0

~

,

~

, t12 , 1~

layer

, 2t0 , 2~

o ,

(CuAu)

-1.00

0

o

o

0

0

0

O

o,,v,,v,, o o o o T/'%;0"98 ./. ,, ^^^^^~/%=n. i

0

O

--0.6

0.2

(Cu)

0

-0.2 ~

, 0~ 0.4',

0.00

0

'

;

'

~

(Cu)

a

o

o

'~176

o

116 ' 2~

layer

o

o

' 2~

'

(CuAu)

b

Fig.12. (a) Plane-averaged stoichiometry profile of the (100) surfaces of Cu3Au for T/Tc =0.8 and 0.9. (b) Short-range order for each plane parallel to the surfaces for T/Tc =0.8, 0.9, 1.1, and 1.2. Layer 0 corresponds to the Cu-terminated surface, and layer 27 to the CuAuterminated surface (a simulation cell, 5 x 5 x 14 in units of the lattice constant of Cu3Au, with two free surfaces (100) was used). According to Ref.36. layer (Fig.11), and their concentration decreases to 32% at 720 K. The surface and bulk disordering temperatures coincide, Ts=Tb=663 K, [52,54,65]. These types of monotonously decreasing surface segregation and T s / T b relationship are consistent with the above model calculations of sec. 3.1.2 when 4 > r > 2 (Table 2). Indeed, rough estimation of r based on V fitted to the bulk transition temperature, and Ah calculated from the constituent surface tensions and size mismatch energy, falls into this range. More insight into equilibrium arrangements of atoms in the near surface region of this alloy was obtained by means of Monte Carlo simulations [36]. As is depicted in Fig.12a, inner layer concentrations oscillate according to the succession of the Cu planes and the CuAu planes. Cu concentrations of subsurface layers converge to 0.75, which corresponds to the average concentration of the alloy and signifying near-surface disordering. Both truncations exhibit surface segregation of Au atoms, which is somewhat higher for the CuAu-terminated surface. Furthermore, while Au segregates mostly to the surface layer, the disorder is manifested in about six layers. The SRO characterized by the Cowley parameter* shown in Fig.12b behaves similarly to the LRO (Fig.12a). Below the transition SRO oscillates between negative values

P Au , where P Au is the CAu probability of finding Au atom at a site in the nearest-neighbor shell around a Cu atom [66]. * The Cowley SRO parameter is defined in this case as o" - 1

103

Fig.13. Concentration profiles for the Cu3Pd alloy at different temperatures (300-700 K) in

different Monte Carlo simulations [40]. (a) Simulation set S 1 (b) simulation set $2. The EAM parameters used were optimized specifically for the Cu-Pd alloy. in CuAu planes (preferred association of unlike atoms), and positive values in the Cu planes (preferred association of like atoms). Order that is maximal in the bulk-like layers gradually decreases near the surface, and for surface atoms with reduced coordination it drops close to the value (o-=-0.2) simulated for the bulk and the surface above T c . The surface fraction of gold (--0.6), predicted in the simulations [32,36] and by electronic theory calculations [67] is somewhat

104

larger than should be in case of perfect bulk truncation (or measured experimentally [53,64], see Fig.11). An opposite segregation trend, namely majority constituent segregation, is exhibited by Cu3Pd(110) surface, as revealed by LEIS [68]. The Cu segregation profile is oscillatory, and second layer ordering gave a (2• 1) LEED pattern. Results of Embedded-Atom-Method Monte-Carlo (EAM-MC) simulations are in good agreement with the experimental findings [40]. Since along the [110] direction, the ordered Cu3Pd bulk consists of alternating pure Cu and mixed CuPd layers, two sets of surface simulations were performed: a simulation set (S1) with pure Cu and another set ($2) with equiatomic CuPd termination (Fig.13). The former simulation set agrees better with experimental observations (note the predicted increase of surface Cu concentration with temperature in the second set). 3.2.2 Surface order in Pt3Sn(111) and CooPt (111) Deviations from bulk-like terminations are quite common in metallic alloys and can involve temperature dependent surface reconstructions, as well as SRO features different from those anticipated in the bulk. First, while the ideal bulklike termination of the L12(lll) surface is the (2• structure, according to LEIS, AES and LEED measurements [69] annealing of sputtered Pt3Sn(lll) leads to Sn-enriched ~ • ~ R 3 0 ~ reconstruction, which gradually transforms at higher temperatures to the bulk-truncated (2• structure. At still higher temperatures, it transforms to a PtSn segregated (2 • 2)' new structure. Monte Carlo simulations combined with the "Macroscopic Atom" Model MAM [38] claim that the ~f3 • ~ structure is associated with preferential sputtering effects and the limited atomic mobility at the lower temperatures. (At higher temperatures the (2• 2) ordered surface was predicted.) On the other hand, recent FCEM study [70] predicted stabilization of the ~ • ,f3R30 ~ structure due to even slight bulk off-stoichiometry (<25% Sn). Moreover, assuming enhanced surface interactions (Vs > V) these calculations indicate stabilization of the (2• 2)' reconstruction at the highest temperate range, just as observed experimentally [69]. The second case of non-bulk lateral ordering involves surface SRO. Scanning tunneling microscopy (STM) and quantitative LEED analysis of the (111) surface of disordered Co3Pt alloy revealed SRO that differs from the type expected in the bulk having this composition (L12). In particular, Pt and Co atoms were found to be locally arranged in monoatomic chains with a (1 • 2) unit cell and nearly equiatomic composition, in a manner similar to the ordered L10 phase (Fig.14). Distortions needed for the tetragonal L10 phase explain why this surface ordering does not extend over larger domains [71 ]. The Co3Pt(111) SRO

105

Fig. 14. STM constant current topograph (10 x 10 nm, 0.5 mV, 3.8 nA) of Co3Pt(111) annealed at temperatures in the range 960-1060 K [71]. SRO appears as small areas with (1 x 2) symmetry in monoatomic chains. The bright spots are Pt sites. parameters, quantitatively estimated by direct analysis of the STM images, indicate much stronger preference for unlike nearest neighbors compared to NiaPt(111) or RhaPt(111) [71-73]. 3.2.3 Segregation characteristics o f aluminide surfaces FCEM calculations for three structurally different groups of equiatomic aluminides (B2(110), B32(110) and L10(111)) further demonstrate the decisive role of the segregation/order energetic factor r (Table 3). Surface-field related contributions, Ah, of the listed surfaces were estimated from the pure A1 and second metal surface tensions [74] and lattice parameters were extracted from volume per unit formula data [75] (it was assumed that elastic strain contributions can be neglected in case of equiatomic composition). The interaction strengths were obtained from the corresponding heats of formation applying NN interaction model to B2 and L10 structures (V = V1 , V n = 0 for

n > 1). The B32 structure exhibits ordering also in the second coordination sphere, since NN and NNN interactions are comparable [76]. Therefore, it was treated assuming uniform interactions (V = V1 = V2 , Vn - 0 for n > 2 ). As can be seen from Table 3, the heat of formation (and the corresponding effective interaction strength) of alloys with B2 structure, except for FeA1, is considerably higher than of alloys with the B32 and the L10 structures (it is least exothermic for aluminides of metals close to group 6 [77]).

106 Table 3 Energetic parameters o f aluminum ordered alloys

Alloy

Structure/ Surface

SeA1

B2(110)

,n, era ,ion

Heat o f formation*, (kJ/mol)

strengthmevV**,

-84.6

218

Surface field Ah, meV 14

r = 0.064

CoAl

.

.

.

.

.

63.8

165

566

3.4

NiA1

.

.

.

.

.

67.3

174

455

2.6

RuA1

.

.

.

.

.

58.2

150

905

6.0

RhA1

.

.

.

.

.

89.3

231

674

2.9

.

.

.

.

.

FeA1 CrA1

B32(110)

28.6

74

660

8.9

-11.7

24

363

15

MoA1

.

.

.

.

.

22.9

47

746

16

TeA1

.

.

.

.

.

19.2

40

830

21

TiAI

-37.2

96

629

6.5

VA1

L 10(111) .

.

.

.

.

20.7

54

695

12.9

MnA1

.

.

.

.

.

23.7

61

124

2.0

* From Ref.75 ** Interaction strength in B2 and L10 is calculated in the NN approximation, while in B32 equal NN and next nearest neighbor (NNN) interactions are assumed.

I

1.0 o= 0.9 3

0.8

2

=o 0.7 1

0.6 0.5 0.4 600

\

I

I

I

I

800

1000

1200

1400

1600

Temperature, K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Fig.15. F C E M calculated variations with temperature in the average AI concentration at the (110) surface o f bee aluminides (B2 structure - solid lines, B32 structure - dashed lines). 1 SEA1, RuAI, RhA1, NiA1, CoAl, 2 - FeA1, 3 - CrA1, 4 - MoA1, 5 - TeAl. The surface concentrations were calculated in accordance with data o f Table 3.

107

i i. ................... iiiiii

IJ 0.80 co

I ."

0.75

.~ .....

t-tO

o E g < o t~ 3:

=

O9

Z

0.70

.."

,.....'"!

.+.

0.65 .. ....

0.60

/

"'" ,.

0.55

.,'"

0.50 .... "

0.45 '

'

,tO

~

,'0

200

~.

. . .

~~

.......

, ............

i. . . . . . . . .

400

,

I

I

,

'

X

',~

(Vt3 X~3)R30 o

I

I

'CT~t'~

~.."

., "d order i '

:9~

/

1_= I

.

a

.

.

.

.

.

.

B(30

600

I

I}

I '

1000

Temperature (~

l0~ C

._o

8=

0,80-

T (100) 0,75- ~ (110) - --11--(210) 0,70 - - e - - (310)

o

0,65 -

E:=

0,60

=

0,55

8

o,5o

co

0,45 -

z

0,40

ca" a~

--0--

(111)

(210) (110

. 0

s~ 200

400

600

800

1000

Temperature (*C)

Fig.16. (a) AES determined near-surface (average) concentration of A1 as function of annealing temperature for the FeAI(111) surface (three datasets). The dotted lines estimate the uncertainty introduced by the error in the matrix factor. The phases, which are observed in LEED after quenching the annealed sample to room temperature, are also shown. (b) Comparison of the segregation curves for all investigated surface orientations. Near-surface concentrations corresponding to bulk terminated surfaces are marked by open circles [77]. These energetic parameters were used in F C E M calculations assuming segregation at the three outmost layers only. As shown in Fig.15, the segregation tendency prevails only in the B32 ordered alloys and the surface concentration decreases with temperature (entropy-driven monotonous desegregation). This behavior is associated with the distinctly high segregation/order factor (sec. 3.1). On the other hand, ordered bulk truncation with surface concentration very

108

slightly increasing with temperature is predicted for most of the B2 ordered alloys. Only in FeA1, with relatively low heat of formation and intermediate r value, there is a subtle balance between segregation and order, leading to a peaked segregation curve in experimentally accessible range of temperatures. Indeed, as measured by AES and LEED, the behavior of the FeAI(ll0) [77] differs substantially from the equivalent surfaces of alloys with NiA1 and CoAl like energetic parameters, which exhibit nearly perfect stoichiometry at the top layer (bulk truncation) [78-80]. The predicted surface segregation increase in FeA1 with annealing temperature was observed [77] also for other surface orientations and was accompanied by surface reconstructions (Fig.16). The calculations for the strictly stoichiometric FeA1 alloy predict somewhat higher segregation levels (Fig.15) compared to the reported formation of incommensurate FeA12 surface alloy on the FeAI(ll0) surface [77]. The discrepancy can be due, at least partially, to a slight deviation from stoichiometry in the measured alloy (see Fig.7). As an example for a third class of equiatomic aluminides, calculations done for three fcc L10 alloys are presented in Fig.17. Again, the segregation behavior is governed mainly by the segregation/order interplay, as expressed by means of r (see Table 3). Only in VA1 with relatively high r value (12.9), the segregation tendency prevails.

1.0 3

0.9

\

0

"~ 0.8

2

,D 0 r

0.7

1

\

0.5 0~

i

i

i

1

1

600

800

1000

1200

1400

1600

Temperature, K

Fig.17. F C E M calculated variations with temperature in the average A1 surface concentration at the (111) surface of fcc L10 aluminides. 1 - MnA1, 2 - TiA1, 3 - VA1. The surface concentrations were calculated in accordance with data of Table 3.

109 4. S e g r e g a t i o n in a b i - p h a s e b i n a r y a l l o y

As discussed in previous sections the involvement of ordering effects in binary alloy surface segregation complicates its theoretical treatment. Unraveling segregation phenomena in multi-component alloys is another challenge. But the situation can become even more complex for segregation in multi-phase alloys, when distinct segregation processes from individual bulk phases are coupled to the temperature dependent phase equilibria (Fig.18) In particular, in many binary alloy systems with ordering tendency bi-phase equilibrium exists between a solid-solution and an ordered compound when the bulk concentration exceeds the solubility limit (Fig.18). As discussed below, besides segregation~RO-SRO effects that can be operative in each phase separately, the variations with temperature in the solid-solution bulk composition can have a dominant effect and also lead to peaked segregation curves. Such a behavior, as measured by means of XPS, was reported previously for fcc-based A1-3%Ag alloy equilibrated between 550 and 770 K [82] (Fig.19a). Below the bulk phase transition (680K)hcp-based Ag2Al-like

I

77~

726

..--..

o

v

Ag

@O

611 $

==

~

I--

567

6:1.

76,.5-~..~..

2-phase 2D equilibrium a(surface) < >- 6(surface) Atomic seg.

Cluster~ seg. /

c~(bulk) 9 > 2-phase 3D equilibrium 0

AO

10

20

30

40

50

60

Atomic Percent AI

70

8(bulk)

80

90

100

AI

Fig.18. Phase diagram of A1-Ag [81]. Insert: schematics of processes pertinent to surface segregation in bi-phase alloys ( a - solid solution, 6;-ordered compound).

110

'

'

I

bI

2.50

13UP

2.40

9 DOWN

z 2.30 ,to

BULK TRANSITION

:

[]uP

i

"

2.20

0.50

'~

-

0.40

i

"q

0.20 570

670

770

TEMPERATURE (K)

Fig.19. The XPS bandwidth of the Ag 4d states in A1-3at.%Ag (top) and the Ag concentration (bottom), deduced from the emission intensity, as a function of temperature [82]. The bulk phase transition lies at 680 K. clusters (~-phase) precipitate in the fcc solid solution a (Fig.18). Evidence for the appearance of (111) surface clusters came from secondary electron imaging (SEI), Fig.20. In addition, changes in the Ag 4d linewidth (Fig.19b) were attributed to varying numbers of Ag neighbors of a given Ag atom, and thus were supposed to reflect the relative extent of clustering vs. Ag dissolved in the A1 matrix. Based on these data, the three regions in the segregation curve (Fig.19a) were tentatively attributed to: i) Segregation enhancement of small Ag2Al-like clusters with increasing temperature; ii) Their gradual dissolution (first-order phase transition), without a change in the overall concentration of Ag atoms in the analyzed volume (610-690K), and iii) Ag atom desegregation at higher temperatures. Recently, an attempt was made to analyze the compositional changes in a quantitative manner and so to elucidate the pertinent mechanism in terms of the

111

Fig.20. Secondary-electronimaging (SEI) pattem obtained from epitaxial Ag on AI(111) heat treated at 410 K. The sixfold symmetry verifies the formation of Ag2AI clusters with hcp structure [82]. two-phase bulk equilibrium as well as the involvement of two distinct segregation routes, that of atomic Ag and of Ag2A1 clusters [83]. In principle, since the formation of ordered phase clusters at lower temperatures is accompanied by a reduction in Ag solute concentration in the bulk of the solid solution, surface segregation from the latter is suppressed. As temperature increases, gradual dissolution of bulk clusters results in increase in bulk and surface Ag concentration of the solid solution matrix. Around the phase transition (crossing the solubility line), when the solid-solution composition becomes constant with temperature (3%Ag), the surface concentration starts to decrease monotonously as is common in random solid solutions (McLeanLangmuir entropy driven desegregation). More quantitatively, the Ag concentration c a of the bulk solid solution can be simply evaluated from the relevant portion of the A1-Ag phase diagram (Fig.18). For a given Ag overall atomic concentration ( c ) , c a increases with temperature (concomitantly with decreasing amounts of the d-phase clusters) according to the solubility line approximate formula c a = A exp -

.

(6)

112

0.3

0.04

0.25

0.03

0.2

0.02

0.15

0.01

O

~D O w o

<

0.1 500

]

I

I

l

I

550

600

650

700

750

800

0.05

b 0.04

0.03

q~'

0.02

~'~ 0.01

0 500

550

600

650

I

i

700

750

800

Temperature, K Fig.21. Segregation temperature dependence in the bi-phase A1-3%Ag alloy: (a) The surface concentration of the solid solution phase (Cas), based on the depicted c a curve. Circles: average surface concentration derived from experimental data of Ref.82 (as in Fig.19a) and c s is the best-fit line. (b) The bulk fraction x 6 and surface fraction X 6 s of the AgzA1 clusters, derived from c a and

c s -cc~ s ,

respectively.

Fitting this equation to AI(Ag) solubility experimental data in the temperature range from 573 to 773 K [84] gives A ~ 62.2 and A ~ 43.0 kJ/mol. The monotonous increase in c a with temperature is depicted in Fig.21 a, and the corresponding decreasing 6-phase fraction (lever rule) in Fig.21 b. Eq.6 with these parameters together with the apparent segregation excess = - 3 8 . 5 kJ/mol and segregation excess entropy = 36.1 enthalpy

AHseg

ASseg

113

kJ/mol/K (as deduced by Lee et al [85] from experimental data on Ag segregation in A1-4.2%Ag) have been used for a Langmuir-McLean type calculation of monolayer segregation levels from the A1-3%Ag solid solution (Cas), including the two-phase region (Fig.21 a), Cas

I. = c----~-ae xI pAHseg - ~ + ASseg ~

1- cas

1-cat

kT

(7)

k

Since ]AHseg]
c s = X6sC6 + (1 - X6s)Cas.

(8)

The values X6s derived from the experimental c s and calculated Cas values (Fig.21b) exhibit a distinct peaked dependence on temperature with maximal surface fraction pertaining to the 6-phase at --70 K below Tb . Questions regarding possible kinetic limitations at the low temperature range [83], the mechanism and driving forces for cluster segregation, and the possibility of clustering/ordering due to segregating Ag atoms (drawn schematically in Fig.18) remain open. This approach can be applied to segregation phenomena in other two-phase alloys, such as reported recently for polycrystalline Fe99Pdl with a small fraction of ordered FePd nuclei [86]. Positive entropy and enthalpy characterize the observed increase in segregation level with temperature. 5. S U M M A R Y

This review was aimed at elucidating some recent developments regarding surface segregation phenomena in metallic alloys exhibiting appreciable compositional order. It comprises three main issues: (i) short range order (SRO) effects in multi-component solid solutions, (ii) the role of long range order, LRO (with emphasis on the underlying energetics), and (iii) preliminary analysis of

114

segregation trends characteristic of bi-phase alloys. For qualitative and quantitative estimation of segregation/order interplay the statistical-mechanical "Free Energy Concentration Expansion Method" (FCEM), that uses simple pair bond energetics but takes into account SRO analytically, appears to be most convenient, and demands much less computational effort than Monte Carlo simulations, examples of which are presented too. In the first issue (sec.2) the FCEM formulas were extended to the case of alloys with any number of components and were applied to the elucidation of effects of interatomic interactions and SRO on surface segregation in Ni-8%A14%Cu as a model ternary solid solution. The calculations revealed strong, SRO enhanced Cu segregation/Al+Ni desegregation in a site competition process that is largely attenuated following a compositional phase transition. The FCEM analytical formulas are applicable to surface segregation in semi-infinite alloys with any number of components, as well as to thin films and nano-particles, allowing systematic studies of SRO effects in these systems. In the second issue (sec.3) several factors pertinent to segregation LRO (and SRO) induced modifications (mainly suppression) are addressed. Besides structural factors such as the surface orientation, the energy balance between the two tendencies plays a decisive role in most phenomena, and can be simply expressed by means of its numerical ratio (the "segregation/order factor", r). According to the FCEM model calculations for alloys with large ratio (e.g., r > -~ 10) segregation should prevail and exhibit a monotonous decrease with temperature typical to random solid solutions, whereas for low r values order is expected to dominate and segregation should be strongly or even completely inhibited. In the most interesting case of alloys with intermediate r values, representing a subtle balance of the two tendencies, a peaked segregation vs. temperature dependence is expected. Depending on the alloy energetics ( r ) and crystal structure, the surface transition temperature can either coincide with the bulk one, or be higher (surface-induced order, SIO), or lower (surface-induced disorder, SID). Experimental studies of equiatomic aluminides and Cu3Au are in general agreement with the predictions. Another factor manifested in the segregation/order interplay involves ordered alloys with slight off-stoichiometry, where the tendency for a perfectly ordered bulk enhances segregation of the element in excess. While both LRO and SRO effects on segregation are determined largely by the energetic balance as reflected by the magnitude of r, the role of LRO is naturally more prominent, and thus received more attention in this review. Yet, as exemplified for the above mentioned ternary solid solution, SRO associated with appreciable solute-solvent interactions can affect considerably surface compositional phase transitions. Likewise, in ordered fcc alloys with both LRO and SRO operative, order-disorder surface phase-transition temperatures are significantly shifted by SRO, thus modifying the individual sub-lattice

115 concentrations and the average surface composition. Another aspect briefly addressed deals with the nature of in plane surface atomic order. Several cases of deviations from the bulk truncated surface order of the long range type (surface reconstruction), and surface modified SRO are demonstrated. Order/segregation interrelations, in their more general sense, can become even more complex in multi-phase alloys, when distinct surface segregation processes from individual bulk phases may be coupled to temperature dependent phase equilibria. Bi-phase systems, comprising of ordered clusters in a solid solution matrix, can exhibit two distinct but interrelated routes, namely, elemental segregation from the solid solution and segregation of small clusters. It is demonstrated for A1-3%Ag (sec.4) that variations with temperature in the solid-solution bulk composition alone can have a dominant effect leading to peaked segregation-temperature curves too. As many metallic alloys of practical importance are both multi-element and multi-phase, extension of this preliminary theoretical analysis seems to be desirable. Finally, it should be noted that compared to a prominent theoretical progress in elucidating trends associated with segregation/order interplay by means of FCEM calculations (or by the generally more accurate Monte Carlo simulations with embedded-atom-method energetics), comprehensive experimental studies of the phenomena are still lacking. It can be anticipated that the growing use of advanced techniques sensitive to the atomistic features of surface structural order and composition will be directed to further unraveling of segregation/order issues.

REFERENCES [ 1] M. Polak and L. Rubinovich, Surf. Sci. Rep., 38/4-5 (2000) 127. [2] J. L. Moran-Lopez and L.M. Falicov, Phys. Rev., B 18 (1978) 2542. [3] M. Polak, J. Deng and L. Rubinovich, Phys. Rev. Lett., 78 (1997) 1058. [4] S. Hofmann and P. Lejcek, Colloque de physique, 51 (1990) C1-179. [5] E. Taglauer, (private communication 1998). [6] P. Weinberger, V. Drchal, L. Szunyogh, J. Fritscher, B. I. Bennett, Phys. Rev., B 49 (1994) 13366. [7] S. Dorfman, V. Liubich, D. Fuks, Intern. Journ. of Quant. Chem., 75 (1999) 927. [8] I. Mirebeau, M. Hennion and G.Parette, Phys. Rev. Lett., 53 (1984) 687. [9] M. Polak, C.S. Fadley, L. Rubinovich, Phys. Rev B (in press). [ 10] D. McLean, Grain Boundaries in Metals, Oxford University Press, London, 1957. [11] J. L. Moran-Lopez and K. H. Bennemann, Phys. Rev., B 15 (1977) 4769. [12] M.J. Sparnaay, Surf. Sci. Rep., 4 (1984) 101. [13] V. Kumar and K. H. Bennemann, Phys. Rev. Lett., 53 (1984) 278. [14] J.M. Sanchez, J.L. Moran-Lopez, Phys. Rev., B 32 (1985) 3534. [15] J.M. Sanchez, J.L. Moran-Lopez, Surf. Sci. Lett., 157 (1985) L297. [ 16] J.M. Sanchez, J.L. Moran-Lopez, Statistical Thermodynamics of Surfaces and Interfaces, in Nanophases and Nanocrystalline Structures, R.D. Shull and J.M. Sanchez Eds., A publication of TMS, Warrendale, Pennsylvania, 1993.

116 [ 17] Y. Teraoka, Surf. Sci., 242 (1991) 113. [ 18] A.V. Ruban, I.A. Abrikosov, D.Ya. Kats, D. Gorelikov, K.W. Jacobsen and H.L. Skriver, Phys. Rev., B 49 (1994) 11383. [19] J.M. Roussel, A. Saul, L. Rubinovich and M. Polak, J. of Phys.: Cond. Matt., 11 (1999) 9901. [20] L. Rubinovich and M. Polak, Surf. Sci. (submitted for publication). [21] F.F Abraham and C.R. Brundle, J. Vac. Sci. Technol. 18(2) (1981) 506. [22] M.S. Dow, M.I. Daskes, Phys. Rev., B 29 (1984) 6443. [23] S.M. Foiles, Phys. Rev., 32 (1985) 7685. [24] S.M. Foiles, Calculation of the surface segregation of alloys using the embedded atom method in: P. A. Dowden and A. Miller Eds. Surface segregation phenomena, CRC, Boston, 1990. [25] S.M. Foiles, M.I. Baskes and M.S. Daw, Phys. Rev., 33 (1986) 7983. [26] J. Tersoff, Phys. Rev., B 42 (1990) 10965. [27] H. Stadler, W. Hofer, M. Schmid and P.Varga, Phys. Rev., B 48 (1993) 11352. [28] H.Y. Wang, R. Najafabadi, D.J. Srolovicz, R. LeSar, Phys. Rev., B 45 (1992) 12028. [29] R. Najafabadi and D. J. Srolovitz, Surf. Sci., 286 (1993) 104. [30] A. Saul, B. Legrand, G. Treglia, Phys. Rev., B 50 (1994) 1912. [31] G. Bozzolo, B. Good, J. Ferrante, Surf. Sci., 289 (1993) 169. [32] M. Hou, M. E1 Azzaoui, Surf. Sci., 380 (1997) 210. [33] K. Binder, Phase transitions at surfaces. In: Cohesion and Structure of Surfaces ED D.G. Pettifor. Elsevier Science B.V., 1995. [34] P. Wynblatt and A. Landa, Comp. Mater. Sci., 15 (1999) 250. [35] B. Good, G.H. Bozzolo and P.B. Abel, Surf. Sci., 454/456 (2000) 602. [36] A. Maidou and H. M. Polatoglou, Phys. Rev., B 60 (1999) 9145. [37] C. Creemers and P. Deurinck, Surf. Interface Anal., 25 (1997) 177. [38] C. Creemers, S. Helfensteyn, Appl. Surf. Sci., 167 (2000) 216. [39] M. I. Baskes, Phys. Rev., B 46 (1992) 2727. [40] P. Deurinck, C. Creemers, Surf. Sci., 419 (1998) 62. [41] U. Bardi, Rep. Prog. Phys., 57 (1994) 39. [42] M. Guttmann, Metall. Trans. 8A (1977) 1383; M. Guttmann and D. McLean, in: Interfacial Segregation, Eds. W.C. Johnson and J.M. Blakely, ASM, Metals Park, OH, 1979, p.261. [43] M.A. Hoffmann and P. Wynblatt, Metall. Trans., 20A (1989) 215. [44] W.C. Cheng and P. Wynblatt, Surf. Sci., 364 (1996) 409. [45] M.A. Krivoglaz and A.A. Smirnov, The Theory of Order-Disorder in Alloys, Macdonald, London, 1964. [46] R V Chepulskii y and V N Bugaev, J. Phys.: Condens. Matter, 10 (1998) 7309. [47] C.H.P. Lupis, J. F. Elliott, Acta Met. 14 (1966) 1019; C.H.P. Lupis, Chemical Thermodynamics of Materials, North-Holland, New York-Amsterdam-Oxford, 1982. [48] G. Treglia, B. Legrand, P. Maugain, Surf. Sci. 225 (1990) 319. [49] O. Kubaschewski and C. B. Alcock, Metallurgical Thermochemistry. Pergamon, Oxford, 1979. [50] M. Polak, M. Talianker, R. Arkush, Surf. Sci., 273 (1992) 363. [51 ] M. Polak, L. Rubinovich, Surf. Sci., 377 (1997) 1019. [52] V.S. Sundaram, B. Farrell, R.S. Alben, and W.D. Robertson, Phys. Rev. Lett., 31 (1973) 1136. [53] T.M. Buck, G.H. Wheatley, and L. Marchut, Phys. Rev. Lett., 51 (1983) 43. [54] E.G. McRae and R.A. Malic, Surf. Sci., 148 (1984) 551.

117 [55] K.D. Jamison, D.M. Lind, F.B. Dunning, and G.K. Waiters, Surf. Sci. Lett., 159 (1985) L451. [56] S.F. Alvarado, M. Campagna, A. Fattah, and W. Uelhoff, Z. Phys., B 66 (1987) 103. [57] H. Dosch, L. Mailander, A. Lied, J. Peisl, F. Grey, R.L. Johnson, and S. Krummacher, Phys. Rev. Lett., 60 (1988) 2382. [58] E.G. McRae and T.M. Buck, Surf. Sci., 227 (1990) 67. [59] H. Dosch, L. Mailander, H. Reichert, J. Peisl, and R.L. Johnson, Phys. Rev., B 43 (1991) 13172. [60] H. Reichert, P.J. Eng, H. Dosch, and I.K. Robinson, Phys. Rev. Lett., 74 (1995) 2006. [61] H. Niehus and C. Achete, Surf. Sci., 289 (1993) 19. [62] H. Niehus, Phys. Stat. Sol., B 192 (1995) 357. [63] F.M. Zhang, B.V. King, and D.J. O'Connor, Phys. Rev. Lett., 75 (1995) 4646. [64] D.H. Oh, H.J. Kang, K.H. Chae, C.N. Whang, B.V. King, D.J. O'Connor and D.W. Moon, Surf. Sci., 477 (2001) L289. [65] Mannori C, Boato G, Canepa M, Cantini P, Mattera L, Terreni S, Europhys. Lett., 45 (1999) 686. [66] J.M. Cowley, J. Appl. Phys. 21 (1950) 24; Phys. Rev., 77, 669 (1950). [67] Y.C. Yong, H.C. Poon, Surf. Sci., 338 (1995) L825. [68] R.H. Bergmans, M. van de Grift, A.W. Denier van der Gon, H.H. Brongersma, Surf. Sci., 345 (1996) 303. [69] W.C.A.N. Ceelen, A.W. Denier van der Gon, M.A. Reijme, H.H. Brongersma, I. Spolveri, A. Atrei, U. Bardi, Surf. Sci., 406 (1998) 264. [70] M. Polak and L. Rubinovich, to be published. [71 ] Y. Gauthier, R. Baudoing-Savois, J.M. Bugnard, W. Hebenstreit, M. Schmid, P. Varga, Surf. Sci., 466 (2000) 155 [72] M. Schmid, H. Stadler, P. Varga, Phys. Rev. Lett., 70 (1993) 1441. [73] E.L.D. Hebenstreit, W. Hebenstreit, M. Schmid, P. Varga, Surf. Sci., 441 (1999) 441. [74] L.Z.Mezey and J. Giber, Jap. J. App. Phys., 21 (1982) 1569. [75] D. Nguyen-Manh, D.G. Pettifor, Intermetallics, 7 (1999) 1095. [76] L.M. Rubinovich, D.M. Stem, E.V. Kozlov, Izv. Vuz. Fiz., 32 (1989) 11 (in Russian). Translation: Sov. Phys. J., 32 (1989) 588. [77] L. Hammer, H. Graupner, V. Blum, K. Heinz, G.W. Ownby, D.M. Zehnerb, Surf. Sci., 412/413 (1998) 69. [78] H.L. Davis, J.R. Noonan, Phys. Rev. Lett., 54 (1985) 566. [79] D.R. Mullins, S.H. Overbury, Surf. Sci., 199 (1988) 141. [80] V. Blum, C. Rath, G.R. Castro, M. Kottcke, L. Hammer, K. Heinz, Surf. Rev. Lett., 3 (1996) 1409. [81 ] T.B. Massalski, et al., Binary Alloy Phase Diagrams, 2 ed., OH, ASM Int.: Materials Park, 1990. [82] M. Erbudak, M. Hockstrasser and E. Wetli, J. Electron Spectr. Related Phenom., 76 (1995) 529. [83] M. Polak and L. Rubinovich, to be published. [84] M. Hansen, K. Anderko, Constitution of Binary Alloys, New York, Mcgraw-Hill, 1958. [85] H.K. Lee, R.W. Hyland, H.I. Aaronson, P.P. Wynblatt, Surf. Sci., 408 (1998) 288. [86] C. Creemers, Surf. Sci., 360 (1996) 10.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces 118

D.P. Woodruff, (Editor)

Chapter 4

Segregation and surface chemical o r d e r i n g an experimental view on the atomic scale M. Schmid and P. Varga Institut ftir Allgemeine Physik, Technische Universit~it Wien, A-1040 Wien, Austria 1. I N T R O D U C T I O N It has been recognised already many years ago that surfaces of solids can differ significantly from the bulk concerning crystallographic and electronic structure. It was the development of adequate and reliable techniques for surface characterisation in the last fifty years which made a systematic study of the properties of surfaces feasible. This knowledge about surfaces is of significant practical interest for example in heterogeneous catalysis where the relevant chemical processes like adsorption, surface diffusion and desorption are almost exclusively determined by the very surface. While pure metal surfaces can differ from the bulk only with respect to crystallographic structure (reconstruction, relaxation), in most cases alloy surfaces also have different composition. Therefore alloy surfaces often exhibit chemical reactivity which is significantly different from that of pure metals and in some cases these surfaces exhibit enhanced catalytic performance. The reason for the change in surface composition is enrichment of one component in the surface by segregation. This has been of some intrinsic interest already in the past because relatively simple macroscopic thermodynamic theories have proved more or less successful in explaining some of the associated phenomena [1,2]. More elaborate models or computer simulations are often needed to deal with systems where the interplay of chemical order and surface segregation becomes important [3,4]. A detailed understanding of the driving force of segregation often needs complete modelling based on density functional theory [5]. From the experimental point of view, methods sensitive to the very first layer are necessary for a proper description. Low energy ion scattering spectroscopy (LEIS) performed with noble gas ions has shown its

119

usefulness in determining the chemical composition within an accuracy of a few atomic percent because of its sensitivity to the very first layer of closepacked surfaces [6]. On more open surfaces, exploiting different shadowing or blocking geometries allows to separate the signal of the first monolayer from that of deeper layers and thereby to determine the concentrations of more than one layer [7-11 ]. The second widely used technique is quantitative low energy electron diffraction (LEED), yielding concentrations of the uppermost two to five atomic layers, together with structural data (relaxations) [12,13]. Using a combination of LEIS results for the first layer and quantitative LEED analyses helps to verify the accuracy of both methods and to avoid possible ambiguities [ 14] of quantitative LEED. By this approach, the crystallographic structure and also the chemical compositions of the first three to five atomic layers have been determined for many alloys [15-19]. Previous studies of surface chemical order on metals were based on diffraction techniques, mostly LEED [ 12,13]. These methods are mainly useful in the case of long-range chemical order, whereas the diffuse scattering intensities resulting from short-range chemical order, if detectable at all, cannot be easily interpreted and analysed. Since these k-space methods average over large regions of the sample, interpretation of their results is limited to statistical criteria. Many alloys do not have long-range chemically ordered phases at all or their order-disorder transition temperature is too low, so that the mobility of the atoms is too small to form long range ordered structure during annealing and cooling down. This is mainly a problem of the bulk, but it can also occur in the surface layer, where diffusion is usually much faster. Coming back to gas adsorption and heterogeneous catalysis, alloys often show so-called synergistic effects of the constituent elements. In a microscopic view, these effects can be often traced back to so-called ensemble- and ligand effects [20,21]. The term ensemble effect refers to the fact that adsorption and bonding of a molecule requires a specific number of unoccupied adjacent sites of a particular atom type. The ligand effect represents the modification of the adsorptive properties of a given site via electronic effects by the neighbours (ligands) of an atom that a given adsorbate binds to. For understanding such phenomena, it is essential to determine the composition and chemical structure of the surface at a given adsorption site. Therefore, it is not enough to know the chemical composition of the surface in average but rather a detailed knowledge of the local composition and local ordering on the atomic level is necessary. In the last decade, STM has proven its power as a unique instrument in that respect, especially if the resolution is high enough that not only atomic resolution is achieved but also different elements can be distinguished in STM images taken in constant current mode (chemical contrast) [22,23].

120

2. C H E M I C A L D I S C R I M I N A T I O N ON B I M E T A L L I C S U R F A C E S W I T H A T O M I C R E S O L U T I O N BY STM Scanning tunneling microscopy has become a powerful technique for the study of solid surfaces. The most commonly used mode of operation is the constant current mode. In this mode, a fixed voltage VT is applied between tip and sample and a feedback loop keeps the tunneling current I T between tip and sample constant by adjusting the tip-sample distance with a piezoelectric element. By scanning the surface and recording the voltage applied to the piezoelectric element controlling the tip-sample distance (height z) a topographic image can be obtained. Considering that the tunneling current not only depends on the tip-sample distance but also on their electronic properties, it is clear that constant current STM images contain both geometric and electronic structure information, which can be quite difficult to disentangle. Under some simplifying assumptions, it can be shown that constant current topographs are actually maps of constant electronic density of states at the Fermi level [24]. For metals one can usually assume that the atoms are situated where STM shows a protrusion, because normally the highest density of states is where the atoms are. Two notable exceptions are the Fe(100) and Cr(100) surfaces where the protrusions in the constant current image can appear at the four fold hollow position of the surrounding metal atoms [25,26]. In general, it is possible to obtain information about the electronic structure of a surface by the spectroscopy mode of a scanning tunneling microscope where the voltage dependence of the tunneling current is studied (STS, scanning tunneling spectroscopy; see, e.g. Ref. 27). In principle, STS should be able to distinguish also between different elements in a metal surface as long as the energy dependence of the local electron density of states [LDOS] of the elements is different. Various spectroscopic modes of operation have been introduced in the past and applied successfully to distinguish between topographical and electronic effects as well as between different elements at a surface as long as no atomic resolution is necessary. Unfortunately, for all spectroscopic modes loss of atomic resolution is inherent especially at metal surfaces because the possible variation in tunneling parameters VT and I T where atomic resolution is achieved is extremely narrow. Nevertheless, discrimination between different metal atoms at many different alloy surfaces has been observed in constant current mode, a phenomenon known as chemical contrast [22]. The basis for such images is a resolution which is high enough to probe small differences between the atoms, e.g., in electron density of states. This means that the residual noise of the STM in z

121

Fig. 1. Mechanisms of chemical contrast in STM (schematic). direction should be small enough (=3• -12 m in our STM). In some cases, chemical contrast can be understood as due to different local density of states (LDOS), i.e., within the framework of the theory of Tersoff and Hamann [24]. LDOS contrast includes cases where one type of atom is significantly larger and/or geometrically higher than the other and, hence, imaged higher by STM. In other cases, chemical contrast obtained in experiment cannot be explained as due to the local density of states and must be attributed to some mechanism beyond the theory of Tersoff and Hamann, i.e., interaction with the tip. The possible reasons for chemical discrimination in constant current images are sketched in Fig. 1 and will be discussed in more detail in the following sections.

2.1. True topographic effect The most simple reason for a contrast between different atom species at the surface is a true topographic effect, i.e., the difference in atom size and/or difference in atomic position (height). An example is the surface alloy formed if Pb is deposited in submonolayer amount on a Cu(111) surface (Fig. 2). Here, we can easily identify the Pb atoms which substitute single Cu atoms in the first layer [28]. The difference in size (metallic radii) of Pb and Cu is large, 37%, or 47 pm in radius. With such a large size difference, it has to be expected that the Pb atoms relax outwards from the Cu surface layer. Unfortunately, exact data for this outwards relaxation are not available for Pb/Cu(111), but a LEED study of an ordered alloy of Pb and Cu on Cu(100) shows a height difference of 65 pm between the cores of the Pb and Cu atoms [29]. Adding the difference in radius to this number would let us expect a difference of apparent height in STM of more than 100 pm (1 A). Experimentally, under moderate tunneling conditions the Pb atoms usually appear as protruding about 4 0 - 60 pm from the Cu surface. The apparent height of the Pb atoms can get as low as 10 pm,

122

Fig. 2. STM constant current topograph (10• nm 2) of a Cu(111) surface with approx. 20 Pb atoms embedded in the first layer [28]. The low tip-sample distance (VT=-0.5 mV, I T = 2.9 nA) necessary for atomic resolution of the Cu lattice leads to a frizzy appearance of the step edge (tip-induced diffusion, left) and to a distorted appearance of some of the Pb atoms.

however, under tunneling conditions where the tip comes very close to the sample and both, the Pb atoms and the Cu lattice, are resolved. For alloys of transition metals the difference in size is much smaller than in this case and the differences of core heights between the alloy constituents are usually only a few pm. For disordered alloys, this is superimposed on height variations of similar magnitude caused by random variations of concentration and, hence, atomic size in the deeper layers. It is therefore unlikely that topographic chemical contrast becomes visible directly in STM. Even if buckling of the surface because of chemical ordering causes a notable difference in height for the different alloy constituents, the only way to identify such a possibility for chemical contrast is comparison with either ab initio calculations or with quantitative LEED experiments. Such a case where the chemical contrast was attributed to different sizes is an ordered CuAu surface alloy on Cu(100). The height differences found experimentally (20-70 pm) are significantly larger than the difference of true geometric height determined by LEED (10 pm) and sometimes also exceed the sum (26 pm) of geometric height and radius difference [30]. We have to conclude that topography, i.e., difference of geometric height

123

and atomic radius, explains the chemical contrast only in a few cases where the atomic sizes differ significantly. Even in these cases other factors such as electronic density of states or tip-sample interaction can contribute to the chemical contrast observed or weaken the contrast expected from pure geometry.

2.2. Difference in local electronic density of states The second reason for chemical discrimination is a pure electronic effect, i.e., different local density of states, following the theory of STM by Tersoff and Hamann [24]. This theory was derived for small tunneling voltages under the assumption that the tip wavefunction responsible for tunneling has s-wave symmetry. Any interaction between tip and surface except the tunneling current is neglected. Under these approximations, the contrast seen by STM is due to the difference in electron density of states near the Fermi edge at the position of the STM tip, i.e., above the different atoms. Such a difference is obvious in alloys between transition metals with a partially filled d shell and the noble metals Au, Ag and Cu because of the huge difference in density of states at the Fermi edge. We therefore expect the noble atoms to appear darker, i.e., lower than the transition metal atoms. This was indeed observed for the (100) and (111) surfaces of AgPd alloys [31 ] as well as for Au3Pd(100) [32]. For alloys made only from transition metals with partially filled d shell chemical contrast is much more difficult to predict because of the smaller difference of LDOS between the pure elements and possible alloying effects affecting the LDOS. To find whether the LDOS is responsible for chemical contrast observed on such an alloy, or why chemical discrimination is not observable one has to perform ab initio calculations of the LDOS by using one of the methods based on density functional theory [5]. We are using results obtained with the FLAPW method (Full potential linearized augmented plane waves) [33], which has the advantage of using exponentially decaying wavefunctions in the vacuum above the surface, where the STM tip is probing the LDOS. This is in contrast to plane-wave codes where a supercell symmetry in z direction is necessary and the rapidly decaying wavefunctions above the surface have to be represented by a superposition of plane waves in the vacuum region of the supercell. To compare the calculations with constant current STM images the LDOS around the Fermi edge is integrated and the corrugation of this value (i.e., difference of electron density of states between hollow and on top sites) is evaluated as a function of the distance from the surface. We have shown at least qualitatively that such an electronic effect is responsible for the chemical resolution achieved on PtCo (Fig. 3) and PtRh(100) surfaces [34]. Quantitatively the calculated values are about a factor 2-3 smaller than the measured corrugation. One way to explain this discrepancy is the possibility of p or d

124

Fig. 3. STM constant current topograph (20x12 nm 2) of the (100) surface of a Pt25Co75 single crystal [37]. Pt atoms appear as protrusions (bright), Co as depressions (dark). Most of the surface shows chemical ordering, the Co and Pt atoms form a c(2• pattern (see white frame enlarged as inset). Unrelated to the chemical ordering, the surface also shows a reconstruction, characterised by a few shifted rows of atoms (arrows). wave functions at the tip, which can result in an increase of the corrugation by a factor of 3 as c o m p a r e d to the s wave functions assumed in the Tersoff H a m a n n theory [35]. For the PtCo(100) and (111) surfaces the calculations have shown that Co atoms have a higher LDOS at the Fermi edge than Pt atoms near their cores. This density decays faster into the vacuum above the Co atoms, and Co atoms therefore appear always darker (smaller) than Pt atoms [22,36]. Whereas one might consider this an effect of different atomic sizes at first glance, comparison with PtNi surfaces shows that such a simple view is not justified. Both Co and Ni atoms are approx. 10% smaller than Pt, nevertheless PtNi(100) and (111) hardly show any LDOS contrast whereas there is significant LDOS contrast on the corresponding PtCo surfaces, with Pt appearing higher in the STM images.

125 For PtRh(100) an alloying effect decreases the LDOS at the Fermi edge for the Pt atoms and therefore the Pt atoms appear darker (lower) than the Rh atoms [34]. Chemical contrast by tip-sample interaction (see next section) will also lead to Rh atoms appearing bright and may therefore enhance this effect.

2.3. Tip-surface interaction If the difference of LDOS above the different atoms of a compound surface is not enough to explain the STM images even qualitatively other mechanisms like tip-sample interaction or special tip conditions have to be considered. Tip-surface interaction is not taken into account in the theory of Tersoff and Hamann and difficult to describe by theory. Current density functional theory codes are well-suited to calculate the forces arising between the tip and sample atoms at small distances, whereas they cannot determine the tunneling current. On the other hand, methods developed for transport properties of solids cannot be easily adapted for the complicated geometries and potentials of the interacting tip and surface. These difficulties, besides the unknown geometric and chemical structure of the tip have so far prevented any full calculation of the tunneling current in the presence of tip-surface interaction. Nevertheless, we believe that tip-surface interaction determines the chemical contrast in all cases where a measured difference in corrugation at a surface can not be explained by a size effect or an electronic effect. Examples for such systems are PtNi(111) [38], PtRh(111) (Fig. 4, Ref. 39) and the quasi hexagonal reconstructed PtNi(100) (Ref. 40, see section 4.5) surfaces. FLAPW calculations of LDOS contrast have indeed shown a negligible corrugation differ -~ ence for the PtNi(111) and PtRh(111) surfaces. In all these cases, the experiments showed chemical contrast only occasionally, indicating that it strongly depends on the composition and/or structure of the tip. From experimental evidence the following model has been proposed which takes into account the different chemical reactivity of the elements in the surface [38,22]. At a small distance (estimated core-core separation of about 4 - 5 A), the forces between the foremost tip atom and the surface will be related to the strength of chemical bonding between the atoms, even if the distance is too large to allow the formation of a strong chemical bond. Both movement (relaxation) of the atoms under this force and possible increase of charge density between tip and sample as a result of this "precursor" of a chemical bond will increase the tunneling current. The result is that surface atoms with higher chemical affinity to the tip atom, i.e., usually the more reactive atoms, appear brighter in the STM image (Fig. 5). It is clear that this mechanism strongly depends on the atom at the tip apex, and we believe that the species at the tip responsible for the chemical o

126

Fig. 4. STM constant current topographs of the (111) surface of a Pt25Rh75 single crystal. (a) 20x8 nm 2, VT=-12 mV, IT= 4.6 nA; (b) 20• nm 2, -2 mV, 0.8 nA. Both images are slightly high-pass filtered to enhance the visibility of the atomic corrugation. The chemical contrast in image (b) is not due the different tunneling parameters but rather depends on the state of the tip.

contrast are typical impurities (adsorbates) on surfaces such as S or CO. This model is confirmed by the fact that the more reactive element always appears brighter in the case of occasional chemical contrast (Rh in PtRh, Ni in PtNi, Cr in PtCr).

127

Fig. 5. Proposed mechanism leading to chemical contrast by tip-sample interaction. It has to be mentioned that tip-surface interaction can happen on all surfaces and can lead to an increase, decrease or even complete reversal of the chemical contrast caused by one of the other mechanisms, but normally such cases can be determined by variation of experimental parameters. 3. S E G R E G A T I O N ON A L L O Y S - - S U R F A C E AND SUBSURFACE COMPOSITION

3.1. Segregation In the context of alloys, segregation is the enrichment of one element on the surface, where it reaches a higher concentration than in the bulk. As the theory of surface segregation is covered in more detail in other chapters of this book as well as a previous book devoted to the subject [41], here we just mention the basics. In thermodynamic equilibrium, the most simple description of segregation is the Langmuir-McLean equation, A Csurface B

Csurface

A Cbulk

= ~

e

-AH/kT

,

(1)

Cbulk

where the concentration of elements A and B are denoted by c A and c B, respectively, and A H is the excess free enthalpy of segregation, i.e., the change of free enthalpy associated with exchanging an A atom in the bulk with a B atom at the surface (excluding changes of configurational entropy, but including, e.g., terms due to vibrational entropy). In cases where the alloy constituents show neither a pronounced tendency towards chemical ordering nor towards demixing, and size effects [42,43] do not play a big role, A H is roughly equal to the difference of surface energies per atom of the two elements, ~/A- ~/B" Fig. 6 shows an example of such a case. As mentioned in the introduction, measurements of surface segregation on alloys are possible by several methods. As Table 1 shows, first-layer concen-

128

100

~ ' ~ x . ~ .~%:

90 -

Pt2sRh75(lll)

'<~'~~

i

LEIS

80 70 m, o

60 50

~

40 30

......... bulk ....i....................-.:..................... |

20 0

,

,

.

I

500

l

,

,

|

1000 temperature [~

,

,

.

,

1500

Fig. 6. Temperature-dependent first layer composition of the (111) surface of a Pt25Rh75 alloy measured with LEIS [ 16]. At high temperatures, where thermodynamic equilibrium is achieved, surface composition follows the Langmuir-McLean equation. Since this alloy comes close to an ideal solution (no significant tendency towards ordering or demixing) and the atomic sizes of Pt and Rh differ by only 3 %, the value of AH is close to the difference of surface energies.

trations derived independently with these methods agree within a few percent. In our experience, based on several different alloys, STM with chemical contrast and LEIS (with single crystal standards) have the best accuracy and agree within 3% or less. Quantitative LEED and MEIS (medium energy ion scattering) have the advantage of determining the composition of several layers, but need more elaborate analysis and have somewhat larger error bars.

3.2 Preferential sputtering and segregation in the altered layer Metal surfaces are usually prepared in surface science by applying sputtering and annealing cycles to get rid of most of the impurities which have a low bulk concentration but diffuse to the surface and reach much higher concentrations there due to segregation. For pure elements such a procedure is unproblematic and can be easily employed to prepare very clean surfaces. For multicomponent materials one has to take care about possible changes in the surface concentration by ion bombardment because the interaction of energetic ions with atoms of the target gives rise to compositional redistribution and selective ejection of different species. Considering only the ejection process itself, the probability of ejecting an atom depends on the strength of bonding to

129

Table 1. First-layer composition of the (111) surface of a Pt25Rh75 single crystal obtained by different methods. Method

Annealing

Measured at

Pt Csurface

Ref.

RT RT 900 ~ RT

71% 70% 66% 1) 69%

[39] [ 16] [ 16] [44]

temperature STM ("atom counting") Quantitative LEED LEIS MEIS

950 ~ 900 ~ 900 ~ 1000 ~

1) Csurface Pt increasesduring cooldown by approx. 5%, resulting in ~-71% Pt at room temperature.

the surrounding atoms and the efficiency of energy transfer in collisions, which is determined by its mass and the cross-section. Besides these effects, the athermal process of displacement mixing (leading to radiation-induced segregation) plays the dominant role near or below room temperature, while radiation enhanced diffusion and Gibbsian segregation are temperature dependent. As almost all ejected particles come from the first layer, radiation-induced and Gibbsian segregation enhance the probability that an ejected atom belongs to the segregating species. All these phenomena are summarised under the term preferential sputtering and result in depletion of the species sputtered preferentially. If a surface is sputtered at sufficiently low temperature to avoid bulk diffusion, atoms of the species preferentially sputtered can reach the surface by displacement mixing and radiation-induced segregation. This leads to a socalled altered layer, which has a composition different from that of the bulk and a thickness close to the penetration depth of the projectiles. Since surface diffusion has a much lower activation barrier than bulk diffusion, annealing a sputtered alloy surface first leads to a local equilibrium between the surface and the immediate subsurface layers, which still belong to the altered layer. Only after the onset of bulk diffusion is reached, usually around 60 - 70% of the melting temperature, the altered layer equilibrates with the bulk and true equilibrium segregation is observed [45]. For alloys of atoms with different size the existence and dissolution of an altered layer can be studied by STM because of the development of a misfit dislocation network between the altered layer and the bulk [46] (Fig. 7).

130

Fig. 7. Evolution of surface and subsurface composition of a sputtered PtNi alloy during annealing. Top: schematic view; bottom: STM images of a Pt25Ni75(111) sample (100x90 nm2 each) corresponding to these states. The misfit dislocations between the altered layer and the bulk are visible as a network of dark lines in the middle STM image (cf. Refs. 45,46).

4. C H E M I C A L

ORDERING OF ALLOY SURFACES

4.1. Bulk chemical order The properties of real alloys usually deviate from the simple model of an ideal solution of its components, where the occupation of lattice sites by the constituents of the alloy in equilibrium is purely random. Instead, real alloys show either a tendency towards forming ordered structures, where usually nearest neighbours of unlike species are preferred (the L11 structure being a notable exception) or towards demixing, where atoms of the same species tend to cluster. In a simple picture, two effects are responsible for this. The first is related to bond strengths. If a bond between alloy elements A and B is stronger than the average of an A-A and a B-B bond, this will favour ordering, whereas

131 a weaker A-B bond will favour demixing. The second effect is the size difference of the alloy constituents, which will cause elastic stress in disordered solid solutions and thereby either drive the elements towards the formation of ordered structures to reduce the stress or it will limit the mutual solubility and lead to demixing. We can distinguish between two types of chemical ordering: In the case of long-range order (LRO), in thermodynamic equilibrium the crystal structure consists of two or more superlattices, each of them of theoretically infinite extent and dominated by one species (anti-site defects occur at finite temperature). A few examples of such ordered phases are shown in Fig. 8. Above a transition temperature Ztrans, only short-range order (SRO) occurs. In this case the average surrounding of an atom still deviates from a random occupation of sites, whereas no long-range correlations between the atom types are present. In practice, diffusion is often too slow to achieve thermodynamic equilibrium within a reasonable time scale (especially in the bulk). Crystals then show only short-range order below the transition temperature. Therefore, in an experimentalist's view, the distinction between LRO and SRO is not so much a matter of equilibrium properties but one rather determines whether the ordering leads to fairly sharp superstructure peaks in a diffraction experiment and takes this as an indication of LRO.

4.2. F u n d a m e n t a l s of surface chemical order Whereas it has been recognised in the early days of surface science that the crystallographic structure of surfaces can deviate from that of the bulk (reconstructions), knowledge of surface chemical order, especially in cases of SRO at the surface, has accumulated only in the recent years. It became clear that there is a strong interplay between surface chemical order and segregation. In a somewhat simplified view, we can distinguish between two cases. In the first case, we can observe a competition between segregation and order, where ordering favours a fixed surface composition (often the bulk composition), whereas the difference of surface energies would favour segregation, i.e., a composition of the surface different from that of the bulk. In this case the stronger one of these two driving forces will determine the actual surface composition. As an example consider the (110) surface of a B2-type ordered bcc alloy (Fig. 8). Changing the surface composition to a value different from the bulk (50% each) will necessarily disrupt the ordering at the surface. In the other case, surface terminations with a composition different from that of the bulk are compatible with the bulk order, e.g., the pure-metal terminated { 100 } surfaces of the L12, L10 and B2 ordered phases. On these surfaces, ordering

132

type

(unit) cell

L12

{ 111 } (2x2)

0 0 0 0

9 9 9

{ 100 } pure O layer orc(2x2)--~

OOOO 000

Cu3Au

I

El0

(lx2)

o o o o

9 9 9 0 0 0 0

CuAu

CuPt

_•-•-•-~

~

pureOorO layer or (lx2)--~

9 O 9 O OOOO 0 0 0 0

pureOorO layer or c(2x2)~

0

9 0 9 9 0 90 0 9 0 9 0 0 0 0

000

Lll

0 0 0 0

0 0 0 0

0

9 9 9

(lx2) 0

0 0 0 0

0 90 0 0 0 0

000 (110)

B2

o 9o 9o

CsC1

OoOoO O O O

90

9 0

9

9

9

(100) pure layer" O O O 9 O 9 O O O or 9 9 9 0

0

0

9

9

9

Fig. 8. Common types of bulk chemical ordering in cubic metals and the corresponding low-index truncated bulk surfaces. The L10 and L11 ordered structures consist of (001) and (111) oriented planes of equal atoms, respectively, and, hence, do not preserve the cubic symmetry, leading to tetragonal (L 10) or rhomboedric (L 11) distortion. The cell shown for the L 11 structure is not a unit cell; the unit cell is twice as long in all three directions.

can actually enhance surface segregation by making a pure-metal termination more favourable than it would be without the tendency towards ordering. Whereas this simple picture often gives a reasonable qualitative description of the situation, a quantitative description must take into account that the ordering tendency may be substantially different between bulk and surface (see, e.g., Ref. 47). Since STM with atomic resolution and chemical contrast offers the unique possibility to determine the SRO of a surface, let us briefly recall how we quantify SRO [39]. One way to do this is the determination of correlations between the chemical species at position r and r + Ar on a surface. For a binary alloy, this can be done by determination of the autocorrelation A(Ar) of the chemical information, given as

133

ziN1 f ( r i )

A(Ar) =

N

f(r i + A r )

~i=1 g(ri)

,

(2)

g ( r i + Ar)

where r i are the atomic positions, N is the number of atoms,

f +~/CB/CA

if r i p o i n t s to a type A atom, if r i points to a type B atom,

(3)

and

g(r/) =

1

if r i p o i n t s to an a t o m p o s i t i o n ,

0

otherwise.

(4)

CA and c s are the concentrations of type A and B atoms, respectively. The denominator of Eq. (2) allows to employ this method for images of finite size and to exclude atoms of uncertain identity from the analysis by setting fir/) = g(ri) = 0. The autocorrelation A(Ar) is equal to zero for a random distribution of the two elements. It is positive if species separated by Ar tend to be of equal type and negative in case of a preference for different atoms separated by At. If Ar is the vector between nearest neighbours (NN), positive A(Ar) means demixing tendency, negative A(Ar) a tendency towards ordering. The maximum positive value that A(Ar) can approach is +1 (in case of complete phase separation where one never finds an unlike atom in a distance Ar), whereas the minimum value is given by Ami n = 1 -

1 . max(cA, Ca)

(5)

Ami n equals -1 only for concentrations of c A = c s = 0.5. Considering the auto-

correlation ANN between nearest neighbours, this value is reached when each atom is surrounded by neighbours of the other species only, e.g. for a square lattice with perfect c(2• long range order. The autocorrelation A(Ar) is related to the probability WAB(Ar) that a pair separated by Ar consists of unlike atoms via the relation WAB(Ar) = 2CACB(1 -- A(Ar))

.

(6)

134 Focusing our interest towards special sites available for adsorption, WAB(Ar) for a nearest-neighbour vector Ar directly yields the fraction of (two-fold) bridge-sites between unlike atoms. Unfortunately, A ( A r ) o r WAB(Ar)do not directly yield the number of three-fold or four-fold hollow sites neighbouring, e.g., only atoms of type A. Nevertheless, we will see that the value of ANNalso gives an indication whether such sites are more or less common than in a random alloy surface.

4.3. Chemical order of close-packed alloy surfaces The most simple case of surface chemical order is that of an alloy with strong long-range order, dominating over surface segregation. An example thereof is the (110) surface of NiA1, a B2-type material known for its high strength at elevated temperatures ("superalloy"). An STM image of this surface is shown in Fig. 9. The image shows perfect LRO at the surface, leading to a surface composition equal to the bulk composition, 50% of each species, in agreement with previous LEED [48] and MEIS [49] data. Obviously, ordering suppresses segregation of A1 to the surface in spite of a difference of surface energies of almost 0.2 eV/atom (~'Al(lll) = 0.531 eV/atom, ~'Ni(lll) -- 0.695 eV/atom [50]). Turning to fcc(111) surfaces, let us first compare the bulk properties of the alloys under consideration (Table 2). Stoichiometric PtFe, PtCo and PtNi form L10 phases in the bulk, characterised by an increase of the number of unlike nearest neighbours (8) with respect to an fcc random alloy with 50% Pt (6 unlike NN). PtCu forms the peculiar L11 ordered phase, where the number

Fig. 9. STM image (10x4 nm2, VT= -2.2 mV, IT = 0.5 nA) of the NiAI(110) surface annealed at 700~ Only one type of atom is visible as protrusion with a corrugation of 20 - 25 pm. As ab initio calculations show [51], this species is A1. The experimental corrugation coincides with the buckling found by LEED (A1 atoms are geometrically 22 pm higher [48]) and is therefore explained as topographic contrast.

135 of unlike N N is the same as in a random alloy (6), indicating no clear preference for A-B bonds versus bonds between equal elements (Fig. 8). Data in the literature for PtRh agree that any ordering or demixing must be weak. As a measure of the strength of the ordering tendency, we can take the transition t e m p e r a t u r e s Ttrans, or better, the ratio of Ttrans and the melting temperature T m. This shows us that the ordering tendency decreases in the sequence PtFe, PtNi and PtCo. Fig. 10 displays S T M images of several PtMe(111) alloy surfaces with surface compositions around 50% Pt. A quantification of SRO in terms of correlations is given in Table 2. The PtFe sample was not a bulk alloy but rather a thin (--- 8 m o n o l a y e r s ) alloy film created by annealing a thin Fe film on Pt(111). The Pt25Ni75 and Pt25Co75 crystals can form L12 bulk ordered phases, but we did not find any indications thereof, presumably because the formation of these phases would require a very slow cooldown.

Table 2. Chemical ordering of various PtMe alloys. Bulk data from Refs. [52,53]. Autocorrelation data A of chemical order given for nearest (NN), second- and third nearest (2NN and 3NN, resp.) neighbours are averaged over the three equivalent directions in the STM images shown in Fig. 10. PtRh(111) data are based on several STM images [39]. Except for PtFe, correlations of neighbours beyond 3NN are near or below the statistical noise (---0.01). Correlation values for two types of perfect LRO are given for comparison. (111) surface investigated

bulk ordering at c A = cB = 0.5 type PtFe PtCo PtNi PtCu PtRh

L10 L10 L10 L11 none

Ztran s

Zm

K

K

1573 1098 910 1085 -

1805 1760 1735 1785 2190

Ztrans/T m

bulk Pt %

surface Pt ANN %

A2NN A3NN

0.87 0.62 0.52 0.61 -

=501) 25 253) =354) 25

502) 52 48 43 69

-0.24 -0.18 -0.12 -0.11 -0.05

-0.11 +0.05 -0.03 +0.02 =0

+0.43 +0.02 +0.08 +0.04 =0

50 25 or 75

-0.33 -0.33

-0.33 -0.33

+1.00 +1.00

perfect (lx2) (L1 o, mixed termination of L11) perfect (2x2) (L12)

1) Surface alloy with --8 layers, average composition of subsurface layers estimated by AES. 2) Data refer to imperfectly ordered surface created by annealing at 530 ~ image (b) in Fig. 10. Data for the sample annealed at 470 ~ image (a), are extremely close to perfect (1• 3) Preferentially sputtered surface annealed at 500~ surface in local equilibrium with an altered layer containing 35-40% Pt [38,46]. 4) Sample was grown as Pt25Cu75, bulk composition given reflects estimated Pt enrichment due to many sputtering/annealing cycles and possible Cu evaporation at high temperatures.

136 The PtFe surface shows a pronounced (1• ordering, which can be more or less perfect depending on details of the preparation, as long as the annealing temperatures are between 350 and 640 ~ (at higher temperatures, Fe diffuses into deeper layers of the Pt single crystal and lower Fe concentrations are found). Auger electron spectroscopy (AES) measurements of the average composition of the first layers show Pt concentrations in the subsurface layers close to 50% in this temperature range. We therefore conclude that the surface and

Fig. 10. STM images of fcc(111) alloy surfaces. The bright atoms are Pt in PtCo and PtCu, but Fe in PtFe and Ni in PtNi. Compositions given are bulk compositions, except for PtFe, which is a surface alloy with --8 PtFe layers obtained by annealing a 4 ML Fe film on Pt(111). The ordering of the PtFe alloy depends on the preparation; frame (a) corresponds to an annealing temperature of 470 ~ frame (b) to 530 ~

137

subsurface compositions are approximately identical, and, hence, segregation is negligible. As Pt segregates to the otherwise similar (111) surfaces of PtCo and PtNi, the absence of Pt enrichment on the PtFe(111) surface can be only explained as a suppression of segregation by strong chemical ordering in the L10 structure, similar to the case of A1Ni(110) discussed above. This view is supported by both the high transition temperature of the order-disorder transformation (87% of the melting temperature Tm) and the large domain of almost perfect (lx2) ordering seen in the STM image obtained after annealing at 470 ~ (Fig. 10). The PtCo, PtNi and PtCu surfaces show SRO with a clear tendency towards the formation of alternate chains of the two species, i.e., local (lx2) ordering. This can be either explained as a tendency towards formation of a truncated L10 or L11 surface (Fig. 8) or simply as a tendency to increase the number of unlike nearest neighbours in the surface. The PtNi surface studied has an average composition in the subsurface layers close to 40% Pt [38,46], still in the range of the L10 structure in the phase diagram. We therefore consider it likely that the ordering observed is a SRO version of the L10 phase. This view is supported by the fact that the correlations of second- and third nearest neighbours have the same sign as for a perfect truncated L10 surface, while their magnitude strongly decreases as one moves from nearest to further neighbours (Table 2). The subsurface layers of Pt25Co75 have a composition quite different from the 50% Pt of a truncated L10 surface. Due to an oscillating segregation profile, the second layer contains only =5% Pt [54]. This shows us that the surface short-range order cannot be simply understood as truncated L10 phase and should be rather seen as a result of the tendency to form unlike nearest neighbours. This view is supported by the fact that already the second-nearest neighbour correlation A2NN differs in sign from the value expected for an L10 phase. Among the alloys of Pt with 3d metals under consideration, the PtCu surface shows the weakest correlations between nearest neighbours. Unfortunately, we have no reliable data for the compositions of the subsurface layers to decide about a possible L11 phase in the upper layers. Given the second and third nearest neighbour correlations and the fact that the bulk L11 phase does not show an increase of the number of unlike nearest neighbours, we consider it likely that the ordering observed is a short-range version of a (lx2), rather than governed by the preference for unlike nearest neighbours. As segregation seems to be very weak in PtCu(111), an L11 phase oriented in such a way that it becomes terminated by a pure Pt or Cu layer will be unfavourable. The only remaining possibility for a bulk-terminated L11 { 111 } surface is (lx2) chemi-

138 cal order (Fig. 8). Comparing the PtCo, PtNi and PtCu surfaces, we therefore conclude that the visually similar surface short-range order may have somewhat different physical background when relating it to the corresponding bulk ordered phases. One reason for the similar appearance of the PtCo, PtNi and PtCu(111) surfaces in Fig. 10 is the comparable degree of SRO, i.e., the existence of only very small domains. For PtNi, it has been shown by simulations that the ordering does not depend significantly on temperature and is, hence, not limited by entropy [55]. It has been rather suggested that the domain size is limited by stress caused by the tetragonal distortion of the L10 structure which does not perfectly fit the underlying fcc lattice of the disordered bulk [38,55]. Regardless whether it is based on a bulk L10 structure or not, the same can be true also for any other (1• ordered phase of an fcc(111) surface, which will usually prefer different lattice constants parallel and perpendicular to the rows. Creating small domains will allow some relaxation of the resulting in-plane stress. Coming to the last alloy of Table 2, the PtRh surface under investigation (Fig. 4) shows only weak short range order, in agreement with the absence of an ordered phase in the bulk. Considering the values of the nearest neighbour correlations of all the surfaces in Table 2, we find a good correlation with the reduced transition temperatures of the bulk ordered phases Ttrans/Tm, with the special case of PtCu falling somewhere out of this trend. As all of the alloys under consideration (except PtRh) also form L12 ordered PtMe 3 phases in the bulk, and some of the bulk compositions investigated are identical or close to PtMe 3, we should also compare the surface ordering observed with the L12 phase. A truncated L12 (111) surface always has the same composition as the bulk and (2x2) ordering (Fig. 8). In other words, preserving the L12 order at a (111) surface is incompatible with segregation. One can therefore expect a truncated L12 (111) surface only in cases where ordering is stronger than segregation, and this is obviously not the case for the Pt25Me75 surfaces studied here. With the weak segregation of PtCu(111) surfaces, the moderate ordering tendency ( T t r a n s / T m - 0 . 6 5 ) of Cu3Pt(111) may be sufficient to obtain such a long-range ordered Cu3Pt(111) surface [56], even though disorder has been also reported for this surface [57]. Interestingly, the correlations A for the (2x2) structure (truncated L12) are identical to those of the (lx2), but one has to keep in mind that the surface concentrations are different (Table 2). Let us finally present an example where segregation is much stronger than ordering. After annealing at 820 K, the (111) surface of a Ag33Pd67 was found to have a composition of Ag95Pd 5 (Fig. 11), quite far from the bulk concentra-

139

Fig. 11. STM image of the (111) surface of a Ag33Pd67 alloy (10• 10 nm2). The difference of apparent height between the Pd atoms (higher; bright) and the Ag atoms is approx. 40 pm [31 ]. Only about 5% of all surface atoms are Pd.

tion [31]. Using the Langmuir-McLean equation (1), this results in a AH value of approx. 0.26 eV/atom, in excellent agreement with the difference of surface energies (0.27 eV/atom [50]). Analysing STM images of this surface shows a tendency towards mixing (ordering); the number of pairs of adjacent Pd atoms is reduced to roughly half the value expected for a random alloy. In the bulk, due to experimental reasons (similar atomic structure factors for X-ray diffraction) it is unclear whether AgPd alloys form ordered phases, and if so, which structures [53]. The enthalpy of formation of AgPd solid solutions (-0.05 eV/atom at 50% concentration, Ref. 53) points to a tendency towards ordering somewhere between PtRh and PtNi. This is in agreement with the tendency towards mixing observed by STM. It is obvious that this comparatively weak tendency towards ordering does not significantly affect segregation. This explains why segregation is dominated by the difference of surface energies.

140 4.4. fee(100) surfaces In contrast to the (111) faces of most ordered fcc structures, the (100) faces allow a termination with a composition different from the bulk composition for both the L10 and L12 structures (Fig. 8). In other words, segregation and chemical order are compatible with each other. Examples for L12 ordered alloys are the prototype Cu3Au, with LRO in the bulk below 663 K (Ttrans/T m = 0.54, Ref. 58), as well as PtCo 3 (Ttrans/Tm- 0.55; Fig. 3, Ref. 37). In both alloys, the 5d element (Au, Pt) segregates to the first layer. The Cu3Au(100) surface adopts the c(2• ordered termination of the L12 structure (Fig. 8), implying a first-layer composition close to CusoAus0, and a second layer of essentially pure Cu. Also the deeper layers follow the bulk L12 order. Even for temperatures above the transition temperature, where the LRO in the bulk is lost, the compositions of the first two layers change only gradually as the temperature is increased [11,58]. The (100) surface of Pt25Co75 (PtCo3) is very similar, with c(2• ordering at the surface (Fig. 3), a second layer of almost pure Co and a continuation of the L12 ordered structure also in the deeper layers [37]. Turning to other ordered phases of the Cu-Au system, we again find that the (100) surface terminations are compatible with both Au segregation and ordering. The L10 and L12 structures of CuAu and Au3Cu, respectively, allow a pure Au termination, and thus segregation of Au (cf. Fig. 8). Thus, in spite of the moderate to weak ordering tendencies (Ttrans/Tm = 0.58 and 0.40, resp.), the compositions of the first and second monolayers measured on these alloys are indeed very close to the bulklike Au-terminated ordered structures [8-10]. Similar to Au3Cu, Au3Pd(100) has L12 structure in the bulk, but a significantly stronger tendency towards ordering (Ttrans/Tm=0.71). Whereas the surface layer of this alloy is pure Au [32], in agreement with both ordering and Au segregation, the composition of the second layer (close to the bulk value) and the absence of any superstructure spots in LEED reported in Ref. 17 are incompatible with a truncated L12 bulk structure. The reason for these unexpected results is unclear so far. Surfaces terminated by (almost) pure monolayers of one material are not restricted to systems with rather strong tendency towards ordering, however. Since the surface energy per atom is higher on fcc(100) than on the closepacked (111) surfaces, also differences of surface energies can be larger, explaining why some (100) faces show much stronger segregation than the (111) faces of the corresponding alloys. Taking the weakly ordering AgPd system mentioned previously as an example, the (100) surface layer of a Ag33Pd67 has a negligible Pd concentration (10 -3 or lower as determined by STM, Ref. 31).

141

This is in agreement with a difference of surface energies of 0.5 eV [50], yielding a Pd concentration in the 10 -4 range. On both the Au3Pd(100) and Ag33Pd67(100) surfaces sputtering at elevated temperatures leads to depletion of the segregating element (Au or Ag) in a near-surface region, from where this element can reach the surface by diffusion. This treatment can lead to a significant Pd enrichment in the subsurface layers (reaching values close to 100%), also increasing the Pd surface composition [31,32].

4.5. Site-specific segregation Up to now we have considered unreconstructed, defect-free low-index surfaces, where all surface atoms have the same geometric environment. In the real world, large defect-free terraces of low-index surfaces are the exception rather than the rule, and in nanometer-sized metal particles (clusters) such as those found in industrial catalysts a significant fraction of all surface atoms sit at steps, edges or corners and therefore have lower coordination than those in the terraces. There are many indications that such sites are more reactive than terraces [59]. On alloys, a key question in this respect is the composition of such sites, i.e., whether segregation differs between such special sites and flat terraces. Whereas well-defined STM studies of small clusters are extremely difficult, we could obtain chemical contrast at steps in a few cases. Fig. 12 shows two PtRh alloy surfaces with steps. Whereas the step of the (100) surface consists of almost exclusively Pt [60], the step on the (111) surface has almost the same composition as the remaining surface [39]. As step atoms on both, the (100) and the (111) surface, have 7 nearest neighbours, the difference in segregation to the two steps is unexpected and has not found any explanation so far. A more easily explicable case of site-specific segregation has been observed on a reconstructed Pt40Ni60(100) surface [40]. Similar to the Pt(100) surface, the alloy surface reconstructs with a close-packed (pseudohexagonal or 'hex') first monolayer. As the hex layer does not fit on the square bulk lattice below, some rows of atoms are in on-top sites, and, hence, geometrically highest, others in fourfold hollow sites (lowest). The other rows of surface atoms fall in between. The coordination numbers of the surface atoms range between 7 (on-top) and 10 (hollow sites of the bulk lattice). In the "normal" STM images of this surface, Pt and Ni atoms are indistinguishable and the corrugation is purely geometric (Fig. 13a). Chemical contrast, related to tip-surface interaction, leads to a superposition of the topography and the chemical information, where the Ni atoms appear brighter (higher) than the Pt atoms (Fig. 13b). As the geometric height does not vary significantly along the rows of atoms, it

142

Fig. 12. STM images of the PtsoRhso(100 ) [60] and PtzsRh75(100) [39] alloy surfaces with steps. Arrows point at Rh atoms (bright) at the step edge.

is easy to disentangle the geometric and chemical information. The result is a high Ni concentration (up to 100%) in the on-top sites, whereas the hollow-site rows consist of pure Pt. These results are in agreement with computer simulations based on embedded atom method potentials optimised for the PtNi system [40]. The reason for this behaviour is the different dependence of the binding energy on coordination between the 3d and the 5d transition metals. For the 5d metal Pt a low coordination of 7 is very unfavourable, while a small reduction of the coordination number from 12 (bulk) to 10 does not cost much energy. Ni, being a 3d metal, has a more linear relationship between loss of coordination and energy. These trends are also responsible for the fact that only the 5d metals Ir, Pt and Au show missing-row reconstructions on their (110) surfaces [61] as well as for the face-dependent segregation of PtMe (Me = Ni, Co, Fe) alloys which show Pt segregation to the (111) and (100) surfaces (coordination numbers 8 and 9, respectively), but Pt depletion in the first monolayer of the unreconstructed (110) surface (coordination number 7).

143

Fig. 13. STM images of the reconstructed Pt40Ni60 (100) surface showing (a) the geometric structure and (b) a superposition of geometry and chemical contrast [40]. Image (b) shows that the rows of atoms in hollow sites (some of them marked "hollow") consist of 100% Pt, while a large number of Ni atoms (brightest atoms of the rows) is found in the rows of on-top atoms. Line scans along two rows marked "A" and "B" in the lower image are shown on the right-hand side.

144 5. I M P L I C A T I O N S F O R A D S O R P T I O N ON A L L O Y S In this section we are going to present three examples demonstrating the main factors which determine the strength of adsorption on alloy surfaces. We will demonstrate that knowledge of the surface composition and chemical ordering is essential for understanding adsorption. In some cases it is necessary to study both the alloy with chemical contrast and the adsorbate on the atomic scale.

5.1. Chemical Affinity In the most simple view, bonding of adsorbates on alloy surfaces is simply governed by the chemical affinity between the adsorbate and the alloy constituents. If the mobility of the adsorbate on the surface is high enough, it usually selects a binding site on the more reactive (less noble) element. An example for this behaviour is adsorbed oxygen on a Ag33Pd67(111) surface shown in Fig. 14 [62]. The STM shows the oxygen atoms as dark spots, while Pd atoms appear bright as in Fig. 11. The remaining part of the surface consists of pure Ag, not resolved at the tunneling conditions used for Fig. 14. We find that the O atoms sometimes change their position on the surface, always hopping from one Pd atom to another. Thus, the oxygen adatoms are bound to Pd atoms most of the time; diffusion over the Ag surface is too fast for observation by STM at room temperature. Trapping of the O atoms by Pd in the surface is easily explained by Pd being the less noble (more reactive) metal.

Fig. 14. Two subsequent STM images (30x30 nm2, VT=-I.2 V, IT= 1.3 nA) of the same area of the Ag33Pd67(111) surface with adsorbed oxygen (dark). The Pd atoms betray their identity by appearing bright only when not binding to an O atom. [62]

145

5.2. The ensemble effect Many adsorbate atoms are bound in hollow sites between three or four surface atoms (this is probably also true for the O atoms in the example above, but we could not verify it). In such a case, the preferred sites will usually be these where all three or four neighbours of the adsorbate belong to the more reactive alloy constituent. This is a simple case of an ensemble effect, where a certain ensemble of surface atoms forms the sites with strongest bonding. The availability of such sites strongly depends on the chemical ordering of the alloy surface. As an example, consider the PlhsRh75 (100) crystal shown in Fig. 15. As mentioned previously, PtRh alloys are close to the division line between mixing (ordering) and demixing. Depending on the annealing temperature, we can indeed prepare (100) surfaces which show either short-range ordering or demixing (Fig. 15, Table 3). Although the reasons for the different ordering behaviour could not be unambiguously determined [39], we can study the consequences thereof. The surface in Fig. 15(a) has a higher concentration of Rh than that in Fig. 15(b). Without any further knowledge of the details of chemical ordering on this surface, one would expect that the number of fourfold hollow sites between four Rh atoms is equal to the fourth power of the Rh concentration, and, hence, higher for the surface with the higher Rh concentration (column "random alloy" in Table 3). Counting the hollow sites surrounded by four Rh atoms in Fig. 15 shows that the reverse is true. The surface with the higher Rh concentration (Fig. 15a) shows a preference of unlike nearest neighbouts, with local c(2• patterns of Pt and Rh atoms. This short-range order leads to a low number of hollow sites surrounded by four Rh atoms, only 0.4% of all fourfold hollow sites. This is only approximately one sixth of the number of such sites which would be present in a random alloy with the same Rh concentration. On the other hand, the surface with the lower Rh concentration shows clustering of Rh atoms, leading to a higher number (1.3%) of fourfold hollow sites surrounded by four Rh atoms than in a random arrangement of Pt and Rh atoms with the same Rh concentration. Thus, even the comparatively weak ordering of these PtRh surfaces can strongly influence the number of certain sites, such as the fourfold hollow sites where an adsorbate can bind to four atoms of the more reactive metal Rh. A somewhat similar case of an ensemble effect was proposed for CO adsorption on a Pt25Ni75(111) alloy surface [63]. Whereas CO adsorbs on-top of Pt atoms [64], on Ni(111) it adsorbs in the threefold hollow sites [65]. The HREELS (high-resolution electron energy loss spectroscopy) peak associated with CO adsorbed on threefold Ni sites almost vanishes if the sample prepara-

146

Fig. 15. STM images of the Pt25Rh75(100) surface after annealing to (a) 600 and (b) 900~ (20• nm 2, VT = -3 mV, I T = 2.2 nA and 4.9 nA, respectively). Rh atoms appear bright [39].

147

Table 3 Surface of a Pt25Rh75(100) single crystal after two different preparation procedures (Fig. 15, Ref. 39). Note that even a small change of surface chemical order can lead to a significant change in the number of the most reactive sites. Annealing temperature

Rh surface concentration

Correlation ANN

600 ~ 900 ~

40 % 26 %

-0.17 +0.15

hollow sites with 4 Rh neighbours random alloy

measured

2.6% 0.5%

0.4% 1.3%

tion results in a surface composition close to 50% Pt. Under these conditions, we have observed the short-range order shown in Fig. 10, where most of the surface shows local (lx2) order. Hence, the number of threefold hollow sites surrounded by Ni atoms only is very low, in agreement to the HREELS result.

5.3. The ligand effect Let us finally present an example of a ligand effect. Recently, adsorption of carbon monoxide on a PtCo surface alloy has been studied by STM [66]. In this work, it was demonstrated for the first time that one can study an alloy surface with chemical contrast and image the positions of the adsorbates in the same surface area, allowing to determine the chemical nature of the binding site of each individual adsorbate molecule. It was found that the CO molecules only adsorb on top of Pt atoms, with a clear preference for Pt atoms having Co neighbours in the first layer (Fig. 16). Even at saturation coverage, no CO was found on Pt atoms surrounded by Pt atoms only, indicating that the binding energy of CO is too low for stable adsorption on these atoms. The binding energy of CO thus depends on the neighbours ("ligands") of the Pt atoms it binds to. Using ab initio calculations, it could be shown that this ligand effect is due to a shift of the d-band energy by the neighbouring atoms. This d-band shift is due to the compression of the surface layer by the higher concentration of large Pt atoms in the surface than in the underlying layers, making the alloy surface as a whole less reactive than the close-packed surfaces of the pure constituents, Pt and Co. The Pt atoms surrounded by Pt only have the least space in the surface, while those with many small Co neighbours have more space and, hence, a narrower and, therefore, higher d-band. The higher d-band energy correlates with a stronger bonding to CO [67]. It should be finally noted that any attempt to determine the surface composition of an alloy by titration with CO must fail in a case like this, where CO molecules adsorb only on some of the Pt atoms in the surface.

148

Fig. 16. STM images of the same area of a PtCo(111) surface alloy showing (a) chemical contrast (Pt atoms appear bright) and (b) the adsorbed CO molecules at saturation coverage. Frame (c) shows a schematic view of the CO positions (circles) on the alloy atoms (small black and grey squares for Co and Pt, respectively) in the lower right quadrant of frames (a) and (b). Frame (d) shows the probability of finding a CO molecule on a given Pt atom as a function of the number of its Co neighbours in the first layer for low CO coverage (0.1 L) and saturation coverage [66].

6. C O N C L U S I O N S Scanning tunneling microscopy with atomic resolution and chemical contrast offers unique possibilities in studying segregation and chemical ordering of alloy surfaces. Chemical contrast in STM can have three different reasons, (a) true topographic effects, (b) different density of states of the alloy constituents, and (c) tip-sample interaction depending on the chemical identity of the atom imaged. The composition and chemical order on surfaces is determined by an in-

149 terplay of ordering and segregation. If the chemical ordering is weak, segregation is mainly determined by the differences in surface energies. On the surfaces of alloys with a strong tendency towards ordering, segregation and ordering can either compete, e.g., in cases where preserving the bulk chemical order requires the surface to assume the bulk composition, or ordering can enhance segregation, e.g. in cases where bulk ordering allows a pure-metal termination. In cases where only short-range chemical ordering occurs at the surface, the trends in surface chemical ordering were found to correspond well to the ordering tendency observed in the bulk. We have also shown that the study of surface composition and chemical order is essential for understanding adsorption on alloy surface. Even weak ordering can lead to significant changes in the availability of some adsorption sites. We could also obtain STM images of an alloy surface with chemical contrast and images of adsorbates in the same surface area, revealing the chemical structure of adsorption sites. We could thereby demonstrate the ligand effect, i.e., the dependence of adsorption strength on the atoms neighbouring an adsorption site.

A CKN O W L E D G E M E N T S The authors would like to thank W. Hofer and R. Koller for carefully reading and correcting the manuscript. This work was supported by the Austrian

Fonds zur FOrderung der wissenschafilichen Forschung. REFERENCES

[]]

T.S. King, in P.A. Dowben, A. Miller (eds.), Surface Segregation Phenomena, CRC Press, Boca Raton, 1990, pp. 27-77. [2] S. Hofmann, in P.A. Dowben, A. Miller (eds.), Surface Segregation Phenomena, CRC Press, Boca Raton, 1990, pp. 107-134. [3] M. Polak, L. Rubinovich, Surf. Sci. Rep. 38 (2000) 127. [4] M.A. Vasiliev, J. Phys. D, Appl. Phys. 30 (1997) 3037. [5] P. Hohenberg, W. Kohn, Phys. Rev. B 136, 864 (1964); W. Kohn, L.J. Sham, Phys. Rev. A 140 (1965) 1133. [6] H. Niehus, R. Spitzl, Surf. Interf. Anal. 17 (1991) 287. [7] P. Weigand, B. Jelinek, W. Hofer and P. Varga, Surf. Sci. 295 (1993) 57. [8] R. Beikler, E. Taglauer, Nucl. Instrum. Meth. Phys. Res. B 161-163 (2000) 390. [9] S. Sch6mann, E. Taglauer, Surf. Rev. Lett. 3 (1996) 1823. [101 E. Taglauer, A. Steltenpohl, R. Beikler, L. Houssiau, Nucl. Instrum. Meth. Phys. Res. B 157 (1999) 270. [111 T. M. Buck, G. H. Wheatley, L. Marchut, Phys. Rev. Lett. 51 (1983) 43. [121 Y. Gauthier, Surf. Rev. Lett. 3 (1996) 1663 and refs. therein. [131 K. Heinz, R. Doll, M. Kottcke, Surf. Rev. Lett. 3 (1996) 1651 and refs. therein.

150 [141 V. Blum, L. Hammer, W. Meier, K. Heinz, Surf. Sci. 488 (2001) 219. [151 E. Platzgummer, M. Sporn, R. Koller, M. Schmid, W. Hofer, P. Varga, Surf. Sci. 423 (1999) 134. [161 E. Platzgummer, M. Sporn, R. Koller, S. Forsthuber, M. Schmid, W. Hofer, P. Varga, Surf. Sci. 419 (1999) 236. [17] S. Speller, M. Aschoff, J. Kuntze, W. Heiland, A. Atrei, U. Bardi, E. Platzgummer, Surf. Rev. Lett. 6 (1999) 829. [18] E. Platzgummer, M. Sporn, R. Koller, M. Schmid, W. Hofer, P. Varga, Surf. Sci. 453 (2000) 214. [19] V. Blum, L. Hammer, W. Meier, K. Heinz, M. Schmid, E. Lundgren, P. Varga, Surf. Sci. 474 (2001) 81. [20] W. M. H. Sachtler, Le Vide 28 (1973) 67. [21] J.A. Rodriguez, Surf. Sci. Rep. 24 (1996) 223. [22] P. Varga and M. Schmid, Appl. Surf. Sci. 141 (1999) 287. [23] http://www.iap.tuwien, ac. at/www/surface/STM_Gallery/ [24] J. Tersoff, T.R. Hamann, Phys. Rev. Lett. 50 (1983) 1998; Phys. Rev. B 31 (1985) 805. [25] W.A. Hofer, J. Redinger, A. Biedermann, P. Varga, Surf. Sci. 466 (2000) L795. [261 M. Sporn, E. Platzgummer, M. Pinczolits, W. Hebenstreit, M. Schmid, W. Hofer, P. Varga, Surf. Sci. 396 (1998) 78. [27] R.M. Tromp, J. Phys. Cond. Matter 1 (1989) 10211. [28] C. Nagl, O. Haller, E. Platzgummer, M. Schmid, P. Varga, Surf. Sci. 321 (1994) 237. [29] Y. Gauthier, W. Moritz, W. H6sler, Surf. Sci. 345 (1996) 53. [30] D.D. Chambliss, S. Chiang, Surf. Sci. 264 (1992) L187. [311 P.T. Wouda, M. Schmid, B.E. Nieuwenhuys, P. Varga, Surf. Sci. 417 (1998) 292. [32] M. Aschoff, S. Speller, J. Kuntze, W. Heiland, E. Platzgummer, M. Schmid, P. Varga, B. Baretzky, Surf. Sci. 415 (1998) L1015. [33] E. Wimmer, H. Krakauer, M. Weinert, A.J. Freeman, Phys. Rev. B 24 (1981) 864.(2001) 97. [34] W. A. Hofer, G. Ritz, W. Hebenstreit, M. Schmid, P. Varga, J. Redinger, R. Podloucky, Surf. Sci. 405 (1998) L514. [35] C.J. Chen, Phys. Rev. Lett. 65 (1990) 448. [36] P. Varga, E. Lundgren, J. Redinger, M. Schmid, Phys. Stat. Sol. (a) 187. [37] Y. Gauthier, P. Dolle, R. Baudoing-Savois, W. Hebenstreit, E. Platzgummer, M. Schmid, P. Varga, Surf. Sci. 396 (1998) 137. [38] M. Schmid, H. Stadler, P. Varga, Phys. Rev. Lett. 70 (1993) 1441. [391 E.L.D. Hebenstreit, W. Hebenstreit, M. Schmid, P. Varga, Surf. Sci. 441 (1999) 441. [401 W. Hebenstreit, G. Ritz, M. Schmid, A. Biedermann and P. Varga, Surf. Sci. 388 (1997) 150. [41] P.A. Dowben, A. Miller (eds.), Surface Segregation Phenomena, CRC Press, Boca Raton, 1990. [421 F.F. Abraham, N.H. Tsai, G.M. Pound, Surf. Sci. 83 (1979) 406. [431 M. Schmid, W. Hofer, P. Varga, P. Stoltze, K. W. Jacobsen, J. K. NCrskov, Phys. Rev. B 51 (1995) 10937. [44] D. Brown, P.D. Quinn, D.P. Woodruff, T.C.Q. Noakes and P. Bailey, Surf. Sci., article 12576 (2001 ), in press. [451 P. Weigand, W. Hofer and P. Varga, Surf. Sci. 287/288 (1993) 350.

151 [46] M. Schmid, A. Biedermann, H. Stadler and P. Varga, Phys. Rev. Lett. 69 (1992) 925; M. Schmid, A. Biedermann, C. Slama, H. Stadler, P. Weigand and P. Varga, Nucl. Instrum. Meth. Phys. Res. B 82 (1993) 259. [47] I. Meunier, G. Tr6glia, B. Legrand, Surf. Sci. 441 (1999) 225. [48] H.L Davis, J.R. Noonan, Phys. Rev. Lett. 54 (1985) 566. [49] S.M. Yalisove, W.R. Graham, Surf. Sci. 183 (1987) 556. [50] L. Vitos, A.V. Ruban, H.L. Skriver, J. Koll~, Surf. Sci. 411 (1998) 186. [51] K.H. Hansen, J. Gottschalck, L. Petersen, B. Hammer, E. L~egsgaard, F. Besenbacher, I. Stensgaard, Phys. Rev. B 63 (2001) 115421. [52] P. Eckerlin, H. Kandel, in: K.-H. Hellwege and A.M. Hellwege (eds.), LandoltB6rnstein - Numerical Data and Functional Relationships in Science and Technology, New Series, vol. III/6B, Springer, Berlin, 1971. [53] B. Predel, in: O. Madelung (ed.), Landolt-B6rnstein - Numerical Data and Functional Relationships in Science and Technology, New Series, vols. IV/5a-j, Springer, Berlin, 1991-1998.

[54] Y. Gauthier, R. Baudoing-Savois, J.M. Bugnard, W. Hebenstreit, M. Schmid, P. Varga, Surf. Sci. 466 (2000) 155. [55] H. Stadler, W. Hofer, M. Schmid and P. Varga, Phys. Rev. B 48 (1993) 11352. [56] G.R. Castro, U. Schneider, H. Busse, T. Janssens, K. Wandelt, Surf. Sci. 269/270 (1992) 321. [57] Y.G. Shen, D.J. O'Connor, K. Wandelt, R.J. MacDonald, Surf. Sci. 328 (1995) 21. [58] H. Reichert, P.J. Eng, H. Dosch, I.K. Robinson, Phys. Rev. Lett. 74 (1995) 2006. [59] S. Dahl, A. Logadottir, R.C. Egeberg, J.H. Larsen, I. Chorkendorff, E. T6rnqvist, J.K. NCrskov, Phys. Rev. Lett. 83 (1999) 1814 and Refs. therein. [60] P. T. Wouda, B. E. Nieuwenhuys, M. Schmid, and P. Varga, Surf. Sci. 359 (1996) 17. [61] J.K. NCrskov, K.W. Jacobsen, P. Stoltze, L.B. Hansen, Surf. Sci. 283 (1993) 277. [62] P.T. Wouda, M. Schmid, B.E. Nieuwenhuys, P. Varga, Surf. Sci. 423 (1999) L229. [63] A. Pantf6rder, J. Skonieczny, E. Janssen, G. Meister, A. Goldmann, P. Varga, Surf. Sci. 331-333 (1995) 824. [64] H. Hopster, H. Ibach, Surf.Sci. 77 (1978) 109. [65] L. Becker, S. Aminpirooz, B. Hillert, M. Pedio, J. Haase, D. L. Adams, Phys. Rev. B 47 (1993) 9710; R. Davis, D.P. Woodruff, P. Hofmann, O. Schaff, V. Fernandez, K.M. Schindler, V. Fritzsche, A.M. Bradshaw, J. Phys. Cond. Matter 8 (1996) 1367. [66] Y. Gauthier, M. Schmid, S. Padovani, E. Lundgren, V. Bu~, G. Kresse, J. Redinger, P. Varga, Phys. Rev. Lett. 87 (2001) 036103. [67] B. Hammer, Y. Morikawa, J. K. NCrskov, Phys. Rev. Lett. 76 (1996) 2141.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces 152

D.P. Woodruff, (Editor)

Chapter 5

Surface alloying and de-alloying of Pb on single-crystal Cu surfaces Gary L. Kellogg Sandia National Laboratories, Albuquerque, NM 87185-1415

1. I N T R O D U C T I O N Metal-on-metal surface alloy formation is a scientifically interesting phenomenon with important technological applications. This is particularly true of those materials combinations in which alloying occurs at the first atomic layer but not in the bulk [1]. Confinement of intermixing to the topmost layer can lead to changes in surface atomic structures that progress from ordered surface alloys to ordered surface overlayers with increasing coverage. Understanding the mechanisms involved in surface alloying and de-alloying thus provides a significant challenge for experimental and theoretical surface scientists. A well-studied example of this type of system is Pb adsorbed on single crystal surfaces of Cu. It is surprising that intermixing occurs at all in this system based on the immiscibility of Pb in bulk Cu and the large lattice mismatch between Pb and Cu (bulk lattice constant of Cu = 0.36 nm, Pb = 0.45 nm). Yet, for all three low-index single-crystal surfaces, i. e., (111), (100), and (110), there is clear evidence that there are coverage regimes where the surface alloy is the thermodynamically stable atomic structure. The formation of a surface alloy in combination with the inability of the bulk to absorb Pb atoms leads to interesting structural transitions on these surfaces. Because the atomistic processes involved in these transitions are accessible to investigation with the modem tools of surface science, from a pure science standpoint, Pb on Cu serves as an excellent model system to investigate the properties of twodimensional systems. These properties include nucleation and growth, twodimensional phase transitions, surface melting, and even pattern formation resulting from competing short-range and long-range interactions. Section 3 provides an overview of these investigations.

9

153

Through much of the history of ultrahigh vacuum (UHV) surface science, scientific curiosity was the primary motivation for investigations of metal-onmetal systems such as Pb on Cu. Semiconductor surfaces and gas-metal interactions received much more attention due to their relevance to microelectronics and catalysis. In recent years, however, UHV surface science studies of metal-on-metal systems have become more interesting from a technological standpoint. For the system of Pb on Cu, the motivation stems primarily from the possibility of producing "man-made" materials with extraordinary magnetic properties. Layers of alternating magnetic and nonmagnetic elements lead to new phenomena such as antiferromagnetic coupling [2], oscillatory magnetic coupling [3] and oscillatory giant magnetoresistance (GMR) [4]. These effects are being exploited in the development of magnetic recording devices. There is now ample evidence that device performance is strongly affected by the microstructure at the interface. Investigations of Pb on Cu are important in this respect because it has been shown that Pb can be used as a surfactant to induce layer-by-layer growth in Cu(111) homoepitaxy and suppress the twinning of Cu spacer layers in Co-Cu multilayer growth. A series of papers by Camarero and co-workers [5-10] describes the atomisfic basis for these effects in detail. Egelhoff and co-workers [ 11 ] have show that the GMR performance in spin valves is improved when Pb is used as a surfactant. The exchange of Pb into the Cu(111) surface layer (surface alloying) is a key factor in explaining Pb's surfactant properties. Soldering is another technological area for which detailed understanding of the interactions between Pb atoms and Cu is important. Although soldering is a mature technology, it is becoming clear that the fundamental processes that control wetting and spreading of solders on surfaces are not well understood [ 12, 13]. Thermodynamic models that describe wetting and spreading behaviour at large length scales simply fail as the size of solder joints go to smaller and smaller dimensions. Pb on Cu is an excellent model system to study the dynamics of wetting and spreading because Pb is a key component of most commercial solders. Moreover, scientific issues common to many solder systems manifest themselves clearly for Pb on Cu. For example, on macroscopic samples, Pb does not wet Cu, but alloys of Pb and Sn (i.e., solders) do. What is the microscopic basis for this? Does surface alloy formation play a role? These questions provide another strong motivation to investigate the atomic-level behaviour of Pb on Cu surfaces. Cu-Pb "alloys" also have a long history of application as bearing materials in the automotive and aircraft industries [14-17]. The materials are referred to as alloys, although the Pb occurs as interdendritic or structurally independent particles. It is possible to retain Pb up to 40wt% in Cu by mixing the components above the liquidus temperature and chill casting. These materials have both good fatigue strength and seizure resistance. It is generally believed

154

that Pb acts as a solid lubricant to reduce the friction and wear of the mating surfaces [ 16, 18]. An evaluation of the mechanical and wear properties of Pb-Cu bearing alloys is reported by Pathak and Tiwari [19]. They find that both the wear and frictional heating are strongly influenced by the Pb content. Pb is also a critical component of bearing bronze alloys. Pb has the unique properties of lowering coefficient of friction and aiding machinability. Although the connection between atomic-scale studies of Pb-Cu surface alloying/de-alloying and the properties of Cu-Pb bearing alloys is not as direct as the applications mentioned above, a more fundamental understanding of Pb-Cu interactions will eventually be needed if one wishes to develop a physics-based approach to the improvement of these materials. In this chapter, the static and dynamic properties of Pb overlayers on singlecrystal surfaces of Cu are reviewed. In section 2 a brief description of the experimental and theoretical probes used to obtain this information is given. In section 3 properties of the overlayers are discussed. Here, a strong emphasis is placed on the atomic structures of the oveflayers and the changes that take place as the Pb coverage is changed. It is remarkable how small changes in coverage cause dramatic changes in atomic structure--often transforming the surface from alloy to overlayer phases even at room temperature. For this part of the review, I rely heavily upon the STM work of Nagl and co-workers [20-23] and in particular the Ph.D. thesis of Dr. Christian Nagl [24]. A tabulation of the overlayer structures is given at the end of section 3. This section also contains discussion of the dynamic properties of the films including the nature of transitions between the phases and the order-disorder (melting) transitions that occur at elevated temperatures. The chapter concludes with energetic and kinetic considerations on why surface alloying and de-alloying takes place for the Pb/Cu system and some remarks concerning future prospects.

2. EXPERIMENTAL AND THEORETICAL TECHNIQUES 2.1. Experimental A variety of experimental probes have been used to investigate the adsorption of Pb on single-crystal Cu surfaces and its subsequent alloying and de-alloying behaviour. The methods used for atomic structural characterisation include low energy electron diffraction (LEED), spot-profile-analysis LEED (SPA-LEED), Auger electron spectroscopy (AES), x-ray diffraction, scanning tunnelling microscopy (STM), low energy electron microscopy (LEEM), and thermal energy atom scattering (TEAS). A brief description of these techniques as they relate to studies of Pb on Cu is given in the following paragraphs. The references provided along with the descriptions offer a more detailed account. Low energy electron diffraction (LEED) is a commonly used probe of surface atomic structure [25-28]. In its simplest form, one determines the

155

periodicity of the atoms on a solid surface by viewing the pattern of diffracted electrons resulting from an incident beam of electrons. The atomic arrangement of adsorbed species, regardless of whether they are incorporated into the surface or reside on top of the surface, can be deduced from the positions of the diffraction spots, as long as the structures are periodic. Thus, when Pb forms an ordered structure on a Cu single-crystal plane with a periodicity that differs from the bulk Cu, extra diffraction spots appear in the LEED pattern. The extra spots, however, do not distinguish between atoms on top of the surface and atoms embedded into the first layer. More detailed information on atom positions and inter-layer distances can be obtained from measurements of the spot intensities as a function of electron energy (LEED I-V analysis). The measured I-V curves are matched to calculated intensity curves for different test structures. The atom positions and interlayer spacings are varied in the test structures until agreement with the measured profiles is achieved. The reliability factor or R-factor defines level of agreement that is achieved. In the technique of spot-profile analysis LEED (SPA-LEED) [29, 30] intensity variations across LEED diffraction spots are measured. The technique provides information on both periodic and non-periodic arrangements of superstructure domains, terraces or facets, and strained regions. The interpretation of spot profiles is simplified due to the validity of the kinematic approximation, which is not the case for LEED I-V analysis. Deviations from simple structures resulting from defects produce characteristic modifications of the spot profile, so that a quantitative evaluation is possible. SPA-LEED has proven itself to be useful in the study of dynamic phenomena on surfaces such as phase transitions. Several examples of this capability are discussed in section 3 for Pb on Cu. In Auger electron spectroscopy (AES) [31], the intensity of electrons scattered from a surface is measured as a function of their kinetic energy. Characteristic peaks in the energy distribution identify the elements present within the near surface region. AES is most often used in conjunction with LEED to characterise the composition and atomic structure of a surface or film. Knowing both the periodicity and relative composition of an overlayer is indispensable to deducing its atomic structure. Another diffraction-based technique that has been used to investigate Pb oveflayers on Cu is x-ray diffraction. To achieve the surface sensitivity necessary to study submonolayer-levels of adsorbates, the x-rays that impinge on the surface are at grazing incidence. (At normal incidence, the absorption length of CuK~ radiation in Cu is approximately 20~tm whereas at 0.4 ~ which is the critical angle for total internal reflection, the penetration length is reduced to -- 5 nm.) Experimental set-ups make efficient use of the properties of synchrotron radiation, which is tightly collimated in the vertical direction, but poorly collimated horizontally. The grazing incidence geometry allows the

156

diffuse scattering direction to be along the poor collimation direction resulting in higher count rates. Analysis of the diffraction peaks provides detailed information on the atomic structure of the oveflayer with respect to the underlying crystal. A description of this technique and its application to studies of Pb oveflayers on Cu(110) can be found in a paper by Brennan and co-workers [32]. Scanning tunnelling microscopy (STM) [33, 34] is widely used for investigations of the local atomic structure of surfaces. A probe "tip" is scanned across the surface revealing the positions of individual atoms. With its ability to achieve atomic resolution and, in most cases, distinguish between chemical species, the STM has provided key insights into the nature of alloy formation on surfaces. Both the static and dynamic properties of surface alloys can be probed with the STM. For the system of Pb on Cu, STM measurements were first to show the existence of surface alloy phases unambiguously and identify many of their structural properties [20-22, 35]. Low energy electron microscopy (LEEM) [36] is a real-space imaging technique that is typically used in conjunction with LEED. Like the transmission electron microscope (TEM), the LEEM is a non-scanning or parallel imaging microscope. Parallel imaging permits rapid data acquisition compared to scanning microscopes and allows one to view dynamic events in real time (video recording rates). The electron source and imaging optics are similar to those of a TEM. However, instead of transmitting high-energy electrons through a thinned sample, in the LEEM high-energy electrons are decelerated and reflected (scattered) from the surface of the sample. They are then re-accelerated to produce an image. The use of low energy electrons makes the technique extremely surface sensitive. The lateral spatial resolution of present-day LEEMs is 4-8 nm depending on the specific design, and the resolution perpendicular to the surface is sufficient to resolve single atom high steps. An especially useful contrast mode for the study of 2-D overlayer structures is achieved by imaging electrons from selected LEED beams. Brightfield images from the (0,0) beam at specific energies or dark-field images from non-integral beams provide large intensity differences between regions with differing atomic structures. Structural analysis of Pb overlayers on Cu has also been carried out using thermal energy atom scattering (TEAS) [37-40]. In this technique an incident beam of monoenergetic atoms or non-reactive molecules is scattered from a surface. The use of light particles (typically He) at low energies (10-300 meV) causes the scattering to be predominantly elastic. Also, because of the low energies, the atoms or molecules probe only the outermost layer of the surface in a non-destructive manner. When He atoms are used the technique is known as either TEAS or helium atom scattering (HAS), the difference depending on the research group. TEAS/HAS is particularly sensitive to the presence of

157

imperfections on a surface and thus is an excellent probe of ad-particles and defects. The sensitivity to ad-particles makes it possible to investigate 2-D lattice gases on a surface and their condensation into 2-D islands, even at very low coverages. As will be shown, this attribute is particularly useful in the investigation of Pb on the smooth surfaces of Cu in the low coverage regime (section 3.1-3.3). The sensitivity of TEAS to defects such as steps has proven useful in the study of Pb on stepped surfaces of Cu (section 3.4). Early reviews of the TEAS are given by Poelsema and Comsa [40] and Hulpke [39]. A more recent review is given by Farias and Rieder [37]. Throughout this article, I have adopted the currently accepted convention of referencing Pb coverages with respect to the Cu substrate. A complete closepacked layer of Pb on the Cu surface thus corresponds to a coverage of 0.56 ML. This convention is not always followed in the literature. In some cases a complete close-packed layer with the bulk lattice spacing of Pb is defined as 1.0 ML. In those cases, I have converted the coverages to the accepted convention. 2.2 Theoretical Theoretical modeling studies of surface alloying and de-alloying in the Pb/Cu system has been rather limited in comparison to the large amount of experimental information that exists. Total energy calculations using effective medium theory (EMT) have been carded out to help interpret STM images of Pb overlayers on Cu(111) and Cu(110) [23]. EMT is a semi-empirical theory in which the potential energy of a system is described by the sum of the embedding energy of an atom in a homogeneous electron gas and system-dependent correction terms [41]. This density-dependent total energy approach is also used in the Embedded Atom Method (EAM) [42] in which the parameters of the model potential are fit to experiment. The energy function of EMT is similar to that of the EAM, but, rather than fitting to experiment, the parameters are calculated from first-principles. For Pb on Cu EMT potentials have been derived by Stoltze [43]. Because it is an effective medium approach, EMT is capable of handling large systems of interacting particles as well as finite temperature and time-dependent problems. The diffusion of individual Pb atoms on the Cu(110) surface has been investigated by molecular dynamics simulations (section 3.4). In this case the interaction potential is derived from a phenomenological model similar to that used in the tight binding method [44, 45]. This potential satisfactorily describes bulk and surface properties of noble and transition metals except for the surface energies. In the Pb/Cu(110) studies the fight binding functional form is used to describe the Pb-Pb and Pb-Cu interactions. Parameters for both the pure metal and cross interaction potentials are obtained from fits to experimental values. Only recently have first-principles calculations based on density functional theory (DFT) [46] been applied to the Pb/Cu system. DFT has gained

158

considerable popularity due to the success of the local density approximation (LDA) (see, for example Ref. [47]), which appears to be accurate, not only for solids, but also for the adsorption of atoms and molecules on surfaces. Computational demands limit the size of the system that can be studied in comparison to the effective medium approaches, but the results are generally believed to be more reliable and have greater predictive capabilities. The availability of several DFT packages has increased the application of DFT to problems in surface physics and chemistry. Unpublished studies of Pb on single-crystal Cu surfaces [48, 49] mentioned later in this article employ the VASP simulation package [50]. 3. ATOMIC STRUCTURE, SURFACE ALLOYING AND DE-ALLOYING 3.1. Pb on C u ( l l l )

Although ultrahigh vacuum experimental investigations of Pb on Cu(111) have been going on for nearly 30 years, surface alloy formation in this system is a relatively recent discovery. In the first report of ultrahigh vacuum investigations of submonolayer Pb film growth on single-crystal surfaces of Cu, Henrion and Rhead [51 ] stated, "No evidence is found for mixed Pb-Cu layers that could be ascribed to surface alloying." Despite missing the surface alloy formation (see below), their LEED investigations provided the starting point for much of the ultra-thin Pb/Cu film growth that has followed. Their results indicate that Pb grows on Cu(111) in the Stranski-Krastanov [52] mode, i.e., the growth is two-dimensional up to monolayer-level coverages, followed by threedimensional island growth. They identified a p(4x4) pattern in their LEED images, which they interpreted as pseudo-(111) Pb, although the coverage at which this structure forms was not given. They noted that melting temperature of this dense monolayer is 558 K, which is considerably lower than that of bulk Pb (600 K). In a subsequent study, Barth6s and Rhead [53] combined LEED studies with Auger electron spectroscopy measurements to determine the coverage at which the p(4x4) structure appears. Rather than Stranski-Krastanov growth, their results suggest that Pb on Cu(111) follows Frank-van der Merwe (layer-bylayer) growth--an interpretation supported by the LEED and Auger measurements of Argile et al. [54]. Later, however, Ocal et al. [55] presented Auger data showing convincingly that the growth is, in fact, Stanski-Krastanov. Additional interesting behaviour for the Pb on Cu(111) system was obtained in SPA-LEED investigations by Meyer et al. [56, 57]. They showed that the "4x4" superstructure is actually an incommensurate overlayer over most of the coverage regime where it exists. The lattice of the overlayer is not fixed, but depends on coverage. TEAS experiments by Hinch et al. [58] in the early 1990s indicated that growth at temperatures below 300 K is qualitatively different than

159 growth at room temperature and above. They found that Pb films grow layer by layer at 140 K and exhibit a preference for particular thickness values, implying a quantum size effect. In none of these early studies was there any suggestion of surface alloy formation in the system of Pb on Cu(111). The situation changed in 1994 when Pb structures on Cu(111) were imaged in the STM. In their landmark paper, Nagl et al. [20] reported that, despite the immiscibility of Pb in bulk Cu, Pb atoms deposited on clean Cu(111) surfaces form a surface alloy, even at room temperature. Their images showed that upon initial deposition at 300 K, Pb atoms are adsorbed substitutionally in the Cu surface in narrow bands or "seams" along single-atom-high steps. An STM image illustrating this adsorption behaviour is shown in Fig. l(a.) With further addition of Pb, the region of incorporation widens to 2-4 nm. Annealing the low Pb-coverage surfaces to temperatures of 473 K and above produces substitutional Pb atoms over the entire surface (Fig. 1(b)), indicating the presence of Pb adatoms on the terraces that move too fast to be imaged. The conversion of these adatoms to substitutional atoms indicates that the surface alloy is the thermodynamically stable form of Pb on Cu(111) in the low-coverage regime. The surface alloy is mostly disordered with a few localised regions of (~/3x ~/3)R30 ~ and p(2x2) periodicity. Fig. 1(c) shows an STM image of a Cu(111) surface with 0.22 ML of Pb after annealing to 423 K. Local ordering can be seen in several regions of this image.

Fig. 1 Atomically resolved STM images of Pb atoms incorporated in a Cu(111) surface. (a) Low coverage of Pb atoms deposited at room temperature appear near steps. 10xl0 nm2. (b) Pb atoms deposited at room temperature and annealed to 423 K. The coverage is 0.03 ML. The atoms are randomly distributed over the surface and decorate the steps. 20x20 nm2. (c) Same as (b) but with 0.22 ML deposited Pb. 10xl0 nm2. Figure from Ref. [20] courtesy of P. Varga. When Pb is deposited at room temperature to coverages higher than 0.11 ML, the excess Pb (beyond that accommodated in the surface alloy) forms 2-D hexagonal close-packed islands that attach to the surface alloy seam. For

160

coverages in the range from 0.11 and 0.22 ML, annealing to 423 K causes the Pb in the close-packed islands to convert back to the surface alloy phase. The close-packed islands in this coverage range are therefore metastable. However, if the coverage is higher than 0.22 ML, hexagonal close-packed overlayer islands are present even after annealing. Fig. 2(a) shows an STM image of a surface covered with 0.31 ML Pb after annealing to 423 K. A narrow band of the overlayer phase is present between the step and the surface alloy phase. At a coverage of 0.56 ML, the whole surface is covered by the close-packed Pb film. The thermodynamically stable phases are therefore a random, substitutional surface alloy at coverages from 0 to 0.22 ML and a two-dimensional Pb overlayer phase at a coverage of 0.56 ML. The overlayer phase is the same p(4x4) phase identified in the LEED studies mentioned above.

Fig. 2 (a) STM image of 0.31 ML Pb deposited on C u ( l l 1) at room temperature and annealed to 423 K. 30x30 nm 2. (b) STM image of Pb on Cu(111). After annealing a 0.22 ML Pb film to 573 K, the surface was cooled to room temperature and an additional 0.22 ML

was deposited. In the two rectangular regions the contrast is enhanced to make the moir6 pattern more apparent (50x50 nm2). Figure from Ref. [20] courtesy of P. Varga. Further insight into the mechanism of surface alloying de-alloying is provided by the STM image shown in Fig. 2(b) [20]. The image is from a surface on which a thermodynamically stable, maximum-coverage, surface alloy phase is obtained by depositing 0.22 ML Pb and heating to 573 K. The surface is then cooled to room temperature, and an additional 0.22 ML of Pb is deposited. In addition to the surface alloy, Fig 2(b) reveals two structures with moir6 patterns indicative of the hexagonal close-packed phase. The different brightness levels of these two regions imply that the structures correspond to close-packed Pb regions of different heights. (The apparent height of the bright close-packed regions with respect to the dark close-packed regions is exactly the height of a Cu step). The fractional coverages of the different regions imply a de-alloying mechanism in which deposited Pb atoms replace Cu atoms in

161

regions of the surface alloy near steps to form the darker appearing close-packed overlayers. The displaced Cu atoms move out onto the terrace and replace Pb atoms to form the brighter close-packed islands. Based on the good agreement between measured areas of the different regions and the areas expected from the proposed mechanism, Nagl et al. [20] conclude that the close-packed islands lie on regions of pure Cu. The two-dimensional overlayer phase has interesting structural and dynamic properties and has been the subject of numerous investigations. The SPA-LEED studies by Meyer et al. [56, 57] mentioned above indicate that the Pb film grows in a (111) orientation with a lattice constant slightly larger than bulk Pb until a complete overlayer is formed. Further deposition leads to lateral compression of the monolayer by up to 3%. In a subsequent SPA-LEED study, MUller et a1.[59] repeated these measurements with more precision and confirmed that the lattice constant of the overlayer varies with coverage from 0.497 nm at half monolayer coverage (0.28 ML) to 0.480 nm at full monolayer coverage (0.56 ML). A plot of the variation in lattice constant of the Pb oveflayer with coverage is shown in Fig. 3. 0.50 r--

a Pb-Bulk

f--

o-~~ r

c 0.49 0

(5

" 0.48 _6

4/3 acu

13_

0

0.5

1.0

Coverage [ML]

1.5

Fig. 3 The lattice constant of the Pb overlayer vs. coverage. One M L of Pb is defined by the coverage of maximum compression; further Pb has to grow in a second layer. The two horizontal lines show the lattice constant of bulk Pb and a 4/3 commensurate layer. Figure

from Ref [59] courtesy of M. Henzler. The same authors found that the melting temperature of the film is coverage dependent as well, varying from 550 K at half-monolayer coverage to more than 700 K at coverages greater than a monolayer. These values bracket the melting temperature of bulk Pb (600 K) and the film melting temperature equals the bulk melting temperature when the film lattice constant equals the bulk lattice constant. A detailed study of the melting process was carried out and interpreted

162 in terms of the two-dimensional melting theories put forth by Kosterlitz and Thouless [60], Halperin and Nelson [61-63], and Young [64]. They suggest that below the melting temperature the overlayer can be described as a "floating solid" with quasi long-range order. Additional evidence for the formation of a Pb/Cu(111) surface alloy at coverages below 0.21 ML was provided by thermal energy atom scattering measurements of de Beauvais, et al. [65]. According to this study the maximum density of the alloy corresponds to one Pb atom for every 4.6 Cu atoms (i.e., 7 Pb for every 32 Cu). At higher coverages, large commensurate Pb islands with a p(4x4) structure appear. The transition between the two phases is first order. The growth and dynamics of ultra-thin Pb films on C u ( l l l ) was further characterised by Braun and Toennies [66] using HAS. They found that the growth mode is altered by the presence of T1. A Pb20%TL alloy forms a second monolayer, which is stable at elevated temperatures. With pure Pb, only one layer is stable. In very recent studies, MUller et a1.[67] used an in-situ combination of LEED, Auger spectroscopy and STM to show that the p(4x4) overlayer phase is a hexagonally close-packed and vertically buckled Pb layer. The Pb layer induces considerable restructuring of the first three Cu layers, but no intermixing of Pb and Cu atoms is observed. Their results indicate that the buckling amplitude of the first Cu layer is larger than that of the Pb overlayer and the height of Pb atoms in atop sites over Cu atoms is lower than for all other Pb. This "inverse corrugation" is consistent with STM images and EMT calculations by Nagl et al. [23] and with unpublished DFT calculations by Bartelt [49]. A schematic representation of the interlayer spacings and buckling amplitudes is shown in Fig. 4. MOiler et al. [67] suggest that the structural flexibility of the substrate may be related to Pb's surfactant effect in homogeneous and heterogeneous epitaxy on Cu(111).

Fig. 4 A cross-section along the mirror plane showing Pb atoms on Cu(111). Atoms cut by the plane and atoms touching it are displayed. The "inverse" buckling is evident. Figure from Ref. [67] courtesy of K. Heinz.

163

The two thermodynamically stable structures of Pb on Cu(111) (viz., the surface alloy phase and the oveflayer phase) are a two-dimensional, two-phase system with remarkable dynamic properties. LEEM investigations by Plass et al [68] show that as the overlayer phase grows on the alloy phase at constant temperature, stable domains assemble themselves into two-dimensional patterns with feature sizes of the order of tens of nanometers. Figure 5 is a sequence of low energy electron microscope (LEEM) images showing the growth of the overlayer structure (bright) at the expense of the surface alloy structure (dark) during Pb deposition at 673 K. During the growth of the overlayer phase the pattern evolves from circular islands with an average diameter of 67 nm to stripes and then to circular holes within the Pb oveflayer matrix. This sequence of domain patterns, typically referred to as "droplets," "stripes," and "inverted droplets," is predicted to be a general property of domain formation on solid surfaces [69-72]. Although there have been numerous observations of droplet and stripe patterns in widely varying physical and chemical systems, the Pb on Cu(111) system provides the first clear observation of the expected sequence of domain patterns with changing area fraction. Moreover, the LEEM results agree well with the positions of the droplet to stripe pattern transitions predicted for competing short-range attractive and long-range dipolar repulsive interactions on surfaces [72]. Plass et al. [68] propose that the long-range interaction arises from a difference in surface stress between the two. From an analysis of the droplet trajectories they determine that the stress difference is approximately 1 ev//~ 2 (-16 N/m). This number is considerably larger than typical stress values for clean surfaces (1-10 N/m) [73] indicating that there is still not a fundamental understanding of pattern formation, even though thermodynamic models suggest the interpretation is straightforward.

Fig. 5 LEEM images showing domain pattern evolution as a function of Pb coverage at 673 K. The images correspond to Pb coverages of (a) 0.33, (b) 0.39, and (c) 0.48 ML. The evolution from droplets to stripes to inverted droplets is evident. Figure from Ref. [68]. An obvious question is why the predicted progression of patterns manifests itself so clearly for Pb on Cu(111), but not for other adsorption systems. The fact that Pb forms a surface alloy, but not a bulk alloy is most likely an important

164

factor. Another reason that domain patterns can occur in this system is the large mobility of the droplet islands. For example, the islands observed in Fig. 5(a), which contain many thousands of atoms, can move as far as a micron before being incorporated into the more tightly packed droplet structure. This large island mobility allows patterns with hundred-nanometer periodicity to form. It is therefore possible that self-assembly is thermodynamically favoured in other thin-film systems, but kinetic constraints keep patterns from forming. LEEM investigations by Plass et al. [74] have provided additional information on the Pb/Cu(111) self-assembly system. Their studies show that S (a common contaminant on Cu surfaces) has a significant effect on the kinetics of self-assembly. Thus, the temperature at which one can observe the full evolution from droplets to stripes to inverted droplets changes with S concentration. In contrast, S has little effect on the thermodynamics of pattern formation. They also find that a similar evolution of patterns occurs when the Pb surface alloy phase is replaced with a Sn surface alloy phase. The ability to substitute one alloy phase for another and preserve self-assembly should prove useful in defining the fundamental interactions underlying the process. During the growth of the Pb overlayer phase, Plass et al also observed that vacancy islands on the surface increase in size and steps retract indicating that Cu is consumed during the formation of the overlayer phase. This observation is consistent with the conclusion of Nagl et al. [20] that in the mixed phase the Pb overlayer islands reside on top of pure Cu, i.e., Cu atoms are needed for the phase transition to take place. This type of information is essential for developing an atomic-scale picture of the self-assembly process. Deposition of Pb on Cu(111) beyond 0.56 ML results in the growth of 3-D Pb islands. The equilibrium shape of 3-D crystallites on Cu(111) has been investigated by STM [75, 76]. The (111) facet exhibits a three-fold symmetry. An analysis of the equilibrium crystal shape near the facet yields shape exponents that have an azimuthal dependence and vary from 1.4 to 1.7. This variation suggests a structure-dependent step interaction behaviour. LEEM investigations of the growth and melting transitions of 3-D Pb clusters have been reported by Luo et al. [77]. In their studies, kinetic growth and equilibrium cluster shapes were distinguished. They found a systematic variation in cluster melting temperature and explained the behaviour using a simple thermodynamic model involving finite size and interface effects. From a comparison of the model to measured undercooling during recrystallization and cluster contact angle, they were able to estimate the difference of cluster/substrate interface tension between liquid and solid clusters. The spreading of 3-D Pb islands on Cu(111) has been investigated by Moon et al [78]. They prepared 3-D Pb droplets with sizes ranging from 5 to 20 ~tm by first depositing a continuous film approximately 0.5 ~tm and then dewetting the film by heating to a temperature just above the Pb melting point. Cooling the

165 surface produced 3-D Pb particles residing on top of the p(4x4) overlayer. To study spreading onto the bare Cu(111) surface, the p(4x4) layer between the particles was removed by ion sputtering. The spreading of the droplets was measured by AES line scans. They found that a Pb film spreads away from the Pb particle in a similar fashion to precursor wetting films observed in liquids. The time dependence of the precursor film growth is quite similar to that observed in more complex molecular systems. The diffusion coefficients extracted from the measured concentration profiles were found to be high at both high and low Pb coverage and are a minimum at 0.5 ML. The coverage dependence is consistent with the changing atomic structures as a function of coverage. In particular, the low diffusivity at intermediate coverage was attributed to incorporation of Pb into the surface alloy phase. 3.2. Pb on Cu(100) The growth of Pb films on Cu(100) is similar to C u ( l l l ) in that one observes surface alloying at low coverages and de-alloying to overlayer structures at higher coverages. Also, as for Pb on Cu(111), 3-D island growth is observed after completion of the most dense overlayer structure (StranskiKrastanov growth). Submonolayer growth in the two systems, however, differs in several respects. Whereas the surface alloy phase that forms on Cu(111) is disordered at all coverages, a well-ordered surface alloy phase forms on Cu(100) at 3/8 ML. At higher coverages, there are two, rather than one, Pb overlayer structures on Cu(100). Also, during de-alloying of the surface alloy phase on Cu(100) adatom islands grow (steps expand), whereas during de-alloying on Cu(111) vacancy islands grow (steps retract). Finally, the evolution of selfassembled, nanoscale domains observed for Pb on Cu(111) is not seen for Pb on

Cu(100). From the early LEED work of Henrion and Rhead [51 ] and H6sler and coworkers [79-81], it is known that there is a progression of ordered superstructures for Pb on Cu(100) with increasing Pb coverages. Examples of LEED patterns from more recent work [82] are shown in Fig. 6. At very low

Fig. 6 LEED pattems showing the progression of superstructures during deposition of Pb on Cu(100) at 400 K. Figure from Ref. [82].

166

coverages, the structure is disordered and the LEED pattern is (lxl). At 3/8 ML a sharp c(4x4) diffraction pattern forms followed by a c(2x2) pattern at 0.5 ML and a c(5{2x~/2)R45 ~ structure at 0.6 ML. Although the earlier LEED studies interpreted the c(4x4) superstructure as chains of Pb on top of the Cu(100) substrate, Auger electron spectroscopy measurements by Sepulveda and Rhead [83] along with more recent LEED I-V analyses by Gauthier et al. [84] and STM investigations by Nagl et a1.[22] provided compelling evidence that the c(4x4) phase is not an overlayer phase, but a surface alloy consisting of Pb chains within the first layer of Cu. The addition of Pb to the c(4x4) phase causes the topmost surface layer to de-alloy and produces a c(2x2) overlayer structure at 0.5 ML. LEED I-V analyses by HOsler et al. [79], low energy ion scattering (LEIS) studies by Platzgrummer et a1.[85], and STM analyses by Nagl et a1.[22] agree that the c(2x2) phase is an overlayer with Pb atoms residing on top of the Cu(100) surface layer in the four-fold hollows. Further deposition causes the structure to disorder resulting in a diffuse ( l x l ) LEED pattern. At 0.6 ML, a third superstructure appears with a c(5~/2x{2)R45 ~ diffraction pattern. LEED I-V analyses by HOsler and Moritz [80] identified this phase as a dislocated c(2x2) overlayer. The Pb atoms reside in the four-fold hollows, but the overlayer is compressed to allow for the extra coverage (Fig. 4c). In a later study using TEAS, Sanchez and Ferret [86] deduced basically the same structure, except that two Pb atoms of the unit cell protrude higher than the others. Once the c(5~/2x~/2)R45 ~ overlayer has been achieved, further deposition of Pb results in the growth of three-dimensional Pb islands. Top-view models illustrating the structures of Pb on Cu(100) are shown in Fig. 7.

Fig. 7 Schematicrepresentation (top view) of the three ordered structures of Pb on Cu(100). (a) c(4x4) surface alloy structure, (b) c(2x2) overlayer structure. (c) dense c(5~/2x~/2)R45~ structure. (second-layer Cu is light grey; top layer Cu is darker grey; Pb is black) Figure 8 shows STM images from Nagl et al. [22] corresponding to the three ordered phases of Pb on Cu(100). The c(4x4) phase shown in Fig. 8(a) is a labyrinth-like structure formed by two rotational domains of the chain structure indicated in Fig. 7. Superposition of a Cu lattice onto the lattice of the c(4x4) structure indicated that the bright atoms are in bridge sites. This conclusion,

167

however, does not agree with LEED I-V measurements of Gauthier et al. [84]. Their analysis indicates that the comer atoms of the unit cell are in hollow sites. Nagl et al. [22] suggest that this discrepancy might be due to the different temperatures that the experiments were carried out. The STM imaging is performed at room temperature, whereas the LEED measurements are at 150 K.

Fig 8. STM images of the three ordered structures of Pb on Cu(100). (a) the c(4x4) surface alloy structure 10xl0 nm2, (b) the transition of the c(2x2) overlayer structure to the c(5~/2x~/2)R45~ structure 10xl0 nm2, (c) the dense c(5~/2x~/2)R45~ structure 5x5 nm2. Figure from Ref. [22] courtesy of P. Varga. With increasing Pb coverage, small domains of the c(2x2) structure appear in the STM images. These domains grow and replace the c(4x4) structure with a c(2x2) structure at a coverage of 0.5 ML. Further increase of the Pb coverage from 0.5 ML to 0.6 ML results in a continuous phase transition to the c(5~/2x~/2)R45 ~ structure. This transition proceeds by insertion of antiphase domain boundaries into the c(2x2) structure. Figure 8(b) shows an STM image of the c(2x2) structure with antiphase boundaries indicated. Figure 8(c) shows the completed c(5x/2x~/2)R45 ~ structure at a Pb coverage of 0.6 ML. The remarkable self-assembly and pattern formation that takes place during surface de-alloying for the system of Pb on Cu(111) (discussed in section 3.1) is not observed for Pb on Cu(100). Nevertheless, the dynamics of surface alloying and de-alloying for Pb on Cu(100) is interesting in its own fight. Real-time LEEM investigations of changes in surface morphology that occur during the initial growth of the c(4x4) surface alloy phase and transitions from the c(4x4) to c(2x2) phase and from the c(2x2) phase to the c(5x/2xx/2)R45 ~ phase are reported by Plass and Kellogg [82]. They find that when the sample is held at elevated temperatures (390-420 K) during deposition, step motion and island nucleation and growth are apparent at a length scale of tens of nanometers. Representative LEEM images from the initial stages of Pb deposition on Cu(100) at 410 K are shown in Fig. 9. Prior to deposition the only feature in the

168

image is the step at the bottom of Fig. 9(a). During deposition, the step progresses outward and two-dimensional islands nucleate and grow (Fig. 9(b,c)). There is compelling evidence that both the step progression and island growth are due to displaced Cu and not the deposited Pb itself. As shown above, STM images from Nagl and co-workers [22] and Robert and co-workers [35] unambiguously show Pb atoms substituted into the surface layer at low Pb coverages. In the same investigations, displaced Cu is seen at the steps and in small, nucleated islands. It has also been noted that Pb islands can be deposited on Cu(100) when the sample is cooled below room temperature and these islands have a totally different appearance in the LEEM from those in Fig. 9 [82].

Fig. 9 LEEM images showing the changes in surface morphology that take place during initial deposition of Pb on Cu(100) at 410 K. (a) 0.05 ML, (b) 0.15ML, (c) 0.23 ML. The field of view is 4.5 lxm. Figure from Ref. [82].

During the initial stages of Pb deposition and Cu island growth, there is no evidence for ordered superstructures. The Pb is substituted randomly into the Cu(100) surface. LEEM images recorded during continued Pb deposition show that the ordered c(4x4) phase first grows out from upper sides of steps. This observation is consistent with the STM investigations mentioned above. Because the c(4x4) phase is a surface alloy in which every other row of Cu atoms is replaced by Pb atoms, it is necessary to displace 0.5 ML of Cu to form a complete c(4x4) layer. The amount of displaced Cu measured in the LEEM investigation (the sum of the area of the islands and the area of the material added to the steps) ranges from 0.50 to 0.56. consistent with the c(4x4) surface alloy structure. The addition of more Pb to the c(4x4) surface alloy causes the surface to dealloy resulting in a c(2x2) overlayer phase at 0.5 ML. LEEM images recorded during transition from the c(4x4) to c(2x2) surface structures are shown in Figs. 10(a) and 10(b). In these images the conditions are chosen such that the c(2x2) phase appears bright. The starting surface is pure c(4x4) with small islands

169

remaining from the growth of the c(4x4) phase (not shown). The sequence of images indicates that the c(2x2) phase nucleates at the steps and proceeds out on to the terraces in both directions. It is obvious from the images shown in Figs. 10(a) and 10(b) that the islands on the surface grow rather than shrink during the conversion. This behaviour is consistent with a model in which the rows of Cu between the rows of Pb in the c(4x4) structure are expelled during deposition and replaced by Pb. The displaced Cu moves to the steps and island edges and causes them to expand. It is also obvious that the islands take on more of a square or rectangular shape, rather than circular, indicating a lower formation energy for steps along the <110>-type directions in the c(2x2) phase.

Fig. 10 LEEM images showing the changes in surface morphology that take place during the conversion of the Pb/Cu(100) c(4x4) surface alloy phase to the c(2x2) overlayer phase at 400

K. (a) Cu islands become square and grow as the surface de-alloys. (b) The c(2x2) overlayer phase (bright) is nearly complete. The field of view is 2.5 ktm. Figure from Ref. [82]. A full conversion from the c(4x4) to c(2x2) phase should result in 0.5 ML of displaced Cu. Previous STM measurements are consistent with this expectation [35]. However, LEEM measurements taken with the sample at 400 K, indicate that there is only 0.23 ML of added material during the conversion. To account for this difference, Plass and Kellogg [82] propose a new model for the higher temperature c(2x2) phase in which some of the Cu remains randomly alloyed in the c(2x2) overlayer structure. This model is at odds with LEED I-V analysis of H6sler and co-workers [79], who concluded that the c(2x2) structure is a pure overlayer phase. The discrepancy is still not resolved, but may be due to the different manner in which the c(2x2) phase was prepared in the two studies. In the LEED I-V study the overlayer was prepared by depositing excess Pb and flashing the surface repeatedly to 700 K until the c(2x2) pattem was its strongest. In the LEEM study the c(2x2) structure evolved from the c(4x4) structure during Pb deposition at 400 K. LEEM images taken during conversion of the surface from the c(2x2) to c(5~/2xx/2)R45 ~ phases at 400 K provide further evidence that Cu is incorporated

170

into the c(2x2) phase. As in the other transitions, the c(5~/2x~/2)R45 ~ phase begins at the step edges and proceeds outwards onto the terraces. Analysis of the images indicates that there is both step progression and island growth as the surface transforms. If the c(2x2) phase were a pure overlayer phase, there would be neither movement of steps nor island growth. The amount of added material during the conversion from the c(2x2) to c(5~/2x~/2)R45 ~ phase is approximately 0.05 ML, roughly consistent with the area consumed by deposited Pb to produce the c(5~/2x~/2)R45 ~ phase. In addition to step progression and island growth, line features appear during conversion from the c(2x2) to c(5~/2x~/2)R45 ~ phase. When the surface is heated to 425 K, these features re-shape into what appears to be two-dimensional Pb-covered Cu islands. This observation suggests that even more Cu is expelled during the conversion process. Adding the line feature area to the island and step progression areas gives a total added material coverage of--0.20 ML, accounting for all the extra Cu alloyed in the c(2x2) phase. All three surface phases of Pb on Cu(100) undergo order-disorder transitions at elevated temperatures. The temperatures at which the structures disorder were first reported by Sanchez and co-workers [86, 87] using TEAS. They found that the c(4x4) phase disorders at 545 K. The critical exponent is near zero indicating a first-order transition. Their data also suggests first-order transitions to disordered structures for the c(2x2) and c(5~/2x~/2)R45 ~ phases at temperatures of 498 K and 490 K, respectively. Kellogg and Plass [88] investigated the thermal behaviour of the phases using both LEED and bright-field LEEM. They found that the c(5~/2x~2)R45 ~ phase exhibits the sharpest behaviour with a first-order phase transition at 490 K, consistent with TEAS. The LEED/LEEM results for c(2x2) and c(4x4), however, differ significantly from TEAS. The c(2x2) phase, in particular, shows no evidence for nucleation and growth of the ordered structure as the temperature is lowered through the transition. Moreover, analysis of dark-field images from the half-order LEED spots as a function of temperature indicates that the transition takes place over a broad range of temperature between 390 K and 550 K. This behaviour is more consistent with a second-order phase transition. A much sharper transition is observed when the c(2x2) phase is mixed with the c(4x4) phase. In this case the transition temperature is found to be dependent on the ratio of the two phases, but always lower than the value reported previously. It is possible that a mixed phase was examined in the TEAS study, leading to the conclusion that the c(2x2) order-disorder transition is first order. In the case of the c(4x4) structure, LEEM observations [88] imply a firstorder phase transition. Bright-field LEEM images clearly show a nucleation and growth process as the temperature is lowered through the transition. However, the LEEM results for the value for the temperature of the transition (--513 K) are

171

considerably lower than that reported in the TEAS work. LEEM measurements indicate that the transition temperature depends on whether or not the c(4x4) phase is complete, and whether or not the surface is subsequently annealed to coarsen the domains. More extensive measurements are required to resolve the discrepancy for the transition temperature of c(4x4) surface phase. From the models shown in Fig. 7 it is evident that both the c(4x4) and the c(5~/2x~/2)R45 ~ surface structures have two-fold symmetry. These two surface phases therefore have orientational domains in perpendicular directions. The dark-field imaging procedure available in the LEEM allows one to view the size of the orientational domains and how they coarsen as a function of temperature [88]. For structures produced by Pb deposition at room temperature, the domain size is smaller than the spatial resolution of the LEEM (7.5 nm). When the surface is heated, the domains increase in size for both phases, but at significantly different temperatures. Whereas the c(4x4) requires temperatures of approximately 673 K (which is higher than the two-dimensional melting transition), domains of the c(5~/2x~/2)R45 ~ phase coarsen at temperatures as low as 425 K. The reason for the difference is clear. The c(4x4) phase is a surface alloy. To change the domain structure requires rearrangement and transport of Cu as well as Pb. In contrast, the c(5~/2x~/2)R45 ~ phase is an overlayer structure. To coarsen this phase only requires transport of Pb on the Cu surface. Deposition of Pb beyond a coverage of 0.6 ML results in the growth of 3-D islands. STM, SEM, and LEEM images [24, 89-91] show that shape of the 3-D islands is asymmetric, elongated along the direction of the dislocation lines in the c(5~/2x~/2)R45 ~ structure. Their dimensions are typically a few hundred nm laterally and 15 nm high. The mean island density is approximately 2.5 ~trn-2. The reason for the asymmetry is that the islands grow on the two-fold symmetric c(5~/2x~/2)R45 0 overlayer phase. A direct correlation between the shape of 3-D Pb islands and the orientation of the c(5~/2x~/2)R45 ~ overlayer has been observed by LEEM. Figure 11 shows a LEEM image of 3-D islands on a Cu(100) surface. The energy of the electrons impinging on the sample is adjusted to give different intensities for the two domain orientations of the c(5~/2x~/2)R45 ~ surface overlayer. The correlation between the orientation of the islands and the domain structure of the 2-D Pb overlayer is evident. The spreading of 3-D Pb islands on Cu(100) has been investigated by Prrvot et al. [92] using Rutherford backscattering analysis and AES. For the AES measurements, the 2-D Pb layer between the islands is removed by sputtering at 150 K. The results indicate that the diffusing species is a Pb atom moving over the dense 2-D Pb layer. Analysis of the concentration profiles yields the activation energy governing the process (0.97_+0.05 eV), which is the sum of the formation energy of the adatoms and the activation energy of surface diffusion across the layer. They also find that the mean domain size and domain size ratio of the orientational domains on the c(5~/2x~/2)R45 ~ structure play an important

172

role in the spreading. They suggest that the domain boundaries may act as barriers to the diffusion of Pb atoms.

Fig. 11 LEEM image showing the domain structure of a coarsened c(5~/2x~/2)R45~ overlayer and subsequently deposited 3-D Pb islands. The island orientation correlates directly with the domain structure. The field of view is 4.5 lam. Figure from Ref. [91].

3.3. Pb on C u ( l l 0 ) As is the case for the lower-index crystal planes, Pb deposition on Cu(110) produces a rich variety of overlayer phases. With increasing Pb coverage, one finds a surface alloy lattice gas at low coverage, a c(2x2) overlayer at 0.5 ML, a p(4xl) overlayer at 0.75 ML, a p(9xl) overlayer at 0.778 ML and a p(5xl) overlayer at 0.8 ML. Structures for these phases have been proposed from studies using LEED [51], Auger spectroscopy [83], grazing incidence x-ray scattering [32, 93-95], TEAS [96], and STM [21]. The following description is based mainly on the STM work of Nagl and co-workers [21 ]. The lowest coverage phase of Pb on Cu(110) could not be imaged in the STM, but TEAS experiments of de Beauvais et al. [96] were able to identify the atomic structure. These authors showed that, like Cu(111) and Cu(100), low coverages of Pb on Cu(ll0) form a lattice-gas surface alloy. Pb atoms, embedded in the topmost layer of Cu, diffuse so fast that imaging in the STM is not possible. The first ordered structure to appear with increasing Pb coverage is a c(2x2) overlayer at 0.5 ML. TEAS experiments by de Beuvais et al. [96] and STM experiments by Nagl et al. [21] show conclusively that the c(2x2) phase is a simple overlayer structure. An example of an STM image of the c(2x2) Pb overlayer structure on Cu(110) is shown in Fig. 12(a). The next ordered phase to appear as the Pb coverage increases is a p(4xl) superstructure. The transition from the c(2x2) to the p(4xl) structure is first order and the p(4xl) covers the entire surface at 0.75 ML. Figure 12(b) shows a close-up STM image of a p(4xl) surface. The surface is made up of grooves that run perpendicular to the [ 110] rows of the Cu(110) surface. The minimum spacing between two grooves is four Cu lattice sites. This gives rise to the p(4xl) diffraction pattern. Note that between the grooves in the regions identified as p(4xl) there are only two rows of Pb atoms. However, the

173

coverage at which this structure appears is 0.75 implying three Pb atoms per four Cu atoms. This leaves an extra row of Pb unaccounted for. As indicated in the models shown in Fig. 13 the extra atoms are found within the grooves. The p(4xl) structure is a thus combination of a surface alloy and a surface overlayer. Regions with larger separations between the grooves in Fig 12(b) are modifications of the p(4xl) structure and are designated p(8xl) and p(12xl). STM images obtained by Nagl et al. [21] show that the p(8xl) structure exhibits the same inverse corrugation as was seen for Pb on the Cu(111) surface.

Fig. 12 (a) STM image of the c(2x2) Pb overlayer on Cu(110). An antiphase domain boundary is visible in the lower right comer. (10xl0 nm2). (b) STM image of the p(4xl) Pb overlayer on Cu(ll0) at a coverage of 0.75 ML. Two modifications of the p(4xl) structure result in the local formation of p(8xl) and p(12xl) structures. (10xl0 nm2). (c) STM image of the p(9xl) Pb overlayer on Cu(ll0) at a coverage of 0.775 ML. (10xl0 nm2). Figures from Ref. [21] courtesy of P. Varga. When the Pb coverage is increased to 0.775 ML, a new structure appears in addition to the p(4xl) and its modifications discussed above. Here, every seventh row of Pb is substituted into the Cu surface layer. The structure is p(9xl) with a local coverage of 7/9 (0.778) ML. An STM image of the p(9xl) structure is shown in Fig. 12(c) and a model in Fig. 14(a). The final structure to form with increasing coverage is a p(5xl)phase. This structure entirely replaces the p(9xl) phase at 0.80 ML. A model of the p(5xl) phase is shown in Fig. 14(b). It is interesting that the p(4xl), p(9xl), and p(5xl) atomic structures determined from the STM images all disagree with structures based on earlier LEED, x-ray scattering and TEAS studies. In the earlier studies all p(nxl) structures were believed to be overlayer structures, whereas the STM images provide convincing evidence that the Pb is substitutes into the Cu substrate. In the case of the p(5xl) structure, Nagl et al. [21] re-analysed the earlier x-ray diffraction study of Marra et al. [93] and found that it is possible to reconcile the

174

STM and x-ray scattering results. The STM studies also cast doubt on the existence of an incommensurate phase suggested by the x-ray scattering studies. According to the x-ray scattering studies, this "floating hexagonal layer" is produced by annealing the p(5xl) structure to 623 K. In the STM studies, Pb desorption is observed at this temperature and the succession of p(nxl) phases is observed in reverse order. Nagl et al. [21] refer to this phase as "discommensurate" structure made up of a random sequence of unit cells of different atomic structures.

Fig. 13 Model of the atomic structure of Pb/Cu(ll0)-p(4xl). (a) clean Cu(ll0) surface, (b) top view of the p(4xl) structure (c) side view of the p(4xl) structure. (Pb overlayer atoms are

black; Pb surface alloy atoms are dark grey). At coverages greater that 0.80 ML, 3-D Pb islands grow on the p(5xl) overlayer. STM images by Nagl [24] show that the islands are strongly elongated along the <100>-type directions, i.e., parallel to the groves of the p(5xl) overlayer. Thus, like 3-D islands on Cu(100), there is a correlation between the symmetry of the Pb overlayer and the shape of the 3-D islands. Contrary to the (111) and (100) surfaces, Pb islands on the (110) surface do not have (111) planes parallel to the surface. Nagl [24] suggests that this may be due to the grooved structure of the p(5xl) 2-D layer. Measurements of single-atom diffusion barriers for Pb Cu(110) have been reported by Pr6vot et al. [97]. They used Rutherford backscattering spectrometry (RBS) to measure the spreading of a 3.5 mm square Pb deposit across the surface at temperatures in the 500-800 K range. The activation energy determined from Gaussian fits to decaying concentration profiles is close to 0.6 eV along both the in-channel and cross-channel directions. The ratio of the diffusion constant in the two directions is 2.4. The results were interpreted in terms of a model in which the embedded Pb atom exchanges with a Cu adatom,

175

jumps along the close-packed (in-channel) direction, and re-inserts itself into the surface plane. The model predicts that the diffusion becomes more anisotropic at lower temperatures because the adatom makes more jumps before re-insertion. The mechanism is supported by molecular dynamics simulations based on empirical tight binding potentials [98].

Fig. 14 Models of the atomic structures of (a) Pb/Cu(110)-p(9xl) and (b) Pb/Cu(110)-p(9xl). (Pb overlayer atoms are black; Pb surface alloy atoms are dark grey).

3.4. Pb on stepped surfaces of Cu Investigations of Pb on the stepped surfaces of Cu provide a systematic way to study the influence of defects on the statics and dynamics of overlayer formation. Stepped surfaces are also interesting in that the observed phenomena (self-assembly, phase transitions, etc.) have a one-dimensional character to them. Although studied less extensively than the flat surfaces, LEED/AES and SPA-LEED studies have been reported for the adsorption of Pb on the (211), (311), (511), (711) (510), and (10,10) planes of Cu. Along with their early studies of Pb on Cu(100) and (110), Sepulveda and Rhead [83] investigated Pb adsorption on Cu(511) and (711) by LEED/AES. They reported that adsorbed Pb layers on the stepped surfaces tend to be disordered due to high mobility. A p(4xl) LEED pattern is observed on both the (511) and (711) planes, but the LEED pattern is never very sharp. Barth6s and Rhead [53] subsequently obtained LEED/AES and thermal desorption data for Pb on the (711), (511), (311) and (211) stepped surfaces. They found no evidence that the (311) and (211) stepped surfaces have disordered surfaces at lower coverages, in contrast to the (711) and (511) surfaces. Later, Barth6sLabrousse [99] extended these studies to the Cu(510) and Cu(10,10) surfaces. LEED/AES results indicated that these surfaces are topologically unstable with Pb adsorption at ambient temperature. For the (10,10) surface (210), (710), and (510) facet orientations are identified with increasing Pb coverage. For the (511) surface decomposition into the (210) and (100) facets is followed by restoration of the initial (510) structure. In 1991 Wollschl~iger et al. [100] used SPA-LEED and AES to investigate the epitaxial growth and thermal behaviour of Pb on Cu(311). They found that

176

the first monolayer forms three different structures with increasing coverage: a lattice gas, a row structure and a close-packed overlayer. At high annealing temperatures the close-packed oveflayer changes to the row structure due to loss of Pb either by desorption or diffusion into 3-D islands. The Pb overlayers on Cu(311) undergo reversible order-disorder phase transitions at a temperature lower than the Pb bulk melting temperature. Their SPA-LEED analysis suggests a dislocation-mediated melting of the Pb overlayer and one-dimensional melting behaviour due to the corrugation of the substrate. For all of the stepped surfaces examined, deposition beyond a complete Pb monolayer produced 3-D island growth consistent with the same Stranski-Krastanov mode observed on the smooth surfaces. More recently, a systematic SPA-LEED investigation of Pb adsorption on Cu(211) and Cu(511) has been reported by Hofmann et al. [101]. Their highresolution patterns in combination with AES coverage measurements reveal a series of new structures on these stepped surfaces. On Cu(211) a streaky c(2x2) pattern is observed at a coverage of 0.27 ML. An atomic model in which linear chains of Pb atoms lie parallel to the steps is proposed to explain this pattern. At higher coverages the SPA-LEED pattern changes and new streaks appear at the 3/4 positions. At 0.34 ML, the new streaks co-exist with those appearing at lower coverages. At 0.37 ML, the streaks vanish and are replaced by sharp p(4/3 x 1) pattern. The proposed structure for this phase involves close-packed rows of Pb running parallel to the steps. At a coverage of 0.45 ML a phase designated p(4/3 x disorder) is observed. This structure gives way to the densepacked hexagonal overlayer structure c(4/3 x 1) at 0.62 ML. Similar to (211), initial deposition of Pb on Cu(511) produces parallel streaks in the SPA-LEED pattern in addition to the substrate peaks. Designated p(2x disorder) these peaks increase in intensity with increasing Pb coverage, becoming most pronounced at a coverage of 0.37 ML. Again, the streaks indicate chain growth parallel to the steps. Further increases in Pb coverage produce new streaks at the 3/4 positions. At 0.39 ML these streaks co-exist with the lower coverage streaks at the 1/2 positions. Between 0.48 and 0.56 ML a structure similar to the p(4/3 x 1) is observed for the (211) surface. The observations indicate that Pb forms dense-packed linear chains on Cu(511). Annealing the monolayer-covered surfaces on both Cu(211) and Cu(511) to 600 K causes the substrate surface to reconstruct with the size of the terraces growing by a factor of 8/3. It is worth noting that all structures proposed for the stepped surfaces of Cu are overlayer structures. No models involving surface alloy formation have been introduced. However, Pb/Cu surface alloy formation on the flat surfaces was proposed only after there was strong evidence for their existence from STM studies. It is therefore not inconceivable that surface alloy structures or mixed

177 overlayer/alloy are present on the stepped surfaces as well and that STM investigations will be required to reveal their form. Table 1 Structures of Pb overlayers on Cu Crystal plane Cu(111)

Cu(100)

Cu(110)

Cu(211)

Cu(311)

Cu(511)

Cu(711) Cu(511)

Coverage range 0<0.22 0=0.22 0.22<0<0.56 0=0.56 0>0.56 0<0.375 0=0.375 0=0.5 0=0.6 0>0.6 0<0.5 0=0.5 0=0.75 0=0.778 0=0.8 0>0.8 0=0.27 0=0.37 0=0.45 0=0.62 0>0.62 0<0.28 0.28<0<0.48 0.48<0<0.56 0>0.56 0=0.37 0=0.56 0>0.56 unknown unknown

Periodicity

Alloy/Overlayer Comments

Random Random Random/p(4x4) p(4x4)

Alloy Alloy Alloy/Oveflayer Overlayer

Random c(4x4) c(2x2) c(5~/2x~/2)R45 ~

Alloy Alloy Oveflayer Oveflayer

Random c(2x2) p(4xl) p(9xl) p(5xl)

Alloy Oveflayer Alloy/Oveflayer Alloy/Overlayer Alloy/Overlayer

c(2x2) p(4/3 x 1) p(4/3 x disorder) c(4/3 x 1)

Overlayer Oveflayer Oveflayer Oveflayer

Random (4xl)

Oveflayer

Close-packed

Oveflayer

(2 x disorder) Square p(4/3x 1)

Oveflayer Overlayer

Mobile Pb Complete Co-exist Complete 3-D Islands Mobile Pb

3-D Islands Mobile Pb Mixed Mixed Mixed 3-D Islands

3-D Islands Lattice gas Referred to centered unit mesh 3-D Islands

p(4xl) p(4xl)

3-D Islands LEED not sharp LEED not sharp

178

3.5. Summary of Pb surface alloy and overlayer structures on single-crystal surfaces of Cu Table I provides a summary of the 2-D Pb structures observed for the various single-crystal planes of Cu. The structures correspond to Pb deposition at room temperature. Different structures are known to form at temperatures below 300 K. In cases where there have been conflicting interpretations of a given periodicity, the most recent structure is listed. The trends in Table I are self-evident. At very low coverages, Pb forms random lattice gas structures. The Pb atoms are substituted into the Cu surface on the (111), (100), and (110) crystal planes. Most likely, this is the case for the stepped surfaces as well. As the Pb coverage increases, the lattice gases condense into stable surface alloy structures that can be either random, as for the (111) face, or ordered, as for the (100) and (110) faces. Further increases in the Pb coverage results in de-alloying and various ordered overlayer structures. Upon completion of the most dense 2-D overlayer structures, 3-D islands grow. 4. CONCLUDING REMARKS The phase diagrams for the various atomic structures of Pb on single-crystal surfaces of Cu have been mapped out in considerable detail using a variety of experimental probes. In fact, few other surface alloy systems have been characterised so completely. The evolution of atomic structures with coverage for the Pb/Cu system displays considerable complexity that, in all likelihood, exists for surface alloy systems in general. The present challenge is to define the atomic-scale interactions that give rise to the observed evolution of surface structures with coverage. The progression from surface alloy to surface overlayer phases arises from an energetic competition between Pb-Cu, Pb-Pb, and Cu-Cu bonds at various separations. To understand the phase diagrams, a comprehensive description of these fundamental interactions is required. This type of information is just starting to become available. For example, an unpublished, first-principles calculation by Feibelman [48] indicates that the gain in energy achieved by Pb adatom when it embeds itself in the top layer of Cu(111) is 0.38 eV. Thus, at very low coverages, a surface alloy phase is expected. The next step is to define the energetics as the coverage is increased and other bonding configurations present themselves. A formalism to predict surface alloy phases and their stability from surface energy calculations has been developed by Christensen et al. [1]. Using both density functional and effective medium theory, they determine the segregation energy and surface mixing energy for combinations of the transition and noble metals, but the Pb/Cu system is not included in their calculations.

179

Characterisation of the phase diagrams for Pb on Cu surfaces is facilitated by the fast kinetics of Pb-Cu intermixing, which allows equilibrium structures to form at relatively modest temperatures. In STM studies of Pb on Cu(111), for example, one can observe surface alloy formation at -400 K on time scales of minutes to hours [20]. Understanding the kinetic process by which Pb adatoms embed themselves into close-packed surface layers is therefore just as important as understanding the thermodynamics of intermixing. As in the case of thermodynamics, relating the kinetics of alloying to atomic processes is still generally at an early stage [102-104]. The spontaneous formation of surface vacancies is not a viable kinetic pathway for a Pb atom to embed itself into Cu(111) at 400 K because the formation energy of a vacancy on a C u ( l l 1) surface is too large [48, 105]. Nagl [24] suggests that the most likely kinetic process involves Pb atoms inserting themselves into the terraces at the steps. This pathway is consistent with the STM observations shown in section 3.1, where the Pb/Cu(111) embedded Pb atoms first form as a narrow band along the step. Once inserted, a diffusion mechanism is required to move the atoms away from the steps. Recent STM investigations by Swartzentruber [106] have determined that, at 300-400 K, the diffusion of Pb atoms embedded in Cu(111) takes place predominantly by a surface-vacancy-mediated process. The embedded atom does not move until it encounters a mobile surface vacancy. When the Pb atom and surface vacancy becomes neighbours, exchange processes produce atom displacements. The same mechanism is operative for the diffusion of In and Pd atoms embedded in Cu(100) [107, 108]. It is this type of detailed information that is needed to build a fundamental understanding of technologically important processes involving thin films of Pb on Cu surfaces. Ultrahigh vacuum surface science investigations of Pb on Cu are also beginning to have a technological impact. The use of Pb as a surfactant in promoting layer-by-layer growth in the fabrication of magnetic thin-film structures, for example, has certainly benefited from our current understanding of Pb-Cu surface interactions. Without prior knowledge of the properties of Pb on Cu, it is doubtful that one would have considered introducing Pb into Co-Cu multi-layer device fabrication. Likewise, our current understanding of the mechanisms by which Cu atoms deposited on top of a Pb surfactant layer make their way to the underlying surface and move to steps [7, 109] should help to define the conditions to optimise layer-by-layer growth and improve device performance. For soldering applications, coupling science to technology appears to be imminent. A key issue in understanding the dynamic processes involved in wetting and flow is what is the nature of wetting line. Is there intermixing of species? Does the spreading of a precursor monolayer restrict the flow? What are the roles of defects and impurities? The detailed information we now have on submonolayer atomic structures and how they evolve with

180 coverage and temperature will be required to sort out what is taking place at the wetting line during the spreading of solders on surfaces. Over the past 30 years, much has been learned about the properties of Pb on single-crystal surfaces of Cu. Yet, our understanding of the system is still far from complete. It seems that each time a new probe is used to study Pb on Cu, an exciting new discovery is made. For example, the first L E E D measurements showed an unusual progression in superstructure patterns with increasing Pb coverage. Later, STM and other surface probes identified these structures as surface alloys at certain coverages and overlayers at others. Now, LEEM is showing the remarkable dynamics that takes place as the different surface phases interact. Driven by both scientific curiosity and potential technological impact, there is little doubt that continued investigations of Pb/Cu surface alloying and de-alloying will be a fruitful area of scientific investigation for many years to come.

ACKNOWLEDGMENTS I am greatly indebted to Prof. Peter Varga of the Technical University of Vienna for Christian Nagl's Ph.D. thesis and for the STM images included in this article. I thank Norm Bartelt, Peter Feibelman and Brian Swartzentruber for many helpful discussions and a critical reading of this manuscript. Sandia National Laboratories is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the U. S. D O E under Contract DE-AC0494AL85000.

REFERENCES [ 1] See, A. Christensen, A. V. Ruban, P. Stoltze, K. W. Jacobsen, H. L. Skriver, J. K. Norskov, and F. Besenbacher, Phys. Rev. B 56 (1997) 5822 and references therein. [2] A. Cebollada, J. L. Martinez, J. M. Gallego, J. J. Demiguel, R. Miranda, S. Ferrer, F. Batallan, G. Fillion, and J. P. Rebouillat, Phys. Rev. B 39 (1989) 9726. [3] J. J. de Miguel, A. Cebollada, J. M. Gallego, R. Miranda, C. M. Schneider, P. Schuster, and J. Kirschner, Journal of Magnetism and Magnetic Materials 93 (1991) 1. [4] S. S. P. Parkin, R. Bhadra, and K. P. Roche, Phys. Rev. Lett. 66 (1991) 2152. [5] J. Camarero, L. Spendeler, G. Schmidt, K. Heinz, J. J. Demiguel, and R. Miranda, Phys. Rev. Lett. 73 (1994) 2448. [6] J. Camarero, T. Graf, J. J. deMiguel, R. Miranda, W. Kuch, M. Zharnikov, A. Dittschar, C. M. Schneider, and J. Kirschner, Phys. Rev. Lett. 76 (1996) 4428. [7] J. Camarero, J. Ferron, V. Cros, L. Gomez, A. L. V. deParga, J. M. Gallego, J. E. Prieto, J. J. deMiguel, and R. Miranda, Phys. Rev. Lett. 81 (1998) 850. [8] J. Camarero, J. J. deMiguel, T. Graf, R. Miranda, W. Kuch, M. Zharnikov, A. Dittschar, C. M. Schneider, and J. Kirschner, Surf. Sci. 404 (1998) 346. [9] J. Camarero, et al., Applied Physics a-Materials Science & Processing 69 (1999) 553. [10] W. Kuch, et al., Journal of Magnetism and Magnetic Materials 170 (1997) L13. [11] W. F. Egelhoff, et al., J. Appl. Phys. 80 (1996) 5183.

181 [ 12] F. G. Yost, F. M. Hosking, and D. R. Frear, The Mechanisc of Solder Alloys Wetting and Spreading (Van Nostrand Reinhold, New York, 1993). [ 13] F. G. Yost, "Fundamentals of Wetting and Spreading with Emphasis on Soldering", in The Metal Science of Joining, M.J. Cieslak, et al., Editor (The Minerals, Metal, and Materials Society, Warrendale, PA, 1992) p. 49. [14] H. N. Bassett, Beating Metals and Alloys (Edward Arnold, London, 1937). [15] P. D. Liddiard and P. G. Forrester, Foundry Trade Journal 92 (1952) 509. [16] P. G. Forrester, Metall. Rev. 5 (1960) 527. [17] G. C. Pratt, Int. Metall. Rev. 18 (1973) 62. [ 18] F. P. Bowden and D. Tabor, The Friction and Lubrication of Solids (Clarendon, Oxford, 1953). [19] J. P. Pathak and S. N. Tiwari, Wear 155 (1992) 37. [20] C. Nagl, O. Hailer, E. Platzgummer, M. Schmid, and P. Varga, Surf. Sci. 321 (1994) 237. [21] C. Nagl, M. Pinczolits, M. Schmid, and P. Varga, Phys. Rev. B 52 (1995) 16796. [22] C. Nagl, E. Platzgummer, O. Hailer, M. Schmid, and P. Varga, Surf. Sci. 333 (1995) 831. [23] C. Nagl, M. Schmid, and P. Varga, Surf. Sci. 369 (1996) 159. [24] C. Nagl, Ph.D. Thesis, Technischen Universitat Wien, 1996 [25] G. Ertl and J. Ktippers, Low Energy Electrons and Surface Chemistry, ed. H.F. Ebel (Verlag Chemie, Weinheim, 1974). [26] L. J. Clarke, Surface Crystallography: An Introduction to Low Energy Electron Diffraction (John Wiley, New York, 1985). [27] D. P. Woodruff and T. A. Delchar, Modem Techniques of Surface Science (Cambridge University Press, Cambridge, 1994). [28] M. A. van Hove, W. H. Weinberg, and C. M. Chang, Low Energy Electron Diffraction (Springer, Berlin, 1986). [29] M. Henzler, in Dynamical Phenomena at Surfaces, Interfaces and Superlattices, F. Nizzoli, K.H. Rieder, andR.F. Willis, Editor (Springer, Berlin, 1986) p. 14. [30] M. Henzler, Applied Physics a-Materials Science & Processing 34 (1984) 205. [31 ] J. A. Venables, in Introduction to Surface and Thin Film Processes(Cambridge University Press, Cambridge, 2000) p. 76. [32] S. Brennan, P. H. Fuoss, and P. Eisenberger, Phys. Rev. B 33 (1986) 3678. [33] G. Binnig, H. Rohrer, C. Gerber, and E. Weibel, Phys. Rev. Lett. 49 (1982) 57. [34] H.-J. Gtintherodt and R. Wiesendanger, ed. Scanning Tunneling Microscopy I Springer Series in Surface Science, ed. Vol. 20. 1992, Springer Verlag: Berlin, Heidelberg. [35] S. Robert, S. Gauthier, F. Bocquet, S. Rousset, J. L. Duvault, and J. Klein, Surf. Sci. 350 (1996) 136. [36] E. Bauer, Rep. Prog. Phys. 57 (1994) 895. [37] D. Farfas and K. H. Rieder, Rep. Prog. Phys. 61 (1998) 1575. [38] G. Comsa, Surf. Sci. 300 (1994) 77. [39] E. Hulpke, ed. Helium Atom Scattering from Surfaces Springer Series in Surface Science, ed. Vol. 27. 1992, Springer: Berlin. [40] B. Poelsema and G. Comsa, Scattering of Thermal Energy Atoms from Disordered Surfaces.ed. G. H6hler. Vol. 115. (Springer Verlag, Berlin, 1989). [41] K. W. Jacobsen, J. K. Norskov, and M. J. Puska, Phys. Rev. B 35 (1987) 7423. [42] M. S. Daw and M. I. Baskes, Phys. Rev. B 29 (1984) 6443. [43] P. Stoltze, Journal of Physics-Condensed Matter 6 (1994) 9495. [44] V. Rosato, M. Guillop6, and B. Legrand, Phil. Mag. A 59 (1989) 321. [45] B. Loisel, D. Gorse, V. Pontikis, and J. Lapujoulade, Surf. Sci. 221 (1989) 365. [46] W. Kohn and L. Sham, Phys. Rev. 140 (1965) Al133.

182 [47] R. O. Jones and O. Gunnarsson, Reviews of Modem Physics 61 (1989) 689. [48] P. J. Feibelman, private communication [49] N. C. Bartelt, private communication [50] G. Kresse and J. Furthmuller, Phys. Rev. B 54 (1996) 11169. [51] J. Henrion and G. E. Rhead, Surf. Sci. 29 (1972) 20. [52] E. Bauer, Appl. Surface Sci. 11/12 (1982) 479. [53] M. G. Barth/~s and G. E. Rhead, Surf. Sci. 80 (1979) 421. [54] C. Argile, M. G. Barthes-Labrousse, and G. E. Rhead, Surf. Sci. 138 (1984) 181. [55] C. Ocal, E. Martinez, and S. Ferrer, Surf. Sci. 136 (1984) 571. [56] G. Meyer, M. Michailov, and M. Henzler, Surf. Sci. 189 (1987) 1091. [57] G. Meyer, M. Michailov, and M. Henzler, Surf. Sci. 202 (1988) 125. [58] B. J. Hinch, C. Koziol, J. P. Toennies, and G. Zhang, Vacuum 42 (1991) 309. [59] B. H. Mtiller, T. Schmidt, and M. Henzler, Surf. Sci. 376 (1997) 123. [60] J. M. Kosterlitz and D. J. Thouless, J. Phys. C 6 (1973) 1181. [61] D. R. Nelson, Phys. Rev. B 18 (1978) 2318. [62] D. R. Nelson and B. I. Halperin, Phys. Rev. B 19 (1979) 2456. [63] B. I. Halperin and D. R. Nelson, Phys. Rev. Lett. 41 (1978) 121. [64] A. P. Young, Phys. Rev. B 19 (1979) 1855. [65] C. de Beauvais, Y. Girard, C. Perard, B. Croset, and B. Mutaftschiev, Surf. Sci. 367 (1996) 129. [66] J. Braun and J. P. Toennies, Surf. Sci. 368 (1996) 226. [67] S. MUller, J. E. Prieto, C. Rath, L. Hammer, R. Miranda, and K. Heinz, Journal of Physics-Condensed Matter 13 (2001) 1793. [68] R. Plass, J. A. Last, N. C. Bartelt, and G. L. Kellogg, Nature 412 (2001) 875. [69] H. M. McConnell, Annu. Rev. Phys. Chem. 42 (1991) 171. [70] M. M. Hurley and S. J. Singer, J. Phys. Chem. 96 (1992) 1951. [71] C. Sagui and R. C. Desai, Phys. Rev. E 49 (1994) 2225. [72] K. O. Ng and D. Vanderbilt, Phys. Rev. B 52 (1995) 2177. [73] H. Ibach, Surf. Sci. Rep. 29 (1997) 195. [74] R. Plass, N. C. Bartelt, and G. L. Kellogg, Journal of Physics: Condensed Mater in press (2002) [75] K. Arenhold, S. Surnev, P. Coenen, H. P. Bonzel, and P. Wynblatt, Surf. Sci. 417 (1998) Ll160. [76] S. Surnev, K. Arenhold, P. Coenen, B. Voigtlander, H. P. Bonzel, and P. Wynblatt, Journal of Vacuum Science & Technology a-Vacuum Surfaces and Films 16 (1998) 1059. [77] E. Z. Luo, Q. Cai, W. F. Chung, and M. S. Altman, Appl. Surf. Sci. 92 (1996) 331. [78] J. Moon, J. Lowekamp, P. Wynblatt, S. Garoff, and R. M. Suter, Surf. Sci. 488 (2001) 73. [79] W. H6sler and W. Moritz, Surf. Sci. 117 (1982) 196. [80] W. H6sler and W. Moritz, Surf. Sci. 175 (1986) 63. [81] W. H6sler, W. Moritz, E. Tamura, and R. Feder, Surf. Sci. 171 (1986) 55. [82] R. Plass and G. L. Kellogg, Surf. Sci. 470 (2000) 106. [83] A. Sepulveda and G. E. Rhead, Surf. Sci. 66 (1977) 436. [84] Y. Gauthier, W. Moritz, and W. H6sler, Surf. Sci. 345 (1996) 53. [85] E. Platzgummer, M. Borrell, C. Nagl, M. Schmid, and P. Varga, Surf. Sci. 413 (1998) 202. [86] A. Sanchez and S. Ferrer, Phys. Rev. B 39 (1989) 5778. [87] A. Sanchez, J. Ibanez, R. Miranda, and S. Ferrer, J. Appl. Phys. 61 (1987) 1239. [88] G. L. Kellogg and R. A. Plass, Surface Review and Letters 7 (2000) 649.

183 [89] F. Bocquet, Ph.D. Thesis, University of Paris VII, 1997 [90] S. Robert, Ph.D. Thesis, University Paris VII, 1995 [91] G. L. Kellogg and R. Plass, Surf. Sci. 465 (2000) L777. [92] G. Pr6vot, C. Cohen, J. M. Guigner, and D. Schmaus, Phys. Rev. B 61 (2000) 10393. [93] W. C. Marra, P. H. Fuoss, and P. E. Eisenberger, Phys. Rev. Lett. 49 (1982) 1169. [94] K. S. Liang, K. L. Damico, C. H. Lee, and E. Y. Sheu, Phys. Rev. Lett. 65 (1990) 3025. [95] C. H. Lee, E. Y. Sheu, K. S. Liang, K. I. Damico, and W. N. Unertl, Applied Physics aSolids and Surfaces 51 (1990) 191. [96] C. de Beauvais, D. Rouxel, B. Bigeard, and B. Mutaftschiev, Phys. Rev. B 44 (1991) 4024. [97] G. Pr6vot, C. Cohen, J. Moulin, and D. Schmaus, Surf. Sci. 421 (1999) 364. [98] G. Pr6vot, C. Cohen, D. Schmaus, and V. Pontikis, Surf. Sci. 459 (2000) 57. [99] M.-G. Barth~s-Labrousse, Surf. Sci. 177 (1986) 338. [100] J. Wollschlager, E. Z. Luo, and M. Henzler, Surf. Sci. 257 (1991) 274. [101] F. Hofmann, U. Svenson, and J. P. Toennies, Surf. Sci. 371 (1997) 169. [102] E. Lundgren, B. Stanka, G. Leonardelli, M. Schmid, and P. Varga, Phys. Rev. Lett. 82 (1999) 5068. [103] A. K. Schmid, N. C. Bartelt, and R. Q. Hwang, Science 290 (2000) 1561. [104] A. K. Schmid, J. C. Hamilton, N. C. Bartelt, and R. Q. Hwang, Phys. Rev. Lett. 77 (1996) 2977. [105] H. M. Polatoglou, M. Methfessel, and M. Scheffler, Phys. Rev. B 48 (1993) 1877. [106] B. S. Swartzentruber, private communication [ 107] R. van Gastel, E. Somfai, S. B. van Albada, W. van Saarloos, and J. W. M. Frenken, Phys. Rev. Lett. 86 (2001) 1562. [108] M. L. Grant, B. S. Swartzentruber, N. C. Bartelt, and J. B. Hannon, Phys. Rev. Lett. 86 (2001) 4588. [109] J. Ferr6n, L. G6mez, J. M. Gallego, J. Camarero, J. E. Prieto, V. Cros, A. L. V. deParga, J. J. deMiguel, and R. Miranda, Surf. Sci. 459 (2000) 135.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and AHoy Surfaces 184

D.P. Woodruff, (Editor)

Chapter 6

Surface alloys and alloy surfaces: the platinum-tin system Sylvia Speller ~ and Ugo Bardi b

Research Institute for Materials, University of Nijmegen, Toernooiveld 1, 6525 ED Nijmegen, The Netherlands b Dipartimento di Chimica, Universith di Firenze, Via G. Capponi 9 - 50014 Firenze, Italy, bardi @unifi.it 1. I N T R O D U C T I O N The effect of the surface structure and composition on the properties of metals and alloys has been recognized since the early times of metallurgy. Considering noble metals, the use of diffusion phenomena in bimetallic systems has a history that goes back to very early times in human history, when "fire gilding" techniques using mercury - gold alloys (amalgams) were developed. In recent times, noble metal alloys are extensively used in fields such as heterogeneous catalysis, where a large number of catalysts are multimetallic and contain at least one noble metal species [1, 2, 3]. The properties of the surface of noble metal alloys also find application in a variety of different fields, at first sight not related to surface science, for instance in thermal barrier systems where platinum aluminides are used as intermediate materials acting as corrosion barriers in high temperature environments [4]. The general field of the surface structural properties of bimetallic systems has been reviewed by Bardi [5] and more recently for the specific field of ordered systems by Vasiliev [6]. The theoretical factors leading to ordering and reconstruction at binary alloy surfaces have been recently reviewed by Treglia et al [7]. Another general review on the properties of bimetallic surfaces was reported by Rodriguez [8] and a general assessment of the factors leading to surface alloying has been published by Christensen et al [9]. The present paper is dedicated to a review of the state of the art of the knowledge of the surface composition and atomic level structure for platinum alloys and for one specific noble metal alloy, the Pt-Sn system, which has been extensively studied during the last few years.

185

In general, noble metals alloyed with metals of the left rows of the periodic table produce compounds with highly negative enthalpy of formation which very often form ordered intermetallic compounds, a fact already noted some time ago [10] and which more recently has been related to electron exchange among the species involved [11]. These compounds form a class of materials which gives rise to a wealth of surface phenomena. Among these phenomena we can cite the formation of ordered 2D ("two-dimensional") phases, "sandwich" surface layers, and long range undulations formed by lattice mismatch in turn due to the variations of the lattice constant resulting from composition variations. Platinum alloys and bimetallic systems are especially important in catalysis. Platinum-tin alloys are specifically important as catalytic electrode materials for direct methanation fuel cells (DMFC) [12] and as catalysts for naptha reforming and hydrogenation/dehydrogenation reaction of hydrocarbons, where the addition of tin to Pt supported catalysts decreases coke formation and increases lifetime and selectivity [ 13]. The Pt-Sn system has a highly negative enthalpy of alloying and gives rise to ordered bulk phases [14]. One of these phases (Pt3Sn) is cubic (symmetry Pm3m, AuCu3 type). This phase, as well as the phases obtained by depositing tin on pure platinum surfaces, have been examined by the full array of the available structure sensitive surface science techniques: diffraction methods (e.g ion scattering ALISS, low energy electron diffraction LEED and X-ray photoelectron diffraction, XPD) and scanning tunneling microscopy (STM). By using these techniques it has been possible to determine the atomic coordinates of the species present within the first few layers of the surface for different orientations and chemical compositions. It has been also possible to evidence complex phenomena of surface reconstruction and of formation of mesoscopic surface features. The present paper will summarize and review the results obtained and compare them with the data available for other ordering alloys. 2. METHODS

A rapid survey of the methods utilized for the study of binary alloys, and specifically for the Pt-Sn system will be reported here. In the present review, we consider only studies performed in conditions of ultra high vacuum (UHV), where bimetallic Pt-Sn surfaces are stable. It is known that in air and in general in the presence of oxygen at pressures larger than ca. 10 .6 Torr, tin alloyed with platinum tends to oxidize and de-alloy to form oxide phases, a phenomenon that will not be treated here. In all studies considered here the samples examined were either single crystal alloys of PtaSn composition prepared by melting and zone refining in

186

vacuum, or surfaces obtained depositing metallic tin on single crystal pure Pt substrates. The deposition was performed in UHV conditions using thermal evaporation sources. The substrates always needed a specific cleaning procedure in vacuum, which was obtained by noble gas ion bombardment (or "sputtering") for the purpose of removing adsorbed impurity and oxide layers and for flattening the surface, i. e. reducing the surface roughness. The sputtering treatment was normally followed by annealing, still in vacuum, for the complete smoothing and equilibration of the surface under study. In the study of the surface phases of the Pt-Sn system, as well as of other binary systems, a variety of experimental methods are available. Surface spectroscopies based on ion or electron interaction with the surface provide composition information with a depth resolution that can go from a few atomic layers (X-ray photoelectron spectroscopy, XPS and Auger electron spectroscopy, AES) to single atomic layer resolution. The latter can be obtained by low energy ion scattering (LEIS) a method which has been extensively used for the study ot the Pt-Sn system. Since surface spectroscopic methods are rather well known we will not review them in detail here. In terms of structural i n f o r m a t i o n , that is the determination of the atomic coordinates of at least some of the species in the surface region, several methods are available. We can class these methods in two main groups: scattering techniques (ion and electron) and scanning probe techniques. In several cases and specifically for the Pt-Sn system - a combination of these methods can provide the complete determination of the crystallographic parameters of the first 2-3 atomic planes of a surface. The surface phases observed in the Pt-Sn system are normally termed in reference to their in-plane periodicity, as observed most often by LEED. For a description of this method and for notes about how the surface crystallographic conventions need to be somewhat modified when applied to the field of alloy surfaces, see the appendix to the present paper. Here, we will briefly describe the surface structural methods extensively used for the Pt-Sn system. - E l e c t r o n d i f f r a c t i o n , Low energy electron diffraction (LEED) is the oldest and still the most widely applied crystallographic technique used for the determination of the structure of ordered solid surfaces. It is based on the diffraction process that a monoenergetic electron beam (ca. 20-500 eV) undergoes when interacting with a surface. On well ordered, single crystal surfaces long range interference of the scattered electrons leads to the formation of a diffraction pattern from which the surface unit mesh can be determined. More information can be extracted from the intensity of each spot as a function of the electron energy and comparing the results with a theoretical calculation. This procedure is defined in the literature as 'dynamical' or "quantitative" LEED analysis, or

187

as LEED I-V analysis, where "I-V" stands for "Intensity vs. Potential" (see, for instance, [15]). From a comparison of measured and calculated I-V curves it is possible to determine a model of the surface structure with an accuracy of the determination of interatomic distances that may be of the order of 0.01 /~. The sensitivity to composition in LEED is highest when the atomic species have significantly different scattering factors. In favorable cases (e.g. Pt-Ni, [ 16], the sensitivity to composition has been estimated to extend to a depth of approximately 5/~ from the surface. When the difference in atomic n u m b e r and hence in the electron density - is not so large, the sensitivity to composition worsens considerably. However, a composition profile for the first 2-3 atomic layers from the topmost surface could still be obtained for transition elements in adjacent rows of the periodic table, such as the Pt-Sn system considered here [ 17]. In LEED it is also possible to analyse the angular distribution of the intensity of single diffracted beams (spot profile analysis LEED, SPA-LEED) in order to obtain information on the domain structure of the surface under study, as it has been done for the Pt3Sn(111) surface [18]. Some studies by electron diffraction of the Pt-Sn system have also been performed by reflected high energy electron diffraction (RHEED) which uses a grazing incidence high energy electron beam. This method can provide information on structural features, such as mesoscopic multilayer "islands" which are difficult to study by LEED. - P h o t o e l e c t r o n d i f f r a c t i o n m e t h o d s , PD. These methods are based on the photon stimulated emission of core level electrons from the atomic species in the surface region. These electrons undergo scattering when interacting with the atoms around the emitter. Interference effects cause a variation in the intensity of electron emission as a function of angle or of energy. Measuring this variation it is possible to obtain information about the local structure around the emitting atom. A common set up for PD uses a conventional photon source in the soft X-ray domain (A1 Kc~ or Mg Kc~). In this version, the photoelectron intensities are measured for variable angles and the technique goes under the name of XPD (X-ray photoelectron diffraction). As the electrons examined are of relatively high energies (several hundreds of eV), the scattering process is dominated by what is called the 'forward focusing' effect [19]. This effect enhances the intensity of electrons emitted along directions that correspond to densely packed atomic rows and it may be exploited to obtain an immediate qualitative interpretation of the data. Calculations assuming varying degrees of approximation [19] can be used to fit XPD data to a detailed surface structural model. The technique has a larger probing depth than LEED and has the further advantage of being species sensitive, but it is scarcely sensitive to the structure of the topmost surface layer. XPD techniques have been extensively used for the Pt-Sn system, mainly for the study of phases obtained depositing tin on bulk

188

Pt substrates. All the results reported here for the Pt/Sn system were obtained using a multichannel hemispherical electron analyzer and a conventional, non monochromatized, Mg K a or A1 K a photon source. Unless otherwise specified, the experimental data were analyzed by means of the single scattering cluster spherical wave (SSC-SW) model. Ion scattering, IS. This term indicates a family of related techniques of which the relevant ones here are low energy ion scattering (LEIS) and alkali ion scattering spectroscopy (ALISS). The former, LEIS, is normally used for surface composition analysis with a depth resolution of the order of a single atomic layer, the latter (ALISS) can also provide structural information. Both LEIS [20, 18] and ALISS [21, 22, 23] have been extensively used in the study of Pt-Sn system. In ALISS a beam of low energy ions (typically Li +) is directed at the surface. The backscattered ions are analyzed in energy and angle. Alkali metals here have a definite advantage over noble gas ions (e.g. He + commonly used in LEIS) in the fact that the neutralization probability is much lower. Typically, for helium only about 1-10% of the ions is not neutralized, whereas for lithium the fraction is as high as 50-80% As obvious, the lower neutralization cross section leads to a much better signal to noise ratio. In ALISS (and in LEIS as well) the energy loss due to the elastic collision of the ion with species of the surface is characteristic of the mass of the target atoms and therefore provides compositional information. Structural information, that is information about the relative position of the scatterers at the surface can be obtained using the "shadow cone" associated with the target atom. Calculating the theoretical shadow cone at a given ion energy it is possible to determine a structural model of the surface by a polar angle scan [24]. The use of this method specifically for alloy surfaces and surface alloys has been reviewed by O'Connors et al [25]. - Scanning Probe M e t h o d s (SPM). Scanning probe techniques are based on the interaction of a sharp tip with the surface under study. This interaction may involve the passage of current by tunneling effect (STM, scanning tunneling microscopy), the measurement of the force between the tip and the surface (AFM, atomic force microscopy) or other physical phenomena. In many cases these techniques provide atomic resolution images of the topmost layer of the surface under study. Of these, STM is at present the most used for the study of metal and alloy surfaces, and atomic resolution has been attained in studies performed in vacuum on samples cleaned by standard treatments. For the studies discussed here on PtaSn surfaces an Omicron STM-1 system has been used housed in an UHV system. The base pressure in the STM chamber was kept in the 10 -11Torr range. With respect to surface roughness STM is rather 'touchy' in comparison to structural analysis methods such as LEED. LEED is a relatively long range -

189

method averaging over the coherence length (ca. 200 A) of the electrons whereas STM is a short range order method with atomic resolution. On a rough surface good tip-surface contact is hard to achieve and, in the worst case, tip damage is likely to occur. With respect to surface cleanliness STM is as sensitive as field ion microscopy (FIM), i. e. very low levels of impurities are detectable albeit not identifiable. Furthermore, rather low levels of impurities, in some cases below the detection limits of AES, lead to a deterioration of the STM tip which may pick up such impurities. The Pt-Sn samples used in the studies discussed here were found to give rise to no impurity segregation phenomena, hence to provide a relatively "easy" system. The final control of the sample quality is, however, the STM topography. All STM topographs reported here were measured at room temperature in the constant current mode. The lateral coordinates were calibrated using the atomically resolved surfaces of Si(111)(7x7) and Pt(ll0)(lx2). The vertical scale was calibrated at atomic steps on surfaces. In the case of the STM studies of low-index Pt3Sn surfaces the results obtained are in some respects textbook examples: compared to other alloys the chemical contrast observed by STM is large ( z p t - z s n ,-~ 50 pm). Much lower values (of the order of 10 pm) were observed with other Pt-alloys [26]. Also the sign of the corrugation varies with the alloy composition: Pt is measured as a protrusion in the Sn-Pt and Co-Pt systems, but as a depression in PtNi and PtRh. The chemical contrast can have different origins. In most cases a difference in the electronic density of states at the two species is responsible. Although chemical discrimination is achieved on very local scale the chemical contrast is partly disturbing because the STM "topography" is polluted by the electronic effects, a similar effect is observed on semiconductor surfaces, eg. Si(111). However, the variation of the contrast with gap voltage is relatively low with Pt-Sn surfaces. Another reason for chemical contrast can be a different interaction of the tip with the two elements on the surface. Additionally, this interaction can be mediated by a molecule or adsorbate at the tip which can be observed by means of sudden contrast changes during the scans. Since the characterization of the tip is a long-standing problem in STM the details of this kind of contrast formation are not known at present. However, the Pt-Sn contrast is never observed together with tip changes, such that one may attribute the contrast to the differences in the electronic density of states [27]. One is tempted to relate the tendency of a system to chemical order with the corrugation amplitude in the chemical contrast, because systems with lower chemical order, most often also show lower chemical contrast (e.g. Au3Pd, [28]) and vice versa. It is interesting to note that STM is not the only method leading to local chemical contrast. Also in field ion microscopy a local tunnel process is exploited, however in a much

190 larger field and chemical discrimination was obtained quite early with Pt-Co surfaces [29]. 3. THE PLATINUM-TIN SYSTEM Pt and Sn form highly exothermic bulk alloys. The phase diagram of the Pt-Sn system is described in detail in [14] (Fig. 1). Two stable intermetallic phases exist: Pt3Sn and PtSn. Of these, Pt3Sn has an enthalpy of formation of-50.2 KJ/mol, a melting point of 1675 K and a cubic face centered structure which is sometimes described in the literature using the metallurgical notation L12. The structure of Pt3Sn is the same as that of the "prototypical" ordered binary alloy, Cu3Au, the first binary alloy to have been studied for its surface properties in ultra-high vacuum conditions. The PtSn phase is hexagonal and ordered ( P 6 3 / m m c ) with an enthalpy of formation reported as -58.6 KJ/mol and a melting point of 1549 K. Other phases with a definite stoichiometry are reported to exist [14] but only the Pt3Sn phase has been studied in terms of surface properties. 1769 1700 r . . --,'

.

m

'~'"~ 1402 ",, ~ // 1370....|. ~

1300

%,..

liqui( uid 1276

..... i0 ,7

9o0

0 5OO , 100

I !

998 .... !

l

0.5

!

XSn

Figure 1" Phase diagram of Pt-Sn after Ref. [ 14].

|

231.96

191

3.1 Low index surfaces of the Pt3Sn alloy The possible "bulk termination" structures for the Pt3Sn ordered alloy are shown in Fig. 2. We summarize here the results obtained for the low index surfaces of the Pt3Sn alloy as a function of the annealing temperature. In the following, we will examine in detail the results for each face.

Figure 2: Structure of the Pt3Sn alloy and low index terminations, after Ref. [30].

3.1.1 Pt3Sn(lll) This surface is the most extensively studied in the Pt3Sn system. It shows interesting phenomena of bulk-surface equilibrium in the interplay of the two surface phases observed" the (V~ x x/3) R30~ the p(Zx2), with the former stable only in the absence of subsurface tin. Here, we will report in some detail the results of the studies performed. The first reports on the surface structure of Pt3Sn were based on qualitative LEED observations and on LEIS results [20, 30, 31, 32]. In these initial studies only the presence of the p(2x2) "bulk periodicity" phase was reported. The atomic structure of this surface was studied by Atrei et. al [ 17] by quantitative LEED and found to correspond indeed to a simple bulk termination model. The other possible termination, the (x/3 x V/-3) R30 ~ was reported and studied in detail by Atrei et al. [33] who also determined the atomic structure by means of quantitative LEED and found it to correspond to a single layer surface alloy.

192

Figure 3" Schematic overview of the surface morphology and the surface composition and altered layer composition on PtaSn( 111) as a function of the anneal temperature and history .The figure follows both the increasing temperature trajectory (left side, going down) and the cooling trajectory (right side, going up). (L) represents the average domain size of the dominating surface reconstruction, and (T) represents the average terrace size. The shape of the reconstructed domains is drawn arbitrarily. The numbers for composition reflect the outermost layer composition (top) and the altered layer composition (bottom). From Ref. [18]. The interplay of the two phases on the Pt3Sn(1 1 1) surface has been object of an extensive study carried out by Ceelen et al. [18] who used mainly a combination of LEIS and SPA-LEED, also carrying the sample at higher temperatures than those attained in the previous studies. A wealth of temperature dependent phenomena was observed in this study concerning bulk-surface chemical equilibrium, domain size variation and phase transitions. The main conclusions that can be drawn from these combined structural and compositional studies is that the ( v ~ x x/~) R30 ~ reconstruction is stabilized by the depletion of tin in the subsurface layers and that this depletion is caused by a the combination of sputtering and high temperature annealing (Fig. 3).

193

Figure 4: (17/~) 2 high-resolution STM image of the Pt3Sn(111) surface (Ut=0.9V, It=l.0nA) and hard-sphere model of the (x/-3 • V/3) R30 ~ structure, as derived by crystallographic LEED [34]. Due to some drift the image is slightly elongated in the vertical direction. Pt corresponds to regions of high tunnel current (bright areas), Sn corresponds to regions of low tunnel current (dark areas). From Ref. [35].

12,

,

, P t 3 S n (total)

10

Sn (local)

.........

P t (local)

..........

6 v

0 4

o ::::::::::::::::::::::::: I

I

-2

-1

I

0 E-EF(eV)

!i.: I

1

2

Figure 5: Total (per unit cell) and local (per space-filling atomic spheres of equal size at both Pt and Sn sites) densities of states of Pt3Sn, calculated by the tight-binding linear muffin-tin orbitals method. At positive (sample-) bias voltages the unoccupied states above E F are imaged in the STM. From Ref. [27].

194

A highly detailed picture of the structure of the Pt3Sn(111) surface in its two possible phases could be obtained by STM [35]. The STM studies were preceded and complemented by studies carried out in the same vacuum chamber by means of LEED, AES and RHEED. In these studies, the depletion of Sn in the near surface layers of the sample resulting by room temperature ion bombardment was confirmed by AES, in agreement with previous studies [34, 18]. Depending on the annealing temperature and annealing time the LEED patterns shows an increasing admixture of the p(2x2) pattern which is the final pattern after annealing to 1000 K. Here, the AES results confirmed the equilibration of the surface composition to reach the expected bulk value. The STM topographs taken in constant current mode of the (x/~ x x/~) R30 ~ surface confirm the the LEED analysis [33] (Fig. 4) [35]. The bright spots in the topograph are regions of high tunnel current and can be identified as Pt atoms. This interpretation is supported by calculations of the local density of states (LDOS) of Pt3Sn (Fig. 5)[27]. The LDOS of Sn electronic states is considerably lower at the Fermi-edge compared to the LDOS of Pt. Hence, since STM samples the LDOS essentially, Pt sites should give rise to a larger tunneling

Figure 6: STM-images of the (x/~ x v/3) R30 ~ structure on the Pt:~Sn(l 11) surface observed after annealing to 600K, a) Ut=0.1V, It=0.5nA. The inset shows a (530A) 2 terrace with the quasi-hexagonal honeycomb-network. The main image is a close-up view of the inset's lower left region, size (236,~) 2. Both the atomic structure and the height modulation due to the honeycomb-network are visible. The irregular line running from the lower left to the upper right corner is a domain wall separating two different (x/~ z v/3) R30~ defect in the upper right region. From Ref. [35].

It shows a

195

current. The contrast between Pt and Sn atoms in these topographs is independent of the tunneling conditions, for variable tunneling currents from 0.5 to 3.0 nA and gap voltages from +(0.1 to 0.9) V. A large area scan of the same (111) sample surface as in Fig. 6 reveals a further feature [35], i. e. the formation of the so called 'honeycomb' network (Fig. 6). The honeycomb network can be attributed to misfit dislocations due to the Sn depletion in the near surface region as it will be discussed more in detail later on. After the thermal equilibration obtained by annealing at high temperature a good p(2x2) LEED pattern develops and large terraces are observed with STM (Fig. 7). However, the terraces are mixed, i. e. some terraces are indeed p(2x2) but there remain a few (v/3 x x/3) R30 ~ patches with their honeycomb network. The p(2x2) areas are atomically flat. The p(2x2) areas are decorated with features too large to be single atoms. In a smaller area scan (Fig. 8) a height scan is taken across one of these white features. The height is approximately 2 and the width of the order of 10 ~ corresponding to a sizeable atomic cluster mono-atomic in height. The clusters are weakly bound, evidence for that is the dragging of clusters across the surface as seen in Fig. 8 for two examples at the left hand side (fuzzy double protrusions). If we assume the islands containing three atoms each and counting the islands leads to an estimate of the atomic

Figure 7" STM-image of Pt3Sn(111) of the mixed p(2x2) and (x/3x v/3) R30 ~ structure taken after annealing to 1000K, size (44/~) 2, Ut=0.9V, It= 1.0nA. The image has been differentiated to enhance contrast. The small adatom islands mark the p(2 x 2) domain whereas in the lower right corner (v/3 x x/~) R30 ~ areas with the honeycomb-network remain. The larger clusters may be due to residual contaminants, below the AES-detection-limit. From Ref. [35].

196

Figure 8" STM-image of the boundary region of a p(2x 2) (upper left half) and a (v/-3xv~) R30 ~ domain on Pt3Sn(lll) (lower right half, not atomically resolved), size (293~) 2, Ut=0.9V, It=l.0nA. The grayscale has been restricted to the flat surface, to enhance visibility of the p(2x2) atomic structure. The height of the adatom-fe ature on the p(2x2)- region equals approx. an atomic step (section profile). Note the two adatom feat ures in the left region, which have been moved by the tip. From Ref. [35].

concentration of 1%. Quantitative LEIS data [30, 18] indicate an increase of Sn surface concentration under the annealing conditions used for the scan reported in Fig. 8. Hence we conclude that the white features are Sn islands. They result from the surplus Sn present in the (V/-3 x x/3) R30~ which is Sn rich compared to the p(2x2) structure. Obviously not all of the excess Sn atoms, ca. 8% can diffuse back into the bulk of the crystal but are stranded on the surface. The identification of the clusters as Sn is supported by recent O adsorption experiments [36]. Upon oxygen adsorption the clusters disappear and become part of the Sn-O overlayer structure formed. The p(2x2) is the bulk truncated structure which develops under annealing via the (V/-3x V/-3) R30~ This phase transformation was also observed

197

in SPA-LEED studies [18]. Extended annealing annealing times (> 30 min) at 1100 K causes the p(2x2) structure to dominate completely. Higher temperature annealing causes an additional structure, defined as (2x2)', to appear. The development of the surface morphology and of the surface composition as derived from the LEIS and SPALEED study [18] is summarized in Fig. 3

3.1.2 Pt3Sn(O01) The Pt3Sn(001) surface shows a more complex behavior than that of the (111). In the early studies by qualitative LEED a c(2x2) bulk termination structure was observed, but also a a "streaked" LEED pattern was reported [20, 30, 32]. The structure of the c(2x2) phase was found by quantitative LEED to correspond to a simple bulk termination model, as expected [ 17]. The later STM studies [27] showed that the streaked pattern originates from mesoscopic features, "pyramids" on a relatively flat surface. In the initial reports on this surface, LEIS and qualitative LEED results for the Pt3Sn(100) surface were reported together with those for the Pt3Sn(111) one [30, 32]. The preparation of the (001) follows the identical recipe used for the (111). Ar + sputtering in cycles with annealing first at moderate temperatures followed by the 1000 K annealing to get a 'perfect' surface. An ordered, bulk termination "c(2x2)" phase was reported after extended annealing (see appendix for notes on the nomenclature used), but also "streaks", i.e. "extra" spots in the LEED pattern which move between the main spots when varying the electron beam energy, were reported to form after annealing at intermediate temperatures. The c(2x2) phase was analyzed by quantitative LEED [17] and a good agreement with the experimental data could be obtained by a simple bulk termination model where the uppermost layer is the "Sn-rich" plane, in agreement with the LEIS data. The LEED analysis indicated also an upward buckling (0.22-1-0.08~) for the tin atoms in the uppermost plane. The STM results confirmed the LEED ones and permitted to clarify some structural elements that the analysis based on electron diffraction could not solve [27]. The STM studies were performed on samples treated in the same way (i.e. ion bombardment and annealing) as the (111) surface. Also here, the STM data were preceded and complemented by parallel AES, LEIS and LEED studies carried out in the same vacuum chamber. Evidence for Sn depletion due to preferential sputtering was reported and after the annealing at 1000 K the Pt/Sn ratio was found to be close to the bulk value by AES. The nature of the facets observed in LEED is revealed immediately looking at the STM images (Fig. 9). The (100) surface is "decorated" with square or rectangular pyramids with a typical base width of 300 to 400 A. The height is in the range of 40 to 50 A. Occasionally, larger pyramids are found. The

198

Figure 9: STM images (a-c) and marble models (d) of the Pt:3Sn(001)-surface after low temperature annealing, a)Overview, scan width (1600A) 2, U:j=0.6V, b=l.0nA, b)( 104)-facet on the side of a pyramid near the top. Scan width

(100A) 2, U:/=0.2V, It=l.0nA.

c)(102)-facet on the

side of a pyramid near the base. Scan width (120,~) 2, U,j=0.4V, b=l.0nA, d)Marble models of the (104)-facet (left panel) and the (102)-facet (right panel). For better visibility the models correspond to a chemically ordered bulk (Pt atoms light grey, Sn atoms dark grey), whereas the real pyramids are substitutionally disordered in the bulk. The unit cells seen by STM are indicated. From Ref. [27].

199

pyramids are sitting on a flat plane, so the structure is not 'hill and valley' as on the Pt3Sn(110) (described in the next section). A closer look to the slopes of the pyramids (see Figure) allows the identification of the facets as { 104} and { 102} respectively. The orientation of the pyramids is parallel to the [ 100] and [010] directions of the surface, the facets are oriented perpendicular to these directions. Marble models (Fig. 9) show these facets. The {102} is, of course, identical to those found on (110). The difference to (110) is quite obvious, due to the fourfold symmetry, compared to the twofold symmetry of (110), the (001) surface can form { 10n} facets in two directions, which can lead to hillocks or holes with rectangular or squared shape.

Figure 10" STM-images of a pyramid on Pt3Sn(001). Left panel) Flat top. Two unit cells with a centered atom are indicated as examples. Usually no centered atom is visible. Scan width (80/k) 2, Ug=0.5V, It= 1.0hA. Right panel)Examples of 'beaded' triple rows on top of pyramids. The distance between the rows is mostly uniform but sometimes larger than shown here. Scan width (96 x 100/k) 2, Ug=0.9V, It= 1.0nA. From Ref. [27].

Between the pyramids the surface is not in the 'final' c(2x2) structural state, but shows a row structure parallel to the [100] and [010] directions, i.e. there are two domains of this structure (Fig. 9a). These row structures are also found on top of mostly the rectangular type pyramids (Fig. 10, right panel). The rows are made of three atomic rows, presumably Pt (see LDOS argument) with a local (100) symmetry, a square with one atom in the middle. The lateral distance of the atoms in direction of the rows is approximately 4 A. Occasionally one of the middle atoms is missing giving the rows a 'beaded' appearance. This reconstruction seems to be an obvious way of the (001) surface to handle the Sn deficiency in the s u r f a c e - quite different from the other two cases, (111) and

200

(110) respectively. The square pyramids tend to have a different top structure (Fig. 10, left panel). Here we observe the c(2x2) symmetry but with a surplus of Pt. As in the row structure the center of the 4 x 4/~2 square is occupied by a Pt atom, where in the annealed structure Sn has to be, and are, naturally, not visible as a protrusion with the STM.

Figure 11- Experimental (left panel) and schematic (right panel) RHEED pattern of the Pt3Sn(001)-surface after low temperature annealing. The main features are transmission spots lying on horizontal lines rather than Laue-circles. The Laue-circles are indicated in the schematic pattern. From Ref. [27].

Electron energy is 12keV, direction of incidence is along [100].

Before discussing more in detail the structure of the fully annealed surface we look at the RHEED results obtained from the pyramid decorated surface (Fig. 11). The RHEED pattern shows a lattice constant of the pyramids of 4.1 + 0.3 A in good agreement with the bulk Pt3Sn lattice constant. However, no half order spots are observed. Such spots only appear after the 1000 K anneal indicating long range chemical order as expected for flat Pt3Sn. We therefore conclude that the pyramids have no long range chemical order, as can be expected from the Sn deficit, i. e the pyramids are substitutionally disordered. It is an interesting notion that the height of the pyramids corresponds to approximately 10 atomic layers or the tin depletion range found by XPD in case of the P t ( l l l ) - S n surface [37]. At any rate, the pyramidal form of the stress relief is the most "aesthetic" of all three surfaces under study here. A question arises with respect to the { 102} facets: why are they favored? A simple argument can be brought forward in relation to the fact that no such facets are observed on (111). The (111) surface contains no [100] rows which are the closets packed rows on (102). We find [100] rows on (110) and (001), so (102) planes intersect with these two surfaces sharing parallel atomic chains. Against { 10n} facets with odd n speaks the difference of the surface energies of Pt and Sn of 2.7 J/m 2 and 0.62 J/m 2 respectively. This difference favors facets where

201

Figure 12: STM images of the PtaSn(001)-surface after high temperature annealing (1000K). Left panel)Overview, all steps are double steps running along the [100] and [010] directions. Scan width (1700/~) 2, Ug=0.9V, It=l.0nA. Right panel) Close up view showing the remaining monoatomic rows and the substrate.The apparent height of the rows is 1A. The square unit cell of the substrate shows no centered atoms, since only Pt is imaged (see text). Some defects are seen in the upper part of the image. Scan width (160~) 2, Ug= 10mV, It= 1.0nA. From Ref. [27].

Sn rows are exposed (Fig. 9e). The { 102} facets may be preferred with respect to { 10n} with even n > 2, because { 102} affords the steepest slope with the smallest steps. On {104} three row wide Pt steps occur which are less stable than terraces mixed with Sn (Fig. 9d). Note that due to the lack of chemical order within the pyramids the facets are not forced to even step height unlike the (100) and (110) surfaces of the well-annealed surface. Still, there remain open questions, for instance the balance between pyramid formation and the 'three row' reconstruction of the flat parts of (001), both structures being part of the effort to relieve the stress due to the Sn deficiency. The fully annealed PtaSn(100) surface (Fig. 12) shows in STM the expected c(2x2) structure determined by LEED. All steps observed are double steps, i. e. all terraces have the identical chemical composition and structure. The pyramids tend to 'melt' away during the annealing, no Oswald type ripening effects are seen, i. e. growth of larger pyramids paid for by the small ones. Large pyramids last longer than small once, real 'big' ones are still found after extended annealing periods. Assuming that only Pt is imaged there are no protrusions in any center of the basic squares of the structure as are found on top of the pyramids, i. e. no excess Pt. What remains are single, occasionally double, atomic rows the chemical nature of which can not be determined from STM imaging. If

202

we carry on with the LDOS argument these rows ought to consist of Pt atoms. If we make an analogous conclusion to the (111) case the atoms could be Sn, left over from the initial sputtering and annealing effects. Adsorption experiments may shed light on this open question.

3.1.3 PtaSn(11 O) The Pt3Sn(110) surface is especially interesting in view of the fact that few studies of this orientation have been reported for intermetallic systems and also in view of the fact that many fcc metals tend to undergo surface reconstruction, e. g. Au(110) and Pt(110) form the 'missing' row (lx2) structure [38] whereas Ir(110) forms a mesoscopic hill and valley structure with (331) facets [39, 35]. The first study by qualitative LEED on the Pt3Sn(110) was reported by Haner et al. [31]. A complex behavior was reported, with a 3x 1 phase forming during the initial stages annealing process, to be replaced later with a (lx2) structure (bulk truncation). The final, and apparently stable, pattern was described as "rhombic" or "quasi-hexagonal" with a periodicity in matrix notation

(1 0) 1/2 3/2

"

The LEIS results [20] showed that the outermost plane of this surface, as the other low index Pt3Sn surfaces, contain tin in concentrations larger than in the bulk. In a combined LEED, LEIS, AES and STM study the sputtering and annealing effects have been recently clarified [40]. The AES data resemble those of the(111) surface and after sputtering with 600 eV Ar ions the surface is Sn depleted. With increasing annealing temperature the Pt signal reduces and levels of at approximately 70 atomic %. The LEIS data in the same annealing range show a rather different behavior depending on the crystallographic direction too. After sputtering, the Pt concentration is approximately 50%. Annealing to 500 K causes an increase of the Pt concentration to 60% for both crystallographic directions, i. e. for scattering along [110] and [001] respectively. In the temperature range between 600 K and 900 K the surface becomes Sn rich, before, at 1000 K, an equilibration of the surface concentration at approximately 50 and 60% is reached for the two respective crystallographic directions. We can assume that at most the two outermost layers contribute to the Pt LEIS signal [41]. Therefore, when scattering along [110] two layers contribute to the Pt signal, 50% from the topmost layer and about 10% from the second layer. The lower signal from the second layer is due to the remaining depletion in the second layer (AES) and due to the enhanced neutralization of the He ions used for scattering from the second layer. For scattering along [001] the signal of the second layer is reduced by additional blocking. The LEED pattern for intermediate annealing contains (lxl), (2x l) and facet beams. The facet beams

203

show the proper 'wandering' when changing the electron beam energy. Sometimes these spots smear out into streaky features as reported earlier. The fully annealed surface is clearly (lx2). The structure of the surface and the identification of the "extra" beams observed in L E E D is straightforward when looking at the STM topographs (Fig. 13). What do we see? The main features are steps and/or facets running perpendicular to the [110] surface direction. There are 'up' and 'down' regions, that is the surface has a mesoscopic hill and valley structure (Fig. 13 b). From the height scan as in Fig. 13 b the slope of the facets can be determined as + 18.4 ~ with respect to the (110) plane. This angle is the crystallographic angle to (102) planes, i. e. the facets observed are { 102) with a distance of 1.5 a o / x / 2 - 4.24

Figure 13: STM image of the (102} facets on the Pt3Sn(110) surface after anneal to 715 K, 154 A,-0.15 V, 2.5 nA (a), height scan between A and B along [110] (b) and sphere model of a non-bulktruncated { 102} facet, that is in accordance with the data. From Ref. [40].

204

/~ between adjacent [00i] rows. The shortest possible period of the facets can be 8.5/~ which is e. g. observed in Fig. 13 b. Based on these findings we can construct a marble model of the faceted surface (Fig. 13 c). The hill and valley structure must be the result of the tensile stress induced in the (110) surface due to the depletion of Sn during sputtering. The stress relief is anisotropic with tipples perpendicular to the [001] direction thus creating { 102} facets. The orientation of these facets is in accordance with the LEED observations. As a consequence of the choice of these facets the [001] rows exposed on the facets are all monoatomic, i. e. either Pt or Sn. The exposure of the Sn atoms of these rows on then facets is the most plausible explanation for the Sn surplus observed by LEIS at the intermediate annealing stage. The corrugation of the hill and valley structure reaches approximately 4 to 5 atomic layers or about 1/3 of the Sn depleted region. As in the case of the (110) surface, higher temperature annealing causes the growth of larger terraces and the gradual disappearance of the { 102} facets (Fig. 14). The terraces are bordered by steps of 2.8 ~ in height or multiples thereof, i. e. composed of double steps (Fig. 14 c and d). Double steps are the consequence of the surface termination by only one type (as on Pt3Sn(100), Fig. 12, left panel). The slope of the steps in [00-1] direction is +18.4 ~ again. So we find here { 102} facets as in case of the intermediate annealing state in the hill and valley structure. The slope of the double steps in [110] direction is 22.5 ~ which is smaller than the expected 35 ~ with respect to the (110) planes for { 111 } facets. The { 111} facets are expected from the marble model constructed for the step structures observed (Fig. 14 e). We suspect the 35 ~ are too large an angle for the STM tip to follow. Additionally there is always the possibility of electronic smoothing due to the Smoluchowski effect. Further details resolved with smaller scanning areas of the step structures (Fig. 15) support the identification of the step directions and the interpretation using the marble model of Fig. 14 e. The atomic corrugation of a fourfold step is, for example, clearly resolved in Fig. 15 b. Since we never observe 'uneven' steps we have an additional strong argument for the termination of the crystal. Final support for the mixed termination comes from high resolution STM images with different bias voltages (Fig. 16 a, b). Knowing the orientation from the crystal and having the STM piezos calibrated it is obvious that the apparent surface lattice constant is larger along [110] directions than along [001] directions. As in case of the (111) and (100) surfaces we can safely assume that the bright spots in the STM images are Pt atoms. The contrast of these spots is also hardly dependent of the bias voltage applied between tip and sample. At negative bias, i. e. when probing the filled states, the Pt atoms appear brighter or larger than at positive bias. These findings are consistent with the interpretation of the STM images with

205

Figure 14: STM images of coexisting faceted structures and flat terraces on the Pt3Sn(110) surface, Tanne~l=920 K, 300 A, 0.4V, 0.8 nA (a) flat surface T~nne~l=920 K, 300 ~, 0.5 V, 0.8 nA (b) with height scan between A and B along [110] (c) and height scan between C and D along [001] (d). Sphere model of double steps (e). Note that the minifacets along the [110] double steps are { 111} oriented and the multiple minifacets along the [001] steps are { 102} oriented as found on the real surface. At the { 102} the structure model deviates from the bulktermination, in accordance with the data. From Ref. [40].

206

Figure 15: STM images of merging double steps on the Pt3Sn(110) surface, 500 ~, 0.45 V, 0.8 nA (a) and 200 ~, 0.40 V, 0.8 nA (b). The [001] steps form double, fourfold and sixfold steps whereas the [il0] steps are predominantly double. From Ref. [40].

Figure 16" STM images of the PhSn(110) surface, a) 120 ~,, 0.5 V, 0.8 hA. Pt atoms are visible as protrusions (open circles), Sn atoms are invisible (filled circles), b) 100 ~, +0.4 V (lower part) -0.4 V (upper part), 0.8 nA. The Pt atoms appear bigger when measuring the empty states (lower part). The contrast is higher when the filled states are measured (upper part). The big bump in the middle is presumably a contamination. From Ref. [40].

207

help of the LDOS data. In turn, this means that the holes are real Sn vacancies rather than an electronic tip effect. It is, furthermore, interesting to note that in case of the (111) surface we find a surplus of Sn on the well annealed surface, whereas the (110) tends to be depleted of Sn.

3.2 Surface alloys obtained depositing tin on platinum surfaces The term "surface alloy" is somewhat generic and may refer to a variety of different systems. Here, we apply it to those systems where ultra-thin metal layers (i.e. a few atomic layers thick) are deposited on a bulk metal surface and where the system is subsequently annealed in vacuum in order to obtain alloying in a surface region a few atoms thick. In these conditions it is possible to obtain single atomic layer binary phases, or multilayer surface alloy phases (also termed "epitaxial alloys" (for a general discussion of these surface alloys, see [5]. Relatively to the subject of the present paper, two Pt-Sn systems have been studied Sn-Pt(111) and Sn-Pt(100). The behavior and the structural properties of these systems will be discussed in detail in the following.

3.2.1 Sn-Pt(lll) The first study of Sn deposition on Pt(111) was reported by Paffet and Windham in 1989 [42] and a subsequent one on the same system was published by Campbell in 1990 [ 1]. In both studies, two LEED patterns were observed after annealing" a 2x2 and a (x/3 x x/3) R30 ~ Both superstructures were interpreted in terms of incorporation of the tin layer in the first platinum layer, but only a qualitative examination of the LEED pattern was performed. Subsequently the results of low energy alkali ion scattering spectroscopy ALISS [43, 21 ] could be quantitatively interpreted as due to ordered, single atomic layer surface alloys. The ion scattering results have been confirmed and expanded by a quantitative LEED study [34]. The atomic structure of both phases corresponds exactly to that of the topmost layer of the phases with the same periodicity observed on the on Pt3Sn(111). The LEED and ALISS results for the Sn/Pt(111) system were confirmed by a recent STM study reported by Batzill et al. [44]. Even though atomic resolution was not attained in this study (only the surface unit mesh could be observed), the results are closely comparable to the atomically resolved ones obtained on the Pt3Sn(111) surface [35]. The formation of multilayer surface alloys has also been investigated in the Sn-Pt(111) system, where Galeotti et al. [37] reported the formation of ordered, epitaxial alloyed Pt-Sn phases. The deposition of amounts of Sn up to 5 monolayers (ML) at room temperature led to disordered or anyway non-epitaxial tin films. Annealing the deposited films led to interdiffusion and to the formation of various alloy phases (Fig. 17). Alloying was detectable in XPS from the

208

Sn 3d5/2 XPD

(Azimuthal Angle, degrees)

Room T deposition 4-5 M L Sn

.................e

70

LEED

..: . . . . . . . lxl

Pt =k

9' . s ,

gk 9 ' "~='

Azimufh=l

'1~"

Angle

"1~'

(dog.)

Pt Sn alloy

~~::i:::):~":':ii:}i!"!"!;!!i!"" .

Annealed

---600K

.

.

.

.

.

.

.

.

.

!

.

~.~._.

--

~..~.

'-'1

2x2

,

. ~ . , ,

. . . .

Azimuthal

,..

angle

,,

~ ;

Cdeg.)

Pt Sn Annealed -- 800 K

8=70 ~ 3x 3 R30 ~

Pt

Azimuthal

Figure 17"

Main results obtained by combined LEED and XPD measurements on the

Sn/Pt(111) system. The left row is a schematic representation of the surface structure. The center row shows the XPD results for the Sn3ds/2 peak. The absence of oscillations in the pattern indicates either a disordered surface ("as deposited") or a single atomic layer (after high temperature annealing) where "forward scattering" effects cannot play a role. The right row shows the LEED results corresponding to the structural models described in the text. From [37].

shift of the Sn core level peaks 0.3 eV with respect to the "as deposited" Sn film. The formation of multilayer surface alloys could be clearly evidenced by XPD after depositing amounts of tin in the range of 3-5 MLs and annealing at temperatures ranging from 400 to 600 K. In LEED, this phase showed a (2x2) translational symmetry. Because of the forward focussing effect, the observation of strong oscillations in the XPD curves for Sn implies that in this phase a significant fraction of tin atoms are located below the surface. A further result that can be derived from the XPD data is that the Sn atoms are located in the same local environment of the Pt atoms. Furthermore, the similarity of the XPD results indicates that the near-surface structure of the Sn/Pt(100) system is the same as that of Pt3Sn(111) sample. The identity of the two phases is confirmed by calculations performed for a bulk truncation model of the Pt3Sn(111) surface. Since the LEED results clearly show long range ordering, it is possible

209

to arrive to a univocal model for the (2x2) phase that involves the formation on the surface of an ordered alloy multilayer of the same structure as that of the bulk, ordered Pt3Sn(111) intermetallic compound. The difference in the lattice parameter in Pt3Sn and pure Pt is small and the unit mesh for the ideal bulk truncated structure of the (111) plane of the alloy can be described as (2x2), indexing the diffraction spots with respect to the Pt(111) surface. After annealing the (2x2) multilayer surface alloy at 1000 K for several minutes, a (v/-3 • v~) R30~ pattern was observed again. In these conditions, the XPD azimuthal curves for the Sn 3d are flat, as those for the (V/-3x v/-3) R30~ obtained starting from Sn coverages of the order of 1 ML (Fig. 17). This result indicates that for extended annealing a '2-dimensional' alloy is formed again and that at this temperature tin atoms diffuse from the surface into the bulk to a depth that cannot be probed by the photoelectrons. This transformation is schematically described in Fig. 17, together with an illustration of the significant LEED and XPD results. The well characterized and stable surface phases observed on the Sn-Pt(111) have provided researchers in the chemisorption and catalysis field with a substrate of great interest for studying the properties of bimetallic interfaces. Simple "probe" gases such as CO have been studied after adsorption on this system [45] as well as a variety of organic molecules such as acetylene [46], cyclohexane and benzene [47, 48], butane and isobutane [49], methanol, ethanol and water [50]. Several surface reactions of the above gases were also studied.

3.2.2 Sn-Pt(100) The first study on this system was published by Paffett and Whindham [42] together with the results for the Sn/Pt(111). After deposition of an amount of Sn of ca. 3 ML and subsequent annealing, two periodicities were observed in LEED" a c(2x2) and a (3v/2 x ~/~) R45 ~ These surfaces were studied from a quantitative structural viewpoint by Li and Koel [23] by ALISS. The experimental setup and the methods used was similar to that used for the Sn/Pt(111) system. Here, the clean Pt substrate surface starts reconstructed, showing in lead the well known "streaks" which have been indexed in terms of a (5x20) periodicity. The formation of a c(2x2) phase was observed after depositing 0.5 ML of tin and annealing in the range 400-700 K. In this range the ALISS polar angle scan was interpreted in terms of an overlayer of tin atoms, i.e. not a surface alloy. At higher temperatures (T>ca. 750 K) considerable structural changes were observed. In this case, the ALISS results clearly indicated the formation of a substitutional Pt-Sn alloy of the same structure as the bulk termination of Pt3Sn(100). In this phase, buckling of the Sn atoms was foind to be very small (0.17-0.22 A). The data do indicate the presence of this substitutional alloy in

210

the topmost surface layer, however, evidence was observed for the presence of tin in the deeper layers. The alloyed c(2x2)-Sn structure on Pt(100) was found to be unstable and to quickly transform into the (3x/2 x v/-2) R45~ which was found to be stable up to annealing temperatures of 1000 K. It was not possible to propose a complete model for this phase, however the ALISS results remained very similar to those for the c(2x2) phase. It was therefore suggested that the the local structure of the (3x/2 x v/2) R45~ the same as that of the c(2x2). Indeed the c(2x2) periodicity can also be written in an equivalent manner as (x/2 • v/2) R45 ~ The "extra" 3x/2 periodicity observed for the Sn-Pt(100) surface can be due to a specific step arrangement or periodic domains of pure Sn atoms every three lattice spacing along the [ 100] azimuth.It appears that the formation of the (3x/2 x V/-2) R45~ accompanied by the disappearance of tin atoms from the subsurface region. No STM results have been reported so far for the Sn-Pt(100) system so it is not possible at present to know if the metastable pyramids observed on the Pt3Sn(100) surface are present also on the surface alloy. Chemisorption and catalysis studies are also lacking for the Sn-Pt(100) system which has not been found as attractive as the Sn-Pt(111) because of the lack of stability of the c(2x2) phase and for the difficulty of quantitatively characterizing the (3x/2 x x/~) R45~ 4. DISCUSSION Among ordered bimetallic systems, the Pt-Sn one can be considered at present as the most in-depth studied not only for its surface structural properties, but also for its reactivity and catalytic properties. A comparable detailed knowledge exists only for a few other cases, among platinum alloys we can cite the Ni-Pt and Co-Pt systems, examined for their catalytic properties and the Pt-Ti system studied for their electrocatalytic properties [5]. Sparse data relative to the surface properties of several other Pt alloys exist (e.g. Fe3Pt and Cu3Pt [3] and Pt3Mn [51 ]. All these data available pertain to fcc phases either random substitutional or ordered compounds. Data exist also for other cubic ordered alloys which are isostructural with the Pt3Sn compound, e.g. Ni3A1 [52, 53] and AuzPd [28] and finally the Au-Cu system, which has been object of interest as the "prototypical" L12 or P m 3 m ordered system in the Cu3Au composition [54,551. If we consider also the availability of theoretical studies on the surface segregation and equilibration phenomena [7] the Pt-Sn system can be seen as the most thoroughly characterized in a whole class of alloys, that of "ordering" al-

211

loys, i.e. alloys which tend tend to form ordered bulk intermetallic compounds with a highly negative enthalpy of formation. We'll see in the following that the surface structural behavior of alloys in this class appears to be similar for the known cases, but that the Pt-Sn system shows a complex series of surface reconstructions not observed on other alloy systems. Regarding the high bonding energy of some Pt alloys system we note that already in the 60s Leo Brewer [ 10] had put forward a simple model (sometimes referred to as the "Engel-Brewer model") which could be used for a qualitative prediction of the strength of the intermetallic bond. The Brewer model predicted charge transfer between different metallic species in reason of the different electronegativity. It is well known how ionic compounds (e.g. NaC1) form by the reaction of elements of the far left and far right row of the periodic table. Something analogous takes place with the transition elements, with the elements of the IVB and VB rows forming highly exothermic alloys with elements of the VIIIB row (e.g. Pt-Ti, Pt-Zr, etc). Conversely, alloys of elements of the same row tend to have small enthalpy of formation and therefore to form random solid solutions or compounds which have a low temperature of order-disorder transition. A classic example here is the Cu3Au alloy which has a transition temperature of 663 K. Indeed this transition has been the main motive of interest which led to the first LEED surface studies on a bimetallic system to be performed on this compound, which can be by now considered a "classic" [56, 57, 54, 58, 59, 55, 60]. On the contrary Pt3Ti, for instance, is an ordered compound in the whole range of temperatures below the melting point and has a highly negative enthalpy of formation o f - 19.5 Kcal/mole [61 ]. In recent times, the electronic structure of transition metal alloys has been studied with more advanced methods. The basic Engel-Brewer model has been confirmed when the intermetallic bond has been correlated to a shift in the overlayer local d-electron band and a simultaneous dip in the noble metal (e.g. Pt) d-electron local density of states (LDOS) at the Fermi level. These models, however, do not directly apply to the platinum-tin system since tin is not a transition element. However, tin is an electropositive element and so, according to the Engel Brewer model, the properties of Pt-Sn alloys in terms of enthalpy of formation could be expected to be comparable to those of the strongly exothermic alloys of platinum. It has been found that in Pt- non transition metal alloys, the same dip in the LDOS observed in Pt-transition metal allos is caused by the hybridization of d-electrons with the p-electron band [62]. According to Pick [63] the electronic structure of noble metal/non transition metal alloys is therefore very similar to that of noble metal/transition metal alloys. This electronic structure leads to a series of consequences, not the least interesting one the change in reactivity towards adsorbates, a subject which will not reviewed

212

here for lack of space, but which has been studied in detail for the Pt-Sn system.

4.1 Surface atomic structure of bulk Pt3Sn alloys In most - but by no means all - studies of binary alloy systems reported so far, qualitative LEED data indicate that the surface unit mesh corresponds to what expected from truncation of the bulk lattice [5]. The observation of the "expected" pattern in LEED in itself is no proof that the surface atomic structure is actually the bulk truncation one. Furthermore, in the case of ordered intermetallic compounds, the 'bulk termination' model is not normally univocal since the planes stacked along a specific crystallographic direction do not necessarily have all the same composition. In the case of fcc Cu3Au (L 12) ordered compounds (Fig. 1) all the crystallographic directions, except the (111 / have an ...ABAB... stacking with - for instance in the case of Pt3Sn - a plane of pure Pt alternating to a plane of composition PtSn. Both terminations correspond to 'bulk truncation'; and in both cases the composition of the outermost plane is different from the average one of the bulk. The experimental observations by LEIS of a number of bimetallic systems have shown that in the preferred termination may be either "mixed" or "pure" depending on the chemical species present. Quantitative surface crystallographic methods (especially dynamic LEED) have confirmed the LEIS results. The cases where the atomic structure of the topmost layer corresponds to that of a "mixed" bulk crystallographic plane For the L 12 phase (fcc, Cu3Au type) has been reported, among other cases, for instance for Cu3Au(100) ([56, 57, 54, 58, 59, 55, 60] and Ni3Al(100) [52, 53] systems which have the same structure and termination as the Pt3Sn(100) [17, 27]. In all these cases, obviously, the presence of different degrees of outward relaxation ("buckling") for the different chemical species present has been reported. Other bulk isostructural compounds show a "pure" termination instead of a mixed one. This behavior was observed in the case of the Pt3Ti(100) surface, a result obtained independently from LEIS [64] and LEED data [65]. Also the Pt3Ti(111) surface was found to be enriched in Pt [64, 66]. This behavior, which is in sharp contrast with that of the isostructural Pt3Sn case, may be related to the difference in the relative sizes of the atomic species involved (Ti and Sn). It may also be worth to consider the possibility that it could be attributed to differences in bulk composition. The Pt3Ti sample used in the crystallographic studies [65] had a nominal 3:1 P t / T i atomic ratio, but there are elements suggesting that a sequel of successive treatments of ion bombardment and annealing led to a depletion in titanium of the selvedge region [67]. The irreversible depletion in the light element in the surface of a bulk alloy as the effect of extended ion bombardment was reported for NiAI(100) [68],

213 Table 1: Summary of the structures observed on Pt3Sn surfaces after annealing at moderate and high temperature 600 K - 800 K 1000 K -1100 K (111) (x/3x x/3) R30~ mesoscopic sub- p(2• adatom islands surface dislocation network (001) multiple row structure, pyramids bor- c(2 • 2) , double steps, single atomic ad dered by { 102 } and { 104} facets rows (110) hill-and-valley-like structure with { 102} (2 • 1), double steps, holes at Sn positions facets

Pts0Fe20(111) [69] and PtsoCo2o(lO0) [70]. Theoretical calculations based on the broken bond model [71] indicate that Pt segregation in Pt3Ti is expected for an excess of platinum in the bulk with respect to the 3:1 stoichiometric ratio. Hence, the actual bulk composition, as opposed to the nominal one, may have an effect on the surface composition and structure of an alloy. For the case of Pt3Sn, there are elements indicating that the "as prepared" Pt3Sn single crystal samples used in the surface studies reported here were slightly "Sn-rich" in comparison to the nominal composition, for instance the observation of excess tin on the topmost layer of the Pt3Sn(111) surface which appeared as "white spots" in the STM scans [35, 40]. The effect of the several cycles of ion bombardment and annealing may have progressively reduced this excess of tin. Although these phenomena are an indication of a complex behavior of the Pt3Sn system (and in general of bimetallic alloy materials), their effect on the topmost surface composition should not be overestimated. Indeed in the case of systems obtained by depositing tin on pure platinum substrates, the excess of platinum is an obvious condition. Nevertheless, two-dimensional surface phases containing tin have been observed (as it will be discussed more in detail later) indicating that there are chemical factors which lead to stabilize tin in the outermost layer independently of the bulk composition. These factors, conversely, appear to destabilize the presence in the topmost layer of such elements as Ti, Co, and Ni. Summarizing, the "mixed" termination is by no means to be taken for granted in all Pt-M system. It does, however, seem to be the general case for the Pt-Sn system. Although the observation of bulk truncation phases in the Pt3Sn(hkl) case is not surprising, the wealth and complexity of the reconstructions observed is remarkable, as well as the interplay of the factors which lead to the transitions observed among them. A list of the phases observed for the Pt3Sn system is provided in Table 1. Surface reconstruction, that is a surface mesh that is not the same as the bulk mesh along the surface plane, has been observed also for other alloys. The random substitutional Pt alloys Pts0Co20 (001) [72], and PtsoNis0 (100) [ 16] show a "pseudo-hexagonal" reconstruction similar (but not identical)

214

to the one observed on pure platinum surfaces and by some other pure transition metals [73]. In both cases the composition of the outermost layer appears to be pure platinum. Conversely, small amounts of deposited metals (e.g. zirconium on Pt(100) [67]) destabilize the Pt reconstruction, reverting the surface to the "expected" 1x 1 structure. Reconstructions similar to the ones observed on the Pt-Sn system have been observed in some other cases of binary alloys. For instance for Cu-AI(111) [74] the quantitative LEED analysis [75, 76] showed that the topmost layer is a mixed plane of the same structure of the reconstructed Pt3Sn(111) surface. Also a (x/~ x v ~ ) R30 ~ has been observed for the (111) surface of the random substitutional A1-6.5at% Li alloy, [77] (Quantitative crystallographic data not available). Nothing comparable to the "pyramidal" structures observed by STM on the Pt3Sn(100) system has been reported so far for other alloy systems. The theoretical interpretation of these results is still in progress but the main elements leading to stabilize some reconstructions seem to be well established. Foiles [78] used the EAM method to study the stability of surface ordered phases low index surfaces of dilute Cu-Au (111) alloys. The calculations indicate a domain of Au bulk concentrations (from ca. 0.001 at% to 5 at%) that produce surface segregation and the formation of stable p(2x2) and (x/~ x x/~) R30~ alloys respectively on the (100) and (111) planes. The theory in this case seems to quantify intuitive considerations based on two facts: i) that the Au-Cu bond is energetically favorable and ii) that Au has a larger radius than copper. These two conditions lead to different tendencies; the first to have Au stay in the bulk to maximize the number of Cu neighbors, the second to squeeze Au atoms from the bulk to the surface where outward relaxation can be energetically favorable. The interplay of the two tendencies leads to an intermediate condition where Au atoms form a single layer phase where they increase the intermetallic bond distance by relaxing outwards. These consideration can help to understand why this kind of reconstruction occurs for dilute, random substitutional alloys. The case of Pt-Sn is more complex and whenever the concentration of the minority metal in the bulk is not negligible, and especially in the case of ordered intermetallic compounds, it is necessary to consider that heterogenous bonds occur in the interaction of the first layer with the underlying one. Consider the Pt3Sn(111) case, here the highest packing periodicity in the topmost plane, the (v/-3 x v/-3) R30 ~ see the structure shown in Fig. 4, leads necessarily to a number of Sn-Sn nearest neighbors between the topmost and of the second layer (assuming that the latter would maintain the expected bulk structure). Since Sn-Sn bonds are less energetically favorable than Sn-Pt ones, the forma-

215

tion of the (x/~x x/~) R30~ should be unfavorable and indeed it is observed on PtaSn only when the substrate is strongly depleted in tin as the result of a ion bombardment [33]. As a rule of thumb, the segregating species is the material with the lower melting point or cohesion. Obviously, the surface is much more driven out of the equilibrium situation when the preferentially sputtered species is identical with the segregating one, as in Pt-Sn alloys. Then, the segregation can take place only after the composition has been restored, i.e. at a quite late stage, at high annealing temperature. This gives rise to compromise structural stages with the formation of several metastable structures. These metastable states are characterized by stress compensation features (dislocations, pyramids, and ripples) because the altered composition of the surface region leads to reduced lattice constants. Indeed, a quantitative study by means of Monte Carlo simulations lead to the conclusion that the PtaSn( 111)-(v~x v~) R30~ is a consequence of a restricted, local equilibrium in the surface region [79]. Such behavior is in contrast to alloy surfaces where the segregating and the preferentially sputtered species differ, e.g. Au75Pd25 [28]. A thermal equilibrium can be even completely out of reach if the sublimation energies differ largely. The latter was observed with Fe-A1 alloy surfaces where at the temperature that is necessary to restore the surface composition severe evaporation of A1 takes place [80]. With Pt-Sn surfaces no significant evidence of Sn for sublimation has observed: However, on the PtaSn(110) surface mobile monolayer-deep depressions have been observed at Sn-positions in the topography which are most likely vacancies left after sublimation of Sn atoms. The high cohesion of heterogeneous bonds prevents that Pt atoms jump in these vacancies at Sn positions. Although these simple considerations help to frame in a general logic the behavior of these bimetallic surface, there are at present no such simple models to explain the more complex "mesoscopic" reconstructions, such as the "pyramids" observed on Pt3Sn(100) or the hill and valley structure observed on PtaSn(110). These phenomena are obviously related to the tendency of the system to relax in-plane stress, in turn resulting from the different atomic radius of the elements involved in the presence of concentration gradients. This relaxation appears to take place on the (111) oriented plane simply by an outward relaxation of the tin atoms. On the other two low index surfaces, instead, it takes a more complex route leading to reconstruction phenomena (pyramids on the (100) and "hill and valley" on the (110)) which are so far unique to the Pt-Sn system.

4.2 Defects and disorder on Pt3Sn alloy surfaces The field of atomic scale defects on alloy surfaces is one that has recently received a strong inpulse by STM studies. Nevertheless, also classic crystallo-

216

graphic techniques can be used to study defects. Ordered step arrays of alloy surfaces can be studied by LEED (PtaTi(510) [81, 82], by LEIS (A1Ni(111), [83], and it has been shown how it is possible to detect a stacking fault by XPD during the growth of a metal overlayer (Ag deposited on Pd(111) [84]). Quantitative LEED crystallography has also been used to study the effect of ion bombardment on the composition of alloy surfaces (the case of FeAI(100), [85]). However, STM has the unique capability of imaging defects in real space. So it is possible, for instance, to observe the step distribution and height on the surface (one of the first reports in this field was on the NiAI(111) surface [86]). Later on [87, 88] it was observed by STM that ion bombardment of the Pt25Ni75(111) surface leads to the formation of a pattern of shallow ditches (some 0.2-0.5 A deep) that have been attributed to the dislocations generated by the lattice mismatch of the top layers and the bulk ones. The top layers are enriched in Pt by ion bombardment and hence have a different lattice constant. These dislocations in sputtered alloys may provide diffusion pipes for implanted atoms to reach the surface. Diffusion of metal atoms in the surface region at relatively low temperatures has however been proven to be related to the presence of defects, such as the "pinholes" observed by STM at the Co/Cu(100) interface [89] The study of the Pt3Sn(111) surface by STM has expanded and clarified this area. Here the mesoscopic "honeycomb" structure reported in [35] is something that finds a parallel only in the case of the Pt-Ni system [87, 88]. In both cases, the surface develops mesoscopic features which are due to lattice dislocations in turn due to the composition gradient in the direction perpendicular to the surface. In the case of the Pt3Sn(111) system, the depletion in the subsurface which is associated with the formation of the (V~ • v/3) R30~ leads to a lattice constant in that region which can be expected to approach the Pt bulk lattice constant of 3.92 A. This value is lower than the Pt3Sn bulk lattice constant of 4.00 A. This mismatch of the lattice constants causes tensile stress which is obviously relieved by misfit dislocations. Additionally, stress relief may be the cause of the slight buckling of the Sn atoms on (111) as observed the quantitative LEED analysis [34]. A direct determination of the Burgers vector of the dislocation is not possible since none of them reach the surface. However from the directions of the walls of the honeycombs along 112 we conclude that the Burgers vectors must be parallel to the surface ~1 (110). Good alignment of the walls of the network is obtained after annealing slightly above 600 K. The half-width of the walls as obtained from a corresponding cross section is of the order of 30 to 40 A. From this width the depth of the dislocation cores can be estimated to be approximately 15 layers [90]. 15 layers is also the range of Pt enrichment found in previous LEED studies [33] so the results of the

217

different methods used, LEIS, LEED, XPD, AES and STM, lead to a consistent interpretation of the metastable phase of the Pt3Sn(111) surface. At present the case of Pt3Sn and Pt~Ni~_l are the only two cases reported of STM observations of misfit dislocations resulting in mesoscopic surface features, however it is certain possible that new cases will be discovered as different alloy systems are studied.

4.3 Multilayer and single layer surface alloys Both single layer and multilayer surface alloys can be prepared in the PtSn system by depositing ultra-thin Sn layers and annealing in vacuum to obtain equilibration. The first case where structural data were reported about a similar phenomenon was for the A1/Ni system [91 ], where the formation of an eptiaxial Ni3A1 layer was observed when depositing A1 on Ni(100). Other case known where this occurs are the Au-Cu(100) [92] and the Pd-Cu(001) [93] systems. In other cases, such as Co-Pt(111) [94], only multilayer surface alloys are known to form, although alloying appears to be limited to the outermost 2 surface layers only. So far, the structure of most of these surface phases turned out to be the one that maximizes the number of heterogeneous pairwise interactions. Qualitatively, the expectation is that such phases would be stabilized by a strong intermetallic bond and hence, exist for elements that form ordered bulk alloys, or anyway alloys with a negative enthalpy of formation. The general explanation for the existence of single layer surface alloys appears to lie in the balance of tendencies that are usually opposite" that of maximizing the number of energetically favorable intermetallic bonds, and that of minimizing surface energy. The maximization of the number of bonds, alone, would necessarily lead to long range bulk diffusion and to the formation of a dilute bulk alloy. However, placing the minority component within the topmost surface layer only may be energetically favorable in several ways; for instance relieving strain effects due to size differences. As already discussed for the case of diluted bulk alloys, the stability of single layer alloy phases can be theoretically predicted, for instance by the EAM theory [78] or by the TBIM approach [95, 96, 97, 98]. In the case of the Cu/Au(111) system the EAM theory predicts that a gold atom placed within the first atomic layer in the c(2x2) phase is 0.14 eV more stable than as an adatom. The stability of the W(100) c(2x2)-Cu phase has been explained in terms of the energetic contribution of the lattice strain of the overlayer to the overall energy of the system [99]. The case of the incorporation of gold atoms in the Ni (110) plane (Fig. 7) could be theoretically explained in the framework of the EMT theory (Effective Medium Theory) [100], that indicates that the surface energy of the Ni(110) surface is lower when Au is incorporated into the first layer. It could be shown that the

218

cohesive energy of the system has a minimum when Au is surrounded by a low number of Ni neighbors (6-7), as it occurs in a flat surface layer. Similar factors are at play in the case of the Pt-Sn system as discussed by [7]. The stable phase at the Sn/Pt(111) interface after extended thermal treatment at high temperature is the (V/-3 x v/-3) R30 ~ layer surface alloy. Its stability can be explained in terms of the surface free energy and the atomic size of Sn and Pt, tin is expected to segregate onto the surface of platinum. On the other hand, a high surface concentration of tin is not a stable situation due to the reduction of the number of favorable Pt-Sn bonds. The single layer Pt(111)(v/-3 x v/-3) R30~ phase results from the balance of these two contributions, since this phase maximises both the surface concentration of Sn (1/3 of a ML) and the number of Pt-Sn bonds (6 Pt first nearest neighbors). The formation of the ( ~ x v/-3) R30 ~ alloy by annealing at 1000 K of indicates that diffusion of Sn into the bulk is effective at such a temperature and that equilibrium can be achieved. The conditions of formation of this alloy on the pure Pt(111) surface parallel exactly those of the Pt3Sn(111) compound. In the latter case, the (x/~ x V~) R30~ can be prepared only after a depletion in tin of the subsurface layers is obtained by ion bombardment so that, eventually, the two systems have the same composition and structure over the first few atomic layers from the surface. In terms of multilayer surface alloys, the deposition of multi-atomic layers of tin on a platinum substrate can lead to the formation of multi-layer surface alloys. The observation of a well defined periodicity in LEED for the Sn/Pt(111) system and the parallel indications of the presence of tin in the subsurface in amount corresponding to approximately 25 at% indicates that we have a true ordered compound which extends for several atomic layers [37]. This behavior appears to be similar to that of the Co-Pt system [94], although in the case of Sn-Pt it was not possible to evidence the same kind of sharp alloy/substrate interface reported for Co/Pt(111). The possibility of obtaining a compound with negative enthalpy of formation is surely a factor favoring the formation of a multilayer homogeneous alloy in this sytem however, in this as in other systems, kinetic factors may be more important, and in particular factors related to the presence of grain boundaries in the deposited film. The bulk diffusion vacancy mechanism at the temperatures at which multilayer alloy phases have been observed to form are orders of magnitude too slow to cause a significant deep layer diffusion. For instance, the diffusion depth for the case of the Fe-Cu system was estimated as 10 .3 A in the conditions in which a multilayer surface alloy was observed [101]. Egelhoff [102, 103] found that surface mixing in the Cu/Ni system occurs rapidly at temperatures for which the bulk diffusion coefficients lead to predict parameters such as one atomic "hop" (site exchange)

219

every 1010 years. Clearly, other mechanisms are at play in this area and the only possible conclusion is that diffusion proceeds in these conditions from the substrate into the deposit, exploiting surface defects and imperfection in the deposited film. Substrate diffusion into the deposit has already been experimentally observed for relatively thick In films on Ag [104]. In 1989 Egelhoff [103] predicted that for very thin deposited layers such diffusion would occur via "pits" on the surface, and such pits have been indeed recently observed by STM in the Co/Cu system [89, 105]. ICISS has also provided evidence that diffusion in the Fe/Cu(100) system occurs only in a very small fraction of the area of the surface [ 106]. Although the diffusion coefficient of Sn in Pt is not known, considering the bulk diffusion coefficient of other metals in platinum Sn diffusion into the Pt substrate should be negligible in a such temperature range [37], so that the mechanism of alloying appears to be dominated here, too, by surface diffusion of Pt atoms through defects of the Sn film. However, the mechanisms of diffusion in these systems, as well as in the Pt-Sn one is something that still needs to be studied in detail. 5. C O N C L U S I O N The present review has attempted to summarize the experimental observations available for the surface structure of the Pt-Sn system for both single crystal Pt3Sn samples and for systems obtained depositing and thermally equilibrating tin onto pure Pt surfaces. In many ways, the results obtained for this alloy indicate structural phenomena comparable with those available for other bimetallic system. Several of these results can be explained in terms of well known properties of compounds with a negative enthalpy of formation, which tend to form structures which maximize the number of heterogeneous pairwise interactions. At the same time, other factors related at least in part to atomic size tend to influence the surface structure by stabilizing or de-stabilizing mixed topmost layer. In the case of Pt-Sn these factors lead to the formation of stable and well characterized surface phases, such as the ( ~ x v/-3) R30~ which can be obtained starting from either single crystal Pt3Sn or from the deposition of Sn on pure Pt(111). This phase is one of the best known and understood "model" for gas-solid interactions which examine how chemisorption, gas phaser catalytic and electrocatalytic reactions can be affected by sterical factors, site availability, and at the same time by electronic deinsity variations resulting from the intermetallic bond. In this area, the behavior of the Pt-Sn system sharply contrast with that of other platinum -metal systems (with the second metal, for instance, Co, Ni, Ti) where there exists a strong tendency for platinum to segregate and to form what may be called "skin" alloy surfaces [5].

220

Although simple, flat surface phases are observed, the Pt-Sn system is also remarkable for the complexity of mesoscopic phenomena observed, such as the "pyramids" formed on the Pt3Sn(100) surface. These phenomena are obviously related to the high surface energy of the system, which is possibily the intermetallic compound with the largest enthalpy of formation studied so far for its surface properties. No comparable phenomena have been observed in other bimetallic systems so far. The field of alloy surfaces has undergone remarkable advances in the last few years, in large part pushed by the application of atomic resolution realspace imaging techniques. The wealth of observations on the Pt-Sn system can be considered as a starting point for a more complete assessment of this vast field. APPENDIX: NOTES ON NOMENCLATURE Some nomenclature problems general to alloy surfaces and specific for the PtSn system will be briefly reviewed in this section, a more detailed discussion can be found in [5]. The first point to be considered is the form of writing of the alloy composition. In metallurgy it is customary to write the elements of an alloy in order of decreasing atomic fraction. This custom contrasts with the recommendation for intermetallic compounds of the international union for pure and applied chemistry[ 107]. In the IUPAC rules, elements in intermetallic compound should be ordered in the same way as in inorganic compounds, that is following columns in the periodic table from the bottom up, and rows from left to right. This rule is somewhat cumbersome to follow and it is almost never used for alloys. In most cases (and in the present paper) the metallurgic convention is used and it is probably the best way, that is writing, "Pt3Sn" rather than the IUPAC style "SnPt3" Elements in "systems" in general can be written simply in alphabetic order (e.g. "the Pt-Sn system"). Another nomenclature problem is related to the definition of surface periodicities. In surface studies the periodicity of the surface unit mesh should be described using the Wood notation [108]. According to this notation, a surface phase is described according to its periodicity referred to that of the substrate. That is, a surface phase which has a unit mesh twice larger than that of the substrate and aligned in the same direction is defined as a "2x2" In the case of binary alloys, when an ordered intermetallic compound (such as Pt3Sn) is cut along a surface plane, the resulting 'bulk truncation' or 'expected' periodicity should be described as a l xl according to the Wood convention. Nevertheless this is practically never done in the literature for binary alloy systems; it is preferred instead to index the surface mesh in terms of a superlattice mesh referred

221

to one of the two pure components (platinum in the case of Pt3Sn). This notation is formally incorrect since what is described as a 'surface mesh' is in reality the periodicity of the bulk lattice, not that of the surface or selvedge. Nevertheless, the 'superperiodicity' notation is almost impossible to avoid in order to describe, for instance, the order-disorder (2x2++ l x l ) transition that occurs in Cu3Au. Otherwise one would have to modify the periodicity notation for the overlayer depending on the order/disorder state of the substrate which would lead to considerable confusion when comparing, for instance, identical structures formed starting from intermetallic bulk compounds or instead by deposition of tin metal on a bulk platinum substrate. REFERENCES [ 1] Campbell C. T., Ann. Rev. Phys. Chem. 41 (1990) 775. [2] Nieuwenhuys B.E., in The Chemical Physics of Solid Surfaces and Heterogeneous Catalysis, Eds. King, DA and Woodruff, DP), Elsevier, Amsterdam, 1993, Vol. 6, pp. 185-224. [3] Wandelt K., in "Surface Science, Principles and Applications, Springer Proceedings in Physics, Vol 73, R.E How, R.N. Lamb and K. Wandelt eds, Springer Verlag, Berlin Heidelberg, 1993, 209-226. [4] Prasad B.D., Sankaran S.N., Wiedemann K. E., Glass D.E. Thin Solid Films 345 (1999) 255. [5] Bardi U., Rep. Prog. Phys. 57 (1994) 939. [6] Vasiliev M.A. J Phys. D Appl. Phys. 30 (1997) 3037. [7] Treglia, G., Legrand B., Ducastelle F., Saul A., Gallis G., Meunier I., Mottet C., Senhaji A. Computational Materials Science 15 (1999) 196. [8] Rodriguez, J. A., Surf. Sci. Rep. 24 (1996) 223. [9] Christensen A., Ruban V., Stolze E, Jacobsen K.W., Skriver H.L., Norskov J.K., Besenbacher F., Phys. Rev. B 56 (1997) 5822. [10] Brewer L. in "Phase stability in Metals and alloys" Eds. E Rudman, J,. Jaffee and R.I. Jafee, McGraw Hill New York, 1967. [ 11] Hayer E. Bros J.R, J. Alloys Com. 220 (1995) 193. [12] Abdel-Rahim M.A., Khalil M.W., Hassan H.B., J. Appl. Electrochemistry 30(10) (2000) 1151. [ 13] Burch R., J. Catal. 71(1981) 348. [ 14] Anres, P, Gaine-Escard, M.. Bros, J.P., Hayer E., Journal of alloys and compounds 280 (1998) 158. [15] Van Hove M. A., Weinberg W. H., Chan C. M., Springer Series in Surface Science, Berlin -Heidelberg 1986. [16] Gauthier Y., Baudoing R., in "Surface segregation and related phenomena, Eds. P.A. Dowben, A. Miller, CrC press, Boca Raton, 1990, p. 169. [ 17] Atrei A., Bardi U., Rovida G., Torrini M., Zanazzi E., Ross P. N., Phys. Rev. B46 (1992) 1649. [ 18] Ceelen, W.C.A.N., Denier van der GonA.W., Rejime M.A., Brongersma H.H., Spolveri I., Atrei A., Bardi U., Surf. Sci. 406 (1998) 264.

222 [19] Fadley C. S., Prog. Surf. Sci. 16 (1984) 275. [20] Bardi U., Pedocchi L., Rovida G., Haner A. H., Ross R N., In Fundamental aspects of heterogeneous catalysis; H.H. Brongersma, R.A. van Santen eds, Plenum Press, New York, 1991, 393. [21] Overbury S. H., Ku Y., Phys. Rev. B46 (1992) 7868. [22] Overbury S. H., van den Oetalaar R. J. A., Zehner D. M., Phys. Rev. B 48 (1993) 1718. [23] Li Y., Koel B.E., Surf. Sci. 330 (1995) 193. [24] Niehus H., Achete C., Surf. Sci. 289 (1993) 19. [25] O'Connor D.J., Shen Y.G., Zur Muhlen E., Zhu L., Macdonald R.J., Surf. Rev. Letters 3 (1996) 1847. [26] Varga E, Schmid M., Appl.Surf.Sci. 141 (1999) 287. [27] Hoheisel M., Kuntze J., Speller S., Postnikov A., Heiland W., Spolveri I., U. Bardi, Phys. Rev. B 60 (1999) 2033. [28] Aschoff M., Speller S., Kuntze J., Heiland W., Platzgummer E., Schmid M., Varga E, Baretzky B., Surf. Sci. 415 (1998) L1051. [29] Tsong T.T., Mtiller E.W., Journ. Appl. Phys. 38 (1967) 3531. [30] Haner A. N., Ross E N., Bardi U., Catalysis Lett. 8 (1991) 1. [31] Haner A. N., Ross E N., Bardi U., 1991 The structure of Surfaces III; S.Y. Tong, M.A. Van Hove, K. Takayanagi, X.D. Xie eds, Springer Verlag, Berlin Heidelberg, 24 276281. [32] Haner A. N., Ross E N., Bardi U., Surf. Sci. 249 (1991) 15. [33] Atrei A., Bardi U., Zanazzi E., Rovida G., Kasamura H., Kudo M., J. Phys. Condens. Matter 5 (1993) L207. [34] Atrei A., Bardi U., Wu J. X., Zanazzi E., Rovida G., Surf. Sci. 290 (1993) 286. [35] Kuntze J, Speller S., Heiland W, Atrei, A., Spolveri I., Bardi U., Phys. Rev. B 58 (1998) R 16005. [36] Hoheisel M., Speller S., Heiland W., Atrei A., Bardi U., Rovida G., submitted to Phys. Rev. B. [37] Galleotti M., Atrei A., Bardi U., Rovida G., Torrini M., Surf. Sci. 313 (1994) 349. [38] Maclaren J. M., Surface Crystallographic Handbook (Dordrecht) (1987). [39] Koch R., Borbonus M., Haase O., Rieder K.H., Phys. Rev. Lett. 67 (1991) 3416. [40] Hoheisel M., Speller S., Kuntze J., Atrei A., Bardi U., Heiland W., Phys. Rev. B63 (2001) 245403. [41 ] Niehus H., Heiland W., Taglauer E, Surf. Sci. Rep. 17 (1993) 213. [42] Paffett M.T, Windham R.G., Surf. Sci 208 (1989) 34. [43] Overbury S. H., Mullins D. R., Paffett M. E, Koel B. E., Surf. Sci. 254 (1991) 45. [44] Batzill M., Beck D.E., Koel B.E., Surf. Sci. 466 (2000) L821. [45] Xu C., Koel B.E., Surf. Sci. Lett. 304 (1994) L505. [46] Xu C., Peck J.W., Koel B.E., J. Am. Chem Soc. 115 (1993) 80. [47] Xu C., Tsai Y.L., Koel B.E., J. Phys. Chem. 98 (1994) 585. [48] Xu C., Koel B.E., Surf. Sci. 304 (1994) 249. [49] Xu C., B.E. Koel, Paffett M.T., Langmuir 10 (1994) 166. [50] Panja C., Saliba N., Koel B.E. Surf. Sci. 395 (1998) 248. [51 ] Gallego S., Ocal C., Mendez J., Torrelles X., Soria E, Surf. Sci. 482-485 (2001) 1303. [52] Sondericker D., Jona E, Marcus P. M., Phys. Rev. B 33 (1986) 900. [53] Sondericker D., Jona E, Marcus P. M., Phys. Rev. B34 (1986) 6770. [54] Potter H. C., Blakely J. M., J. Vac. Sci. Technol 12 (1975) 635.

223 [55] Nakanishi S., Kawamoto K., Fukuoka N., Umezawa K., Surf. Sci. 261 (1992) 342. [56] Sundaram V. S., Farrel B., Alben R. S., Alben,, Robertson W. D., Phys. Rev. Lett. 31 (1973) 1136. [57] Sundaram V. S., Alben R. S., Robertson W. D., Surf. Sci. 46 (1974) 653. [58] Buck T. M., Wheatley G. H., Marchut L., Phys. Rev. Lett. 51 (1983) 43. [59] McRae E. G., Malic R. A., Surf. Sci. 148 (1984) 551. [60] Stuck A., Osterwalder J., Schlapback L., Poon H. C., Surf. Sci. 251/252 (1991) 670. [61] Meschter P.J., Worrell W.L., Metall. Trans. A 7(1976) 299. [62] Pick S., J Phys. Cond. Matter 5 (1993) 6581. [63] Pick S., Surf. Sci. 436 (1999) 220. [64] Paul J., Cameron S. D., Dwyer D. J., Hoffmann E M., Surf. Sci. 177 (1986) 121. [65] Atrei A., Pedocchi L., Bardi U., Rovida G., Torrini M., Zanazzi E., Van Hove M. A., Ross E N., Surf. Sci. 261 (1992) 64. [66] Chen W., Paul J. A. K., Barbieri A., VAn Hove M. A., Cameron S., Dwyer D. J., J. Phys, Condens. Matter 5 (1993) 4585. [67] Bardi U., Ross E N., Somorjai G. A., J. Vac. Sci. Technol. A2 (1984) 40. [68] Mullins D. R., Overbury S. H., Surf. Sci. 199 (1988) 141. [69] Beccat E, Gauthier Y., Baudoing-Savois R., Bertolini J. C., Surf. Sci. 238 (1990) 105. [70] Bardi U., Atrei A., Rovida G., Cortigiani B., Rovida G., Torrini M., Surf. Sci. 282 (1993) L365. [71 ] Spencer M. S., Surf. Sci. 145 (1984) 145. [72] Bardi U., Atrei A., Ross P. N., Zanazzi E., Rovida G., Surf. Sci. 211/212 (1989) 441. [73] Van Hove M. A., Koestner R. J., Stair E C., Biberian J. E, Kesmodel L. I., Bartos I., Somorjai G. A., Surf. Sci. 103 (1981) 189. [74] Baird R. J., Eberhardt W., J. Vac. Sci Technol. 18 (1981) 538. [75] Baird R. J., Ogletree D. F., Van Hove M. A., Somorjai G. A., Bull. Am. Phys. Soc. 29 (1984) 222. [76] Baird R. J., Ogletree D. F., Van Hove M. A., Somorjai G. A., Surf. Sci. 165 (1986) 345. [77] Esposto E J., Zhang C. S., Norton E R., Timsit R. S., Surf. Sci. 290 (1993) 93. [78] Foiles S. M., Surf. Sci. 191 (1987) 329. [79] Creemers C., Helfensteyn S., Appl. Surf. Sci. 167, (2000) 216. [80] Meier W., Blum V., Hammer L., Heinz K., J Phys. Cond. Mat. 13, (2001) 1781. [81] Bardi U., Santucci A., Rovida G., Ross E N., Proceedings of the ICSOS-2, Springer Verlag, Berlin, Heidelberg, New York, London, Paris Tokyo, 1987, 147-151. [82] Bardi U., Ross P. N., Rovida G., Surf. Sci. Lett. 205 (1988) L798. [83]. Overbury S. H., Mullins D. R., Wendelken J. E, Surf. Sci. 236 (1990) 122. [84] Eisenhut B., Stober J., Rangelov G., Fauster T., Phys. Rev. 47 (1993) 12980. [85] Wang C. E, Jona E, Gleason N. R., Strongin D. R., Marcus E M., Surf. Sci. 298 (1993) 114. [86] Niehus H., Raunau W., Besoche K., Spitzl R., Comsa G., Surf. Sci Lett. 225 (1990) L8. [87] Schmid M., Biedermann A., Stadler H., Slama C., Varga E, Appl. Phys. A55 (1992) 468. [88] Schmid M., Biedermann A., Stadler H., Varga E, Phys. Rev. Lett. 69 (1992) 925. [89] Schmid A.K., Atlan D., Itoh H., Heinrich B., Ichinokawa T., Kirschner J., Phys. Rev. B. 48 (1993) 2855. [90] Stalder R., Sirringhaus H., Onda N., von K~inel H., Appl. Phys. Lett. 59 (1991) 1960.

224 [91] Lu S. H., Tian D., Wang Z. Q., Li Y. S., Jona E, Marcus R M., Solid state Comm. 67 (1988) 325. [92] Naumovich D., Stuck A., Greber T., Osterwalder J., Schlapbach L., Surf. Sci. 269/270 (1992) 719. [93] Lu S. H., Wang Z. Q., Wu S. C., Lok C. K. C., Quinn J., Li Y. S., Tian D., Jona E, Marcus P. M., Phys. Rev. B37 (1988) 4296. [94] Galeotti M., Atrei A., Bardi U., Cortigiani B., Rovida G., Torrini M., Surf. Sci. 297 ( 1993) 202. [95] Teraoka Y., Surf. Sci. 232 (1990) 193. [96] Teraoka Y., Surf. Sci 235 (1990) 208. [97] Teraoka Y., Surf. Sci 235 (1990) 249. [98] Teraoka Y., Surf. Sci 238 (1990) L453. [99] Singh D., Krakauer H., Surf. Sci. 216 (1989) 303. [100] Jacobsen K. W., Norskov J. K., Puska M. J., Phys. Rev. B35 (1987) 7423. [ 101 ] Thomassen J., May F., Feldmann B., Wuttig M., Ibach H., Phys. Rev. Lett. 69 (1992) 3831. [ 102] Egelhoff W.F., J. Vac. Sci. Technol. A7 (1989) 2060. [103] Egelhoff W.F., Steigerwald D.A., J. Vac. Sci. Technol. A7 (1989) 2167. [104] Wesche R., Fink R., Krausch G., Platzer R., Voigt J., Wohrmann U., Schatz G., Thin Solid Films 190 (1990) 153. [ 105] Girgel J., Kirschner J., Landgraf J., Shen J., Woltersdorf J., Surf. Sci. 310 (1994) 1. [106] Detzel Th., Memmel N., Phys. Rev. B 49 (1994) 5599. [107] IUPAC, J. Am. Chem. Soc. 82 (1960) 5525. [108] E.A.Wood, Crystal Orientation Manual, Columbia University Press, New York and London, 1963.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces D.P. Woodruff, (Editor)

225

Chapter 7

Alkali-Aluminum surface alloys David L. A d a m s a and Jesper N. Andersen b

aInstitute of Physics and Astronomy, University of Aarhus DK-8000 Aarhus C, Denmark bDepartment of Synchrotron Radiation Research, Institute of Physics, Lund University, S-223 62 Lund, Sweden 1. I N T R O D U C T I O N 1.1. B a c k g r o u n d A little more than a decade ago at the beginning of the 1990's the title of the present chapter would have seemed to be a contradiction in terms. At that time, the adsorption of alkali metals on aluminium surfaces was the experimentalist's emulation of the theorist's vision of the adsorption of alkalinium on jellium. In this vision, a reconstruction of the substrate is by construction impossible. In fact, with a few exceptions, the adsorption of alkali metals on metal surfaces in general was presumed to occur without serious perturbation of the substrate. Notwithstanding this incomplete understanding of the structural nature of alkali metal adsorption, important theoretical treatments by Gurney (of the adsorption of the alkaline-earth metals) [1], and later by Lang [2] and Lang and Williams [3], still stand as landmarks in the development of the theory of adsorption. History, of course, repeats itself, so that alkali metal adsorption on aluminium surfaces is once again an active testing ground for theoretical models. However, whereas the earlier interest in alkali metal adsorption on aluminum was largely in its use as a model system for understanding the surface electronic structure, current interest is now extended to investigations of the geometrical structure of quite complicated surfaces. In the present chapter we describe a number of unique surface alloys formed between alkali metals and aluminum. (We refer the reader who is interested in alkali metal adsorption on metal surfaces in general to an excellent review [4].) The discovery of these surface alloys in the early 1990's came as a complete surprise, not least because the alkali metals, with the exception of Li, are im-

226

miscible with aluminum. This unexpected finding spurred a large experimental and theoretical interest in surface alloys, and led in particular to the identification and characterisation of a number of alkali-aluminum surface alloys. The investigation of these alloys has significantly improved our understanding of the mechanisms governing alloy formation on surfaces. In particular, it has become evident that surface alloys can form between materials that are immiscible in the bulk. The characterisation of these alloy structures resulted from a very fruitful interplay between experiment and theory. Experimental investigations provided accurate and detailed determinations of the geometrical structure, and the simple electronic structure of aluminum and the alkali metals made feasible a full optimisation of the geometrical structure of the surface alloys by ab initio calculations. In addition to the basic interest in these novel surface alloys, the studies of A1-Li surface alloys are of relevance to the understanding of the properties of A1-Li bulk alloys, which are of significant technological interest because of their unusual mechanical properties. Li-dilute A1-Li alloys are used in the aerospace industry because of their high strength and low density compared to other A1 binary alloys [5, 6]. The desirable properties of A1-Li alloys are thought to be related to the formation of microcrystalline precipitates of the metastable A13Li phase, which contribute to the stiffness of Li-dilute alloys both by pinning down defects and by their own large Young's modulus. The relationships between the surface and bulk alloys are of considerable interest for this system. 1.2. Present work In this Chapter, we focus on alkali-aluminum surface alloys where the geometrical structure has been determined in detail. As can be seen from Table 1, which contains a list of the adsorbed phases formed by adsorption of alkali metals on aluminium surfaces, this limitation is not a serious restriction, since studies exist for a quite a number of low index aluminum surfaces and alkali metals. Although most of the structures of the phases listed in Table 1 have been determined by low energy electron diffraction (LEED), the crucial, first observation of substitutional adsorption for alkali-aluminium systems was made in a combined surface extended x-ray fine structure (SEXAFS) and density functional theory (DFT) study of the A I ( 1 1 1 ) - ( , / 3 x x / 3 ) R 3 0 ~ phase formed by adsorption of 1/3 ML Na at room temperature by Schmalz et al [7] in 1991. The structure of the AI(111)-(4 x 4 ) - N a phase was also determined by SEXAFS. The alkali-aluminum surface alloys are formed by deposition of the alkali metal onto a low index aluminum surface. If the deposition is made at low temperatures (typically below 140 K), the alkali metal simply forms an adsorbed

227 Table 1 Adsorbed phases formed by adsorption of alkali metals on aluminium surfaces at different coverages (0) and substrate temperatures (T). 'v@ is short for (~/3 x ,/~)R30 ~ '2~/~' is short for (2V~ x 2V~)R30 ~ and 'V~' is short for (~/5 x ~/5)R26.6 ~ Structures have been determined for all the phases listed in the table, except for the A1(111)-(2~/3 x 2~/3R30~ Al(100)-c(2 x 2)-K and AI(110)-c(4 x 2)-Rb phases. Surface 0 (ML) T (K) Li Na K Rb Cs A1(111) 1/4 1/3 1/3 9/16 1/2

100 300 100 300

1/5 1/2 1/2 1

250 100 300 400

1/2 3/4

300 300

(2x2) q~

v/3 (4x4) (2 x2)

~/3

c(2x2) c(2x2)

c(2x2)

v/3

(2x2) 45 2~/3

AI(IO0)

c(2x2) c(2x2)

A1(110) c(2x2)

c(2x2) (4 x 1)

c(4x2)

layer and no alloying occurs, although significant perturbation of the substrate structure can occur even after adsorption at low temperature. However, if such an adsorbed alkali layer is annealed, or if the alkali deposition is performed at higher temperatures, alloying does occur and an Al-alkali surface alloy forms. Quite surprisingly, the temperatures needed for such alloying are at or even below room temperature. This indicates, firstly, that the activation barriers for alloying are small and, secondly, that the mobility of the alkali and aluminum atoms are large even at relatively low temperature, since the formation of some of the surface alloy structures requires a considerable mass transport across the surface. Since the bulk of the studies reviewed here were carried out in the authors' laboratories using LEED and core-level photoemission spectroscopy (CLS), we preface our discussion with short accounts of the LEED and CLS methods in our implementation. These are illustrated by brief accounts of their application to the clean AI(111), (100), and (110) surfaces, which serve to define the starting points for the studies of alkali metal adsorption on these surfaces.

228

2. EXPERIMENTAL METHODS 2.1. L E E D measurements The LEED measurements were carried out in a Vacuum Generators (VG)/zmetal ultra-high vacuum chamber, fitted with an Omicron reverse-view LEED optics. LEED intensity measurements carried out before 1997 were made using a video-LEED system [8] based on a video camera with an image intensifier. The control program included an automatic gain control to circumvent the limited (5-6 bits) dynamic range of such cameras caused by their large dark current [9]. More recent measurements [10] were made with much greater precision using a Princeton slow-scan, Peltier-cooled CCD with an intensity resolution of 16 bits. In both systems the digital image of the LEED pattern on the fluorescent screen of the LEED optics at a given electron energy was analysed to obtain the intensities of the diffracted beams, by summing pixel intensities in the diffracted spots in the pattern. 2.2. L E E D analysis The determination of surface structure was carried out by comparison of experimental LEED intensity-energy spectra with spectra calculated using the dynamical theory of LEED, using computer programs [11, 12] derived from the layer-doubling and combined-space programs of Pendry [13] and of Van Hove and Tong [14]. Atomic scattering matrices for A1 and the alkali metals were calculated using phase shifts calculated from the muffin-tin band-structure potentials of Moruzzi et al [15], and were renormalised for the effects of thermal vibrations. The isotropic vibrational amplitudes are defined by the timeaverage displacement given by U 2 - - U 2 + U 2 "1- U 2 - - 3u 2, where u 1,2,3 are the time-average values of the projection of u on three orthogonal axes [16]. The complex electron self-energy E -- V0 + i g / m w a s taken to be independent of energy. Structural refinement was carried out using an iterative procedure [12], in which the disagreement between experimental and calculated intensities, as measured by an R factor, is minimised as a function of the structural and nonstructural variables. The R factor is a normalised X 2 function defined [17-19] as:

R1__ _ {'hk,i?k,i (I~;,i) /lcalN Ih~k i /lex lcalll2/h~k i 2 h~ki {'hk,i) 2 9 k,, O'hk . \ O'hk ,] . \ O'hk ,/

(l)

in terms of the experimental I~{,i and calculated intensities "hk,ilcat,where the index i runs over the electron energy, and Crh~, the root-mean-square experimental uncertainty of the beam hk, obtained [ 18] via comparison of measurements for

229

symmetry-equivalent beams. Implicit in this definition of R is the use of the same scaling constant between the experimental and calculated intensities f o r all beams.

2.3. The surface structures of clean A l ( l l l ) , (100) and (110) The geometrical parameters of the low-index A1 surfaces are summarised in Table 2. As can be seen from the table, the surface structures of AI(111) and Table 2 The surface geometries of clean AI(111), AI(100), and AI(110). The vertical spacings between the i'th and j'th layers are denoted dij (~), and the RMS vibrational amplitudes are denoted ui (]~). The estimated uncertainties on dij and ui are typically -t-0.02/~. The fifth column lists the bulk values of the interlayer spacings at the measurement temperature of 100 K. The final column gives the value of the R factor for the comparison of experimental and calculated LEED spectra. Surface d12 d23 d34 dbulk U1 U2 U3 R AI(111) AI(100) AI(ll0)

2.36 2.06 1.31

2.33 2.04 1.51

2.33 2.01 1.37

2.329 2.017 1.406

0.13 0.17 0.17

0.08 0.14 0.17

0.08 0.10 0.12

0.009 0.021 0.038

(100) correspond to almost perfect truncations of bulk crystals [17, 20, 21]. For the (110) surface, however, significant relaxations of the first three interlayer spacings are found [22-24]. The experimental values - 8 . 1 % , +5.5%, and - 3 . 8 % for the relaxation of the interlayer spacings are in reasonable agreement with values o f - 6 . 8 % , +3.5%, and - 2 . 0 % calculated by Ho and Bohnen [25]. For all three surfaces, enhanced vibrational amplitudes are found for A1 atoms in the first few layers. The particularly large vibrational amplitudes of second layer A1 atoms in the AI(110) surface have been attributed [26] to the relatively unhindered movement normal to the surface of A1 atoms in the second layer of this open surface. A comparison [17] of experimental and calculated LEED intensity spectra for clean AI(111) is shown in Fig. 1. Visual inspection indicates an almost perfect agreement, and the overall R factor for the comparison of 0.009 sets the standard against which the determinations of the surface structures of the adsorbed alkalis must be judged.

2.4. Core-level measurements The use of core-level photoemission spectroscopy for investigating the geometrical structure of overlayers and surface alloys is based on the fact that

230

,A

a) = (0,T) R 0.003

A

Vv d) (0,2) R = 0.017 X 2.3

,~

e) (O,2) R = 0.039 X 5.4 -'---v

T

100

r

~

200

300

400

Energy (eV)

Fig. 1. Comparison of experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for clean AI(111) at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

the core-level binding energy of an atom depends on its local surroundings [27]. The change in binding energy with respect to some agreed-upon standard ("chemical shift") allows the local surroundings of the various elements in the near-surface region to be monitored. Important qualitative information can thus be obtained rather directly from core-level photoemission. For instance, a simple counting of the number of components in the core-level spectra gives the number of inequivalent atoms present in the surface region. (Strictly, counting of the components gives only a lower limit to the number of inequivalent atoms, since the shifts in binding energy may be too small to be resolved experimentally.) The magnitude of the chemical shifts may be used for estimating the nearestneighbour coordination since, for the metallic systems of interest here, the chemical shifts are to a good approximation dominated by the nearest neighbours. Thus, for example, the shift in the A1-2p core-level binding energy provides information on the coordination of the A1 atom to adsorbed alkali atoms. At the most basic level, analysis of the magnitude of the chemical shifts may be done in a finger-printing fashion, simply by comparison with spectra measured for samples where the geometrical structure and hence the coordinations are known. A more general way is to establish a relation between the local en-

231

vironment of an atom and its core-level binding energy, as has been carried out in previous work on other alloy systems [28]. The basic idea of this approach is that the total shift in binding energy for a particular atom is given by a sum over partial shifts induced by the atom's nearest neighbours, including vacuum for an atom at the surface. The individual partial shifts may be calculated from thermo-chemical properties, or may be estimated from experimental measurements of shifts in core-level binding energies for well-characterised interfaces [28]. Once these partial shifts have been worked out, the above methodology allows back-of-the-envelope estimates of the coordinations giving rise to particular shifts in core level binding energies. In the present context it can be noted that coordination to alkali metals (A1) lowers the core-level binding energy of A1 (alkali metals). Although simple and fast estimates of the shifts in core-level binding energies are still of great value, they are currently being supplemented if not superseded by full ab initio calculations of the shifts. This has become feasible during the last decade due to major advances in computer codes for ab initio calculations of total energy, and advances in computer hardware. The calculations are based on density functional theory with the core-level binding energies being calculated as the difference in total energy between the initial (non-ionised) system and the final (ionised) system with a core hole localised on one of its atoms. We refer the reader to a recent review [29] for more details. Here it is sufficient to note that such calculations reproduce experimental shifts with high precision; typical errors being 50 meV or smaller. Measurement of shifts in binding energy, which are often quite small ( ~ 100 meV or smaller), demands high resolution. High count rates are needed in order to reduce measurement times on the often quite reactive alkali-aluminum surface alloys, which implies the need for a large photon flux and an efficient electron energy analyzer. In addition, the possibility to vary the photon energy is most helpful in order to optimise the ionisation cross-sections of the different core levels and in order to vary the probing depth and thereby distinguish bulk and surface components. These requirements may all be fulfilled by the use of a high-resolution synchrotron radiation beam line for provision of the photons, in combination with a high-throughput and high-resolution electron energy analyzer for detection of the emitted photoelectrons. The majority of the CLS measurements on the surface alloys of interest here have been carried out using beam lines at the storage rings MAXLAB and ASTRID [30, 31], which allow total energy resolutions of ~ 50 meV at photon energies around 100 eV while still maintaining count rates of about 104 cps or better. More recent measurements have been performed at a new beam line I311 at MAXLAB [32], capable of total energy resolutions of "~ 10 meV at count rates in excess of 105 cps at

232

photon energies around 100 eV. Use of this very high resolution has enabled the resolution of very close-lying components, including for example the detection [29, 33] of a ~ 28 meV surface core-level shift of the AI(111) surface. High experimental resolution is a necessary but unfortunately not sufficient condition for the measurement of small energy shifts and the resolution of close-lying spectral components; the intrinsic widths of the core levels in question must also be sufficiently small. The A1 2p core level meets this requirement, with a lifetime FWHM < 30 meV and with very small broadening due to vibrational and other thermal effects [29, 33]. In the case of the alkali core levels, the intrinsic lifetime and thermal broadenings are much larger and it is these broadenings and not the experimental resolution which define the level of the binding energy shifts which may be resolved. Consequently, not much is gained for these levels by using a resolution better than -~ 100 meV. In core-level photoemission the intensity of the emitted photoelectrons from a given element is proportional to the amount of that element present. Thus, provided that a proper account is taken of the small mean free paths of low energy electrons in solids, and of photoelectron diffraction effects [34], the relative intensities of chemically-shifted components for a given atom yield the relative concentrations of the atom in inequivalent environments. In order to derive the relative intensities of close-lying spectral components, a decomposition of the experimental spectra is necessary. This can be carried out by fitting core-level spectra with a number of distinct components corresponding to the number of inequivalent atoms, together with a constant or linear background, using a nonlinear minimisation program [35]. The line-shapes of the individual spectral components are obtained as the convolution of a Doniach-Sunjic function [36] representing the intrinsic line-form, with a Gaussian representing broadening due to the combined effects of the instrumental resolution and thermal or static disorder. Each peak is described by five parameters: the binding energy, intensity, the full-widths at half maximum (FWHM) of the Lorentzian and Gaussian components, and the so-called MND singularity index ol [36-38]. In passing we note that recent investigations [39, 40] have shown that a description of phonon effects on the core-level line shape as simply causing an extra Gaussian broadening is not always correct. However, inclusion of a more correct description of the phonon broadening effects does not significantly alter the conclusions of the present work. An A1-2p core-level spectrum for clean AI(100) [33] is shown in Fig. 2. The spectrum contains two spin-orbit components due to emission from surface and bulk A1 atoms. The 2p3/2 component from bulk A1 occurs at a binding energy of 72.72 eV, whereas the corresponding 2p3/2 component from surface A1 atoms is shifted by --~ 90 meV to lower binding energy. The spin-orbit splitting is 0.41

233

Fig. 2. A1-2p core-level spectrum measured at 100 K for a clean AI(100) surface.

eV for each component. 3. A D S O R P T I O N ON A I ( l l l ) In this section we describe the structures of the surface alloys formed by adsorption of the alkali metals on AI(111) at room temperature. This is prefaced by an account of the ordered phases formed by adsorption at low temperature, since for several systems, these phases undergo order-preserving phase transitions to the room temperature phases. As discussed in Sec. 6, these phase transitions shed some light on the mechanisms of formation of the surface alloys. As listed in Table 1, adsorption of 1/4 ML Rb and Cs at low temperature leads to the formation of (2 x 2) phases. Further adsorption of Rb and Cs, or adsorption of K, leads to the formation of (~/3 x ~/3)R30 ~ phases at 1/3 ML coverage. Li is exceptional in not forming an ordered phase at low temperature, whereas Na forms a (4 x 4) phase after adsorption of 9/16 ML. Adsorption of 1/3 ML X = Li, Na, K and Rb at room temperature leads to the formation of AI(111)-(~/3 x ~ / 3 ) R 3 0 ~ phases, whereas adsorption of Cs leads to a (2~/3 x 2~/3)R30~ phase. These phases correspond to saturation coverage of K, Rb and Cs. Further adsorption of Li and Na occurs, which for the latter leads to a new (2 x 2) phase. As described in Sec. 6, orderpreserving phase transitions occur between the (~/3 x ~ / 3 ) R 3 0 ~ Rb, and Cs phases formed at low temperature and the surface alloy structures formed at

234

room temperature.

3.1. A1(111)-(2 • 2 ) - R b and Cs phases formed at 100 K Adsorption of 1/4 ML Rb or Cs at 100K leads to the formation of wellordered (2 x 2) phases [41]. The structure of the A I ( 1 1 1 ) - ( 2 x 2 ) - R b phase is shown in Fig. 3. As can be seen from the figure, Rb atoms are adsorbed in

Fig. 3. Hard-sphere scale model of the AI(111)-(2 x 2 ) - R b structure, where Rb atoms are adsorbed in on-top sites, a)top view, in which the unit cell is marked, b) side view, shown as a central projection on the [1 12] plane through the dashed line in a). The directions of vertical displacements of A1 atoms are indicated by arrows.

on-top sites on a rumpled first A1 layer. The A1 atom lying directly beneath a Rb atom is displaced towards the bulk with respect to the remaining three A1 atoms in the (2 x 2) unit cell of the first A1 layer. Minor relaxations of the substrate also occur in the second A1 layer, where one of the A1 atoms in the (2 x 2) unit cell is displaced towards the surface. Minor, radial displacements of the remaining three A1 atoms in the unit cell of this layer are also found, away from an axis through the adsorbed alkali atoms, but are of the order of the uncertainties. The structure of the AI(111)-(2 x 2 ) - C s phase is essentially the same as that shown in Fig. 3 for Rb. It can be noted that adsorption in on-top sites is unusual, but not unprecedented. Adsorption in on-top sites has also been found in the C u ( 1 1 1 ) - (x/3 x x/3) R 3 0 ~ and N i ( 1 1 1 ) - (x/3 x x/3) R 3 0 ~ phases [42, 43]. The detailed results of the LEED analyses for these two systems are given in Table 3. Apart from the large rumpling of the first A1 layer, a remarkable

235

Table 3 The surface geometries of the AI(111)-(2 x 2)-Rb and Cs phases formed by adsorption at low temperature, which contain Rb and Cs atoms in on-top sites. The interlayer spacings between the i'th and j'th layers, measured from the outer surfaces in the case of rumpled layers, are denoted dij (/~) and the vibrational amplitudes are denoted Ui (/~). d01 (A) is the vertical spacing from the alkali layer to the outer surface of the first, rumpled A1 layer, r (A) is the effective hardsphere radius of the adsorbed alkali atom. u0 (A) is the vibrational amplitude of an adsorbed alkali atom. Azl (A) is the vertical spacing between the subplanes in the first rumpled layer. Az2 (A) is the vertical spacing between the subplanes in the second rumpled layer. The final column gives the value of the R factor for the comparison of experimental and calculated LEED spectra. Alkali

dol

r

Az 1

d12

AZ2

d23

uo

u1

u2

R

Rb Cs

3.11 2.97

2.00 1.88

0.22 0.28

2.17 2.12

-0.03 -0.04

2.31 2.32

1.11 1.65

0.18 0.17

0.11 0.10

0.051 0.088

feature of the results shown in the table are the exceptionally large vibrational amplitudes of the adsorbed alkali atoms, which most likely are predominantly parallel to the surface. A comparison of experimental LEED spectra with spectra calculated for the structure of the A I ( 1 1 1 ) - ( 2 x 2 ) - R b phase shown in Fig. 3 is given in Fig. 4.

3.2. A1(111)-(~/3 x ~/3)R30~

Rb, and Cs phases formed at 100 K

Adsol~tion of 1/3 ML K, Rb or Cs at 100K leads to the formation of wellordered (~/3 x ~/3)R30 ~ phases [9, 44-47]. The structures of these phases are shown in Fig. 5. As can be seen from the figure, alkali atoms are adsorbed in on-top sites on a rumpled first A1 layer, as in the A l ( 1 1 1 ) - ( 2 x 2 ) - R b structure shown in Fig. 3. The A1 atom lying directly beneath an alkali atom is displaced towards the bulk with respect to the remaining two A1 atoms in the ( , / 3 x ~/3) R30 ~ unit cell of the first A1 layer. The detailed results of the LEED analyses for these three systems are given in Table 4. By comparison with the results given in Table 3 for the corresponding A 1 ( 1 1 1 ) - ( 2 x 2 ) - R b and Cs phases, it can be seen from Table 4 that the vibrational amplitudes of the adsorbed alkali atoms are very considerably reduced on increasing the coverage form 1/4 ML to 1/3 ML. However, the enhanced vibrations of A1 atoms in the first layer are again retained on adsorption.

3.3. A I ( l l l ) - ( 4 x 4 ) - N a phase formed at 100 K Adsorption of 9/16 ML Na on AI(111) at low temperature leads to the formation of a well-ordered (4 x 4) phase. The structure of this phase has been shown by SEXAFS [48] to consist of an epitaxial, quasi-hexagonal Na layer on

236

a) (0,1) R - 0.033 X 1.0 b) (0,1) R = 0.030 X 1.1 "'..,.

c) (],1/2) R = 0.037 X 2.5

d) (1/2,1) R = 0.024 X7.1

.,..a

o

e)

(1/2,3/2) R = 0.057 X 5.3

f)

(3/2,3/2) R = 0.041 X 9.0

100

200 Energy (eV)

300

400

Fig. 4. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for AI(111)-(2 x 2)-Rb at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

Fig. 5. Hard-sphere scale models of the AI(111)-(~/3 x ~/3)R30~ structures, a) K, b) Rb, c) Cs. Alkali atoms are adsorbed in on-top sites. Side view, shown as a central projection on the [ 112] plane tilted by 10 ~ with respect to the plane of the paper.

an unperturbed substrate. The N a - N a bond length was found to be 3 . 7 0 / ~ . N a atoms are adsorbed in three different sites with Na-A1 bond lengths of 2.8 ~ , but with different vertical spacings from the first A1 layer, such that the Na layer

237 Table 4 The surface geometries of the A l ( l l l ) - ( ~ / 3 x ~/3)R30~ Rb, and Cs phases formed by adsorption at low temperature, which contain K, Rb and Cs atoms in on-top sites. The interlayer spacings between the i'th and j'th layers, measured from the outer surfaces in the case of rumpled layers, are denoted dij (~) and the vibrational amplitudes are denoted ui (~). do1 (~) is the vertical spacing from the alkali layer to the outer surface of the first, rumpled A1 layer, and u0 (~) is the vibrational amplitude of an adsorbed alkali atom. r (*) is the effective hard-sphere radius of the adsorbed alkali atom. AZl (/~) is the vertical spacing between the subplanes in the first rumpled layer. The final column gives the value of the R factor for the comparison of experimental and calculated LEED spectra. Alkali

d01

r

AZl

d12

d23

/r

/,/1

/,/2

R

K Rb Cs

2.94 3.09 3.16

1.79 1.93 2.02

0.28 0.27 0.29

2.19 2.20 2.19

2.33 2.33 2.33

0.25 0.22 0.25

0.17 0.18 0.15

0.08 0.08 0.08

0.051 0.066 0.061

is not strictly planar.

3.4. A l ( l l l ) - ( ~ / 3 x ~/3)R30~

Na, K, and Rb phases formed at 300 K

A d s o r p t i o n of Li, Na, K, and Rb at r o o m t e m p e r a t u r e leads to the f o r m a t i o n of (~/3 x ~ / 3 ) R 3 0 ~ phases, w h i c h are first o b s e r v e d in the L E E D pattern at a c o v e r a g e of about 1/6 M L and fully d e v e l o p e d at 1/3 M L . T h e structure [18] of the (~/3 x ~ / 3 ) R 3 0 ~ p h a s e is s h o w n in Fig. 6. As can be seen f r o m the figure, N a atoms are a d s o r b e d in 6-fold c o o r d i n a t e d v a c a n c i e s ("substitutional sites"), f o r m e d by displacing 1/3 M L A1 atoms f r o m the first layer of the substrate. T h e displaced A1 atoms are p r e s u m e d to be re-

Fig. 6. Hard-sphere scale model of the AI(111)-(~/3 x ~/3)R30~ structure formed by adsorption of 1/3 ML Na at room temperature, where Na atoms are adsorbed in substitutional sites, a) top view, showing the unit cell. b) side view, shown as a central projection on the [112] plane through the dashed line in a).

238

adsorbed at surface steps. The determination of this structure in a combined SEXAFS and DFT study [7] in 1991 was the starting point for the work described in this Chapter, and was significant in a number of respects. Firstly, the fact that the structure was found both from experiment and theory gave extra credibility to the very unexpected result. Secondly, this was one of the first successful DFT studies of an adsorption system and thereby demonstrated the maturity and applicability of the theory. Thirdly, it demonstrated that surface alloys could be formed between materials that are immiscible in the bulk. Finally, it indicated that surface alloy formation was a potential driving force for the reconstruction of the substrate on adsorption. It was nevertheless important to verify this unusual structure by independent experimental methods. This was achieved by a normal-incidence x-ray standing wave (NIXSW) [45, 49] study in 1992, a LEED study [17] in 1994, and in an STM study [50] in 1995. Fig. 7 shows a comparison of experimental LEED spectra with spectra calculated for the AI(111)- (x/3 x ~ / 3 ) R 3 0 ~ structure with Na atoms adsorbed in substitutional sites.

aR'=<0ol L

c) (T/3,]'/3) R = 0.062

/

100

200

300

400

Energy (eV)

Fig. 7. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for the AI(111)-(~/3 x ~/3)R30~ phase formed by adsorption at room temperature, measured at normal incidence and 100K. The beam hk indices, R factors, and scale factors are shown in each panel.

239

Further LEED studies of the (~/3 x ~/3)R30 ~ phases formed by Li, K, and Rb showed that these alkalis also form substitutional surface alloys by adsorption at room temperature. The structures of these phases are shown in Fig. 8 As can be seen from the figure, the structure formed by Li is almost perfectly

Fig. 8. Hard-sphere scale models of the structures of the (~/3 x ~/3)R30 ~ phases formed by adsorption of a) Li,_b) Na, c) K, and d) Rb on AI(111) at room temperature, shown as central projections on the [112] plane tilted by 10~ with respect to the plane of the paper.

substitutional. The effective hard-sphere radius of the adsorbed Li is 1.46 ~ as compared to that of A1 of 1.43 ~. In the corresponding structures formed by the larger alkalis, the adsorbed alkali atoms naturally sit somewhat above the surface because of their larger size, but they are located nevertheless in the same substitutional site occupied by adsorbed Li. The detailed geometries determined by LEED for the (~/3 x ~/3)R30 ~ phases are given in Table 5. It is interesting to note that the effective hard-sphere radii of the adsorbed alkalis are only ~ 5% less than the bcc metallic values. By contrast, as listed in Table 4, the corresponding hard-sphere radii of alkali atoms adsorbed in on-top sites in the (~/3 x ~/3)R30 ~ phases formed by adsorption at low temperature are 20% less than the bcc metallic values. It can also be seen from Table 5 that the adsorption leads to a small, 2-3% contraction of the first interlayer spacing in the substrate. Somewhat surprisingly, the enhanced vibrations of first layer A1 atoms in the clean AI(111) surface are not reduced by the adsorption of the alkalis, except for K.

240 Table 5 The surface geometries of the substitutional Al(lll)-(~/3 x ~/3)R30~ Na, K, and Rb phases. The second column of the table lists the effective hard-sphere radii r (/~) of the adsorbed alkali atoms, and the third column lists the corresponding values rbcc for the bcc metallic phases. For these systems, relaxations of A1 atoms parallel to the surface are found to be less than the estimated uncertainties. Thus the geometry can be specified in terms of the interlayer spacings dij (~) between the i'th and j 'th layers, and the RMS vibrational amplitudes ui (~). d01 is the distance between the alkali layer and the first A1 layer, and u0 is the vibrational amplitude of an adsorbed alkali atom. The final column lists the R factor values from the LEED analyses. Alkali r rbcc dol d12 d23 uo Ul U2 R Li Na K Rb

1.46 1.78 2.15 2.31

1.52 1.86 2.27 2.47

0.41 1.47 2.16 2.41

2.28 2.27 2.27 2.27

3.5. A1(111)-(2~/3 • 2~/3)R30~

2.31 2.32 2.32 2.32

0.39 0.23 0.21 0.22

0.16 0.13 0.11 0.14

0.10 0.10 0.10 0.10

0.032 0.042 0.062 0.051

phase formed at 300 K

Adsorption of Cs on AI(111) at 300 K is exceptional in a number of respects. As in the case of adsorption of K or Rb, adsorption of Cs saturates at 1/3 ML, but a (2~/3 x 2 ~ / 3 ) R 3 0 ~ rather than a (~/3 x ~ / 3 ) R 3 0 ~ phase is formed. CLS measurements [51] show evidence for the presence of substitutionallyadsorbed Cs, in that A1-2p spectra contain a shift of - 3 5 0 meV, identical to that found for the corresponding (~/3 x ~ / 3 ) R 3 0 ~ phase. However, Cs4d spectra contain two components, indicating unequivocally that two different types of adsorbed Cs atoms are present, with a relative abundance of roughly 3:1. Extensive analysis of LEED measurements for this system carried by the authors [52] have so far resulted in pure frustration. In the light of the CLS measurements, models involving Cs atoms adsorbed in two inequivalent sites have been tested, with a total of 4 Cs atoms in the (2~/3 x 2~/3)R30 ~ unit cell, corresponding to 1/3 ML coverage. These have included models with three Cs atoms adsorbed in substitutional sites in a honeycomb, pseudo (~/3 x ~/3)R30 ~ structure together with one Cs atom in a normal adsorption site, and the converse. Models involving Cs atoms adsorbed in substitutional sites in a (2 x 2) motif, together with Cs atoms in normal adsorption sites, have also been tested. These and many other tests have produced nothing resembling agreement with the experimental LEED measurements. A possible resolution of this dilemma is the possibility [53] that the (2~/3 x 2~/3) R30 ~ LEED pattern is in fact the result of averaging over domains of lesser symmetry. We have recently learned that there is in fact some evidence from an

241

STM study which supports this suggestion [54]. The relatively large computing requirements for models involving perhaps twelve atoms per unit cell, and having less than three-fold rotational symmetry, were a barrier to investigating such models in the past, but our current computer resources makes their evaluation feasible, so we plan to test such models soon.

3.6. A1(111)-(2 x 2 ) - N a phase formed at 300 K Adsorption of Na, unlike adsorption of K, Rb, or Cs, proceeds beyond 1/3 ML coverage to yield a (2 x 2) phase at a coverage of 1/2 ML. In the early work of Porteus [55], it was suggested that this phase consisted of three domains of (2 x 1) periodicity, containing close-packed rows of chemisorbed Na atoms. Later, Hohlfeld and Horn [56] pointed out that the close-packed arrangement was physically unlikely and suggested instead that the structure consisted of a double layer of Na atoms, with each layer having a (2 x 2) periodicity. Based on CLS measurements, it was suggested by one of the present authors [57] that the model of Hohlfeld and Horn should be modified to include an A1 layer sandwiched between two Na layers. Na-2p core level spectra for the (2 x 2 ) - N a phase were found to contain two components, which were attributed to emission from the separate layers of the Na double layer in the model of Hohlfeld and Horn. The existence of an A1 layer sandwiched between these two Na layers was suggested by the presence of an A1-2p component shifted by --~ 500 meV towards lower binding energy, indicative of the presence of A1 atoms in very Na-rich surroundings. A quite different model involving two layers of NaA12 on a bulk A1 layer was proposed by Kerkar et al [49] based on a NIXSW study. Finally, Brune et al [50] concluded from an STM study that the structure consisted of a double Na layer with the Na atoms of the lower layer located in substitutional sites. The structure of the (2 x 2 ) - N a phase as determined in a combined LEED, SEXAFS, and DFT study [58], and shown in Fig. 9, contains some of the features of the models proposed previously and, in particular, confirms the qualitative conclusions of the CLS study [57]. As can be seen from the figure, the structure consists of an Na-A1-Na sandwich, on a reconstructed A1 layer with a (2 x 2) vacancy structure. Na atoms in the lower layer of the sandwich are located in the substitutional sites of this vacancy layer. Na atoms in the upper layer of the sandwich and A1 atoms in the middle of the sandwich are located in fcc sites and hcp sites, respectively, in the vacancy layer. A later photoelectron diffraction study [59] is reported to be in agreement with the structure as described here. An interesting feature of this structure is the fact that it contains a full complement of A1 atoms, in the sense that the A1 atoms which have been substituted

242

Fig. 9. Hard-sphere scale model of the structure of the Al(111)-(2 x 2 ) - N a phase formed by adsorption of 1/2 ML Na at room temperature, a) top view. b) side view tilted by 10~ with respect to the plane of the paper.

by Na atoms in the lower layer of the sandwich, seemingly reappear in the middle layer of the sandwich. Thus, viewed in isolation, it could be imagined that the mechanism of formation involves only local rearrangements of atomic positions. However, the (2 • 2 ) - N a phase is formed by the addition of 1/6 ML Na to the (~/3 x ~/3)R30~ phase, in which Na atoms have substituted 1/3 ML A1 atoms in the first layer. Thus the formation of the (~/3 x ~/3)R30~ phase from the (2 x 2 ) - N a phase requires not only the addition of 1/6 ML Na, but also the addition of 1/3 ML A1. The conclusion is therefore inescapable that, whereas formation of the (~/3 x ~/3)R30~ phase involves diffusion of 1/3 ML displaced A1 atoms to surface steps, the formation of the (2 x 2 ) - N a phase involves a back-diffusion of 1/3 ML A1 atoms from surface steps into the (2 x 2 ) - N a structure. Thus the mechanism of formation involves a large-scale mass transport across the surface. A detailed specification of the geometry of the (2 x 2 ) - N a structure is given in Table 6, in which the results of LEED, SEXAFS, and DFT studies are compared. As can be seen from the results in the table, a near-quantitative agreement exists between the results of the three studies. Equally important is the fact that the LEED and DFT calculations clearly discriminated against the models, mentioned above, proposed previously. Furthermore, both analyses discriminated against a simple variation of the structure shown in Fig. 9 in which Na atoms in the upper layer are adsorbed in hcp sites and A1 atoms in the middle layer of the sandwich are adsorbed in fcc sites. A comparison of experimental LEED spectra with spectra calculated for the structure of the AI(111)-(2 x 2 ) - N a phase shown in Fig. 9 is given in Fig. 10.

243

Table 6 Comparison of experiment and theory for the surface geometry of the AI(111)-(2 x 2 ) - N a phase. Interlayer spacings are denoted dij (A) and are referenced to the layer midpoints. The LEED and DFT analyses also indicate the presence of small (~ 0.04 ~) lateral displacements of A1 atoms in the second A1 layer (the layer with (2 x 2) vacancy), and in the third A1 layer, and small (~ 0.05 *) vertical displacements of A1 atoms in the fourth A1 layer. Method

LEED SEXAFS DFT

d12

d23

d34

d45

d56

0.85 0.75 0.72

0.55 0.70 0.62

1.52 1.50 1.46

2.25

2.28

2.20

2.32

A

/\\

j

A~,

a) (0,T)

~=0.028

-o.o7o

,y~V/

X 13.0

~ ~, 1 \ . .

~

~

;5" "" "

/"\.,

,

c) (0,1/2) R = 0.038

g d) (T/z,T) .~ .

~

ra~ (D

:

A

o.o,2

X 33.1

~/2,o)

........ I"

= 0.020 X 5.8

,~

g) (3/2,3/2) R = 0.033 X6.5 r

A /

100

fk.

/

~J

&

&

k~..,.."'~ "

.~

z;

\ L _~k.-~"

.

~ "'" 200

300

400

Energy (eV)

Fig. 10. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for AI(111)-(2 x 2 ) - N a at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

3.7. Ternary surface alloys formed by coadsorption of Na and K, Rb, or Cs on A I ( l l l ) at 300 K C o a d s o r p t i o n of 1/4 M L N a with 1/4 M L X = K, Rb or Cs at r o o m temperature leads to the f o r m a t i o n of w e l l - o r d e r e d (2 x 2 ) - N a / X phases [60].

244

The basic structure of these phases was suggested by CLS measurements. It was found that Na-2p core-level spectra contained only one component, in contrast to the corresponding spectra for the AI(111)-(2 x 2 ) - N a phase where the two Na layers give rise to two separate Na-2p components, as described in Sec. 3.6. Furthermore, it was found that the Na-2p component in the ternary alloys had a binding energy very similar to that of the inner Na layer in the A l ( l l l ) - ( 2 x 2 ) - N a phase, thereby pointing to a structure for the ternary (2 x 2) alloy phases containing an inner Na layer and an outer K, Rb or Cs layer. This qualitative model of the structure was confirmed by a detailed LEED analysis. The structure of the AI(111)-(2 x 2 ) - N a / K phase is shown in Fig. 11. Similar structures are formed by coadsorption of Na with Rb or Cs. As can be seen from the figure, the structure is the same as that of the AI(111)-(2 x 2 ) - N a

Fig. 11. Hard-sphere scale model of the structure of the AI(111)-(2 x 2 ) - N a / K phase formed by coadsorption of 1/4 ML Na and 1/4 ML K at room temperature, a) top view. b) side view tilted by 10 ~ with respect to the plane of the paper.

phase, except that the upper layer of Na in Fig. 9 has been replaced by K. Similar conclusions apply to the corresponding structures formed by coadsorption of Na with Rb or Cs. An interesting feature of these structures is that the order of the adsorption is immaterial. An X-A1-Na sandwich is always formed even if Na is adsorbed after adsorption of K, Rb, or Cs, which can probably be attributed to the relatively larger adsorption energy of Na. The detailed geometry of the AI(111)-(2 x 2 ) - N a / K structure is given in Table 7. A comparison of a subset of the experimental LEED spectra with spectra calculated for the structure of the AI(111)-(2 x 2 ) - N a / K phase shown in Fig. 11 is given in Fig. 12.

245

Table 7 The surface geometry of the A l ( l l l ) - ( 2 x 2)-Na/K phase formed by coadsorption of 1/4 ML Na and 1/4 ML K at room temperature. Interlayer spacings are denoted dij (/~). The final column gives the value of the R-factor for the comparison of experimental and calculated LEED spectra. d12

d23

d34

d45

d56

R

1.37

0.65

1.40

2.26

2.28

0.072

a) (0,]')

#.A ~,

A

/.\

R

R = 0.070 x13.o

A

\..7.7/V\.Vr ~ '

~

~, c~ I i ........ ~

-

0.028

^

i %, c) (0,1/2) R = 0.038

9 v,..l

d) (T/2,T) t /& d)=(10!251~ 9

r~

f) (3/2,0)

,,~..

R =0.020

~

g) (3/2,3/2) R = 0.033 X6.5 r

-TA / ~ "

x5.8

x~

100

',,. / "

..z.,A

~.. .......

~

j

300

400

~x 200

Energy (eV)

Fig. 12. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for AI(111)-(2 x 2)-Na/K at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

4. A D S O R P T I O N

O N AI(100)

As can be seen from Table 1, ordered phases have been found for adsorption of Li, Na, and K on AI(100). The only ordered phase formed at low temperature for which the structure has been determined is the c(2 x 2 ) - N a phase formed by adsorption of 1/2 M L Na. Adsorption of 1/5 M L Na at just b e l o w room temperature leads to the formation of a (~/5 x ~ / 5 ) R 2 6 . 6 ~ phase. Adsorption

246

of 1/2 ML Li or Na at room temperature leads to the formation of c(2 x 2 ) - L i and Na phases. Finally, adsorption of 1 ML Li at 400 K leads to the formation of a c(2 x 2 ) - 2 L i phase, containing 2 Li atoms in the c(2 x 2) unit cell. As described in Section 6, an order-disorder phase transition occurs between the (~5 x ~/5)R26.6~ phase and a (1 x 1 ) - N a phase, and an order-preserving phase transition occurs between the c(2 x 2 ) - N a phases formed at low temperature and room temperature, respectively.

4.1. AI(100)-c(2 x 2 ) - N a phase formed at 100 K Adsorption of 1/2 ML Na at low temperature leads to the formation of an ordered c(2 x 2) phase. The structure of this phase has been shown by SEXAFS [61] and LEED [20] to contain Na atoms adsorbed in simple four-fold hollow sites on an essentially unreconstructed substrate. The detailed geometry, as determined in the LEED study, is given in Table 8. Table 8 The surface geometry of the AI(100)-c(2 x 2)-Na phase formed by adsorption of 1/2 ML Na at low temperature. Interlayer spacings are denoted dij (/~), where d01 is the Na-A1 spacing. The effective hard-sphere radius of the adsorbed Na atoms is denoted r (/~). RMS vibrational amplitudes are denoted ui (/~), where u0 (~) is the vibrational amplitude of adsorbed Na atoms. dol

r

dl2

d23

uo

Ul

u2

2.57

1.84

2.03

2.02

0.30

0.15

0.13

4.2. Al(100)-(~/5 x ~/5)R26.6~

phase formed at 240 K

Adsorption of 1/5 ML Na at 300 K leads to no changes in the (1 x 1) LEED pattern of the clean surface. However, subsequent cooling to 240 K leads to the formation of a well-ordered (~/5 x ~ / 5 ) R 2 6 . 6 ~ phase. As described further in Sec. 6.3, a reversible, order-disorder phase transition occurs between the (~/5 x ~/5) R26.6 ~ and (1 x 1) phases. The structure of the (~/5 x ~ / 5 ) R 2 6 . 6 ~ phase as determined by LEED and DFT calculations [62] is shown in Fig. 13. The structure turns out to be very similar to that of the corresponding Al(100)-(~/5 x ~ / 5 ) R 2 6 . 6 ~ phase, as determined by LEED and photoelectron diffraction [63]. As can be seen from the figure, Na atoms are adsorbed in substitutional sites formed by displacing 1/5 ML A1 atoms from the first A1 layer. It is interesting to note that the structure has only p4, four-fold rotational symmetry, whereas the LEED pattern and intensities exhibit p 4 m m symmetry. Thus the structure lacks the mirror-plane symmetry of the substrate, and therefore two degenerate domains related by

247

Fig. 13. Hard-sphere scale model of the structure of the AI(100)- (~/5 x ~/5) R26.6~ phase formed by adsorption of 1/5 ML Na. a) top view, showing the unit cell. b) side view, shown as a projection on the (011) plane through the dashed line in a). Arrows indicate the directions of

relaxations of A1 atoms. mirror-plane symmetry must be present, one of which is shown in Fig. 13a. The LEED analysis involved averaging over the intensifies of these domains. A further interesting feature of this structure is the pattern of relaxations of A1 atoms, which is also a consequence of the p4 symmetry. As can be seen from Fig. 13a, the four A1 atoms which are nearest-neighbours to a Na atom move radially away from, and rotate about, the Na atom. Small vertical displacements are also found for one of the five A1 atoms in the (~/5 x ~/5)R26.6 ~ unit cell relative to the other four atoms in the unit cell, in the second A1 layer. The detailed geometry of the AI(100)- (~/5 x ~ / 5 ) R 2 6 . 6 ~ phase is given in Tables 9 and 10. A comparison of a subset of the experimental LEED spectra with spectra calculated for the structure of the AI(100)- (~/5 x ~ / 5 ) R 2 6 . 6 ~ phase shown in Fig. 13 is given in Fig. 14.

4.3. Al(100)-c(2 x 2 ) - L i and Na phases formed at 300 K Adsorption of Li or Na at room temperature on AI(100) leads to the formation of c(2 x 2) phases. These phases grow as islands, which consolidate to wellordered c(2 x 2) structures at 1/2 ML coverage. Early LEED studies [64, 65] of the c(2 x 2 ) - N a phase led to the conclusion that it contained Na atoms adsorbed in four-fold hollow sites on an unreconstructed substrate, as found more recently for the corresponding c(2 x 2 ) - N a phase formed by adsorption at low temperature, as described above in Sec. 4.1. However, SEXAFS [61] and CLS [66] measurements showed unequivocally that the structure of the room temperature phase differed from that of the low temperature phase. Several structural models were found to be compatible with the SEXAFS

248 Table 9 The surface geometry of the Al(100)-(~/5 x ~/5)R26.6~ phase formed by adsorption of 1/5 ML Na at 300 K and cooling to 240 K, as determined by LEED and DFT. Vertical spacings between the outer surfaces of the i'th and j'th layers are denoted dij (/~). d01 (~) denotes the Na-A1 layer spacing. The effective hard-sphere radius of the adsorbed Na atoms is denoted r (A). Radial positions of A1 atoms in the first layer with respect to an axis through an adsorbed Na atom, and rotation angles, are denoted 11 (/~) and 01 (o), respectively. The corresponding bulk values are Ibulk = 2.85 ,~, and Obulk = 0 ~ respectively. Vertical displacements between sublattices in the substrate layers are denoted Azi (/~). The calculated adsorption energy is denoted Ead (eV atom-l). The final column gives the value of the R-factor for the comparison of experimental and calculated LEED spectra. Method dol r ll 01 d12 Az2 d23 Ead R LEED DFT

0.91 0.81

1.75 1.76

3.04 3.08

1.5 1.7

2.04 2.03

0.05 0.07

1.98 2.00

0.068 - 1.72

Table 10 The vibrational amplitudes ui (,~) of atoms in the i'th layer in the Al(100)-(~/5 x ~/5) R26.6~ structure, as determined by LEED. u0 (/~) is the vibrational amplitude of adsorbed Na atoms. uo

u1

u2

u3

Ubulk

0.31

0.23

0.14

0.13

0.12

measurements, but it was concluded that the most likely structure consisted of Na atoms adsorbed in on-top sites on an unreconstructed A1 substrate, but with a c(2 x 2 ) - A 1 layer adsorbed on top of the Na layer [61 ]. However, later L E E D [20] and D F T [67, 68] studies clearly showed that the correct structure consists of Na atoms adsorbed in substitutional sites on a reconstructed A1 substrate. A L E E D analysis [21] of the c(2 x 2 ) - L i phase indicates that the structure is very similar to that of the c(2 x 2 ) - N a phase, with Li atoms adsorbed in substitutional sites, as shown in Fig. 15. The detailed geometries of the A 1 ( 1 0 0 ) - c ( 2 x 2 ) - L i and Na structures are given in Table 11. A comparison of experimental and calculated L E E D spectra for A I ( 1 0 0 ) - c ( 2 x 2 ) - L i structure is shown in Fig. 16.

4.4. A l ( 1 0 0 ) - c ( 2 x 2 ) - 2 L i p h a s e f o r m e d at 400 K Further adsorption of Li beyond 1/2 M L coverage at room temperature causes a gradual decrease in intensity of the fractional-order spots in the L E E D pattern of the c(2 x 2 ) - L i phase and some increase in background intensity. However,

249

___~ _

i

b) ( 2 , 1 ) R = 0.070 X 13.0

~

/

'

a) (0,1) R = 0.028 ........ X 1.0

~

i

...-~__

-

''/ N

i

~ ' "

c) (0,1/2) R = 0.038 X 1.3

r~ .l,.a

4

.

,

~

1.

.

.

.

.

.

.

.

.

--

--i

---r---

o0,1

xd) (1/2,1) ~ "

.

.....

/

\

..-

.

_.--....

=

~D

R = 0.052 X 33.1 I

I

nip

"

I

r

[

.

.

.

.

I

"

"

--

I

- - --

.

R = 0.020 X 5.8 |

(3/2,3/2) R = 0.033 X 6.5 g)

i

i

lOO

I

200 Energy (eV)

i

i

300

i

i

400

Fig. 14. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for A1(100)-(~/5 x ~/5)R26.6~ at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

Fig. 15. Hard-sphere scale model of the structure of the Al(100)-c(2 x 2 ) - L i phase formed by adsorption of 1/2 ML Li at room temperature, a) top view, showing the unit cell. b) side view, shown as a central projection on the [010] plane through the dashed line in a).

250

Table 11 The surface geometries of the AI(100)-c(2 x 2 ) - L i and Na phases formed by adsorption of 1/2 ML Li and Na at room temperature. Interlayer spacings are denoted dij (/k), where d01 (/k) is the alkali-A1 spacing. The effective hard-sphere radius of the adsorbed alkali atoms is denoted r (~). RMS vibrational amplitudes are denoted ui (/~), where u0 (/~) is the vibrational amplitude of adsorbed alkali atoms. The final column gives the value of the R factor for the comparison of experimental and calculated LEED spectra. alkali

Li Na

d01

r

d12

d23

uo

u 1

u2

R

0.32 1.10

1.44 1.63

1.90 1.84

2.03 2.04

0.24 0.40

0.24 0.22

0.17 0.15

0.055

a) (1,0) R = 0.055

~

_-._ |

~l.0 i

b) (T,T)

/ \\ i

\\.

c) (~,i) R=0.092

R=OO38

~

X 1.0 AA tx~/~:t • ~

/A

"

Xl9O

cg~

....

~

.........

d) (1/2,1/2) R = 0.047 X 3.3 e) (1/2,5/2) R = 0.036 X 10.2

!

~

.

.

.

.

.

.

.--2e.

. . . . . . . . . . . . . . . . . . .

f) (3/2,3/2) R =0.179 X 34.4 !

100

200

300

400

Energy (eV)

Fig. 16. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for Al(100)-c(2 x 2 ) - L i at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

a w e a k c ( 2 x 2) L E E D pattern is still present after adsorption o f 1 M L Li. CLS m e a s u r e m e n t s [69] revealed that adsorption o f Li b e y o n d 1/2 M L coverage leads to the formation o f a new Li binding state, as e v i d e n c e d by the presence of a s e c o n d Li-1 s peak in L i - l s core-level spectra for coverages above 0.5 ML. This

251

can be seen from Fig. 17, which shows Li-ls core-level spectra measured at 1/2 ML and 1 ML coverage. It can be noted that the binding energy of 54.60

Fig. 17. Li-ls core-level spectra measured after adsorption at room temperature, a) 1/2 ML Li. b) 1 MLLi.

eV of the narrow peak in Fig. 17a is close to the binding energy of 54.70 eV for bulk Li [70]. Since A1 is not expected to induce any large shifts in corelevel binding energy of Li [71], it was natural to consider the possibility that the narrow peak in Fig. 17b was due to emission from Li atoms in a more bulk-like environment. This possibility was confirmed by measurements of the peak intensities in Li-ls and A1-2p spectra as functions of the emission angle for a number of different photon energies, as illustrated in Fig. 18. The strong attenuation of the peak at lower binding energy with increasing emission angle, and hence with increasing surface sensitivity, indicates that this peak is due to emission from Li atoms adsorbed in an underlayer. Comparison of the relative attenuation with emission angle of peaks in the Li-ls and A1-2p spectra led to the conclusion that the structure consisted of two mixed A1/Li layers separated by a pure A1 layer. It was further suggested that the structure corresponded to the first three layers of the metastable, bulk A13Li alloy which has a Cu3Au-type structure, as shown in Fig. 19a. It was not possible to confirm the model shown in Fig. 19a by LEED because of the imperfect order of the Al(100)-c(2 x 2 ) - 2 L i phase formed by adsorption at room temperature. Attempts to improve the order by annealing led to diffusion of the Li underlayer into the bulk. However, it proved to be possible to prepare a very well-ordered c(2 x 2 ) - 2 L i phase by adsorption of Li at 400

252 |

|

,

|

I emiS~gl~

5ft.0

I

'

5;.0

'

5~.0

Binding Energy (eV)

Fig. 18. Li- 1s core-level spectra as a function of emission angle after adsorption of 1 ML Li on AI(100) at room temperature

Fig. 19. a) A13Li-type model proposed [69] for the A1(100)-c(2 x 2 ) - 2 L i phase formed by adsorption of 1 ML Li at room temperature, b) A13Ti-type structure of the Al(100)-c(2 x 2 ) - 2 L i phase formed by adsorption of 1 ML Li at 400 K [72]. Side view, tilted 20 ~

K. A LEED and DFT analysis [72] of this phase led to the structure shown in Fig. 19b. As can be seen from Fig 19b, the structure differs from that proposed pre-

253

viously, shown in Fig. 19a, in the registry of the two mixed A1/Li layers. It turns out that the registry is the same as that in the A13Ti bulk alloy rather than the expected A13Li alloy. The detailed geometry of the Al(100)-c(2 x 2 ) - 2 L i phase as determined by LEED and DFT is given in Table 12. As described later in Sec. 7.6, the DFT calculations also led to the prediction of an unusual surface structure for the bulk A13Li alloy. A comparison of experimental and calculated Table 12 The surface geometry of the A1(100)-c(2 x 2 ) - 2 L i phase formed by adsorption of 1 ML Li at 400 K, as determined by LEED and DFT. Vertical spacings between the outer surfaces of the i'th and j ' t h layers are denoted dij (A). AZl (A) is the vertical separation between Li and A1 atoms in the first layer. The effective hard-sphere radius of Li atoms in the first layer is denoted r (/~). The calculated adsorption energy is denoted Ead (eV atom -1). The final column gives the value of the R-factor for the comparison of experimental and calculated LEED spectra. Method AZl r d12 d23 Ead R LEED DFT

0.38 0.32

1.45 1.44

1.82 1.85

1.97 1.98

0.053 -2.27

LEED spectra for the Al(100)-c(2 x 2 ) - 2 L i structure is shown in Fig. 20.

5. ADSORPTION ON Al(ll0) As can be seen from Table 1, well-ordered phases have been found for adsorption of Li, Na, and Rb on AI(110) at room temperature. Adsorption of 1/2 ML Li or Na leads to the formation of c(2 x 2) phases, whereas adsorption of 1/2 ML Rb leads to the formation of a c(4 x 2) phase. Adsorption of 3/4 ML Na leads to the formation of a (4 x 1) phase.

5.1. A1(110)-c(2 x 2 ) - L i and Na phases formed at 300 K Adsorption of Li or Na at room temperature on AI(110) leads to the formation of c(2 x 2) phases. These phases grow as poorly-ordered (3 x 2 ) - 2 L i or (3 x 2 ) - 2 N a islands, which consolidate to well-ordered c(2 x 2) structures at 1/2 ML coverage [73, 74]. LEED analyses indicate that the structures of the c(2 x 2 ) - L i and Na phases are very similar, with Li or Na atoms adsorbed in substitutional sites, formed by displacing 1/2 ML A1 atoms from the first layer of the substrate. A hard-sphere model of the c(2 x 2 ) - N a structure is shown in Fig. 21. As indicated in the figure, adsorption of Li or Na in the substitutional site leads to quite large vertical displacements of A1 atoms in the third and fifth A1 layers, which can be understood in terms of a simple, point-ion, frozen back-

254

"

"A

m h.-.. \

Ra) (0,0.078T) =

b) (0, 2)

1\

X3.4 --'T

--------'---'l"

"""

c) (1,2) R = 0.054 x 19.4

"~" .~ d~

'~ r~ ~ ,". " "~';4 A / A v u w \

/~k~f~.

.,

A

d) (1/2, ]'/2)

o9o,, "Ua = .

.-. IA ;~

"" t~'~ /~

,

e) (3/2, 1/2) R = 0.069

f) (3/2, 3/2) R = 0.058 X ll.5 "'1

T

"

~

'

r

g) (1/2, 5/2) R=0.161 X 74.4 T

100

[ I

~ r

"f

~

~

~

T

200

300

400

Energy (eV)

Fig. 20. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for AI(100)-c(2 x 2 ) - 2 L i at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

ground model [73]. The detailed geometries of the AI(110)-c(2 x 2 ) - L i and Na structures are given in Table 13, and the vibrational amplitudes are given in Table 14 A comparison of experimental and calculated LEED spectra for A I ( 1 1 0 ) - c ( 2 x 2 ) - N a structure is shown in Fig. 22. 5.2. A I ( l 1 0 ) - ( 4 x 1 ) - 3 N a phase formed at 300 K Further adsorption of Na on AI(110) leads to the formation of a poorlyordered (3 x 1 ) - 2 N a phase at a coverage of 2/3 ML. This phase co-exists with the c(2 x 2 ) - N a phase. However a well-ordered (4 x 1 ) - 3 N a phase is formed after adsorption of 3/4 ML. A LEED analysis indicates that the unit cell of this phase contains one substitutionally-adsorbed Na atom and two Na atoms adsorbed in low-symmetry sites, displaced by 4-0.33 A from two four-fold coordinated nearest-neighbour sites [75]. A hard-sphere model of the structure is shown in Fig. 23. It is interesting to note that the c(2 x 2 ) - N a phase, as shown

255

Fig. 21. Hard-sphere scale model of the structure of the AI(110)-c(2 x 2 ) - N a phase formed by adsorption of 1/2 ML Na at room temperature, a) top view, showing the unit cell. b) side view, shown as a central projection on the (1 i0) plane through the dashed line in a). Arrows in b) indicate the sense of the rumpling of the third and fifth A1 layers

Table 1 3 The surface geometries of the AI(110)-c(2 x 2 ) - L i and Na phases formed by adsorption of 1/2 ML Li or Na at room temperature. Interlayer spacings are denoted dij (~,), where d01 (,~) is the alkali-A1 spacing. The spacings are given with respect to the outer surfaces in the case of the rumpled third and fifth A1 layers, r (/~) denotes the effective hard-sphere radius of an adsorbed alkali atom. Vertical splitting of the i'th A1 layer is denoted Azi (~). The final column gives the value of the R factor for the comparison of experimental and calculated LEED spectra. alkali

do1

r

d12

d23

AZ3

d34

d45

Az5

R

Li Na

0.38 1.06

1.45 1.62

1.25 1.27

1.39 1.36

0.11 0.14

1.36 1.36

1.40 1.38

0.05 0.06

0.039 0.045

Table 14 The vibrational amplitudes ui (A) of atoms in the i'th layers in the AI(110)-c(2 x 2 ) - L i and Na structures, u0 (/~) is the vibrational amplitude of adsorbed alkali atoms. alkali

Li Na

uo

u1

u2

u3

u4

u5

0.35 0.27

0.17 0.17

0.15 0.13

0.13 0.13

0.09 0.10

0.10 0.11

s c h e m a t i c a l l y in Fig. 21, contains 0.5 M L N a a t o m s in s u b s t i t u t i o n a l sites. T h u s the a d d i t i o n of a further 0.25 M L N a in g o i n g f r o m the c(2 • 2 ) - N a p h a s e to the (4 • 1 ) - 3 N a p h a s e causes a partial lifting of the r e c o n s t r u c t i o n of the substrate.

256

~

~

a) (],0) R = 0.041 ,

.~

-

.

i

_

. |.

.

.

.

.

.

.

X~.6"

' - 7 " - - ' '

.....

|

'

~ i

I

.,.

!

: , ~ . . . t~ 9:

'

1O0

__ !

..

/ x~ ~ I

~ , ~

'

/ix

// \\ Y '~

.

_. . . . . . . .

.....~

~ ~ _ 1 7

'

! ~

300

!

I

""1"

'

R=0.051 L~,.~_ Xll.6

!

200 Energy (eV)

I

d) (3/2,1/2) R = 0.084 X 5.9 . e )(3/2,3/2) R = 0.053 . 1 ........... |

,

'

"'":

....

R - 0.046 X 2.0 c) (1/2,5/2) R = 0.096 X3.5

~

~

' ~

.

!

-

400

Fig. 22. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for AI(110)-c(2 • 2 ) - N a at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

Fig. 23. Hard-sphere scale model of the structure of the AI(110)-(4 x 1)-3Na phase formed by adsorption of 3/4 ML Na at room temperature, a) top view, showing the unit cell. b) side view, shown as a central projection on the (001) plane through the dashed line in a), tilted by 10~ with respect to the plane of the paper.

Since the c(2 x 2 ) - N a and (4 x 1 ) - 3 N a phases are m i s s i n g 1/2 M L and 1/4 M L A1 atoms, respectively, it follows that f o r m a t i o n of the latter phase f r o m the

257 Table 15 Atomic coordinates (xyz) (A), interlayer spacings dij (/~), and rms vibrational amplitudes u (/~) determined for the (4 x 1)-3Na structure. The x and y axes lies in the surface plane along the [110] and [001] directions, respectively. The z axis is the outward surface normal. The uncertainties on the x coordinates are about -t-0.09 A, and the uncertainties on the z coordinates interlayer spacings, and the vibrations are about 4-0.04 A. The effective hard-sphere radii of Na atoms in low-symmetry sites and substitutional sites are 1.68/~ and 1.60/~, respectively. The bulk lattice constant at 100 K is a0 -- 4.0341 A. The R factor for the LEED analysis was 0.063 Layer Atom x y z dij u m

Na Na Na A1 A1 A1

3.95 -3.95 0 ~/~a0 2.86 -2.86

ao/2 ao/2 0 0 0 0

A1 A1 A1 A1

1.41 - 1.41 4.29 -4.29

a0/2 a0/2 a0/2 ao/2

A1 A1 A1 A1

0 2.81 -2.81 ~/2a0

0 0 0 0

2.09 2.09 1.01 0.10 0 0

0.30 0.30 0.28 0.18 0.16 0.16 1.27

-

1.27 1.27 1.30 1.30

0.15 0.15 0.16 0.16 1.40

-2.69 -2.75 -2.75 -2.76

0.13 0.11 0.11 0.11 1.38

bulk

A1

0.12

f o r m e r requires not only the addition of 1/4 M L N a atoms, but also the addition of 1/4 M L A1 atoms. As in the case of the formation of the A I ( 1 1 1 ) - ( 2 x 2 ) - N a phase, described in Sec. 3.6, the required A1 atoms are p r e s u m e d to be acquired f r o m surface steps. The detailed g e o m e t r y of the (4 x 1 ) - 3 N a phase is given in Table 15. A c o m p a r i s o n of experimental and calculated L E E D spectra for the A I ( 1 1 0 ) - ( 4 x 1 ) - 3 N a structure is shown in Fig. 24. 6. P H A S E T R A N S I T I O N S

One of the r e m a r k a b l e features of alkali metal adsorption on A1 surfaces is the o c c u r r e n c e of phase transitions involving ordered phases. T h e s e p h a s e transitions occur at or below r o o m temperature. Perhaps most r e m a r k a b l e are the order-preserving, irreversible phase transitions involving the A I ( 1 1 1 ) - ( ~ / 3 x ~/3)R30~ and Rb phases, and the A l ( 1 0 0 ) - c ( 2 x 2 ) - N a phases. F o r these systems, the surface unit cell, and hence the s y m m e t r y of the associated L E E D

258

a) (1,0) R = 0.023 X 1.8

_.," /"....~._'~,_ " :

.

.

.

.

- -

.

'

b) (1,1) R = 0.047 X 2.3 c) (1/4,1) R = 0.028 X 4.9

,~s .-.

;~

-

I

"

I

......

~

9 =

F

d) (I/2,1) R = 0.028 X 6.5 - - ~

_

. . . . . . .

i

i

e) (1/4,2) \

R = 0.043 ,

......

qA

4

~176 A . ,/~"

100

Xl

..

200

..:'"

R = 0.051 X 23.4

300

Energy (eV)

Fig. 24. Comparison of a subset of the experimental (solid lines) and calculated (dotted lines) intensity-energy spectra for AI(110)-(4 x 1)-3Na at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

patterns, are unchanged by the phase transition. A similar, order-order, irreversible phase transition occurs between the A I ( 1 1 1 ) - (x/3 x ~ / 3 ) R 3 0 ~ and A I ( 1 1 1 ) - ( 2 ~ / 3 x 2~/3) R 3 0 ~ phases. However, our understanding of this phase transition is incomplete, because the structure of the latter phase is not yet known, as discussed in Sec. 3.5. Finally, an order-disorder, reversible phase transition occurs for the A l ( 1 0 0 ) - ( x / 5 x ~ / 5 ) R 2 6 . 6 ~ phase. 6.1. A i ( l l l ) - ( ~ / 3 x ~ / 3 ) R 3 0 ~ and Rb Adsorption of 1/3 ML K or Rb on AI(111) produces (x/3 x x/3)R30~ phases with two distinctly different structures depending on the adsorption temperature, as described in Secs. 3.2 and 3.4. The fact that different structures are formed by adsorption at 100 K and 300 K for these two systems was immediately apparent from experimental LEED and CLS measurements before the structures had been determined. For the (~/3 x ~ / 3 ) R 3 0 ~ structures, the conclusion of the LEED and CLS studies concerning the occupation of different sites in the two structures was later confirmed by a study [76] using Second Harmonic Generation (SHG) to monitor the change from on-top to substim-

259

tional sites. To illustrate how easily the occurrence of two different structures can be seen from the experimental measurements, LEED intensity-energy spectra from the two (~/3 x ~ / 3 ) R 3 0 ~ phases are shown in Fig. 25, from which it is evident that the two phases have very different structures.

A

9

'

~,~,

"

........ .....

..." "......

..

"

a) (0,T) .. R=0.580

...

Oo,,76

i

rae~ ,l...a

c)(~,~)

.

X8.6 ------~-~, ~

(D

/['~k,

/\

i\

. A

\A.,.4 \., i \,.,,,

7/_-'o:777

fVI

r d) (1/3,1/3) R = 0.379

...---.....

i-'" \/-.. '~k " \ ...."\

f) (3,3,4,3) R = 1.063 X 13.6 i

e) (2/3,2/3) R=0.824

[\ ~ T

100

;'x .t . t \ "

-

~

.

.

~

_

T

z "

200

300

~176

400

Energy (eV)

Fig. 25. Comparison of experimental intensity-energy spectra for the Al(lll)-(~/3 x ~/3) R30~ phase formed by adsorption at room temperature (solid lines), with spectra (dotted lines) for the corresponding phase formed at 100 K. Both measurements are at normal incidence and 100 K. The beam hk indices, R factors, and scale factors are shown in each panel.

Similarly, Rb-4p core-level spectra for the two A I ( 1 1 1 ) - (~/3 x ~/3) R 3 0 ~ phases are strikingly different, as shown in Fig. 26 for the two structures formed by adsorption at 100 K and 300 K, respectively [51, 77]. The Rb-4p binding energy of the high temperature phase is shifted by 250 meV towards lower binding energy. This lowering of the binding energy indicates a higher coordination of Rb atoms to A1 atoms in the high temperature phase than in the low temperature phase. This was suggested [51, 77], and later confirmed by a quantitative LEED study [9], to be due to a transition from a phase with Rb

260

Fig. 26. Rb-4p CLS spectra for the AI(111)-(~/3 x ~/3)R30~ phases formed a) by adsorption at room temperature, and b) adsorption at 100 K. Both measurements are at 100 K. (Each spectrum contains 4pl/2 and 4p3/2 components, with a spin-orbit splitting of 0.86 eV)

atoms adsorbed in on-top sites to a phase with Rb atoms adsorbed in substitutional sites. The irreversible phase transition was followed by monitoring the changes in Rb-4p spectra that result from annealing to successively higher temperatures [77]. By resolving the spectra into contributions from the low and high temperature phases, the relative amount of Rb in the two phases was determined as a function of temperature. The results of this analysis are shown in Fig 27 and clearly demonstrate that the phase transition starts at about 200 K and occurs very rapidly at temperatures around 250 K. Fig 27 also contains the results of a similar study of the phase transition from the low-temperature AI(111)-(2 x 2 ) - R b phase, in which Rb atoms are adsorbed in on-top sites, to a phase containing islands of the AI(111)-(~/3 x ~ / 3 ) R 3 0 ~ phase, in which Rb atoms are adsorbed in substitutional sites. As can be seen from the figure, this latter phase transition occurs at a lower temperature. Similar results concerning the phase transition have also been obtained by CLS and LEED for the AI(111)- (~/3 x ~/3) R 3 0 ~ system. At a K coverage of 1/3 ML, the transition from on-top to substitutional sites is found [66] to start at "-~ 220 K and be completed just below room temperature. As discussed later

261 1.0 9

~ll

0.8 9

0

0.6 o

~9 0.4

a) 1/3MLRb

0 0 0 0

9 b) 1/4MLRb

or

o.2

o o O

0.0

9

|

100

150

200

9

O~Q ~

250

300

Temperature (K)

Fig. 27. The relative intensity of the 4p3/2 component in Rb-4p core-level spectra due to Rb atoms adsorbed in on-top sites, to that of the corresponding component due to Rb atoms adsorbed in substitutional sites, as a function of annealing temperature, a) Phase transition between low and high-temperature (~/3 x ~ / 3 ) R 3 0 ~ phases, shown as open circles, b) Phase transition between the low temperature (2 x 2 ) - R b phases and islands of a (~/3 x ~ / 3 ) R 3 0 ~ phase, shown as closed circles

in Sec. 7.1, this onset temperature is consistent with a activation barrier of "-~ 0.8 eV as derived from DFT calculations [46, 78]. A later SHG study [76] reported a lower activation barrier of ~ 0.7 eV. However this lower value can probably be attributed to the fact that the coverage used was only 0.24 ML. As described above, CLS measurements for the AI(111)- (~/3 x ~/3) R 3 0 ~ system show that the transition occurs at lower temperature for lower coverage. A possible mechanism for the phase transitions is evident on comparing hardsphere models of the two different (~/3 x ~ / 3 ) R 3 0 ~ structures, as shown in Fig. 28. It is natural to presume from this comparison, that the mechanism is an interchange of position of a Rb atom in an on-top site with the A1 atom directly beneath it, followed by movement of the displaced A1 atom to a surface step where it is readsorbed. 6.2. Al(100)-c(2 • 2 ) - N a Adsorption of 1/2 ML Na on AI(100) produces c(2 • 2 ) - N a phases with two distinctly different structures depending on the adsorption temperature, as described in Secs. 4.1 and 4.3. As in the case ofthe AI(111)-(~/3 • ~ / 3 ) R 3 0 ~ and Rb phases, discussed in Sec. 6.1, the fact that different structures are formed by adsorption at 100 K and 300 K was apparent from experimental CLS and SEXAFS measurements before the structures had been determined [61, 66]. A LEED structure determination showed that the structure formed at low tempera-

262

Fig. 28. Side views of hard-sphere scale models of the structures of the AI(111)-(~/3 x ~/3)R30~ phases formed by adsorption of 1/3 ML Rb at a) 100 K, and b) 300 K.

ture contains Na atoms adsorbed in four-fold hollow sites, whereas the structure formed by adsorption at 300 K contains Na atoms in substitutional sites [20]. The transformation from the low to the high temperature c(2 x 2) phase was followed by CLS [66] in the same way as described for the AI(111)-(~/3 x x/3)R30~ system in Sec. 6.1. It was found that the transition from on-top to substitutional sites starts already at 160 K and is completed at temperatures just below room temperature.

6.3. Al(100)-(~/5 x ~/5)R26.6~ As described in Sec. 4.2, adsorption of 0.2 ML Na on A1(100) at room temperatures followed by cooling to 240 K leads to a disorder-order phase transition from a phase with a (1 x 1) LEED pattern to a (~/5 x ~ / 5 ) R 2 6 . 6 ~ phase [62], with Na atoms adsorbed in substitutional sites. Once the substitutional structure has been formed, the order-disorder phase transition is completely reversible. However, if 0.2 ML Na is first adsorbed at low temperature, islands of c(2 x 2) structure are formed in which Na atoms are adsorbed in four-fold hollow sites [79, 80]. Annealing this phase to room temperature leads to an irreversible phase transition to a phase with a (1 x 1) LEED pattern, with Na atoms adsorbed in substitutional sites. At this point the reversible, disorderorder phase transition between (1 x 1) and (~/5 x ,/5)R26.6 ~ phases can be performed by cooling to 240 K as before. The course of the reversible phase transition has been studied by CLS measurements [62], as shown in Fig. 29. Panel a) of the figure contains A1-2p spectra measured as a function of temperature. Panel b) shows a plot of the F W H M of the Na-induced component, indicated by the dashed line in the spectra of panel a). As is evident from the figure, the F W H M changes considerably

263 i

.

.

.

.

i

a)

I

.

.

.

.

"~

i

.

.

.

.

i

.

,

if . ', (lxl) ,(

~ ~

30(1

9

)

(~/5x~/5)R26.6"

//I----

250

~ 2oo 150

,0o Temperature(K)

73.5

73.0

72.5

72.0

Bindingenergy(eV)

Fig. 29. CLS measurements as a function of temperature, in the region where a reversible phase transition from a (~/5 x ~/5)R26.6~ phase to a (1 x 1 ) - N a phase occurs with increasing temperature, a) A1-2p spectra. The dashed line indicates the component shifted by ~-- - 2 0 0 meV induced by adsorption of Na. b) FWHM of the Na-induced component as a function of temperature,

as the phase transition progresses. The change of the width is accompanied by a small shift in binding energy of the Na induced component. It has been suggested that these changes are due to unresolved components in the A1 2p spectra related to the presence of A1 atoms with different coordinations to Na in the disordered phase above the phase transition temperature. The irreversible phase transition, which occurs on annealing after adsorption of 0.2 ML Na at low temperature, has also been followed by monitoring the A1-2p level after annealing the sample for 30 seconds at successively higher temperatures [62]. These measurements revealed the appearance of an A1 2p component shifted towards lower binding energy by > 400 meV at temperatures around 190 K. This large shift indicates a high coordination of A1 to Na, as found for example in the substitutional Al(100)-c(2 x 2 ) - N a structure. Thus the mechanism of the irreversible phase transition appears in part to include the transformation from Na adsorbed in four-fold hollow sites to Na adsorbed in substitutional sites, with a local A1/Na coordination similar to that of the c(2 x 2 ) - N a structure. At higher temperatures, the Na atoms in substitutional sites disperse to form the disordered (1 x 1 ) - N a phase.

264 7. T H E R O L E OF DFT C A L C U L A T I O N S As noted in the Introduction, the adsorption of alkali metals on aluminium surfaces has been a system of recurring interest in the development of the theory of surfaces. In the last decade, ab initio calculations based on density functional theory have achieved a sufficient sophistication that they enter on an equal footing with experimental approaches to the determination of surface structure for simple systems. This was clearly demonstrated by the successful prediction that the AI(111)-(~/3 x ~ / 3 ) R 3 0 ~ phase consisted of a surface alloy, in the combined SEXAFS and DFT study of this system in 1991 [7]. In the rapid development of the characterisation and understanding of alkali-aluminum surface alloys that followed this very unexpected result, DFT calculations played a crucial role, not least in providing explanations for the unusual structures that were found, and in providing a basis for discussion of the mechanisms and kinetics of formation. At the same time, these systems provided a very valuable testing ground for DFT calculations of surface structures and helped to define the applicability of such calculations. For a more general view of DFT calculations of surface properties we refer the reader to a recent review by Scheffler and Stampfl [81 ]. In the following we give a brief description of the role of DFT calculations in studies of alkali-aluminum surface alloys.

7.1. A I ( l l l ) - ( ~ / 3 x ~/3)R30~ and K As described in Sec. 1.2, the first observation of surface alloy formation for alkali-aluminium systems was made in 1991 in a combined SEXAFS and DFT study of the A l ( 1 1 1 ) - (~/3 x ~ / 3 ) R 3 0 ~ phase formed by Na adsorption on Al(111) at room temperature. The surface structure resulting from the calculations is compared with the experimental results in Table 16. As can be seen Table 16 Comparison of theory and experiment for the surface geometry of the Al(lll)-(~/3 x ~/3)R30~ phase with Na atoms in substitutional sites. The Na-A1 bond length and Na radius are denoted L N a - A l (A) and rUa (/~) respectively. The calculated adsorption energy is denoted Ead (eV atom-l). Method Ref. L Na-Al rNa Ead SEXAFS LEED NIXSW DFT DFT

[7] [17] [45, 49] [7, 78] [82]

3.31 3.21 3.09 3.13 3.12

1.78 1.72

-1.58 -1.51

265 from the results given in the table, a nearly quantitative agreement was found between experiment and theory for this system. Perhaps more important was the convincing discrimination against other structural models. Thus, Neugebauer and Scheffler reported adsorption energies E a d = --1.42 eV, - 1 . 4 1 eV, and - 1 . 2 8 eV for adsorption in hcp, fcc and on-top sites, respectively, as compared to E a d -- --1.58 eV for adsorption in the substitutional site. Shortly afterwards, DFT calculations were also carried out for K adsorption on AI(111) [78, 83, 84]. The surface structures resulting from these calculations are compared with the experimental findings in Table 17. As can be seen Table 17 Comparison of experiment and theory for the surface geometry of the two AI(111)-(~/3 x ~/3)R30~ phases with K atoms in on-top and substitutional sites, respectively. The KA1 layer spacing and K radius are denoted dK-Al (/~) and r/((,~), respectively. The vertical displacement of first-layer A1 atoms lying beneath K atoms in the on-top structure is denoted AZl (A). The calculated adsorption energy is denoted Ead (eV atom-1). Method Ref. site d K-AI Azl rK Ead LEED DFT

[44, 46] [78]

on-top on-top

2.94 3.22

LEED DFT

[44, 46] [78]

subst. subst.

2.16 2.39

0.23 0.16

1.79 1.97

- 1.21

2.15 2.29

- 1.17

from the table, the detailed surface structures for the two (~/3 x ~ / 3 ) R 3 0 ~ phases agree reasonably well, if less quantitatively than for the substitutional (~/3 x ~ / 3 ) R 3 0 ~ phase. It can be noted in particular, that both experiment and theory reveal a increase in the effective hard-sphere radius of adsorbed K atoms in going from on-top sites to substitutional sites, and that a vertical displacement occurs of A1 atoms lying beneath K atoms in the on-top structure. However, unlike the case of Na adsorption, the calculated adsorption energies for the various adsorption sites are essentially degenerate, also with K adsorption in fcc sites where Ead - - - - 1 . 2 0 eV. Perhaps the most important outcome of the DFT studies was the explanation of why alkali metal adsorption on A1 can result in the formation of surface alloys. This can be attributed in part to the exceptionally low energy of formation of a surface vacancy on A1 surfaces, which was calculated to be Evac = 0.41 eV for an A I ( 1 1 1 ) - (~/3 x ~/3)R30 ~ vacancy structure [78]. Thus substitutional adsorption is thermodynamically favoured if the difference in binding energy between adsorption in a vacancy and adsorption in a normal site is > Evac. The

266

formation energy of a vacancy includes the energy, equal to the cohesive energy, regained by readsorption of the ejected A1 atom at a kink site. Eoac is therefore less than the energy E pair required to create a Frenkel pair, where the ejected atom is readsorbed on the flat surface. The second part of the explanation of surface alloy formation is the large binding energy for an alkali metal in a substitutional site, which is in turn explained by the relatively efficient screening of the essentially ionic alkali by the A1 valence electrons. Assuming that similar considerations apply to the phase transition between on-top and substitutional adsorption in the two AI(111)-(~/3 x ~ / 3 ) R 3 0 ~ phases, it was possible to estimate the activation energy for the transition [46]. Thus the proposed mechanism involves a place exchange of a K atom in an ontop site with the underlying A1 atom, followed by a diffusion of the ejected A1 atom to a step where it is readsorbed. Since the energy of surface diffusion was calculated [85] to be only 0.04 eV, it can be assumed that the place exchange is rate-determining, with an activation energy given by E a c t - - E pair - Evac lZ;,subs l:;,on-top ~ad --*-'ad ). In other words, the activation energy is the energy required to create a Frenkel pair, minus the difference in binding energy between adsorption in substitutional and on-top sites. Ignoring the small difference in the calculated adsorption energies leads to a value of Eact "~ 1 . 2 - 0 . 4 1 -- 0.79 eV, compatible with the measured onset temperature of the phase transition of 220 K from CLS measurements [66]. Finally, DFT calculations [83] for different alkali coverages also provided an explanation for another unexpected property of alkali adsorption on aluminum, namely that many of the structure develop by island growth, even when no intermixing occurs with the substrate [51 ]. It was found that depolarisation of the adsorbate dipoles occurs already at quite low coverages, thereby reducing the tendency to form dispersed structures.

7.2. A I ( l l l ) - ( 2 x 2 ) - N a The AI(111)-(2 x 2 ) - N a phase, as described in Sec. 3.1, is a complicated multilayer surface alloy. We regard the successful prediction [58, 86] of the structure of this phase by DFT calculations, as evidenced by the detailed, quantitative agreement with experimental results shown in Table 6, as marking a major advance in the application of DFT theory to surfaces. Determination of the surface structure by DFT involved optimising the geometry for eight, qualitatively different models, including all those previously proposed for this system, leading to adsorption energies in the range - 0 . 1 8 eV to - 3 . 0 8 eV, where the latter value applies to the optimum structure shown in Fig. 9. This structure has been described as a four-layer surface alloy consisting of a N a - A 1 - N a sandwich on a reconstructed A1 layer containing a (2 x 2)

267

vacancy structure. It can alternatively be considered in terms of a substrate consisting of the vacancy layer, with Na atoms adsorbed in the vacancies, and in fcc sites, and the ejected A1 atoms from the vacancy adsorbed in hcp sites. The DFT calculations showed that an alternative possibility obtained from the optimum structure by switching the A1 atoms adsorbed in hcp sites with the Na atoms adsorbed in fcc sites was less favourable by 0.2 eV with respect to the optimal structure. As noted in Sec. 3.1, formation of the (2 x 2 ) - N a structure from the (~/3 x ~/3)R30~ structure by addition of 1/6 ML Na also requires the addition of 1/3 ML A1, which implies that the formation mechanism includes backdiffusion of A1 atoms from surface steps. However, the detailed mechanism of formation is not known. In fact much of the difficulty in solving this structure can be attributed to the difficulty in describing the transition from the (~/3 x ~/3)R30 ~ structure to the (2 x 2) structure in terms of local atomic rearrangements.

7.3. AI(100)-(~/5 x ~/5)R26.6~ As described in Sec. 4.2, the structure of the Al(100)- (~/5 x ~ / 5 ) R 2 6 . 6 ~ phase, formed by adsorption of 0.2 ML Na at 300K and cooling to 250 K, was determined in a combined DFT and LEED study, which yielded quantitative agreement between experiment and theory for the surface geometry, as demonstrated by the comparison shown in Table 9. The calculations showed that not only is adsorption of Na in the substitutional site energetically favoured over adsorption in normal sites, but that the (~/5 x ~/5)R26.6 ~ structure is energetically favoured over an alternative structure containing islands of the substitutional c(2 x 2) structure that is energetically favoured at 0.5 ML coverage of Na. This agreement with experiment is not limited to the geometrical parameters, but also includes the Na-induced shifts in core-level binding energy of the A1 2p level, which can be calculated and used for distinguishing among different structural models. Thus, for the optimum (~/5 x ~/5)R26.2 ~ structure with Na adsorbed in substitutional sites, the calculations yield A1-2p shifts of - 1 9 0 meV for the A1 atoms remaining in the first layer, and - 6 0 meV and - 2 0 meV for the two inequivalent kinds of A1 atoms in the second layer. By contrast, for adsorption in a four-fold hollow site, the calculated A1-2p shifts are - 7 5 meV and - 5 5 meV for the two kinds of A1 atoms in the first layer, and -t-40 meV and -t-65 meV for the two kinds of A1 atoms in the second layer. Comparing these values to the measured shifts of - 2 0 0 meV and - 4 5 meV for the two shifted components seen in spectra from the AI(100)-(~/5 x ~ / 5 ) R 2 6 . 2 ~ overlayer unambiguously excludes a four-fold hollow site and gives strong support to the

268 assignment of the substitutional adsorption site.

7.4. A l ( 1 0 0 ) - c ( 2 x 2 ) - N a As described in Secs. 4.1, 4.1, and 6.2, adsorption of 1/2 M L Na on AI(100) at low temperature and at room temperature leads to the formation of two different c(2 x 2) structures. DFI" calculations [67, 68] played a decisive role in solving the structure of the room temperature phase by showing that previously proposed models were incorrect, and predicting that the structure contained Na atoms in four-fold substitutional sites. This was later cons by a LEED study [20]. Comparison of theory and experiment for the two c(2 x 2 ) - N a phases is given in Table 18, from which a quantitative agreement can be seen. Table 18 Comparison of theory and experiment for the surface geometry of the two Al(100)-c(2 x 2)-Na phases with Na atoms in four-fold hollow and substitutional sites respectively. The interlayer spacings between the i'th and j'th layers are denoted dij (,~,), where d01 (/~) is the Na-A1 spacing. The calculated adsorption energy is denoted Ead (eV atom -1). Method Ref. site dol d12 d23 Ead LEED DFT

[20] [67, 68]

hollow hollow

2.57 2.35

2.03 1.95

2.02 1.97

- 1.59

LEED DFT

[20] [67, 68]

subst. subst.

1.10 1.10

1.84 1.81

2.04 1.99

- 1.70

DFT calculations as a function of Na coverage also indicate that occupation of the four-fold hollow site is in fact energetically favourable at coverages up to 0.2 ML, after which occupation of the substitutional site is favoured [87].

7.5. A l ( 1 0 0 ) - c ( 2 x 2 ) - L i As shown in Table 19, calculations [88] using the full-potential linearised augmented plane wave (FLAPW) method for the A I ( 1 0 0 ) - c ( 2 x 2 ) - L i phase formed by adsorption of 0.5 ML Li at room temperature cons quantitatively the results of the LEED analysis described in Sec. 4.3 where, as shown in Fig. 15, Li was found to adsorb in a four-fold, substitutional site. Adsorption energies were also calculated for Li adsorbed in a four-fold hollow site in a c(2 • 2) structure and in a two-fold substitutional site in a (2 x 1) structure with results of - 2 . 5 1 eV, and - 2 . 5 9 eV, respectively, as compared to the value o f - 2 . 6 3 eV calculated for Li adsorbed in a four-fold substitutional site in a c(2 x 2) structure. An interesting feature of these calculations was the

269

Table 19 Comparison of theory and experiment for the surface geometry of the Al(100)-c(2 x 2 ) - L i phase with Li atoms in substitutional sites. The interlayer spacings between the i'th and j ' t h layers are denoted dij (~), where d01 (A) is the Li-A1 spacing. The calculated adsorption energy is denoted Ead (eV atom -1). Method Ref. d01 d12 d23 Ead LEED DFT

[21] [88]

0.32 0.40

1.90 1.84

2.03 2.06

-2.63

lower formation energy of a vacancy, Evac : 0.27 eV, in a (2 x 1) structure as opposed to the value of Evac : 0.53 eV for a c(2 x 2) structure. However, this was more than compensated for by a considerably larger binding energy for Li in the four-fold substitutional site.

7.6. Al(100)-c(2 x 2 ) - 2 L i DFT calculations have also proved to be most helpful in the analysis of the Li-A1 system where bulk alloys are formed. Freeman and coworkers [89-91] showed that an A13Li alloy with a Cu3Au-type structure is stable at 0 K but metastable at room temperature. In particular, it was shown that the bulk A13Li alloy is more stable in a Cu3Au-type structure than in an A13Ti-type structure [90]. Based on these findings it was proposed [69] that the multilayer A13Li surface alloy formed by adsorption of 1 ML of Li on AI(100) corresponded to a Cu3Au-type structure, containing a mixed A1/Li layer followed by a pure A1 layer, followed by a second mixed A1/Li layer. As described in Sec. 4.4, this proposal was shown to be incorrect in a recent LEED and DFT study [72]. In fact the multilayer A13Li surface alloy adopts an A13Ti-type structure rather than a Cu3Au-type structure. LEED results and DFT calculations agree quantitatively on the details of this structure, as can be seen in Table 12. DFT calculations were particularly useful for this system in that they enabled further studies of systems which are not easy to prepare experimentally. Thus calculations for five-layer surface alloys showed that the lower layers of the surface alloy revert to a Cu3Au-type structure, but that the stacking fault at the surface which converts a Cu3Au-type structure into an A13Ti-type structure (see Fig. 19) is still present. Further calculations of the surface structure of bulk alloys then led to the prediction that the surface structure of the bulk A13Li alloy has a stacking fault, such that it consists of a three-layer A13Ti-type structure on a Cu3Au-type bulk. At the present time these novel theoretical predictions await experimental confirmation. Attempts to prepare thicker surface alloys by deposition of Li on AI(100) indicate that diffusion of Li into the bulk occurs

270

rather than growth of further mixed A1/Li layers. 8. S U M M A R Y AND C O N C L U S I O N S It is an axiom of Surface Physics and Chemistry that an understanding of the surface geometry is a prerequisite for the understanding of many other surface phenomena. For the adsorption of alkali metals on aluminium surfaces, as shown in previous sections, the detailed geometrical structure is now known with high precision for about twenty binary and three ternary surface phases, spanning all the alkali elements as well as the three most close-packed aluminium surfaces. The results of these many studies constitute a large database against which different techniques and theoretical ideas can be tested. The database of surface structures also provides a starting point for studies of other physical properties such as the surface electronic [92] and vibrational [93, 94] structure, and chemical properties such as the catalytic influence of adsorbed alkali metals [95, 96]. The main new phenomenon described in this Chapter is the formation of binary and ternary surface alloys between A1 and the elements Na, K, Rb, and possibly Cs, that are immiscible in bulk A1. The ability of A1 to form surface alloys with alkali metals is explained by DFT calculations in terms of the low energy required for formation of a vacancy in a close-packed A1 surface, and the large binding energy of alkali atoms in this vacancy. Adsorption of Li, which is known to form bulk alloys with A1, also leads to the formation of surface alloys which are, perhaps surprisingly, similar to those formed by the heavier alkalis, at least at submonolayer coverages. However, Li adsorption in general continues beyond the submonolayer regime. For this system, the studies of surface alloys shed new light on the structure of the metastable A13Li bulk alloy. In summarising the types of surface alloys that are found, it is convenient to distinguish between single-layer and multi-layer surface alloys, as listed in Table 20. Quite generally, the single-layer surface alloys are formed by ejection of some fraction of a monolayer of A1 atoms from the first layer of the surface, and occupation by alkali atoms of the vacancies that are thus created. The ejected A1 atoms do not form a part of the structure, but are presumed to be readsorbed at surface steps. These alloys are the two-dimensional analogue of ordered, binary bulk alloys, but their formation is only possible for the larger alkalis because size constraints are lifted at the surface. A general description of the multilayer surface binary alloys formed by Na, and the multilayer surface ternary alloys formed by coadsorption of Na with K,

271

Table 20 Surface alloys formed by adsorption of alkali metals on aluminium surfaces. Single-layer surface alloys Multilayer surface alloys AI(111)- (~/3 x ~/3) R30~ Na, K, and Rb Al(100)-c(2 x 2 ) - L i and Na AI(110)-c(2 x 2 ) - L i and Na

AI(111)-(2~/3 • 2~/3) R 3 0 ~ (?) AI(111)-(2 • 2 ) - N a A l ( l l l ) - ( 2 • 2 ) - N a / K , Rb or Cs Al(100)-c(2 • 2 ) - 2 L i Al(l10)-(4 • 1 ) - 3 N a

Rb, or Cs, is only possible in part. They can be described in terms of adsorption on a substrate whose first layer contains some fraction of a monolayer of vacancies, just as in the case of the single-layer surface alloys. These vacancies are occupied with a first layer of Na. A second layer of Na (or K, Rb, or Cs in the case of the ternary alloys) is then adsorbed on the A1 vacancy layer. However, in the case of the AI(111)-(2 x 2) multilayer surface alloys, this second layer of Na is adsorbed in fcc sites on the A1 vacancy layer, and the A1 atoms ejected from the vacancy layer are readsorbed in hcp sites on the vacancy layer. By contrast, in the AI(110)-(4 x 1 ) - 3 N a multilayer surface alloy, the second Na layer is adsorbed in sites of low symmetry on the vacancy layer, and the A1 atoms ejected from the vacancy layer do not form a part of the structure, but presumably diffuse to surface steps where they are readsorbed. The multilayer surface alloy formed by Li adsorption on AI(100) is exceptional in that substitution of 1/2 ML A1 by Li occurs in both the first and third A1 layers. An unexpected feature of this structure is that the registry of Li atoms in the first and third layers is such that they are staggered along the surface normal direction, as in the A13Ti-type bulk alloy structure, rather than collinear, as in the expected Cu3Au-type bulk alloy structure known to be adopted by the metastable, A13Li bulk alloy. DFT calculations for this system lead to the novel prediction that the A13Li bulk alloy has a stacking fault at the surface, such that it can be described as an A13Ti-type surface on a Cu3Au-type bulk. The detailed geometries of the surface alloys reveal that the bond lengths between the alkali and aluminum atoms are close to the sums of the corresponding metallic radii, suggesting that the bonding is essentially metallic. Several geometrical properties of the surface alloys follow from simple considerations of the size of the atomic radii. For example, in the substitutional A l ( l l l ) - ( ~ / 3 x ~/3)R30~ structure, Li atoms lie almost in the plane of the first A1 layer, whereas in the corresponding Rb structure, Rb atoms are located much further from the surface plane, as shown in Fig. 8. Also, the smaller Li and Na atoms are the only alkali atoms which can occupy sites not directly

272

in the surface layer. Simple considerations of size also indicate that the metallic radius of Cs is too large for Cs to form a substitutional (~/3 x ~/3)R30 ~ structure on AI(111), as can be judged from Fig. 8. Finally, for adsorption in on-top sites, in the metastable, low temperature AI(111)-(~/3 x ~ / 3 ) R 3 0 ~ Rb, and Cs structures, the bond lengths indicate a smaller size of the alkali atoms, which allows a (~/3 x ~/3)R30 ~ structure to be formed also by Cs, as shown in Fig. 5, and which also points to a more ionic bonding than for the substitutional structures. In addition to the strong reconstructions of the A1 substrate in these single and multi-layer surface alloys, further weak reconstructions are found, in which A1 atoms in the first few layers are subject to both vertical and lateral displacements from their bulk positions. Accurate treatment of these weak reconstructions is important in both LEED and DFT analyses in order to achieve agreement with experiment. Thus the alkali on aluminum system exhibits a wide spectrum of adsorbate-induced reconstructions of the substrate. A second new phenomenon reported in this Chapter is the occurrence of order-order and order-preserving irreversible phase transitions between adsorbed phases, in which the adsorbed atoms change sites during the phase transition. In general, the phase transitions occur between metastable phases, in which the alkali atoms are adsorbed in normal sites, to surface alloy phases, in which the alkali atoms are adsorbed in substitutional sites formed by ejecting A1 atoms from the substrate. These phase transitions occur at or below room temperature. Thus, although the formation of surface alloys is an activated process, the activation energy on A1 surfaces is quite low, due to the relatively low formation energy of surface vacancies. In addition to the information concerning the mechanism of formation of surface alloys that can be deduced from studies of the phase transitions, further insight derives from a consideration of the evolution of the surface structures with alkali coverage. For example, for the AI(111)-(2 x 2 ) - N a phase, formed at a coverage of 1/2 ML Na, the mechanism of formation involves a partial lifting of the reconstruction of the substrate. The structure contains a vacancy layer where 1/4 ML of A1 atoms have been ejected. However, these atoms reappear as an adsorbed layer on the vacancy layer. Since the (2 x 2 ) - N a phase is obtained by further adsorption of 1/6 ML on the (x/3 x ~ / 3 ) R 3 0 ~ phase, which contains a vacancy layer from which 1/3 ML A1 atoms have been ejected, formation of the (2 x 2 ) - N a phase also requires the addition of 1/3 ML A1 atoms. As a second example, the AI(110)-(4 x 1 ) - 3 N a phase formed at a coverage of 3/4 ML Na, contains a vacancy layer from which 1/4 ML A1 atoms have been ejected. The A l ( l 1 0 ) - ( 4 x 1 ) - 3 N a phase is formed from the AI(110)-c(2 x 2 ) - N a phase by addition of 1/4 ML Na. However, the

273

c(2 x 2 ) - N a phase contains a vacancy layer from which 1/2 ML A1 atoms have been ejected. Thus formation of the (4 x 1 ) - 3 N a phase from the c(2 x 2 ) - N a phase also requires addition of 1/4 ML A1. From these examples it can be concluded that surface steps can act both as sinks for and sources of A1 atoms in the formation of surface alloys. Finally, we note that the creation of the database of surface structures described in this Chapter was the result of a serendipitous confluence of people and techniques. At the beginning of the 1990's, LEED theory and DFT were sufficiently mature, and sufficient computing power was available, that analysis of quite complicated structures could be carried out. At the same time, powerful experimental methods in both LEED and synchrotron-based CLS and SEXAFS had been developed. With hindsight, the new phenomena were just sitting there waiting to be discovered. That they were discovered owes in no small part to the conviction of Jochen Haase, virtuso of the SEXAFS method, that if SEXAFS showed that the surface of a free-electron-like metal was reconstructed by the adsorption of a large alkali atom, then we sceptics had better believe it! We believe that the rapid advances subsequent to the SEXAFS study [7] were in large part the result of the combined power of the different methods that were used, and the willingness of the groups involved at the Fritz Haber Institute in Berlin, at Lund University, and at Aarhus University, to informally share their results at early stages of their different investigations. ACKNOWLEDGEMENTS The authors would like to thank their colleagues and collaborators Mikael Borg, Jeppe Burchhardt, Jochen Haase, SCren Hoffmann, Edvin Lundgren, Anders Mikkelsen, Wolfgang Moritz, Martin Nielsen, Ralf Nyholm, Jakob Petersen, Matthias Scheffler, and Cathy Stampfl for their contributions to the work reviewed here and for many stimulating discussions. Support of this work by the Danish and Swedish Natural Science Research Councils is gratefully acknowledged.

REFERENCES [1] [2] [3] [4] [5]

R. W. Gurney, Phys. Rev. 47 (1935) 479. N. Lang, Phys. Rev. B 4 (1971) 4234. N. Lang, A. R. Williams, Phys. Rev. B 18 (1978) 616. R. D. Diehl, R. McGrath, Surf. Sci. Rep. 23 (1996) 72. D.M. Gruen, A. R. Krauss, M. H. Mendelsohn, S. Susman, J. Nucl. Mater. 111/112 (1982) 831. [6] P. G. Partridge, int. Materials Rev. 35 (1990) 37.

274 [7] A. Schmalz, S. Aminpirooz, L. Becker, J. Haase, J. Neuegebauer, M. Scheffler, D. R. Batchelor, D. L. Adams, E. BCgh, Phys. Rev. Lett. 67 (1991) 2163. [8] D. L. Adams, S. E Andersen, J. Burchhardt, in: S. Y. Tong, M. A. V. Hove, X. Xide, K. Takayanagi (Eds.), The Structure of Surfaces III, Springer, Berlin, 1991, p. 156. [9] M.M. Nielsen, J. Burchhardt, D. L. Adams, E. Lundgren, J. N. Andersen, Phys. Rev. Lett. 72 (1994) 3370. [ 10] A. Mikkelsen, D. L. Adams, Phys. Rev. B 60 (1999) 2040. [11] D. L. Adams, J. Phys. C 16 (1983) 6101. [12] D. L. Adams, V. Jensen, X. E Sun, J. H. Vollesen, Phys. Rev. B 38 (1988) 7913. [13] J. B. Pendry, Low Energy Electron Diffraction, Academic Press, London, 1974. [14] M. A. V. Hove, S. Y. Tong, Surface Crystallography by LEED, Springer-Verlag, Berlin, 1979. [15] V. L. Moruzzi, J. F. Janak, A. R. Williams, Calculated Electronic Properties of Metals, Pergamon, New York, 1978. [ 16] A. Guinier, X-Ray Diffraction, W. H. Freeman and Company, San Francisco, 1963. [17] J. Burchhardt, M. M. Nielsen, D. L. Adams, E. Lundgren, J. N. Andersen, Phys. Rev. B 50 (1994) 4718. [ 18] M. M. Nielsen, S. V. Christensen, D. L. Adams, Phys. Rev. B. 54 (1996) 17902. [19] M. M. Nielsen, J. Burchhardt, D. L. Adams, Phys. Rev. B 50 (1994) 7851. [20] W. Berndt, D. Weick, C. Stampfl, A. M. Bradshaw, M. Scheffler, Surf. Sci 330 (1995) 182. [21] J. H. Petersen, A. Mikkelsen, M. M. Nielsen, D. L. Adams, Phys. Rev. B 60 (1999) 5963. [22] H. B. Nielsen, J. N. Andersen, L. Petersen, D. L. Adams, J. Phys. C 15 (1982) Lll13. [23] J.N. Andersen, H. B. Nielsen, L. Petersen, D. L. Adams, J. Phys. C 17 (1984) 173. [24] A. Mikkelsen, J. Jiruse, D. L. Adams, Phys. Rev. B 60 (1999) 7796. [25] K.M. Ho, K. E Bohnen, Phys. Rev. B 32 (1985) 3446. [26] N. Marzari, D. Vanderbilt, A. D. Vita, M. C. Payne, Phys. Rev. Lett. 82 (1999) 3296. [27] W. E Egelhoff, Surf. Sci. Rep. 6 (1986) 253. [28] A. Nilsson, B. Eriksson, N. Mgtrtensson, J. N. Andersen, J. Onsgaard, Phys. Rev. B 38 (1988) 10357. [29] J.N. Andersen, C. O. Almbladh, J. Phys. C, to be published. [30] J.N. Andersen, O. Bj6rneholm, A. Sandell, R. Nyholm, J. Forsell, L. Th~nell, A. Nilsson, N. M~rtensson, Synchrotron Radiation News 4 (1991) 15. [31 ] E. UggerhCj, Nucl. Instrum. Methods Phys. Res. B 99 (1995) 261. [32] R. Nyholm, J. N. Andersen, U. Johansson, B. N. Jensen, I. Lindau, Nucl. Instr. and Methods in Phys. Res. A 491 (2001) 99. [33] M. Borg, et al, to be published. [34] D. E Woodruff, A. M. Bradshaw, Rep. Prog. Phys. 57 (1994) 1029. [35] D.L. Adams, J. N. Andersen, in preparation. The program FitXPS used to fit XPS spectra with the convolution of a Doniach-Sunjic line form and a Gaussian is available from the author at ftp://boopic.ifa.au.dk/pub/fitxps/. [36] S. Doniach, M. ~,unjid, J. Phys. C 3 (1970) 285. [37] G. D. Mahan, Phys. Rev. 163 (1967) 612. [38] E Nozi~res, C. T. DeDominics, Phys. Rev. 178 (1969) 1097. [39] J.N. Andersen, A. Beutler, S. L. Sorensen, R. Nyholm, B. Setlik, D. Heskett, Chem. Phys. Lett. 269 (1997) 371. [40] J. N. Andersen, T. Balasubramanian, C. O. Almbladh, L. I. Johansson, R. Nyholm, Phys.

275 Rev. Lett. 86 (2001) 4398. [41] D. L. Adams, M. M. Nielsen, J. Burchhardt, J. N. Andersen, Phys. Rev. B 60 (1999) 11078. [42] S. ~. Lindgren, L. Walld6n, J. Rundgren, E Westrin, J. Neve, Phys. Rev. B 28 (1983) 6707. [43] D. Fisher, S. Chandavarker, I. Collins, R. D. Diehl, E Kaukasoina, M. Lindroos, Phys. Rev. Lett. 68 (1992) 2786. [44] C. Stampfl, M. Scheffler, H. Over, J. Burchhardt, M. M. Nielsen, D. L. Adams, W. Moritz, Phys. Rev. Lett. 69 (1992) 1532. [45] M. Kerkar, D. Fisher, D. E Woodruff, R. G. Jones, R. D. Diehl, B. Cowie, Surf. Sci. 278 (1992) 246. [46] C. Stampfl, M. Scheffler, H. Over, J. Burchhardt, M. M. Nielsen, D. L. Adams, W. Moritz, Phys. Rev. B 49 (1992) 4959. [47] M. M. Nielsen, J. Burchhardt, D. L. Adams, J. N. Andersen, Phys. Rev. B 58 (1998) 12655. [48] A. Schmalz, S. Aminpirooz, J. Haase, D. R. Batchelor, M. M. Nielsen, E. BCgh, D. L. Adams, Surf. Sci. 301 (1994) L211. [49] M. Kerkar, D. Fisher, D. E Woodruff, R. G. Jones, R. D. Diehl, B. Cowie, Phys. Rev. Lett. 68 (1992) 3204. [50] H. Brune, J. Wintterlin, R. J. Behm, G. Ertl, Phys. Rev. B 51 (1995) 13592. [51] J.N. Andersen, E. Lundgren, R. Nyholm, M. Quarford, Surf. Sci. 289 (1993) 307. [52] D. L. Adams, J. N. Andersen, et al, unpublished. [53] A. Mikkelsen, unpublished response to an undergraduate excercise in Surface Physics, Aarhus University (1996). [54] C. Mogensen, S. M. Gray, to be published. [55] J. O. Porteus, Surf. Sci. 41 (1974) 515. [56] A. Hohlfeld, K. Horn, Surf. Sci. 211 (1989) 844. [57] J. N. Andersen, M. Quarford, R. Nyholm, J. J. van Acker, E. Lundgren, Phys. Rev. Lett. 68 (1992) 94. [58] J. Burchhardt, M. M. Nielsen, D. L. Adams, E. Lundgren, J. N. Andersen, C. Stampfl, M. Scheffler, A. Schmalz, S. Aminpirooz, J. Haase, Phys. Rev. Lett. 74 (1995) 1617. [59] R. Fasel, E Aebi, J. Osterwalder, L. Schlapbach, R. G. Agostino, G. Chiarello, Phys. Rev. B 50 (1994) 14516. [60] S.V. Christensen, J. Nerlov, K. Nielsen, J. Burchhardt, M. M. Nielsen, D. L. Adams, Phys. Rev. Lett. 76 (1996) 1892. [61 ] S. Aminpirooz, A. Schmalz, L. Becker, N. Pangher, J. Haase, M. M. Nielsen, D. R. Batchelor, E. BCgh, D. L. Adams, Phys. Rev. B 46 (1992) 15594. [62] M. Borg, A. Mikkelsen, M. Birgersson, M. Smedh, E. Lundgren, D. L. Adams, C. O. Almbladh, J. N. Andersen, to be published. [63] R. Fasel, M. Gierer, H. Bludau, E Aebi, J. Osterwalder, L. Schlapbach, Surf. Sci. 374 (1997) 104. [64] B. Hutchins, T. Rhodin, J. Demuth, Surf. Sci. 54 (1976) 419. [65] M. A. V. Hove, S. Y. Tong, N. Stoner, Surf. Sci. 54 (1976) 258. [66] J. N. Andersen, E. Lundgren, R. Nyholm, M. Quarford, Phys. Rev. B 46 (1992) 12784. [67] C. Stampfl, J. Neugebauer, M. Scheffler, Surf. Sci. 307/309 (1994) 8. [68] C. Stampfl, J. Neugebauer, M. Scheffler, Surf. Rev. Lett. 1 (1994) 222. [69] J. H. Petersen, C. SCndergaard, S. V. Hoffmann, A. Mikkelsen, D. L. Adams, Surf. Sci.

276 437 (1999) 317. [70] D. Vaughan (Ed.), X-Ray Data Booklet, University of California, Berkeley, CA, USA, 1986. [71] E. Lundgren, R. Nyholm, M. Qvarford, J. N. Andersen, D.Heskett, S. V. Christensen, Phys. Rev. B 56 (1997) 12560. [72] A. Mikkelsen, M. Borg, J. H. Petersen, J. N. Andersen, D. L. Adams, Phys. Rev. Lett. 87 (2001) 096102. [73] A. Mikkelsen, S. V. Hoffmann, J. Jiruse, D. L. Adams, Phys. Rev. B 61 (2000) 13988. [74] A. Mikkelsen, J. H. Petersen, S. V. Hoffmann, J. Jiruse, D. L. Adams, Surf. Sci. 487 (2001) 28. [75] A. Mikkelsen, J. H. Petersen, J. Jiruse, S. V. Hoffmann, D. L. Adams, Surf. Sci. (2001) in press. [76] J. Wang, Z. C. Ying, E. W. Plummer, Phys. Rev. B 51 (1995) 5590. [77] E. Lundgren, R. Nyholm, J. Burchhardt, D. Heskett, J. N. Andersen, Surf. Sci. 343 (1995) 37. [78] J. Neugebauer, M. Scheffler, Phys. Rev. B 46 (1992) 16067. [79] E. Lundgren, PhD thesis, Lund University, unpublished (1996). [80] E. Lundgren, A. Beutler, R. Nyholm, J. N. Andersen, D. Heskett, Surf. Sci. 370 (1997) 311. [81 ] M. Scheffler, C. Stampfl, in: K. Horn, M. Scheffler (Eds.), Handbook of Surface Science, Vol. 2: Electronic Structure, Elsevier, Amsterdam, 1999, Ch. 5, pp. 286-356. [82] J.C. Boettger, U. Birkenheuer, S. Krtiger, N. R6sch, S. B. Trickey, Phys. Rev. B 52 (1995) 2025. [83] J. Neugebauer, M. Scheffler, Phys. Rev. Lett. 71 (1992) 577. [84] J. Bornet, J. Neugebauer, M. Scheffler, Phys. Rev. B 49 (1994) 17242. [85] R. Stumpfl, M. Scheffler, Phys. Rev. Lett. 72 (1994) 254. [86] C. Stampfl, M. Scheffler, Surf. Sci. 319 (1994) L23. [87] C. Stampfl, M. Scheffler, Surf. Rev. Lett. 2 (1994) 317. [88] I. Ohsaki, T. Oguchi, J. Phys. Soc. Jap. 69 (2000) 2192. [89] X.-Q. Guo, R. Podloucky, A. J. Freeman, Phys. Rev. B 40 (1989) 2793. [90] X.-Q. Guo, R. Podloucky, J.-H. Xu, A. J. Freeman, Phys. Rev. B 41 (1990) 12432. [91] M. Sluiter, D. de Fontaine, X.-Q. Guo, R. Podloucky, A. J. Freeman, Phys. Rev. B 42 (1990) 10460. [92] C. Stampfl, K. Kambe, R. Fasel, E Aebi, M. Scheffler, Phys. Rev. B 57 (1998) 15251. [93] T. Nagao, Y. Iizuka, T. Shimazaki, C. Oshima, Phys. Rev. B 55 (1997) 10064. [94] S. E. Finberg, J. V. Larkin, R. D. Diehl, Surf. Sci. (2001) in press. [95] C. E McConville, A. B. Hayden, J. Robinson, D. E Woodruff, J. Phys. C 5 (1993) 4677. [96] D. E Woodruff, S. M. Driver, J. Ludecke, R. J. Dixon, P. B. J. Thompson, G. Scragg, B. Cowie, Surf. Sci. 391 (1997) 300.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces D.P. Woodruff, (Editor)

277

Chapter 8

The structure of surface alloy phases on metallic substrates D.P.Woodrut~ and E.Vlieg b

aphysics Department, University of Warwick, Coventry CV4 7AL, UK bNSRIM Department of Solid State Chemistry, University of Nijmegen, Toemooiveld 1, 6525 ED Nijmegen, The Netherlands 1. INTRODUCTION There are now many examples of surface alloy phases on metallic surfaces in which a sub-monolayer coverage of an 'adsorbate' deposited onto the surface occupies substitutional sites (i.e. strictly becomes an 'absorbate') to form a local alloy which is commonly comprised of only a single atomic layer. In some cases this occurs even for substrate-adsorbate elemental combinations which form no bulk alloy, and may be immiscible. While the existence of these surface phases has been widely recognised, the number of quantitative structure determinations is actually quite small, some of which have been reviewed in the past [1], although this volume contains several new such surveys. For surface alloy phases there are two key questions one can ask about the geometry: is the structure truly a substitutional surface alloy rather than an overlayer, and if this is the case, what are the exact atomic positional parameters such as the layer spacings of the constituent atoms relative to the underlying bulk? Linked to this second question is the degree of 'rumpling' of the surface alloy phase: i.e.~the extent to which the constituent atoms of the alloy layer are not strictly coplanar. In a small number of cases there is a further question, notably is there any kind of structural defect (notably a stacking fault) at the surface aUoy/substrate interface? The question of the surface layer rumpling is an interesting one because the effective radii of different metallic atoms differ in their pure bulk solids, but in a surface alloy phase the interatomic distances parallel to the surface are largely fixed at the lattice parameter of the underlying elemental solid. Depending on the symmetry of the reconstructed surface cell, some parallel relaxation is allowed, but this is found to be very small in practice. Evidently, therefore, if

278

the adsorbate atoms which substitute outermost layer substrate atoms to form the surface alloy have a larger metallic radius, one might expect these atoms to lie higher above the underlying substrate. While this is generally true, the degree of rumpling is almost always significantly smaller than that predicted by the simple hard-sphere picture implied by the foregoing argument, and the reason for this is probably attributable to the effects of the valence charge depletion at the surface/vacuum interface. There is, however, a notable exception to this situation for the case of the surface alloys formed by Mn on Cu(100), Ni(100), Pd(100) and Cu(110). On Pd(100) the rumpling amplitude is actually larger than expected on this basis, while on the other surfaces its behaviour is anomalous relative to other adsorbates; this seems to be explained by a quite different surface effect associated with the local magnetic moment on these surface Mn atoms. One special group of surface alloy phases for which these arguments are particularly complex are those formed by alkali atom deposition on A1 surfaces; the very large difference in the effective radius of the alkali atoms in their neutral metallic and ionic charge state makes this simple hard-sphere picture especially inappropriate, and these particular systems are dealt with separately in the chapter by Adams and Andersen. In this chapter we will therefore discuss a series of structurally wellcharacterised surface alloys formed on metal substrates by metallic (or nearmetallic) adsorbates and compare the measured interatomic distances and surface rumpling amplitudes with the estimates of these parameters to be expected on the basis of a simple hard-sphere picture. We group the alloy systems according to some common physical problems or characteristics, or to provide a clear contrast of such phenomena, but then review the issue of effective atomic radii in surface alloys in general. The structural methods used in the individual investigations are varied, and include diffraction methods (quantitative low energy electron diffraction- L E E D - and surface X-ray d i f f r a c t i o n - SXRD), local electron interference methods (scanned-energy mode photoelectron diffraction - P h D - angle-scan photoelectron diffractionXPD - and surface extended X-ray absorption spectroscopy) and various forms of ion scattering methods at low and high energies. These methods have been reviewed extensively and will not be discussed in detail here. In general, we concentrate only on the quantitative results of these structure determinations, although we illustrate the overall methodology by reference to some results from several different methods applied to the first group of surface alloy phases we present. 2. CASE STUDIES

2.1 Cu(111)/Sb and Ag(111)/Sb: interfacial stacking faults Interest in the adsorption of Sb on surfaces, including both metal and semiconductor surfaces, has been heightened by the ability of Sb surface layers

279

to act as surfactants in epitaxial growth; the presence of the Sb causes the growth to occur in a layer-by-layer fashion in growth systems which would otherwise develop island structures, and the Sb is ejected from the growing material to remain on the surface. In the case of metal growth, the first such example was the homoepitaxial growth of Ag (on Ag) [2, 3]. While the surfactant effect also occurs for lower coverages, it was found that at a coverage of 0.33 ML Sb a stable (~/3x~/3)R30~ phase occurs. The inference that the Sb occupies substitutional sites in this surface, thus forming a single-layer Cu2Sb surface alloy in this ordered phase, was first found in the results of density-functional theory (DFT) total energy calculations [4, 5], and some early low energy ion scattering experiments appeared to support this view [6], but full quantitative experimental structure determinations of this surface, and the comparable Cu(111)(~/3x~/3)R30~ surface [7], were only undertaken several years later. The first such investigation, of both of these surface phases, was obtained by surface X-ray diffraction [8], and reached a very surprising conclusion. For both surfaces the data provided a clear indication that not only is the outermost atomic layer a substitutional alloy, but that all the atoms on this alloy layer occupy not the 'fcc hollow sites' (directly above a third layer substrate atoms) as would be expected from simple substitution of surface Ag or Cu atoms by Sb atoms, but that all these alloy layer atoms occupy 'hcp hollows' directly above second layer substrate atoms (fig. 1). In effect, therefore, there is a stacking fault at the alloy/substrate interface. Whereas in an unfaulted surface the usual A-B-C-A-B-C sequence of stacking registry of the close-packed layers which characterise the bulk fcc structure would be terminated by an 'A' layer, the surface alloy actually occupies the 'B' site registry. During metal growth the Sb segregates and the metal atoms return to the correct stacking, while the new Ag atoms in the top layer again have the wrong stacking ('floating stacking fault') [9]. It appears that the interfacial stacking fault occurs only when the Sb surface coverage is sufficient to create the ordered (~/3x~/3)R30~ phase. Rather direct evidence of this comes from the results of an investigation of the intensity of the (0,1,0.3) X-ray reflection from a Ag(111) surface during Sb deposition at a substrate temperature of 250~ shown in fig. 2. At a time around 60 seconds, the shutter of the Sb Knudsen deposition cell was opened, but initially no change in the intensity is observed. This can be attributed to the fact that under these conditions, the Sb atoms simply substitute top-layer Ag atoms, but no other structural changes occur. Since the X-ray scattering crosssections of Sb and Ag atoms are very similar, no significant change in intensity occurs. After a total deposition of 1/3 ML, however, the intensity suddenly increases. A full structural analysis undertaken on the surface at this point showed that a (~/3x~/3)R30~ reconstruction with the interfacial stacking fault had occurred. A similar experiment for Sb deposition on Cu(111) showed very

280 similar behaviour, except that in this case an initial linear intensity increase with deposition time was observed between 0 and 0.3 ML Sb, because the scattering cross-sections of Sb and Cu are very different [8].

Fig.1 Schematic diagram showing the faulted (~/3x~/3)R30~ surface alloy phase formed on both Ag(111) and Cu(111) surfaces. The main structural parameters are defined in the figure and their values are given in Table 1.

281

~0

i000

'

.~.-4

~ .d

500

'

'

'

I

'

'

(0

0.3)

T =

250~

'

'

I

'

'

'

'

I

'

'

'

'

I

'

'

I

i

i

i

i

i

a

t

'

'

off Sb on

or--I

1/3 ML

l

(D

100 ,

,

,

\ ,

I

loo

,

!

i

i

'2 o'

i

'3oo

400

'500

Time (see) Fig. 2 The intensity of the (0,1,0.3) X-ray reflection from Ag(lll) during Sb deposition at a substrate temperature of 250~ The sudden rise in intensity at a coverage of 1/3 ML is associated with the formation of the ordered (~/3x~/3)R30~ faulted surface alloy phase as described in the text. From de Vries et al [8]

Fig. 3 STM images of Ag(lll) with different coverages of deposited Sb. The left-hand image corresponds to an area of 60 x 60 ,~2 with a coverage of 0.12 ML at room temperature. The right-hand image is of an area of 300 x 200 ~2 of an annealed surface with a coverage of 0.08 ML. After van der Vegt et al [ 10]. Scanning tunneling microscopy (STM) provides some insight into the ordering behaviour of the Sb atoms in the surface at low coverage, as shown in fig. 3. In the left-hand panel is shown an STM image of an area of 6 0 / ~ x 60 of the Ag(111) surface after the deposition of 0.12 M L Sb with the substrate at r o o m temperature. While the ordered (~/3x~/3)R30 ~ reconstruction corresponds to a Sb coverage of 1/3 ML, small patches of this reconstruction can be seen to have formed near the step edges at this substantially smaller average coverage. The regular array of white dots in the centre of this picture are identified with the Sb atoms. The grey dots in the upper terrace correspond to isolated Sb atoms. The fight-hand panel shows a lower magnification S T M image of the A g ( l l l ) surface with an average coverage of 0.08 M L Sb after an anneal

282

(displayed area 300 x 250/~2). At this even lower Sb coverage no (~/3x~/3)R30 ~ domains are seen, and a lattice gas of Sb atoms dispersed in the top layer of the substrate is energetically more favourable. Nevertheless, the image does show that Sb atoms do form local regular arrays at the step edges, the other adatoms being visible as grey dots in the terraces [ 10]. Fig. 4 shows an example of the clear evidence found in the SXRD study that the Ag(111)-(~/3xx/3)R30~ structure really does involve a stacking fault at the surface alloy/substrate interface. The solid circles show the results of measurements of the (01) diffracted beam intensity as a function of the momentum transfer perpendicular to the crystal surface, defined by the diffraction index I. Bulk Bragg peaks occur for l values of -4, -1, 2 and 5. Away from the conditions for these bulk reflections, the structure factor is sensitive to the surface structure. Each data point is obtained by measuring the integrated intensity using a so-called rocking scan at that particular location in reciprocal space. The results of calculations for two different structural models, each with optimised structural parameter values, are shown: the solid curve corresponds to the structure with the interfacial stacking fault, while the dashed curve is for a simple substitutional surface alloy. 300

.

.

,

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

0

o ~200

i

/

o/

~100

o- 5........

i--'4 ;3 ,

,

--'2 , _ 1- , diffraction

,

9

,

index

, o ,

:

,

l

Fig. 4 The (01) crystal truncation rod of the Ag(lll)-(x/3x~/3)R30~ structure. The experimental data points, shown as solid circles, are compared with the results of calculations for optimised versions of the faulted (solid curve) and unfaulted (dashed line) surface alloy models. An independent medium energy ion scattering (MEIS) study of the Cu(lll)(x/3x~/3)R30~ reached the same conclusion that the surface alloy does involve occupation of hcp hollow sites shortly after the original SXRD study [11], while the existence of the stacking fault in the

283

Ag(111)(~/3x~/3)R30~ was confirmed by a quantitative low energy electron diffraction (LEED) study subsequently [12]. A second SXRD study on Cu(111)/Sb favoured an unfaulted surface alloy, but did not consider the faulted model [ 13]. The MEIS study also found very clear evidence for the interfacial stacking fault as may be seen in fig. 5. MEIS provides structural information through the 'shadowing' of one atom by another due to the role of elastic scattering of the incident ions. This effect is exploited both for the incident ions for which scattering from near-surface atoms can prevent illumination of sub-surface atoms if the incidence direction corresponds to a shadowing direction in which the atomic positions are aligned, but also in the outgoing trajectories where surface atoms can prevent scattered ions from subsurface atoms reaching the detector. By convention this shadowing in the outgoing direction is referred to as 'blocking'. In a MEIS experiment the incident ion beam is directed in specific crystallographic directions designed to illuminate only a small number of near-surface layers, and the scattered ion intensity is measured as a function of scattering angle yielding so-called 'blocking curves'. Fig. 5 shows a comparison of the experimental blocking curves for two different incidence directions with the results of simulations for the optimised faulted and unfaulted alloy surface structures. Both the shapes of the blocking curves and the absolute scattering yields are clearly described far better by the faulted alloy structure. 3.0-

[11 O]

1.5-

4"0 t

,.

~ ."

2.5-"

. ~ 2.0-

Incidence

,

A-i" " ~,:

9

, |

t'~df

i'~.l

!

i #

i! v

"

.~

Oo

.~

:..

9.

,,,,.." .~.:::.. "

scattering angle

[degrees]

.

9

"-, 9

"

2.0- ,.

untaulted alloy

110

9 o

2.5

f,u.~d=,oy

.... 70, . , o . ; o . , . , . 1,O0. . ,

9 ,r,

\/ .

.....

#, 9

9

9

"

---

[1-1"2] i n c i d e n c e

.

.::~..,

120

130

1

90

.

5

~

100 110 120 scattering angle [degrees]

130

Fig. 5 Comparison of experimental 100 keV I-I+ MEIS blocking curves from the Cu(lll)(~/3x~/3)R30~ surface in two different incident ion directions with the results of simulations for the optimised faulted and unfaulted surface alloy structures. Adapted from Bailey et al [ 11 ].

Table 1 summarises the quantitative structural parameters (as defined in fig. 1) which emerged from the various structure determinations. Because there are several complementary studies of these surface alloy structures we provide a complete listing of all the interlayer spacings to allow comparison of the results

284

of these different methods although our primary interest is in the outermost layer spacing values; the table shows that the agreement between the results of the applications of the different methods is quantitatively as well as qualitatively, good. In addition to the perpendicular displacements, the SXRD study found small lateral displacement towards the Sb atom of the second layer metal atoms of 0.06/~ and 0.01/~ for Ag and Cu, respectively. Owing to the glancing angle geometry, in typical SXRD experiments the parallel momentum transfer is much higher than in LEED, and thus these small lateral displacements are quite accurately determined using SXRD whereas this is less so when using LEED, typically near normal incidence. For the perpendicular displacements, the opposite is typically true [ 14]. substrate

method

zsb2 (A)

zl2 (A)

Z23 (/~)

Ag(111)

DFT [ 161 SXRD [8] LEED [ 12] MEIS [ 15] DFT [16] SXRD [8] MEIS [ 11 ]

2.49 2.53 2.53 2.46 2.56 2.58 2.52

2.47 2.50 2.46 2.43 2.09 1.98 2.05

2.34 2.36 2.34 2.36 2.07 2.08" 2.08"

Cu(lll)

ZSb2-ZI2 rumple (/~) 0.02 0.03 0.07 0.03 0.47 0.60 0.47

Table 1 Comparisons of the optimised surface structural parameters (in /~) for the Ag(111)(x/3x~/3)R30~ and Cu(11l)(x/3xx/3)R30~ faulted alloy phases found in the experimental and theoretical studies. The asterisks denote values assumed to be equal to the bulk layer spacing. The parameters are defined in fig. 1 In an attempt to try to gain some understanding of this surprising result new DFT calculations [ 16] were undertaken; the original calculations for the Ag/Sb system which established the preference for substitutional sites had not considered the possibility of this stacking fault, but the new calculations covered both substrates and included both fcc and hcp overlayer and alloy phase models. The results of these calculations revealed the extremely low stacking fault energy (approximately 3 meV and 40 meV/(lxl) unit mesh for Ag and Cu, respectively) at the clean surfaces on both metals, and also indicated that the faulted alloy structure was energetically preferred, albeit by only a very small amount (approximately 12 meV/(x/3xx/3)unit mesh for both metals) relative to the unfaulted alloy phase (which is approximately 800 meV/unit mesh lower energy than the ovedayer structures). The fact that these stacking faults clearly have very small energies (positive for the clean surface and negative for the surface alloy according to these

285

calculations) is also reflected in the results of a very recent MEIS study of the A g ( l l l ) / S b system [15]. Typically the 0.33 ML Ag(lll)(~/3x~/3)R30~ phase is produced by dosing the surface with an excess of Sb and then heating to a temperature sufficient to drive off the excess Sb and produce the ordered surface phase. The excess Sb, however, appears to diffuse into the sub-surface, so a succession of such preparations can lead to some loading of the subsurface by Sb, despite that fact that this may not be visible in surface-specific experiments. The recent MEIS study found that under some circumstances this could actually lead to the presence of further stacking faults in the outermost 10-20 monolayers, perhaps triggered by the presence of this sub-surface Sb. In the case of the higher Sb coverage (approximately 0.6 ML) Ag(111)(2~/3x2~/3)R30~ phase this MEIS study even found evidence for the formation of an essentially hcp Ag phase in the near subsurface region. (Notice, incidentally, that in the phase diagram of bulk Ag-Sb alloys the 13-phase (also referred to as ~(Ag-Sb) and occurring in the composition range of approximately 10-17 atomic % Sb) does have an hcp structure [17]; the subsurface concentration in the MEIS experiments is not thought to be as high as 10%, but the phase transformation may be stabilised with lower compositions for a surface phase.) One further finding of the recent MEIS investigation is that in preparing the Ag(lll)(~/3x~/3)R30~ phase on different crystals with different histories of Sb dosing, it was found possible to prepare both faulted and unfaulted forms of this surface alloy phase. At coverages below 0.33 ML, SXRD shows that the Sb still occupies substitutional sites, but in this case the Sb atoms are completely disordered [10] and n o stacking fault is present [8]. Low energy angle-scan photoelectron diffraction data obtained at low coverage (0.22-0.26 ML) appear to provide independent evidence of substitutional adsorption in an unfaulted outermost Ag layer [18]. Strictly, these authors did not test the possibility of the faulted structure, but this type of data has previously been shown to be very sensitive to the difference between local fcc and hcp sites [19], so the quality of agreement of their model calculations with experiment should clearly exclude this possibility. These authors appear to have assumed that the Sb and Ag atoms are coplanar in the outermost alloy layer, but also report that tests at different spacings indicate that the height difference between the Sb and Ag atoms is less than about 0.1 A. This indicates that the local rumpling of the disordered Sb atoms is similar to found in the ordered faulted reconstruction. These results strongly suggest that the formation of the stacking fault is dependent on coverage and seems to coincide with the formation of the (~/3x~/3)R30~ phase. The one conflicting piece of evidence is the most recent MEIS experiment which shows that under some circumstances an ordered 013x~/3)R30 ~ phase may be seen in LEED but the average surface, as seen in MEIS, is not faulted.

286

It is possible that this apparent conflict arises from the different way that MEIS and true diffraction methods sample different parts of the surface. The transition to a top layer with a stacking fault and the (x/3x~/3)R30~ periodicity occurs at a Sb coverage around 0.3 ML for both Ag and Cu. This sudden transition resembles the phase transition in the exactly solvable hardhexagon model in statistical mechanics, in which hexagons can be positioned on three equivalent sites, but may not overlap. In this model, ordering is predicted above a critical coverage of 0.28 ML. The qualitative agreement with the Sb induced reconstruction suggests that the main driving force for this ordered phase is a strong repulsive interaction between the Sb atoms [8].

2.2 Ni(lll)/Pb: a case of strongly suppressed surface alloy rumpling In view of the surface stacking fault found in these (~/3x~3)R30 ~ adsorption phases of Sb on Cu(111) and Ag(111) there has been some interest in exploring other similar systems, previously believed to involve a surface alloy phase, to check for the possibility of similar effects. One such surface phase which has been studied by several different methods is Ni(111)(~3x~/3)R30~ An early low energy ion scattering (LEIS) investigation of this surface [20] had concluded that the surface does involve substitutional adsorption of Pb but at that time the possibility of a surface stacking fault was not explored. A more recent MEIS study [21] was undertaken to test this aspect, MEIS being far more sensitive to the subsurface structure and thus to the registry of the surface alloy/substrate interface. Very recently quantitative LEED has also been used to provide further independent structural information on this surface [22]. While these more recent studies were motivated to a significant degree by a desire to establish whether or not this system shows a faulted or unfaulted surface alloy, perhaps the most interesting parameter to emerge is the (small) amplitude of rumpling within the surface alloy layer. As described in the previous section, MEIS provides structural information through the 'shadowing' of one atom by another due to the role of elastic scattering of the incident ions, and the equivalent 'blocking' of scattered ions on the outward trajectory to the detector. By an appropriate choice of the scattering geometry, one becomes sensitive to particular structural features. Scattering experiments in the <112> azimuth (the geometries used in the experiments on Cu(lll)(~/3xx/3)R30~ shown in fig. 5) clearly show that there is no interfacial stacking fault for the Ni(111)/Pb system, while additional measurements using <123> incidence in a <110> azimuth identify the substitutional surface alloy model as favoured over the simple ovedayer model. In comparison to the Ag(111)/Sb and Cu(111)/Sb surface phases, therefore, a key difference in the Ni(111)/Pb system is that while it does correspond to a surface alloy phase, it does n o t involve a surface stacking fault. More detailed inspection of the structural parameters to emerge from this study, however,

287

shows that potentially even more interesting are the values of the effective atomic radii of the atoms in the surface alloy as deduced from the experimental value of the surface rumpling amplitude. In their bulk elemental forms the effective radii of Ni and Pb (i.e. half the value of the nearest-neighbour interatomic distances in these fcc solids) are substantially different, 1.25/~ for Ni and 1.75 ,& for Pb. Clearly a simple model based on touching hard spheres with these radii in which a Pb atom replaces a surface Ni atom will lead to the Pb atom failing to fit into the vacant site and thus sitting with its centre significantly above the surrounding Ni atoms; the actual amplitude of this surface alloy rumpling predicted by this simple calculation is 1.67/~, almost as large as the (111) layer spacing of the Ni substrate (2.04 ~). In fact, the MEIS study finds this value in practice to be only 0.65+0.15 /~, clearly very significantly lower. Inverting this argument to calculate the effective radius of the Pb atoms in the surface alloy leads to a value of 1.33_+0.04 ~. Curiously, the original LEIS study [20] concluded that the rumpling was even smaller (0.2 A) implying an even smaller effective radius for the Pb atoms (1.25 /~) in the surface layer, but did not comment on this. While the MEIS study clearly did favour the surface alloy structural model, this surprisingly large reduction in the effective radius of the Pb atoms led to a more recent re-investigation using quantitative LEED [22], a technique based on wholly different physical principles and thus a valuable independent check. This new investigation confirmed that the unfaulted substitutional surface alloy is the correct structural model and yielded a value for the surface layer rumpling of 0.73+_0.07 ~, in excellent agreement with the results of the MEIS study. The large reduction of the effective radius of the Pb atoms in this surface phase is therefore clearly real. We will return to this effect, which is actually systematic of almost all the systems discussed in this chapter, but is especially pronounced in this example, in section 3. substrate

adsorbate

Ni(lll)

Pb

rumpling amplitude (/~)

0.2 [20]

substrate metallic radius ,(A) 1.246

adsorbate metallic radius (A), 1.750

0.65 + 0.15 [21] 0.73 + 0.07 [22] Table 2 Summary of structural parameters obtained for the Ni(111)/Pb surface alloy system

288

2.3 Mn and non-magnetic metals on Cu(100), Ni(100) and Pd(100): effect of local magnetic moments

Fig. 7 Schematic plan and side view of the structure of the c(2x2) surface alloy phases

formed on a number of fcc (100) surfaces. A small group of systems involving fcc (100) surfaces, first investigated somewhat earlier than the (111) examples described so far, include the c(2x2) surface phases formed by Mn on Cu(100), Ni(100) and Pd(100) and by Au and Pd on Cu(100). In all these cases the surface layer comprises half a layer of the substrate species and half a layer of the adsorbate species to form a substitutional ordered surface alloy layer of 1:1 stoichiometry (fig. 7). Cu/Au

289 and C u ~ d both form a range of bulk alloy phases and have attracted considerable interest as reflected by other articles in this volume, with electronic structure and surface chemical reactivity both being investigated. They may be regarded as reference systems expected to display 'simple' structural properties. The quantitative structure of their c(2x2) ordered surface phases has been studied in early quantitative LEED work and in some more recent investigations. Interest in the structure of Mn on Cu surfaces in particular, and in the solid solution of Mn in bulk Cu, has been mainly motivated by their novel magnetic properties. As a free atom Mn has a large magnetic moment, having a halffilled 3d shell and thus, according to Hund's rules, the largest possible d-state magnetic moment. In bulk Mn metal, however, Mn-Mn hybridisation leads to a large reduction in this local moment. By contrast, Mn atoms at low concentration in Cu retain their large moment and form a spin glass [23]. Ultimately, this triggered an interest in the property of the surface alloy, which was predicted to also retain the large local moment, possibly with ferromagnetic or antiferromagnetic ordering [24]. In fact there is now ample evidence that the large local moment is retained in the surface phase, although there appears to be no magnetic ordering [25, 26, 27, 28]. On the other hand, a similar Ni(100)c(2x2)-Mn surface was found to show ferromagnetic ordering despite the bulk alloy in this case being antiferromagnetic [26, 27]. One further interesting prediction of this theoretical study is that it predicted that the Mn atoms at the surface would have anomalously large effective radii relative to their value in bulk metallic Mn as a consequence of this large local moment, leading to a rumpled surface with the Mn atoms outermost, and this was originally confirmed on Cu(100) by a quantitative LEED study [29], although the effect is actually evident in a much earlier study of the structure of the Pd(100)c(2x2)-Mn surface [30]. Table 3 summarises the key results of the various quantitative structural studies of these five different surface phases together with the related Cu(ll0)c(2x2)-Mn surface alloy, namely the amplitude of the surface alloy phase rumpling, defined as the layer spacing of the adsorbate atoms to the first complete elemental substrate layer below minus the same layer spacing for the substrate atoms in the surface alloy layer (fig. 7). In addition to the results obtained by quantitative LEED [24, 29, 30, 32, 33, 37] there are also values obtained from both a full [36] and simplified [34] application of the scannedenergy mode photoelectron diffraction (PhD) technique [31], and from a MEIS investigation [35]. Notice that in all cases this rumpling parameter is positive (the adsorbate atoms sit higher above the surface) although in some cases the investigations were not strictly able to determine the parity of this rumpling. In particular, the case of Mn on Cu is difficult because the two elements are close in the Periodic Table, thus having very similar electron scattering cross-sections

290 (in LEED) and quite similar mass numbers to determine the recoil energy in MEIS. In this case, however, the PhD method, which is element specific (studying photoelectron emission from the Mn atoms) provided clear confirmation of this parity. In addition to the experimentally-determined surface alloy rumpling amplitudes. Table 3 also shows the values of the metallic radii of the constituent elements, defined as half the nearest-neighbour distance in the elemental metallic (fcc) solid. A more detailed discussion of the significance of the exact values of these radii will be postponed until later in this chapter, and here we will concentrate only of their relative values and the associated rumpling amplitude. substrate

adsorbate

rumpling amplitude (]k)

Cu(100) Cu(lO0)

Pd Au

Cu(lO0)

Mn

Cu(110) Ni(100) Pd(100)

Mn Mn Mn

0.02(3) [32] 0.10 [33], 0.10134], 0.06(4) [35] 0.30(2) [24], 0.39(8) [36], 0.37(6) [35] 0.22(5) [37] 0.25(2) [29] 0.20(5) [30]

substrate metallic radius (~) 1.278 1.278

adsorbate metallic radius (,~) 1.376 1.442

1.278

1.2921

1.278 1.246 1.376

1.292 1.292 1.292

,

Table 3. Summary of the results of the surface layer rumpling found in the quantitative structure determinations of the c(2x2) surface alloy phases formed by Pd, Au and Mn on Cu(100) and by Mn on Ni(100) and Pd(100). Also included is the rumpling amplitude for the Cu(ll0)c(2x2)-Mn surface alloy. Values in parentheses show the estimated errors in hundredths of an/~ngstrom unit, and are omitted when no error estimates are given in the original papers. In a simple hard-sphere model, the lateral periodicity within the (pseudomorphic) surface alloy layer is constrained by the substrate atomic spacing, so if the adsorbate alloying species has a larger radius we expect this adsorbate to lie at a higher layer spacing than the surrounding substrate atoms, being unable to totally accommodate into the vacant site which it substitutes. The larger the adsorbate radius, the larger we expect this rumpling to be. In fact I The choice of the atomic radius for Mn is complicated due to the existence of several different structural forms of bulk Mn and is discussed further in section 3

291

we see that although Pd atoms are larger than Cu, they are almost coplanar in the PdCu surface alloy formed on Cu(100), although the slightly larger Au atoms do show a small but more significant rumpling of the AuCu alloy on this same surface. These two surface alloys thus show the anticipated qualitative trend, although the amplitude of the rumpling is smaller than would be predicted by the simple hard-sphere model. This quantitative effect, already highlighted for the Ni(lll)/Pb case in the previous sub-section, will be discussed further in the following section. Relative to the Pd and Au surface alloys on Cu(100), however, the behaviour of Mn on both Cu(100) and Ni(100) is clearly anomalous. The metallic radius of Mn is essentially identical to that of Pd and significantly smaller than that of Au, yet the surface alloy rumpling amplitude is much larger. The same effect is seen in the c(2x2)-Mn surface alloy formed on Cu(110). Moreover, while the atomic radius of Mn is actually smaller than that of Pd, on Pd(100) the Mn atoms in the c(2x2) surface alloy are rumpled outwards. This is precisely the effect predicted theoretically for these surface alloys; the Mn atoms in these phases have an effective radius significantly larger than in bulk Mn due to their high spin state in the surface alloys, and thus show significantly larger surface rumpling than other nonmagnetic atoms of similar atomic radii, defined relative to their bulk phases. We should briefly remark on data for some further surface phases which have not been included in Table 2. One potentially interesting such system concerns the surface phases formed by Mn deposition on Ag(100), a system potentially very similar to that of Cu(100)/Mn. In fact LEED patterns characteristic of a Ag(100)c(2x2)-Mn surface phase have been reported, but the only quantitative LEED analysis of this surface indicates that it comprises essentially a monolayer of Mn on the Ag(100) surface, with the c(2x2) LEED pattern arising from magnetic, rather than elemental alloy, ordering [38]. There do, however, appear to be other structures with a c(2x2) periodicity in this substrate/adsorbate combination at different coverage or annealing treatments, one of which is suggested to involve an ordered AgMn alloy in the second layer [39]. Furthermore, in the case of the Pd(100)/Mn combination, a second phase was reported after annealing in which the Mn atoms were buckled inwards in the outermost c(2x2) alloy layer, but in this case the authors found multi-layer alloying to form a bulk-like Pd3Mn near-surface phase.

2.4 Surface alloys formed by Sn on Cu, Ni, Pt and Rh surfaces: effect of changing substrate lattice parameter and surface orientation on rumpling amplitude One further group of surface alloy phases which have been studied by quantitative structural methods are those formed by Sn on all three low index faces of Ni as well as Cu(111), Pt(111), Pt(100) and Rh(111). In part these investigations were motivated by the fact that Sn addition to some transition

292

metals provides improved performance in heterogeneous catalysis [e.g. 40]. Most of these quantitative structural investigations were performed by low energy (500-1000 eV) Li § or Na § ion scattering, although the Ni(110)c(2x2)-Sn surface has been studied by a combination of quantitative LEED and MEIS, and additional LEED studies of Pt(111)/Sn have been performed. Table 4 shows the results of the sequence of measurements using low energy ion scattering for the surface alloys, either (2x2) 0.25 ML phase, (x/3xq3)R30 ~ 0.33 M1 phase, or both, on the (111) surfaces of Ni, Cu, Rh and Pt. In this case the common alloying adsorbate species is measured on closefaced fcc (111) surfaces of differing lattice parameters, and the trend in the rumpling amplitude is exactly that which one would expect for a simple hardsphere model; in all cases the Sn atomic radius (whether one takes the value for the tetragonal room temperature I]-phase or the low temperature diamondstructure ct-phase) is larger than the metallic substrate atom which the Sn atoms replace, and the rumpling amplitude is increased as the substrate atomic radius decreases. Here too the quantitative rumpling amplitude is smaller than predicted by simple hard-sphere models based on these radii. substrate

Ni(111) Cu(111) Rh(111) Pt(lll)

adsorbate

Sn Sn Sn Sn

rumpling amplitude (/~) 0.46(4) [41 ] 0.39 [42] 0.29(5) [43]

0.22(3) [44] 0.23(5) [45] 0.30(5) [45]

substrate metallic radius (/~) 1.246 1.278 1.345 1.387

adsorbate radius (]k)

a/f 1.405/1.511 1.405/1.511 1.405/1.511 1.405/1.511

Table 4. Measured rumpling amplitudes of Sn surface alloys on the (111) surfaces of Ni, Cu, Rh and

Pt in a similar format to table 2. Note that Sn atomic radii corresponding to half the Sn-Sn distances are given for both semiconducting diamond-structure tx-Sn and tetragonal [3-Sn which is the stable phase at room temperature. In the case of the Ni/Sn system, data also exists for all three low index faces, forming c(2x2) 0.5 ML surface alloy phases on both Ni(100) and Ni(110), while on Pt(100) there are also data for a c(2x2)-Sn surface alloy, so these systems allow a rather different comparison of rumpling amplitudes, shown in Table 5. Notice that on all these surfaces the Sn atoms must replace Ni or Pt atoms which, at least in one direction within the surface, are separated by their nearest-neighbour distance of twice their metallic radius. A simple

293

hard-sphere model would therefore predict exactly the same rumpling amplitude for all three faces. substrate

adsorbate

rumpling amplitude (/~)

Ni(lll) Ni(lO0) ,Ni(l I0) Pt(ll 1)

Sn Sn Sn Sn

Pt(100)

Sn

0.46(4) [41 ] 0.44(5) [46] 0.40(3) [47] 0.22(3) [48] 0.23(5) [49] 0.30(5) [45] 0.19 [50]

substrate metallic radius (/~) 1.246 1.246 1.246 1.387

adsorbate radius (/~)

1.387

1.405/1.511

1.405/1.511 1.405/1.511 1.405/1.511 1.405/1.511

Table 5. Measured rumpling amplitudes of Sn surface alloys on different orientation surfaces of Ni and Pt in a similar format to table 4.

Of course, the number of Ni or Pt nearest neighbours within the surface alloy layer for each Sn atom differs on the three faces, being 6 on (111), 4 on (100) and 2 on (110). The differences in the rumpling amplitudes shown by the data of Table 5 are actually marginally significant in view of the estimated errors, yet the actual values follow a clear trend to reduced rumpling amplitude for reduced alloy layer coordination. 3. I N T E R - A T O M I C DISTANCES IN SURFACE ALLOYS In the presentation of the structural studies in the previous section there has been frequent mention of the fact that the amplitude of the rumpling of the surface alloy phases is less than one would anticipate from a simple hard-sphere model. This assertion, of course, depends on a knowledge of the appropriate radii for the hard spheres which, in tum, depends on the nature of the anticipated bonding character. For the atoms which comprise metallic solids, and especially those having close-packed structures (fcc in all the relevant cases here for which each atoms has 12 nearest neighbours) the atomic radius value which seems most appropriate is simply half the nearest-neighbour distance in the appropriate elemental metal. Some of the alloying adsorbate species, however, do not form elemental solids of this type. As a starting point in our discussion, and the construction of Table 6, we follow a common procedure of taking the atomic radius from the value of half the interatomic spacing in the elemental solid, but 'corrected' to a coordination number of 12 [51, 52] and take values from a

294

standard text [53]. This leads to large radii: with respect to the radius for a coordination number of 12, the radii are 0.97, 0.96 and 0.88 for coordination numbers of 8, 6 and 4, respectively. Of course, the coordination at the surface is not 12 for any of these systems, but the role of the reduced surface coordination is potentially one of the issues of interest, so this standard normalisation provides a common starting point for our discussion. The radii used are listed in Table 5. Notice that the proper choice for Mn is complicated by the fact that while the stable form of Mn at room temperature is bcc (8-Mn) which is only 8fold coordinated, there are also 12-coordinated forms on Mn. Indeed, there is an fcc phase (y-Mn) with an effective Mn radius of 1.366/~ but this corresponds to a high temperature of 1100~ In I]-Mn, stable between 800~ and 1100~ Mn atoms have 12 near-neighbours with half-distances in the range 1.180-1.335 ,~; the value we have taken derives from the nearest-neighbour distance in the tetragonal pseudo-cell of the quenched y-Mn phase [54]. Of course, not all the elements which form the surface alloys discussed here are 'true metals', so the use of these atomic radii based on close-packed metallic structures may not be appropriate in all cases, and some degree of covalency may be involved. An alternative reference value is therefore the covalent radius. There are significant variations in the values of covalent radii for some elements in different collations [55, 56]. The values listed in Table 6 are 'single-bond covalent radii' as defined by Pauling [57]. element Ag Au Cu Mn Ni Pb Pd Pt Rh Sb Sn

atomic radius (~) 1.445 1.442 1.278 1.292 1.246 1.750 1.376 1.387 1.345 1.590 1.623

covalent radius (/~.) 1.34 1.34 1.17 1.18 1.15 1.54 1.28 1.30 1.25 1.41 1.40

Table 6 The atomic and covalent radii used to evaluate the bond lengths in Table 7.

Table 7 brings together the structural information from the surface investigations described above and compares the experimentally-determined interatomic distances with those predicted by the different atomic radii. For

295

each of the surface alloy phases discussed so far it is straightforward to calculate the nearest neighbour bondlength between the adsorbate and substrate species within the surface alloy from the known lateral periodicity of the underlying substrate and the measured rumpling amplitude. This can then be compared with a 'theoretical' value based on the sum of the atomic radii of the constituent atoms. Substrate

Adsorbate

dA-B (A)

dA-B (/~)

dA-a (A)

dA-B (~k)

('A')

('B')

surface alloy

bulk alloy

atomic radii

covalent radii

Ni(100)

Mn

2.50 [29]

2.54

2.33

Pd(100)

Mn

2.76 [30]

2.57 ~-MnNi 2.74 ~-MnPd

2.67

2.55

Cu(100) Cu(110)

Mn

2.57

2.35

Cu(100)

Pd

2.57-2.59 [24, 35, 36] 2.56 [37] 2.56 [32]

2.65

2.45

Cu(100)

Au

2.72

2.51

Cu(lll)

Sb

2.87

2.58

Ag(lll)

Sb

3.04

2.75

Ni(lll)

Fb

3.00

2.69

"Pt(111) Pt(lOO)

Sn

2.60 Cu3Pd 2.82 Cu3Au 2.62 Cu2Sb 2.94 ~Ag-Sb 2.73 NiPb film 2.83 PtaSn

3.01

2.70

Cu(lll)

Sn

2.90

2.57

Ni(lll), Ni(100), Ni(ll0) Rh(lll)

Sn

2.71-2.78 Cu3Sn 2.61-2.64 Ni3Sn

2.87

2.55

2.67 RhSn

2.67

2.65

Sn

2.56 [33,34,35] 2.60-2.63 [8,11,16] 2.89-2.90 [8,12,16,15] 2.57 [21] 2.59 [22] 3.78 [44] 2.78 [50] 2.59 [42] 2.53 2.53 2.52 2.71

[41] [46] [47] [43]

Table 6 Summary of the measured adsorbate-substrate species nearest neighbour bondlengths in surface alloy phases compared with some reference values of this parameter discussed in the text.

296 Now consider the bondlengths in Table 7. We have already highlighted the anomalous behaviour of Mn as a surface alloy ingredient in producing a corrugation amplitude significantly larger than that produced by other atoms of similar or smaller atomic size, an effect which is most pronounced for the Pd(100)c(2x2)-Mn case in which the Mn atoms lie further from the substrate than the surrounding Pd atoms of the surface alloy despite the smaller atomic radius of Mn. Notice that in the Cu/Mn and Ni/Mn systems the interatomic bondlength is actually quite similar to the sum of the atomic radii, but relative to all the other systems in which the bondlength is significantly shorter, this apparently 'normal' behaviour is actually anomalous. This anomalous behaviour has been attributed to the different spin state of the Mn atoms in these surface alloys and in bulk Mn. In all other systems listed in Table 7 it is clear that the observed surface interatomic bond lengths are systematically shorter than the sum of the atomic radii and are, in most cases, longer than the sum of the covalent radii. In determining which values for the radii might be most appropriate we can take some guidance from knowledge of bulk alloys. Standard texts on crystalline structures distinguish between alloys involving two true metals (which include all the noble and transition metals) and those involving a true metal and a socalled B sub-group element from further to the fight of the Periodic Table including elements in groups IVA and VA such as Sn, Pb and Sb [52]. For example, in the case of Cu-Sb this is reflected also in bulk intermetallic phases; the Cu-Sb bondlength in bulk Cu2Sb is 2.62/~ [54], very close to the actual value found in the surface phase and also to the sum of the covalent radii, and much less than the sum of the atomic radii. In the case of the Ag/Sb case we give the interatomic spacing for the hexagonal-close-packed ~Ag-Sb [17] which is especially relevant in view of the apparent tendency to form the hcplike stacking faults in this surface alloy. Clearly, using interatomic spacings of the actual elements involved from known bulk ordered alloy phases should remove the uncertainty of establishing the appropriate 'bulk' reference, and where these are available they have also been included in the relevant column of Table 7. This actually leads to some surprises. In particular, in the Cu-Au system, which is a case of an allo~, of two true metals, the Cu-Au bondlength in the ordered Cu3Au alloy is 2.82 A, which is actually larger than the sum of the atomic radii and thus substantially larger than in the case of the Cu(100)c(2x2)Au surface alloy phase. Indeed, this surface phase is notable in that there is almost no corrugation of the surface alloy layer despite the large difference in atomic radii (1.28 ,~, and 1.46/~) of the constituents. More generally, Table 6 shows systematically shorter surface interatomic distances than those in the relevant bulk alloy phase. Notice, also, the case of the Ni(111)/Pb system, which we identified earlier as displaying a particularly large difference from the sum of the (metallic) atomic radii. The equilibrium phase diagram of the Ni-Pb

297

system shows very limited (<3%) solid solubility of Pb in Ni [58], but a NiPb phase, which is presumably metastable, was produced by vacuum evaporation, and was found to have the NiAs structure with a=4.15 /~, c=5.28 A [59], leading to an implied Ni-Pb nearest-neighbour distance of 2.73 /~. This is significantly closer to the sum of the covalent radii than the sum of the atomic radii but also much larger than the value measured in the surface phase. Clearly, therefore, there is some additional factor at the surface which tends to lower the corrugation and the associated interatomic bondlength relative to the bulk alloy phase (in cases where such an alloy phase exists), and relative to reasonable predictions based on atomic and covalent radii. A key difference in the surface case, of course, is the reduced coordination of the atoms due to the termination of the solid. According to the same rules used to generate the 12fold coordinated atomic radii from solids in which the coordination is lower, we know that reduced coordination generally shortens the bondlengths. Notice, however, that there is only an extremely small change in bondlength in the Ni/Sn and Pt/Sn systems between the different surface orientations in which there are large changes in coordination (from 6 in a corrugated (111) surface alloy to 2 in a corrugated (110) surface alloy). The bondlength scaling with coordination number used to determine atomic radii in bulk compounds, mentioned earlier in this section, is therefore clearly not appropriate in this case. In the case of a metal surface, the reduced coordination is known to have two effects, both reflecting the depletion of valence charge density around the surface atoms due to the spill-over into the vacuum and the smoothing of the charge fluctuations parallel to the surface. One effect is that the outermost layer of atoms is generally pulled into the solid, relative to the underlying bulk layer spacing. This allows the surface atoms to slightly increase the surrounding valence charge density and approach more closely the optimum value found in the bulk. The contraction is largest (up to 10% or more) for open-packed surfaces such as fcc(110) on which the smoothing parallel to the surface leads to more severe depletion at the surface atoms. On close-packed surfaces, such as fcc(111), the effect is marginal (and indeed is so marginal that it is not even clear what is the sign of the surface layer spacing change). A second consequence of the surface valence charge depletion relates to surface stress. It seems to now be rather well-established that clean unreconstructed elemental metal surfaces are in a state of tensile stress [60]. This means that the surface atoms would prefer to have a shorter interatomic spacing parallel to the surface. In some cases (such as Au(111) and Au(100) surfaces) this effect can lead to a reconstruction of the surface layer to a (more) close-packed overlayer (e.g. [61 ]). However, in most metals the surface atoms that are under tensile stress are locked in the periodic potential of the underlying bulk. Substituting some fraction of the atoms in such a surface by

298

those of another element with a larger radius would reduce the tensile stress. These atoms are therefore drawn more into the surface layer than expected from the bulk radii. For adsorbates with very large radii, the surface stress could reverse and become compressive. In those cases rumpling is expected, but the amplitude of the rumpling should be much less than calculated on the basis of the bulk atomic radii. These arguments might also be extended to suggest that alloying involving smaller adsorbates would only enhance the tensile surface stress and thus not be favourable. Of course, this argument is probably oversimplistic, because the valency of the adsorbate may also be relevant in determining the interatomic separation which provides the optimum charge density in which the surface atoms are immersed. In fact Tersoff has argued previously [62] that surface alloying may be characteristic of a significant mismatch in the size of the alloying elements and not a function of which atom is larger. Our tentative suggestion that large adsorbates will tend to form a substitutional alloy, is actually supported by the (admittedly far from complete) list of substitutional alloys in Table 7: in all these cases (apart from the special case of Mn on Pd(100) discussed above) the adsorbate atomic radius is larger than that of the substrate. We thus expect in general that the effective radius of a substitutional atom is smaller than expected on the basis of bulk properties. It is important to realise that a small reduction in radius can have a large effect on the predicted rumpling. For example, in the Cu(100)c(2x2)-Au phase the corrugation amplitude is less than 0.1 A whereas in Cu(100)c(2x2)-Mn this corrugation is 0.3 A or more, yet the interatomic bondlength difference calculated for these two cases is only 0.03 ,~. The same effect is notable in the cases of Sn on different faces of Ni and Pt; the reduction in corrugation amplitude with reduced surface coordination is quite conspicuous in the data of Table 4, yet the change in bondlength associated with this seen in Table 6 is negligible. Bearing in mind that the driving force for surface structure modification is increase of the valence charge density around the substitutional atom, the energy gain achieved by suppressing the corrugation amplitude is probably significantly more important than the energy cost of very slightly shortening the adsorbatesubstrate species interatomic spacing within the surface alloy layer. 4. M O R E C O M P L E X SYSTEMS In this chapter we have concentrated on surface alloy systems for which there exists rather complete and quantitative structural information, but have also concentrated on relatively simple structures for which a fairly consistent picture with respect to the surface rumpling has been found. One group of such surface phases which we have not discussed are those formed by alkali metal adsorption on aluminium; because there is such a huge difference

299

(approximately 1 ,~) between the metallic and ionic radii of the highly electropositive alkali metals it is difficult to draw any conclusions about surface effects in rumpling amplitudes in these systems, but these systems are described in a separate chapter by Adams and Andersen. Similarly, Pb does form surface alloys on Cu(100) and so might have proved an interesting complement to the Ni(111)/Pb system discussed here, but the Cu(100)c(2x2)Pb surface phase appears to be a simple oveflayer rather than an alloy [63], and while a c(4x4)-Pb phase does exist which is a surface alloy [64], it is not one involving a simple substitution of some of the surface Cu atoms as in the other cases described here. More detailed studies of this system are reported here in the chapter by Kellogg. A few other surface alloys which have been documented (typically with less structural details) are nevertheless of interest in the context of the present chapter. Of particular interest in view of our previous discussion is W(100)/Cu, because in this case we have a substrate comprised of atoms with a smaller atomic radius (1.37 A) than that of the adsorbate (1.278 A). On the basis of the previous arguments concerning the role of atomic size and surface stress in determining surface alloy formation one might expect that this system would not form a surface alloy. Moreover, if a surface alloy does form, we would probably anticipate that there would be no detectable rumpling. Of course, unlike fcc (100), bcc (100) surfaces do not involve touching hard-sphere atoms in the surface plane, so the arguments we have used concerning rumpling amplitudes and effective radii do not apply to this surface. In fact a c(2x2)-Cu surface alloy is found to form on W(100), in which the Cu atoms are located 0.14 ~ above the W layer (i.e. the Cu atoms behave as though their radius is significantly larger than their size in bulk Cu). Another important difference in the behaviour of W(100) relative to the fcc (100) and (111) surfaces we have discussed so far is that at room temperature the clean W(100) surface is actually only marginally stable to a c(2x2) reconstruction which occurs below a phase transition temperature of 250 K. This clean surface reconstruction involves the formation of more closely-spaced zigzag chains of W atoms but lowers the surface symmetry. A similar, but more complex instability is found on Mo(100) on which a c(2x2)-Pd surface alloy phase is formed, also involving slightly smaller (1.376 A) adsorbate atoms than those of the substrate (1.400/~) [65]. The case of AI(100)/Yb, with a very large difference between the atomic radius of the substrate (1.432 A) and adsorbate (1.776/~), and having a stable fcc (100) substrate is also of interest. Based on the atomic radii, a simple pseudomorphic substitutional surface alloy would be expected to show a very large buckling of 1.45 A, whereas a much reduced value of 0.37 A is found experimentally [66]. However, while this phase does, in terms of simple stoichiometry, involve only substitution of every fifth surface A1 atom, the (~/5x~/5)R27~ surface phase formed involves substantial lateral displacements of

300

the A1 atoms in the surface layer which actually enlarge the 'vacancies' which the Yb atoms then occupy (see fig. 8). In fact the A1-Yb nearest neighbour distance is found to be 3.39/~ which is actually slightly larger than the sum of the atomic radii (3.21 /~) but the A1-A1 nearest-neighbour distances in this surface phase (2.658/~) are less than in the clean surface (2.864/~), consistent with the idea that the surface alloy leads to a reduction in the tensile surface stress expected at the clean surface.

Fig. 8. Plan view of the AI(100)(~/5x~/5)R27~ structure showing the Yb atoms (large dark-shaded spheres) surrounded by 8 AI atoms which appear in groups of 4 between the Yb atoms. Note that the AI-AI distance in the surface alloy layer is reduced relative to that of the bulk, and in the diagram the radius of the AI atoms has been reduced from the bulk value to avoid overlap. As a further example of more complex behaviour, consider the case of Bi on Cu(100) [67]. In this system there is a large difference in the atomic radii: 1.70 A for Bi and 1.278 /~ for Cu. Deposition of Bi on Cu(100) does lead to an ordered c(2x2) 0.5 ML surface phase, but this is a simple overlayer and not a surface alloy. However, using SXRD it was found that for Bi coverages below 0.35 ML, a disordered surface alloy is formed in which Bi atoms substitute Cu surface atoms. At higher coverages dealloying occurs and the ordered c(2x2) overlayer phase is formed. In the surface alloy formed at low coverage the amplitude of the layer rumpling was found to be 0.61+0.10/~, much smaller than the value of 1.53/~ which would be predicted on the basis of the atomic radii. This example thus again confirms the general trend of a reduced rumpling

301

amplitude. Following through on our earlier arguments we may infer that the reduction in tensile stress allows the surface to accommodate approximately 0.3 ML of the much larger Bi atoms, but at higher coverages this surface alloy is no longer the energetically most favourable situation. It is interesting to note that the Ni(111)/Pb case discussed earlier involves a similarly large (indeed slightly larger) difference in atomic radii, and this system does form an ordered alloy surface phase; however, on the (111) surface the coverage of the (x/3x~/3)R30 ~ is only 0.33 ML, whereas the c(2x2) phase on the fcc (100) surface requires a higher coverage of 0.5 ML. Of course, the classic example of the role of the tensile surface stress of a clean surface acting as the driving force for reconstruction to a surface of higher atomic density is the so-called 'herringbone' reconstruction of Au(111) mentioned earlier, in which extra rows of Au are added in the surface layer [61]. The addition of A1 to this surface leads to a different reconstruction ('distorted hexagonal') with an density that is enhanced by 10% with respect to a bulk (111) layer [68]. The reconstruction is similar to the one found on A u ( l l l ) at elevated temperatures, in which case the density enhancement amounts to 7%. Bearing in mind that the atomic radius of A1 is almost identical to that of Au (less than 1% smaller), the A1 atoms evidently display a reduced effective radius in this surface alloy, as manifest in reduced rumpling amplitudes in the many systems we have discussed in this chapter. These more complex examples thus provide significant support for the general arguments proposed in this chapter but nevertheless highlight the fact that the problem is potentially quite subtle and that atomic size, while apparently a major factor, is clearly not the only property of relevance in understanding surface alloy formation in general. 5. CONCLUSIONS Recent structural studies of a number of surface alloy phases have revealed some surprising results. Here we have concentrated on just two issues. One is the formation of a stacking fault at the interface of the surface alloy layer and the underlying substrate which has been found for the (~/3xx/3)R30~ phases formed on C u ( l l l ) and Ag(111). A number of other fcc(111) (~/3x~/3)R30 ~ surface alloy phases, such as that formed by Pb on Ni(111), do not show this effect, and it may be rather specific to the role of Sb on these two substrates; for both Cu and Ag there are bulk hcp phases of solid solutions containing a significant concentration of Sb, and this transformation may be effected at much lower concentrations near a surface. A more general issue we have addressed is the magnitude of the rumpling corrugation in a range of substitutional surface alloy phases, which can be expressed in terms of the effective radii of the adsorbate species when

302 accommodated into the surface alloy phase. The case of Mn on Cu(100), Pd(100) and Ni(100) is found to show anomalously large surface rumpling when compared to other alloying atoms of similar or larger atomic radius, and this can be attributed to the high spin state of the Mn atoms in these surface alloys relative to the value in bulk metallic Mn. In all other cases, at least on metal surfaces which are close-packed in at least one direction within the surface, the rumpling amplitudes are smaller (in some cases very significantly smaller) than would be expected on the basis of estimates of the expected adsorbate-substrate species interatomic distances or values found for the same species in bulk alloy phases. This has been attributed to a real surface effect, associated with the valence charge depletion at metallic surfaces which is wellknown to lead, for elemental metals, to outer layer spacing contraction and surface tensile stress. An extension of this general argument suggests that surface alloy formation may be favoured when the alloying 'adsorbate' species has an atomic radius which is larger than that of the substrate atoms, and the limited number of cases studied do seem to support this suggestion, at least in cases of simple substitutional surface alloys.

REFERENCES 1 U.Bardi, Rep.Prog.Phys. 57 (1994) 939 2 H.A.van der Vegt, H.M.van Pinxteren, M.Lohmeier, E.Vlieg and J.M.C.Thornton, Phys.Rev.Lett. 68 (1992) 3335 3 J.Vrijmoeth, H.A.van der Vegt, J.A.Meyer, E.Vlieg and R.J.Behm, Phys.Rev.Lett. 72 (1994) 3843 4 S.Oppo, V.Fiorentini and M.Scheffler, Phys.Rev.Lett. 71 (1993) 2437 5 Fiorentini V, Oppo S and Scheffler M 1995 Appl.Phys.A 60 399 6 T.C.Q.Noakes, D.A.Hutt, C.F.McConville and D.P.Woodruff, Surf.Sci. 372 (1997) 117 7 H.Giordano and B.Aufrey, Surf.Sci. 307-309 (1994) 816 8 S.A.de Vries, W.J.Huisman, P.Goedtkindt, M.J.Zwanenburg, S.L.Bennett, I.K.Robinson and E.Vlieg, Surf.Sci. 414 (1998) 159 9 S.A.de Vries, W.J.Huisman, P.Goedtkindt, M.J.Zwanenburg, S.L.Bennett, and E.Vlieg, Phys.Rev.Lett. 81 (1998) 381 10 H.A. van der Vegt, J. Vrijmoeth, R.J. Behm and E. Vlieg, Phys.Rev. B 57 (1998) 4127 11 P.Bailey, T.C.Q.Noakes and D.P.Woodruff Surf.Sci. 426 (1999) 358 12 E.A.Soares, C.Bitencourt, V.B.Nascimento,V.E. de Carvalho, C.M.C. de Castilho, C.F.McConville, A.V. de Carvalho and D.P. Woodruff, Phys.Rev.B 61 (2000) 13983 13 I.Meunier, J.-M.Gay, L.Lapena, B.Aufray, H.Oughaddou, E.Landemark, G.Falkenberg, L.Lottermoser and R.L.Johnson, Surf.Sci. 422(1999)42 14 E.Vlieg, I.K.Robinson, K.Kem, Surf. Sci. 233 (1990) 248 15 P.D.Quinn, D.Brown, D.P.Woodruff, P.Bailey and T.C.Q.Noakes, to be published 16 D.P.Woodruff and J.Robinson J.Phys.:Condens.Matter 12 (2000) 7699 17 W.B.Pearson A Handbook of Lattice Spacings and Structures of Metals and Alloys

303 (Pergamon, London, 1958); Vol. 2 (Pergamon, Oxford, 1967) 18 H.Cruguel, B.L6pine, S.Ababou, F.Solal, G.J6zEquel, C.R.Natoli and R.Belkhou, Phys.Rev.B 55 (1997) R16061 19 H.H.Farrell, M.M.Traum, N.V.Smith, W.A.Royer, D.P.Woodruff and P.D.Johnson, Surf.Sci. 102 (1981) 527 20 K.Umezawa, A.Nakanishi, T.Yumura, W.M.Gibson, M.Watanabe, Y.Kido, S.Yamamoto, Y.Aoki and H.Naramoto, Phys.Rev.B 56 (1997) 10585 21 D.Brown, P.D.Quinn, D.P.Woodruff, P.Bailey and T.C.Q.Noakes, Phys.Rev.B. 61 (2000) 7706 22 P.D.Quinn, C.Bittencourt and D.P.Woodruff, to be published. 23 see, e.g.N.van der Post, J.A.Mydosh, J.M.van Ruitenbeek and I.K.Yanson, Phys.Rev.B 53 (1996) 15106 24 M.Wuttig, Y.Gauthier and S.Bltigel, Phys.Rev.Lett.70 (1993) 3619 25 W.L.O'Brien, J.Zhang and B.P.Tonner, J.Phys.:Condens.Matter 5 (1993) L515 26 W.L.O'Brien and B.P.Tonner, J.Appl.Phys. 76 (1994) 6468 27 W.L.O'Brien and B.P.Tonner, Phys.Rev.B 51 (1995) 617 28 A.B.Hayden, P.Pervan and D.P.Woodruff, J.Phys.:Condens.Matter7 (1995) 1139 29 M.Wuttig, C.C.Knight, T.Flores and Y.Gauthier, Surf.Sci. 292 (1993) 189 30 D.Tian, R.F.Lin, F.Jona and P.M.Marcus, Solid State Commun. 74 (1990) 1017 31 D.P.Woodruff and A.M.Bradshaw Rep.Prog.Phys. 57 (1994) 1029 32 S.C.Wu, S.H.Lu, Z.Q.Wang, C.K.C.Lok, J.Quinn, Y.S.Li, D.Tian, F.Jona and P.M.Marcus Phys. Rev. B 38 (1988) 5363 33 Z.Q.Wang, Y.S.Li, C.K.C.Lok, J.Quinn and F.Jona Solid State Commun. 62 (1987)181 34 J.G.Tobin, J.C.Hansen and M.K.Wagner J. Vac. Sci. Technol. A 8 (1990) 2494 35 D.Brown, T.C.Q.Noakes, D.P.Woodruff, P.Bailey and Y.Le Goaziou, J.Phys.:Condens.Matter 11 (1999) 1889 36 R.Toomes, A.Theobald, R.Lindsay, T.Geissel, O.Schaff, R.Didszhun, D.P.Woodruff, A.M.Bradshaw and V.Fritzsche J. Phys. : Condens. Matter 8 (1996) 10231 37 C.Ross, B.Schirmer, M.Wuttig, Y.Gauthier, G.Bihlmayer and S.BlUgel, Phys.Rev.B 57 (1998) 2607 38 P.Schrieffer, C.Krembel, M-C.Hanf, G.Gewinner and Y.Gauthier Phys.Rev. B 82 (2000) 2944 39 P.Schieffer, M.C.Hanf, C.Krembel and G.Gewinner, Surf.Sci. 446 (2000) 175 40 A.F.Lee, C.J.Baddeley, C.Hardacre, G.D.Morridge, R.M.Ormerod and R.M.Lambert J. Phys .Chem. B 101 (1998) 2797 41 Y.Ku and S.H.Overbury Surf. Sci. 273 (1992) 341 42 S.H.Overbury and Y.Ku Phys. Rev. B 46 (1992) 7868 43 Y.Li, M.R.Voss, N.Swami, Y.-L.Tsai and B.E.Koel, Phys.Rev.B 56 (1997) 15982 44 S.H.Overbury, D.R.Mullins, M.T.Paffett and B.E.Koel Surf. Sci. 254 (1991) 45 45 A Atrei, U.Bardi, J.X.Wu, E.Zanazzi and R.Rovida Surf.Sci. 290 (1993) 286 46 Y.D.Li, L.Q.Jiang and B.E.Koel Phys. Rev. B 49 (1994) 2813 47 P.D.Quinn, C.Bittencourt, D.Brown, D.P.Woodruff, T.C.Q Noakes and P.Bailey, J.Phys.:Condens.Matter to be published 48 S.H.Overbury, D.R.Mullins, M.T.Paffett and B.E.Koel Surf. Sci. 254 (1991)45 49 A Atrei, U.Bardi, J.X.Wu, E.Zanazzi and R.Rovida Surf.Sci. 290 (1993) 286 50 Y.D.Li and B.E.Koel, Surf.Sci. 330 (1995) 193 51 V.M.Goldschmidt, Z.phys.Chem. 133 (1928) 397

304 52 R.C.Evans, An Introduction to Crystal Chemistry (Second Edition, Cambridge University Press, 1966) 53 N.W.Alcock, Bonding and Structure (Ellis Horwood, 1990): see also http://www.iumsc.indiana.edu/radii.htm 54 R.W.G.Wyckoff, Crystal Structures, Vol. 1 (Second Edition, Wiley Interscience, New York, 1965) 55 R.S.Berry, S.A.Rice and J.Ross Physical Chemistry (Wiley, New York, 1980) 56 http://www.webelements.com 57 L.Pauling The Nature of the Chemical Bond (Cornell Univ.Press, Ithaca, 1939) 58 T.B.Masralski, Ed., Binary Alloy Phase Diagrams (American Society for Metals, 1986) 59 R.R.Bitti, J.Dixmier and A.Guiner, C.R.Acad.Sci.Paris B. 266, 565 (1986) 60 H.Ibach, Surf.Sci.Reports, 29, 193 (1997) 61 A.R.Sandy, S.G.J.Mochrie, D.M.Zehner, K.G.Huang and D.Gibbs, Phys. Rev. B 3 (1991) 4667 62 J.Tersoff, Phys.Rev.Lett. 74 (1995)434 63 W.Hoesler, W.Moritz, E.Tamura and R.Feder, Surf.Sci. 171 (1986) 55 64 Y.Gauthier, W.Moritz and W.Hoesler, SS 345 (1996) 53 65 D.Wu, W.K.Lau, Z.Q.He, Y.J.Feng, M.S.Altman and C.T.Chan, Phys.Rev.B 62 (2000) 8366 66 R.Fasel, M.Gierer, H.Bludau, P.Aebi, J.Osterwalder and L.Schlapbach, Surf.Sci. 374 (1997) 104 67 H.L. Meyerheim, H. Zajonz, W. Moritz and I.K. Robinson, Surf. Sci. 381 (1997) L551 68 B.Fischer, J.V.Barth, A.Fricke, L.Nedelmann and K.Kern, Surf. Sci. 389 (1997) 366

9 2002 Elsevier Science B.V. All rights reserved. Surface Alloys, and Alloy Surfaces D.P. Woodruff, (Editor)

305

Chapter 9

Surface Alloy Formation on Cu{lO0} C.J. Barnes School of Chemical Sciences, Dublin City University, Dublin 9, Republic of Ireland.

1. INTRODUCTION The phenomena of surface alloy formation in which a metallic adsorbate distributes itself throughout the outermost or outermost few surface layers, intimately mixed with the substrate has only recently begun to be addressed in detail. Early work on metal-metal interfaces focussed largely on metallic films in which intermixing was thermodynamically forbidden such as coinage metals on close packed refractory surfaces or simply was not considered as a possibility, with interpretation preferring to assume abrupt interface formation [ 1,2]. In the last decade, it has become increasingly appreciated that intermixing is a widespread phenomena in the early stages of metal-metal interface formation and is not limited to binary systems that are bulk miscible or those forming ordered compounds [3,4]. This review attempts to bring together the work performed to date on a single substrate: Cu{ 100}. The reasons for choice of Cu{ 100} stems first and foremost from the large body of work performed allowing a coherent review of surface alloy formation on a single well defined template. Copper has proved to be a favoured substrate for studies of ultra-thin metallic film growth for many reasons, including the relative ease of cleaning and maintaining surface cleanliness, the high level of crystalline quality and the advantages of a full dband electronic configuration allowing high resolution studies of the surface and bulk electronic structure [5]. Copper, both pure or as a component in a bimetallic combination is widely used industrially within the field of heterogeneous catalysis for reactions as varied as methanol synthesis and in CO oxidation catalysts for pollution control [6]. The Cu{ 100} surface itself is known to be stable towards surface reconstruction, the clean surface undergoing a small damped oscillatory

306

relaxation with the first interlayer spacing uniformly contracted (Adz~E/dbun,= -2.0 _+0.8%) and the second layer uniformly expanded (AdzE3/dbulk = + 0 . 4 % ) with the third and deeper interlayer spacings close to the value in bulk Cu (1.807/~) [7]. In common with other metal surfaces, Cu { 100 } is in a state of tensile stress as has been shown by band mapping of Rayleigh phonons [8]. Effective medium theory (EMT) calculations of Stoltze have provided estimates of the energetics for removal of atoms from a Cu{ 100} surface [9]. The energy required to form an adatom/vacancy pair is calculated within EMT to be 0.98 eV. The high thermal energy of arriving "hot" metal adatoms from Knudsen evaporation sources combined with the liberated chemisorption energy during metal deposition is thus usually sufficient to allow dispacement of top layer Cu atoms for films grown at room temperature allowing surface alloys to be formed. Once copper atoms are free from their terrace binding sites, considerable mobility is possible at 300 K. The activation energy for surface diffusion of Cu adatoms on Cu{100} terraces has been experimentally determined by Breeman and Boerma to be 0.39 eV [10,11]. This activation energy is low enough to allow Cu adatoms formed by dispacement upon surface alloy formation to be highly mobile at room temperature and diffuse away to defect sites such as step edges. In order to provide a framework for comparison of the behaviour of the wide range of metal adsorbates known to form surface alloys on Cu{ 100}, table 1 summarises the surface energies and 12-fold co-ordinate metallic radii of adsorbates dealt with within this review. The surface energies have been taken from the tabulation of Vitos et al. [12] calculated using the full charge density linear muffin-tin orbital (LMTO) method. In cases where elements do not adopt a face-centred-cubic (f.c.c) structure, surface energies are given for similar symmetry surfaces: body-centred-tetragonal { 100} for In and Sn, simple cubic { 100} for Bi and body-centred-cubic { 100} (b.c.c) for Fe and Li. In the cases of metals adopting hexagonal-close-packed (h.c.p) stacking the surface energies correspond to the most stable {0001 } basal plane and as such are underestimates of the true surface energy per adsorbate atom when adopting a local geometry such as that imposed by Cu{ 100}. It should also be remembered that these energies may only be used as guidelines when considering the role played by surface energy in alloy formation as they are specifically calculated for elements in their most stable bulk crystallographic phase and with lattice constants close to experimental equilibrium values.

307 Table 1 Surface energies (,/) and 12-fold co-ordinate metallic radii for selected elements. The surface energy ratio (relative to copper) and the percentage mismatch in metallic radii are also given. Average values for experimentally determined surface energies are given in brackets in J.m-2. The value in brackets relating to the surface energy ratio is calculated based on experimental surface energies in J.m2. Element Ag Au Bi Co Cu Fe In Ir Li Mg Mn Ni Pb Pd Pt Rh Sn

Surface energy eV J.m"z 0.65 0.89 0.36 0.96 0.91 1.27 0.34 1.77 0.38 0.44 1.04 0.97 0.37 1.15 1.38 1.31 0.39

1.20 (1.25) 1.63 (1.50) 0.54 (0.49) 2.77 (2.54) 2.17 (1.81) 2.22 (2.41 ) 0.49 (0.69) 3.72 (3.02) 0.52 (0.52) 0.79 (0.77) 3.10 (1.60) 2.43 (2.41) 0.38 (0.60) 2.33 (2.03) 2.73 (2.48) 2.80 (2.68) 0.61 (0.69)

Tx/TCu

Metallic radius A

rx/rcu x 100%

0.55 (0.69) 0.75 (0.83) 0.25 (0.27) 1.28 (1.40) 1 1.02 (1.33) 0.23 (0,38) 1.71 (1.67) 0.24 (0.29) 0.36 (0.43) 1.43 (0.88) 1.20(1.33) 0.18 (0.33) 1.07 (1.12) 1.26 (1.37) 1.29 (1.48) 0.28 (0.38)

1.445 1.442 1.700 1.252 1.278 1.274 1.663 1.357 1.562 1.602 1.264 1.246 1.750 1.376 1.387 1.345 1.623

+13.1% (+0.17 A) + 12.8% (+0.16/~) +33.0% (+0.42/~)

-2.0% (-0.03A) -0.3% (+0.00A) +30.1% (+0.39 A) +6.2% (+0.08 A) +22.2% (+0.28 A) +25.4% (+0.32 A) -1.1% (-0.01 A) -2.5% (-0.03 A) +36.9% (+0.47 A) +7.7% (+0.10 A) +8.5% (+0.11 A) +5.2% (+0.07 A) +27.0% (+0.35 A)

In particular, they do not account for the strain due to occupation of differing crystal structure and/or lattice constant imposed by the underlying Cu{ 100} substrate. The table also includes 12-fold co-ordinate metallic radii and the ratio of 12-fold co-ordinate radii relative to copper in order to gauge the importance of strain effects [ 13]. This review deals primarily with structural aspects of surface alloy formation on Cu { 100},with emphasis given to systems which form ordered surface alloys in which the structure and layerwise composition is considered to be reasonably well defined. All surface coverages will be quoted with respect to the atomic density of Cu { 100 }-( 1x 1) phase of 1.5 3x 1015atoms cm 2.

308 2. Cu{100}-c(2x2)-X (X=Au,Pd,Mn) SURFACE ALLOYS

2.1. Geometric and Electronic Structure A Cu { 100}-c(2x2)-Au superstructure was first reported by Palmberg and Rhodin upon deposition of 0.5 ML of Au on Cu{100} [14]. Palmberg and Rhodin argued that the c(2x2) periodicity originated from an ordered twodimensional CuAu alloy due to the tendancy for transition metals to form close packed monolayers when adsorbed as overlayers to enhance co-ordination when surface alloying is absent [ 1,2]. Graham compared low energy ion scattering spectroscopy (LEISS) and electron spectroscopic data from the Cu{100}-c(2x2)-Au and that from the surface of a bulk Cu3Au{100} alloy possesing a mixed c(2x2) CuAu layer outermost, adding further evidence that mixing occurs to form a twodimensional alloy layer [15]. Low energy electron diffraction (LEED) observations indicated sharp c(2x2) reflexes at a Au coverage of 0.5+0.1 ML. Streaks in the [011] and [011] directions, were also noted thought to be indicative of strain within the adlayer [15]. Figure 1 illustrates a top and side view of a Cu { 100}-c(2x2) surface alloy.

[011 ]

[001 ]

Key: = adsorbate C ) = top layer Cu ~,.....~= second layer Cu

[OLO]

= third layer Cu [Ol 1]

A1 dz12 dz23

Figure 1. Top (upper) and side (lower) view along the [011] azimuth of a Cu{100}-c(2x2) alloy defining azimuthal directions and important geometric parameters. Within the side view the spacing between adjacent layers has been artificially increased for clarity.

309 Table 2 illustrates the results o f quantitative structural studies on the Cu { 100}c(2x2)-Au phase. Table 2 Summary of experimentally determined geometries of Cu{100}-c(2x2) surface alloys. Positive values for buckling in the outermost mixed layer (A~) indicate adsorbate atoms relaxed towards the vacuum interface while positive values for buckling in the third copper layer (A3) indicate that Cu atoms directly below substitutional adsorbate atoms in layer 1 are buckled outwards towards the surface. The values in brackets below the interlayer spacings indicate the percentage contraction (negative) and expansion (positive) relative to the bulk Cu value of 1.807 A.

Adsorbate

Au

Technique

AI

dZl2

dz23

A3

A

A

A

A

LEED

+0.1

1.88 (+4.0%)

-

-

[ 16]

PhD

+0.10

1.78 (-1.5%)

1.71 (-5.4%)

-

[17]

-

[ 18]

Bulk

-

[26,27]

MEISS

+0.06+0.04

1.88+0.03

(+4.0% a:1.7%)

Pd

LEED

+0.02+0.03

1.807+0.03

1.86+0.02

Ref.

(+2.9+1.1%)

(0.0%+ 1.7%)

Mn

Mg

LEED

+0.30-a:0.02

1.79+0.02 (-0.5%+1.1%)

1.80-a:0.03 (-0.4+1.7%)

+0.02 4-0.03

[37,38]

PhD

+0.39+0.08

1.63+0.08 (-9.8%+4.4%)

1.83+0.08 (+1.3+4.4%)

-

[39]

MEISS

+0.37+0.06

1.72+0.04 1.83+0.02 (-4.8% _-k2.2%) (+1.3+1.1%)

-

[ 18]

TLEED

+0.55+0.10

1.72+0.02 (-4.8~ 1.1%)

0.00 +0.02

[101]

1.81+0.02 (+0.2+1.1%)

310

Three independent studies of this structure by low energy electron diffraction (LEED) I(V) analysis, energy-scanned photo-electron diffraction (PhD) and medium energy ion scattering spectroscopy (MEISS) have been performed [16,17,18]. While the LEED and MEISS studies utilised large data bases, the PhD work relied on a single energy scan. The early LEED analysis of Wang et al [16] clearly ruled out a c(2x2) Au overlayer and confirmed the suggestion of Palmberg and Rhodin that the Cu{100}-c(2x2)-Au structure corresponds to a two-dimensional surface alloy in which Au atoms substitute in an ordered array in the outermost Cu monolayer. All studies are in good quantitative agreement that the outermost layer is buckled by 0.08 A (average of the LEED and MEISS analyses), with the Au tippled outwards as expected on the basis of its 0.16 A larger metallic radius. Substitution of the larger Au atom within the outermost layer leads to a large expansion of the first interlayer Cu spacing of +0.07 A.The early LEED study did not consider the possibility of relaxation of the second and deeper interlayer spacings. The MEISS study yields a second Cu interlayer expansion of +0.05+0.02 A, in contrast to the unusually large contraction o f - 0 . 1 0 A obtained via PhD. While not a full structural analysis, a LEISS study modelling measured scattered azimuthal ion distributions, allowed a tippling in the CuAu mixed layer of between 0.12+0.06 and 0.15+0.06 A depending of the chosen scattering potential utilised [19]. Furthermore, the form of the LEISS scattering pattern from Au indicated imperfections in the form of sub-surface Au, with approximately 15% of Au atoms located in second and/or third layer sites suggesting that the Cu{ 100}c(2x2)-Au alloy is not truly two-dimensional. Embedded atom calculations performed by Foiles confirm the stability of Au atoms in top layer substitutional sites relative to overlayer adsorption. The surface alloy model is lower in energy by 0.14 eV per Au atom [20]. The energy of Au substituted in the surface layer is 0.40 eV lower than that for a bulk substitutional site, indicating a surface localised alloy to be the lowest energy state, although experimentally high temperature annealing of the Cu{100}c(2x2)-Au surface alloy eventually leads to Au dissolution in-to bulk. Monte Carlo simulations for a Cu {100} sample with dilute Au content (0.1at%) leads to strong Au segregation forming a stongly buckled c(2x2) top layer alloy with Au atoms situated 0.18 A above copper [20]. The Cu{100}-c(2x2)-Au surface alloy has been investigated using scanning tunneling microscopy (STM) by Chambliss and Chiang [21,22]. Atomic resolution images verified a c(2x2) periodicity consistent with a twodimensional CuAu alloy. Two inequivalent sites are imaged within the c(2x2) unit cell, with an apparent height difference of 0.2+0.1 A, due to the differing STM response to Au and Cu due to a combination of geometric and chemical contrast effects. A longer range superstructure was also imaged with an average periodicity of 21+2 A in the [011 ] and [0i 1] directions, indicative of a periodic

311

array of defects. The apparent depth of the periodic troughs was about 0.5A, although the imaged depth was found to be tip dependent. Figure 2 illustrates an STM image, showing this longer range periodicity. Chambliss and Chiang suggested that the faults are due to regions in which single rows of Au atoms are laterally displaced yielding a local structure which is hexagonal rather than square. This structure has a number of Au-Au nearest neighbour sites. Tersoff, using a simple pairwise interaction model has illustrated that the occupation of nearest neighbour substitutional sites in surface alloys is energetically unfavourable due to overlap of local strain fields due to the size mismatch between substrate and adsorbate [23]. Alternative models such as a semi-periodic array of Cu rows lacking substitutional Au to minimise the strain within the layer should also be considered. As a considerable proportion of surface atoms are within or directly beside the imaged "faults", further careful investigation by LEED or PhD may determine the nature of the periodic defect array.

Figure 2. Nano-metre faults in the Cu{100}-c(2x2)-Au surface alloy: (a) An STM image in differentiated form (dz/dx). Diagonal streaks illustrated by black wedges appear as shallow troughs. Ridges parallel to [001] are 3.6 A apart; (b) Suggested model in which a Cu3Au {100} mixed layer is compressed from L to L to fit the Cu{ 100} lattice [21].

312

The Cu{100}-c(2x2)-Pd structure first reported by Smith, Norris and Binns [24] was suggested to be due to an ordered CuPd surface alloy by Graham based on LEISS and angle-resolved ultra-violet photo-emmision spectroscopy (ARUPS) investigations [25]. Franco Jona's group in Stony Brook were the first to demonstrate conclusively that the structure corresponds to an ordered CuPd surface alloy by LEED I(V) analysis [26,27]. A small tippling in the bimetallic layer of 0.02+0.03 A was determined with Pd buckled outwards (see table 2). The first Cu interlayer spacing adopts the bulk value. No relaxation in deeper layers in the form of Pd-induced buckling in layer 3 or Pd-induced modification of interlayer spacings were considered, although the level of theory-experiment agreement attained was suggestive that no significant distortions in the third or deeper layers occur. This has been recently confirmed by Kaukasoina and co-workers using LEED I(V) analysis. This analysis using a combination of normal and off-normal data favours a slightly larger top layer tippling of 0.04+0.03 A (with Pd buckled outwards) and an increase in the outermost copper interlayer spacing to 1.86 A (+2.9%) [28]. Pope et al. used MEISS and LEED I(V) analysis to study the surface geometry. The MEISS data was not sufficient to perform a full structural analysis, however upon assuming that the Cu interlayer spacings were maintained at their bulk values, a tippling amplitude of between 0.04 and 0.08 A was preferred in the bimetallic layer with Pd tippled outwards [29]. The LEED I(V) analysis of Pope ruled out a c(2x2) CuPd underlayer alloy in which the mixed CuPd layer was sub-surface, a structure not originally tested by Jona and co-workers. All studies are in qualitative agreement that the correct structure is a c(2x2) top layer CuPd alloy with a small buckling of the outer CuPd monolayer with Pd relaxed outwards. From the theoretical viewpoint, Black used total energy calculations with Finnis-Sinclair potentials of Rafii-Tabar and Sutton to show that a c(2x2) twodimensional CuPd alloy with 50% of the surface covered with p(lx 1) Cu islands (due to displaced Cu) is the lowest energy structure [30]. This structure yields a lower energy than either an ordered c(2x2) Pd overlayer or p(lxl) close-packed Pd islands. The surface alloy was found to be extreemly close in energy to the p(lxl) Pd overlayer, the alloy being only 30 meV per atom lower in energy [30,31]. Pope et al. utilised molecular dynamics within the embedded atom method (EAM) [29]. Again, clear preference was found for a surface alloy over overlayer structures or sub-surface alloy models with a decrease in the Cu interlayer spacing to 1.75 A (-3%) and a tippling in the outermost alloy layer with Pd outermost of amplitude 0.12 A [29]. Kudmovsky, Bose and Drchal later confirmed, via first-principles electronic structure calculations within the local density approximation (LDA) using the tight binding linear muffin-tin orbital (TB-LMTO) method that an ordered c(2x2) surface alloy is a more stable structure than overlayer models [32]. The calculations indicate that nearest neighbour pairwise interactions dominate with an attractive Cu-Pd

313

nearest neighbour interaction of +88.6 meV in the surface alloy, to be compared to a repulsive nearest neighbour interaction between Pd and vacancy sites in the c(2x2) overlayer model o f - 2 6 9 meV. The small repulsive next nearest neighbour interaction between Cu-Pd pairs enhances the tendancy towards c(2x2) ordering within the surface alloy. Bozzolo and co-workers have studied the Cu {100}/Pd surface alloy system extensively thoughout the coverage range 0 to 0.5 ML using the Bozzolo-Ferrante-Smith (BFS) method [34], a theoretical framework suited for treatment of surface strucure and compositional profiles in binary alloys [35]. As in previous total energy studies, the surface alloy geometry with expelled Cu atoms coveting 50% of the mixed top layer c(2x2) CuPd alloy is the favoured structure. The Cu{100}-c(2x2)-Mn structure, first observed by Binns and Norris [36] is the best defined geometry of the three Cu{ 100}-c(2x2) surface alloys, having been subjected to three independent state-of-the-art analyses by LEED, PhD and MEISS [37,38,39,18]. Wuttig and co-workers examined the structure of the Cu{ 100}-c(2x2)-Mn alloy at 0.50• ML by tensor-LEED (TLEED). An unusually large buckling of the mixed layer of 0.30+0.02 A (Mn outermost) with the first two Cu interlayer spacings very close to bulk values was found [37,38]. This is surprising considering the rather similar metallic radii of Cu (1.278 A) and Mn (1.264 A) This large buckling has been shown to be due to magnetism in the 2D mixed layer and in particular the increase in effective size of Mn atoms due to their large local magnetic moment [38]. The large buckling, combined with the importance of the correlation between surface magnetism/structure attracted considerable interest to confirm the LEED structure. Woodruff and co-workers have independently studied this surface alloy using PhD and MEISS [39,18]. Table 2 summarises the geometries obtained, confirming the primary aspect of the geometry: the large buckling in the CuMn layer. However, discrepancies exist particularly in the value of the first copper interlayer spacing which varied significantly from 1.79+0.02 A (LEED) to 1.63+0.08 A (PhD). The MEISS analysis falls between the previous two values at 1.72+0.04 A. It is not clear at present why this difference arises, although the possibility of the presence of Mn sub-surface which is not taken account of in the modelling of experimental data and effecting the retrieved geometry in differing ways for each technique is a possibility. D'Addato and Finetti have studied the geometry by surface X-ray absorption fine structure (SEXAFS) [40] further confirming that the Cu{ 100}-c(2x2)-Mn structure is a surface alloy. Ab-initio full-potential linearized augmented plane wave calculations suggest that the large buckling is caused by ferro-magnetism within the monolayer (a ferro-magnetically ordered layer is lower in energy than a paramagnetic layer by 1.4 eV per Mn atom) [37]. Only in the ferromagnetic case was the large buckling found, due to the larger effective size of a ferromagnetic

314

Mn atom with calculations favouring a buckling amplitude of 0.24 A. The magnetic moment per Mn atom was found to be 3.75 ~tB. A large buckling has also been theoretically predicted for substitution of Cr in Cu{100}. Experimental studies to verify this would be of significant interest. In the cases of Au and Pd, the formation of c(2x2) ordered top layer alloys is quite understandable. These systems have exothermic enthalpies of mixing with the surface alloys themselves corresponding to the surface analog of the {100} surfaces of Cu3Au and Cu3Pd bulk alloys which consist of alternate layers of mixed c(2x2) CuAu(Pd) alloy and pure Cu. The case of Mn is more surprising since Cu and Mn do not form bulk alloys. A general trend for all c(2x2) surface alloys studied to date is that adsorbate-induced buckling in deeper layers is very small, as are variations in second and deeper interlayer spacings. In the cases of the Cu { 100}-c(2x2)-Au(Pd) systems, the buckling is much reduced from that expected based on the sum of metallic radii with the first Cu interlayer spacing being significantly expanded. Brown et al. have noted that for a range of surface alloys (including Cu{ 100}-c(2x2)-Au(Pd)) that the nearest neighbour adsorbate-substrate distances are noticeably smaller than the sum of metallic radii and are closer to the sum of covalent radii [41,42]. The contraction in the nearest neighbour Cu-adsorbate distances for Au and Pd respectively are 0.10 and 0.16 A relative to the sum of metallic radii. This is a manifestation that adsorbate atoms, which have lost nearest neighbours, prefer to sit in regions of high valence electron density within the surface: this provides a driving force for reduction of the adsorbate-substrate bond lengths. Tersoff has illustrated genetically that strain effects for systems with size mismatch can lead to formation of surface alloys even for bimetallic combinations which are not bulk miscible [23]. These systems have been termed "surface confined alloys". Large size mismatches, while preventing bulk miscibility due to the large strain energy of mis-sized atoms placed within the bulk of a solvent, do allow surface alloy formation where the strain energy is much reduced. Most importantly, local strain fields generated around adsorbate atoms substituted in the outermost layer makes cases in which adsorbate atoms occupy nearest neighbour sites energetically unfavourable. A small attractive adsorbate-adsorbate next nearest neighbour interaction promotes c(2x2) long range order (no experimental evidence for c(2x2) island formation at low adsorbate coverge exists at present (see section 2.2)). The electronic structure of the Cu{ 100}-c(2x2)-Pd, Au and Mn systems have been probed by ARUPS. Wang et al. [16] have compared valence band photo-emission spectra using synchrotron radiation from the Cu{ 100}-c(2x2)Au surface alloy with a Cu3Au { 100} bulk alloy terminated by a mixed c(2x2) CuAu monolayer. A surface-induced Au d-band narrowing of 0.45 eV was found for the Cu{100}-c(2x2)-Au surface alloy, despite the smaller lattice

315

constant of the surface alloy which would be expected to promote a larger dband width. The narrowing is a manifestation of the reduced co-ordination of the Au atoms in the surface alloy relative to the bulk alloy. Virtually no dispersion of the Au-derived d-band complex is observed perpendicular to the {100} surface in the photon energy range 24-40 eV in-line with a two-dimensional CuAu layer. This contrasts with the Cu3Au{100} bulk alloy surface which contains chains of Au atoms separated by 3.76 A for which a noticeable dispersion of the Au d-bands normal to the { 100} surface is observed. Graham has also compared the electronic structures of the Cu{ 100}-c(2x2)-Au system and its bulk alloy counterpart using resonance radiation from a noble gas discharge source [15]. At the M point of the Surface Brillioun Zone (SBZ) the electronic structure was found to be very similar with both surfaces exhibiting a strongly shifted and broadened Tamm surface state, indicating the two surfaces to be highly similar. This observation has been utilised to study the growth mechanism of Au on Cu { 100} at coverages (see section 2.2). In the case of Cu{100}-c(2x2)-Pd, ARUPS studies have identified a marked withdrawl of the Pd d-band from the Fermi level due to the absence of Pd-Pd nearest neighbour bonding [25,26,27]. The Pd atoms substituted within the copper surface appears to adopt a closed d-band electronic configuration, hence would be expected to have significantly different chemisorption and reactivity properties with respect to pure Pd surfaces. In agreement with its Cu{100}-c(2x2)-Au counterpart, the Pd d-band emission shows little or no dispersion as a function of photon energy in normal emission ARUPS , consistent with formation of a largely two-dimensionally confined surface alloy. The Cu{100}-c(2x2)-Mn phase has been extensively investigated by ARUPS [42,43]. Early studies reported a Mn related electronic state at a binding energy of 1.3 eV with respect to the Fermi edge assigned to a virtual bound 3d state of Mn atoms which dissapeared for Mn coverages >0.25ML, thus is not a characteristic feature of the c(2x2) surface alloy. The Mn 3d majority spin band position has been determined to be at a binding energy of 3.7+0.3 eV by Huttel et al. and 3.0 eV by Schiller and co-workers with the minority spin states 1.85 eV above the Fermi level giving an exchange splittings of 5.5 eV and 4.8 eV respectively. Huttel et al have also found evidence for formation of new electronic states close to the Fermi level.

2.2. Growth Mechanism of Cu{100}-c(2x2) Surface Alloys Murray et al. have probed the detail of the growth mechanism of Pd on Cu{ 100} by STM [44,45]. LEED observations indicate the onset of ordering occurs at Pd coverages of around 0.25 ML, in the form of circular diffuse c(2x2) beams which increase in intensity and decrease in full-width-at-halfmaximimum as the Pd coverage is increased to 0.554-0.05 ML. STM studies

316 indicate that at low Pd coverages protrusions are formed on ( l x l ) lattice sites with apparent heights between 0.1 and 0.3 A, interpreted as formation of a twodimensional substitutionally disordered alloy. Figure 3(a) illustrates an atomic resolution STM image upon deposition of 0.20 ML of Pd. Adsorbate atoms substitute into top layer Cu lattice sites and are imaged either as protrusions or depressions (figure 3(b)) depending on the tip

Figure 3. A series of STM images for the Cu{ 100}/Pd surface alloy system: (a) a 50x50/~k2 image (0pd=0.20 ML) illustrating the formation of protrusions at top layer lattice sites as indicated by the (lxl) unit mesh superimposed in the upper fight comer; (b) as (a) after a tip change with protrusions imaged as depressions; (c) a 1000xl000/~2 image illustrating the formation of islands on the copper terraces and roughening of step edges; (d) a 1500x1500/~2 image (0pal=0.40 ML) indicating an increase in the density of islands and a widening in their size distribution [44].

317

utilised. Larger scale images indicate the appearance of islands decorating the Cu {100} terraces (figure 3(c)) and roughening of step edges. The island density increases with Pd coverage (figure 3(d)). The height of the islands is similar to the mono-atomic step height of Cu {100} (1.8 A ) , and are assumed to contain predominantly Cu atoms ejected from the terraces by arriving Pd atoms, while the roughening of step edges occur due to a portion of ejected Cu atoms diffusing to and attaching themselves to copper step edges. Statistical analysis of STM images indicate that growth occurs with substitutional Pd atoms avoiding occupation of nearest neighbour sites with an enhanced proportion of Pd atoms with Pd next nearest neighbours compared to that expected based on a totally random site occupation. Statistical analysis of STM images also indicates that the Pd coverage within the outermost layer is less (by between 17 and 25%) than the Pd coverage deposited as calibrated by Rutherford backscattering spectroscopy (RBS) due to burial of a fraction of Pd adatoms by displaced Cu arriving at step edge coveting areas of CuPd alloy or due to Cu islands coveting or "capping" areas of c(2x2) CuPd alloy. At a Pd coverage of 0.55 ML, large areas of defect free c(2x2) CuPd alloy may be imaged with a very low concentration of anti-phase domain wall boundaries, implying that site switching between Cu and Pd atoms occurs readily at 300 K ironing out defects within the alloy islands. However, a significant defect concentration in the form of p(lxl) Cu domains located primarily at upper step edges occurs. The Cu {100}-c(2x2)Pd surface alloy is thus inhomogeneous even at Pd coverages corresponding to maximal perfection as judged by LEED. Diffuse LEED (DLEED) has been applied by Barnes et al. [46] to probe the local geometry of the Cu {100}/Pd system in the Pd coverage range 0.10 to 0.55 ML. Figure 4 illustrates DLEED (figures 4(a) and (b)) and conventional LEED I(V) spectra (figure 4(c)) from the (1/2,1/2) beam position. Normal incidence LEED/DLEED is not highly sensitive to the lateral ordering of atoms within the surface plane [46]. The similarity in spectral structure and peak positions for the low coverage spectra with that from the Cu{100}-c(2x2)-Pd overlayer at 0.55 ML demonstrates that the majority of Pd adatoms occupy top layer substitutional sites throughout the entire coverage range. The top layer Cu and Pd were found to be almost co-planar at all coverages, with a small buckling of the larger Pd atom outwards of between 0.02 and 0.05 A, accompanied by a small Pd-induced expansion of the outermost Cu interlayer spacing [46].

318 9

I

'

I I I I

.,

I

'

I

'

I

'

I

'

I !

(c)

"

I

I

r

(b) I

~

I

"

%

~

I

100

,

I

,

150

(a)

" " __

,

I

,,.

,,..

,

200

I

250

_

,

I

300

Energy (eV) Figure 4. Experimental (full lines) normal incidence DLEED and LEED spectra from the (1/2,1/2) position: (a) 0pd =0.15 ML; (b) 0pd = 0.25 ML and (c) Cu{ 100}-c(2x2)-Pd at 0Pd = 0.55 ML. (a) and (b) correspond to DLEED spectra collected at low Pd coverage where a p(lxl) LEED pattern is observed. The dotted lines represent theoretical fits [46].

The partition of Pd between the outermost and second layer has been studied in detail by Yao et al. using He + and Li + LEISS utilising pure Cu {100} and Pd { 100} samples as standards [ 19]. Figure 5 illustrates the results, showing that for a coverage at which the c(2x2) LEED pattern reaches perfection (0.5-0.6 ML), the top layer composition is 62 at% Cu and 38 at% Pd, consistent with one quarter of the outermost layer being composed of pure Cu. Figure 5 illustrates that a portion of the Pd deposited is incorporated subsurface throughout the entire coverage range up-to and including the completion point of the Cu { 100 }-c(2x2)-Pd surface alloy around 0.5 ML.

319

00 100 '

.,J E

3.0

80

.o

60

E o o

40

(/) 0 o.

Pd Coverage (ML) 0.5 1.0 1.5 2.0 2.5

o

20

0 ~,

0

150 300 450 600 750 Pd Exposure Time (s)

900

Figure 5. First and second layer composition in the Cu{100}/Pd surface alloy system as a function of Pd coverage determined by combined He+/Li+ LEISS [ 19]. As reported by numerous workers using a range of techniques including MEISS, thermal desorption of probe molecules (CO) and polar X-ray photoelectron diffraction (XPD) [47-50], there is significant heterogeneity in the Cu {100}-c(2x2)-Pd alloy with a significant portion of the deposited Pd resides in second layer substitutional sites and domains of pure Cu in the outermost layer. The mechanism of formation of the Cu{100}-c(2x2)-Au surface alloy exhibits many similarities to its Cu{100}-c(2x2)-Pd counterpart. LEED observations indicate a p(lxl) with increasing background up-to coverges of about 0.25 ML with a diffuse c(2x2) appearing between 0.25 and 0.30 ML coverage, increasing in perfection as the coverage is raised up-to 0.50 ML [14,15]. Aspects of the mechanism of formation of the Cu {100}-c(2x2)-Au surface alloy have been probed by monitoring the Tamm surface state located at the M point of the surface Brillioun zone of the clean Cu {100}-(Ix l) surface. The Tamm state consists of dx2.y2 orbitals of top layer Cu atoms, split-off from the top of the Cu d-band by the potential energy discontinuity at the surface. This surface state is delocalised within the surface plane, hence is sensitive to _ . . . _

320

the average Au concentration within the outermost layer [51]. Early ARUPS work by Hansen and co-workers reported a linear shift of the Tamm state binding energy with gold coverage up-to 0.50 ML [52-54], with only a single peak observed at all Au coverages. This suggests that Au is randomly distributed across the surface ruling out island growth of c(2x2) surface alloy. Island growth would necessarily produce two Tamm states at energies characteristic of the clean Cu{100} surface (binding energy of-1.78 eV with respect to the Fermi level) and the Cu{100}-c(2x2)-Au surface alloy (binding energy: -1.55 eV) as HeI ARUPS is easily capable of the resolution required. In a careful follow up study, Thielmann et al. were able to detect a clear deviation from linearity of the Tamm state binding energy in the Au coverage regime 0.3 to 0.5 ML [55], with a final binding energy o f - l . 5 1 eV at 0A~=0.50 ML, in good agreement with that from a bulk Cu3Au{100} alloy terminated by a mixed CuAu layer. The deviation in linearity has been associated with the observation of streaks in LEED due to formation of nano-metre scale defects (see section 2.1). High resolution core level spectroscopy indicates a single Au 4f7/2 core level between Au coverages of 0.125 and 0.5 ML at a binding energy of 84.05+0.03 eV, indicating occupation of a single identical Au site throughout this coverage range. This provides strong evidence for surface alloy formation even at low Au coverages. Again, the measured Au 4f7/2 binding energy is identical with measurements from the surface Au component of a Cu3Au { 100} bulk alloy at 84.05+0.03 eV and may be compared with binding energy of 83.84+0.03 eV for 0.5 ML of Au deposited at 173 K where a p(1 x l) LEED pattern is observed and intermixing thermally inhibited [53]. No evidence of a second core state due to Au atoms within or in the vicinity of the nano-defect sites was found. Thus, a growth mechanism in which Au initially substitutes within the outermost Cu { 100} layer in a quasi-random fashion, spreading homogeneously across the surface and avoiding occupation of nearest neighbour sites due to the large associated strain energy. Local c(2x2) ordering is observed at coverages above 0.25ML, with the order within the layer increasing up-to completion of the c(2x2) overlayer at a coverage around 0.5ML. Essentially, the mechanism of formation of the Cu{ 100}-c(2x2)-Au surface alloy appears highly similar to its Cu{100}-c(2x2)-Pd counterpart, with the possible exception of less of a tendancy for Au to occupy sub-surface sites due to the larger associated elastic strain due to the larger metallic radius of Au. In the case of Cu{100}-c(2x2)-Mn, the mechanism for formation of the surface alloy from low coverage up-to completion of the c(2x2) has been probed in detail by STM by a number of groups including Noh et al. [56], Van der Kraan and van Kempen [57] and more recently by Wuttig, Flores and coworkers [58,59,60]. In the original work of Noh et al., Mn adsorbates were imaged as bright spots at low coverage which appeared to cover 3 or 4 Cu atoms, interpreted as

321

clusters of Mn adatoms. As the coverage was raised, fuzzy images of "disordered" regions of surface were imaged and at a Mn coverage of 0.4 ML, regions of c(2x2) with anti-phase domain boundaries co-existing with disordered regions. Step edge roughening occurred, assigned to displaced Cu atoms. Noh at al postulated that the fuzzy images were due to a particularly high Mn diffusion co-efficient at intermediate Mn coverages, estimated to be >l.5xl0~6cm2s ~ with a diffusion activation energy <0.74+0.06 eV [56]. The apparent corrugation of 0.12 A measured by STM is much lower the geometric corrugation of between 0.3 and 0.4 A [18,37,38,39]. In contrast with the Cu{100}/Pd and Au systems, Noh et al. postulated that surface alloying is delayed until a critical Mn coverage is reached. Van der Kraan and van Kempen [57] reported formation of approximately circular islands of height 1.8 A forming upon deposition of 0.20 ML of Mn. Atomic resolution images indicate that the Cu{100}-(lxl) is maintained both between and on the islands for this and coverages up-to 0.5 ML, suggesting that nucleation of c(2x2) domains does not occur. With increasing Mn coverage the islands grow in size and become more disordered with height differences between 0.3 and 0.5 A between regions maintaining good p(lxl) order and disordered regions. An apparent discrepancy was noted with the earlier study of Noh et al who did not observe circular islands decorating the Cu { 100} surface. This was explained by van der Kraan and van Kempen to be due to a slightly higher substrate temperature used during Mn deposition by Noh et al. [56]. Van der Kraan and van Kempen found that annealing of the Cu{ 100}/Mn films led to disappearence of Cu islands due to diffusion away to step edges. For Mn coverages around 0.5 ML, where a c(2x2) LEED pattern is observed, rectangular islands at a height of 1.8 A above the lower terrace with edges running in the [001] and [010] directions with an internal c(2x2) periodicity are imaged as illustrated in figure 6. Areas between the rectangular islands also adopt a c(2x2) periodicity, implying that Mn is incorporated both into the terraces and the growing islands. A larger height corrugation of 0.3 to 0.4 A is observed by STM in the work ofvan der Kraan and van Kempen in the c(2x2) structure. Several aspects of Noh et al and van der Kraan and van Kempen's interpretation of the growth mechanism were later challenged by Wuttig and coworkers [58,59,60]. In particular, they were able to show that there existed no critical coverage for alloy formation. Protrusions in the Cu {100} surface centred on lattice sites could be imaged at coverages as low as 0.0033 ML due to single Mn atoms alloyed into the Cu{ 100} surface. As the coverage was raised to 0.09+0.01 ML islands could be imaged above the Cu terraces with small protrusions on the islands and on the terraces in the vicinity of island edges indicating alloying into both islands and terraces close to island/terrace boundaries. The c(2x2) ordering sets in at a ban coverage around 0.3 ML with a well ordered c(2x2) alloy forming at 0.5 ML.

322

Figure 6. STM images from the Cu{100}-c(2x2)-Mn surface alloy at a Mn coverage of around 0.7 ML: (a) a 270x270 nm2 image illustrating a dual-level surface consisting of rectangular islands with island edges aligned along [001] and [010] directions; (b) a 16x16 nm2 image illustrating c(2x2) periodicity on both islands and terrace. The black arrow indicates a small disordered region [57]. In agreement with previous STM studies, only one atom, in the two atom per unit cell c(2x2) could be imaged, assumed to be Mn due to the large buckling detected in structural studies [18,37,38,39]. For deposition at room temperature the Mn was not distributed entirely homogeneously at low coverages. The majority of Mn found around island edges and in the vicinity of step edges with more on upper step edges. Furthermore, steps of differing type such as [010] and [011 ] oriented steps had differing distribution of Mn in the vicinity of upper and lower step edges. Annealing to 340 K for 10 minutes led to a homogeneous Mn distribution with no copper islands at low coverage (0.032 ML), indicating that these inhomogeneities are kinetically limited.

3. SURFACE A L L O Y F O R M A T I O N UPON Co, Fe and Ni A D S O R P T I O N A number of other transition metal adsorbates undergo intermixing. Three of these cases will be mentioned: Co, Ni and Fe. There has been considerable work on both the Cu{100}/Co and Cu{ 100}/Fe thin film systems due to the increasing interest in magnetism in systems of reduced dimensionality. The literature on both systems is extensive, hence this section concentrates solely on reports of surface alloy formation.

323

Angle-scanned X-ray photo-electron diffraction [61,62] and STM work [63,64] indicated Co growth on Cu {100} in the form of bilayers. Pentcheva and Scheffler using density functional theory reported that growth of bilayer Co islands capped by a copper monolayer is the lowest energy structure at higher Co coverages [65]. Maximisation Co-Co bond formation was identified as the driving force for the unusual bilayer growth mechanism. Copper capping results in a substantial lowering of surface energy by 0.5 eV per (1 x l) unit cell. At low Co coverages (<0.25 ML) substitutional adsorption of Co is preferred. The substitutional adsorption becomes energetically unfavourable with increasing Co coverage in favour of bilayer island formation. Recent STM studies of the Cu{100}/Co system by Fal3bender et al. report that the growth of the initial two cobalt monolayers depends strongly on growth conditions [66]. Nouvertne and co-workers [67,68] have reported that atomic exchange occurs at the Cu{ 100}/Co interface upon room temperature adsorption, with Co adatoms substituting into the outermost Cu monolayer in a somewhat disordered fashion. Adsorption of 0.06 ML of Co and annealing to 453 K yield interdiffused regions which are imaged as dark patches. The interdiffused regions are a few atomic diameters in width as illustrated in figure 7. Islands above the Cu {100} substrate are also imaged. A height profile indicates the islands to be similar to the Cu{ 100} step height, while the alloyed regions appear as depressions of depth 0.6+0.1 A. It is thought that the islands consist mainly of copper expelled due to Co substitution into the Cu {100} surface, as indicated by combined STM/total energy calculations [68]. Upon deposition of 1 ML of Co followed by annealing to 453 K, c(2x2) ordering has been observed both in LEED and reflection high energy electron diffraction (RHEED). The intensity of the superlattice beams were very weak in both cases, assigned to chemical ordering of Cu and Co in a c(2x2) alloy weakly sensed due to the similar elastic backscattering factors of Cu and Co. Domains of c(2x2) periodicity could be imaged in STM, however the surface was strongly heterogeneous. A multi-layered surface was formed with three layers visible. The size of the single layer domains was small, with the regions containing c(2x2) periodicity being the largest with an average length of about 100 A. The apparent height variation in each layer was found to be 0.6+0.1 A and was attributed to a combination of surface topographical and electronic contrast between Cu and Co. The possibility of formation of Cu capped subsurface Co has been considered theoretically [69]. The energy of Co substituted in-to second and deeper layers is energetically favoured, although the difference in energy compared to Co sustituted into the outermost Cu monolayer is small. Sub-surface nanoparticles of Co have been observed experimentally [70].

324

Figure 7. (a) An STM image aquired at 300 K of 0.06 ML of Co on Cu{ 100} atter annealing to 453 K for 1 hour. The dark regions are imaged 0.6 A below the copper terraces. Bright regions indicate islands of height 1.8 A above the Cu{100} surface ; (b) a line profile illustrating the apparent height difference across A-B in (a) [67]. Photo-emission from adsorbed xenon (PAX) studies indicate that for a Co film of nominal thickness of 2 ML grown at 300 K, approximately 65% of the surface consists of disordered CoxCu~.x alloy with the remaining 35% being pure Co [71 ]. A recent combined SATEED and thermal desorption study using CO as a probe molecule of the room temperature growth of Co on Cu { 100} favoured initial growth of a p ( l x l ) disordered alloy with Co in both the outermost and second layers co-existing with Co clusters at lower Co coverages (<2 ML). A p ( l x l ) epitaxial Co overlayer forms above the intermixed interface as the Co coverage is raised [72]. Blugel has predicted that a c(2x2) CuCo surface alloy is

325

unstable with respect to Cu surface segregation in the Cu {100}/Co bimetallic combination [73]. The Cu{100}/Ni system is also controversial. X-ray photo-electron spectroscopy (XPS) initially provided evidence for intermixing upon growth at room temperature [74,75] while a LEED I(V) analysis of Kim et al. favoured copper-capping of the Ni films for Ni coverages of 1-3 ML [76]. A subsequent LEED study indicated that Ni grows pseudomorphically up-to coverages >11 ML [77]. However, the similar scattering properties of Ni and Cu for low energy electrons is likely to make the LEED analyses rather insensitive to decide whether atomically sharp Cu{100}/Ni interface formation, intermixing or copper capping occurs in the early stages of Ni adsorption. STM studies of the Cu {100}/Ni interface have been carded out. Lindner et al. provide compelling evidence for intermixing [78]. Lindner et al. report that deposition of low Ni coverages (<0.1 ML) leads to roughening of step edges and appearance of dark patches or "holes", rather similar to those observed on the Cu { 100}/Co system and of apparent depth 0.5-0.8/k and 5-8 A in diameter. Islands of height 1.6 to 1.7 A are also imaged. The "holes" are thought to be regions where intermixing has taken place due to differences in the local density of states (LDOS) between pure Cu and intermixed regions [78]. The local composition of the "holes"is unclear. Lindner et al. have suggested two possibilities: clusters of Ni atoms substituted within the outermost Cu monolayer or mixed CuxNi~_x regions. It is also possible the dark regions correspond to areas where Ni has penetrated subsurface. There exists a strong driving force of lowering the systems surface energy by penetration of Ni sub-surface. Higher Ni coverages (about 1 ML) lead to a highly heterogeneous surface with a somewhat random distribution of "holes" coveting about 35% of the surface. The adsorption of Cu {100}/Fe system has been extensively studied due to the excellent lattice matching offering the possibility to grow Fe films with a meta-stable f.c.c, stacking pattern. This created intense interest in modification of magnetic properties for thin films with altered crystal structure. Wuttig, Feldmann and Flores have reviewed the work on the Cu {100}/Fe system up-to 1995 [79]. The Cu {100}/Fe system has been highly controversial with results on the structure, growth mechanism and magnetic properties often differing from laboratory to laboratory due to a strong temperature dependence of the growth mechanism and structure. Early work favoured ideal layer-by-layer growth forming a metastable (lxl) f.c.c film up-to Fe coverages of about 10 ML [79]. Steigerwald, however identified strong deviation from layerwise growth. Polar X-ray photo-electron diffraction suggested the presence of Fe in second and deeper layers at a coverage of 1 ML and intermixing was suggested to occur for room temperature deposited Fe films at low coverages [80]. X-ray induced Auger spectroscopy [81 ] and RHEED indicated formation of Fe bilayer clusters at a Fe coverage of 1 ML [82]. Detzel suggested a tendacy for Cu segregation

326

leading to copper capping of Fe films and formation of a Cu/Fe/Cu type structure [83]. Chambliss and co-workers have extensively studied the growth of Fe on Cu {100} by STM reporting strong deviation from layer-by-layer growth [84-87]. Islands of size 20-40 A of height 1.7:L0.2 A were imaged along with holes of depth 1-2 A [85]. The depth of the holes were noted to vary with bias voltage, suggesting that these features are not purely topographic but due to chemical differences and were suggested to be due to Fe rich or intern~xed regions. Shen et al. using STM have shown that thermal activation of Fe films to 490 K leads to Cu/Fe/Cu structures due to copper segregation [88]. Most recently, Kim et al. using positron-annihilation induced Auger electron spectroscopy deduced that at 1 ML coverage for deposition at low temperature Cu remains present in the outer layer, suggestive of non-2D growth [89]. The copper signal increases at higher temperature suggestive of either interdiffusion or Cu capping of the Fe clusters. The metals Co, Ni and Fe are all rather well matched to copper in terms of metallic radii yet have higher surface energy (see table 1). Thus, in order to minimise surface energy it is likely they prefer to be Cu capped if sufficient thermal activation is provided. While Ni forms a disordered f.c.c, solid solution throughout the composition range, Fe and Co are relatively insoluble in Cu. It is possible that the high Fe-Fe and Co-Co bond energies may lead to clustering and phase separation rather than true surface alloy formation.

4. SURFACE ALLOY FORMATION UPON ALKALI AND ALKALINE EARTH METAL ADSORPTION Alkali metal adsorption on Cu{ 100} has been extensively studied for adsorbates including Li,Na,K and Cs [90,91]. With the exception of Li, alkali metals form overlayer structures on Cu {100}.

4.1. The Cu{100}/Li Surface Alloy: The Coverage Dependent (2xl) --~ (3x3) --~ (4x4) Transition Tochihara and Mizuno have shown that the structure formed upon Li adsorption is strongly dependent on the adsorption temperature [92]. Adsorption at 180 K leads to formation of a c(2x2) structure in the coverage range 0.2<0L~<0.5 ML with complex rotated structures forming at higher coverage upto completion of the first monolayer at 0Li=0.7 ML. The c(2x2) structure has been studied by LEED I(V) analysis [93]. Lithium occupies four-fold hollow overlayer sites at a perpendicular distance of 1.96+0.08 A above the outermost Cu layer with a copper interlayer spacing of 1.81+0.04 A, equal to the bulk value. An effective Li radius of 1.39+0.08 A was deduced which is smaller than

327

the Li metallic radius by 0.17 A. The Cu{ 100}-c(2x2)-Li overlayer converts irreversibly to a (2x 1) structure at temperatures above 200 K. The Cu { 100}-p(2xl)-Li phase is imperfect at room temperature with the superstructure LEED beams being streaked. LEED I(V) analysis, indicates formation of a missing row type reconstruction in which every second Cu row along the [011] (or [011] in the 90 ~ rotated domain) directions is removed [92,94,95]. Lithium atoms are assumed to occupy the missing rows forming a CuxLi~.x surface alloy as illustrated in figure 8. The streaking of superlattice beams has been interpreted as loss of correlation between neighbouring rows of Li atoms. Quenching the temperature to below 300 K leads to removal of the streaking, signalling formation of a more ordered state in which correlation between neighbouring Li rows is established. The first interlayer spacing was contracted to 1.68+0.05/~ (-7+3%) with the second layer spacing expanded to 1.84+0.07 A (+2+4%) [94,95]. As the LEED modelling did not include scattering due to the adsorbates, it was not possible to extract details of the Li adsorption site. As the Li coverage is raised, the p(2xl) transforms to a p(3x3) overlayer. The structure of this phase has been determined by TLEED [96]. The favoured structure corresponds to a complex surface alloy consisting of quartets of Cu atoms capped by a Li adatom arranged in a p(3x3) periodicity with a pyramidlike morphology with additional Li atoms substituted in-to missing Cu rows as illustrated in figure 9.

Top view

Side view

Figure 8. Schematic models of the Cu{ 100}-p(2xl)-Li structure. The open circles indicate Cu atoms and filled circles Li atoms. The p(2xl) unit cell is indicated by dotted lines with Li atoms placed in the missing Cu rows [94,95].

328

Figure 9. (a) Top and side views of the Cu{ 100}-(3x3)-5Li structure. Filled circles represent Li atoms and open circles Cu. The line A-A' in the side view represents the base line from which heights are measured in table 3; (b) top view of the first complete copper layer. Arrows show lateral relaxation directions while the numbering system indicates groups of symmetrically equivalent Cu atoms [96]. Table 3 summarises the surface geometry of the Cu{100}-p(3x3)-5Li phase. The substitutional Li atoms are slightly displaced from 4-fold hollow sites with a distance between neighbouring Li adatoms of 3.15+0.52 A,. The additional Li adatoms present are adsorbed in four-fold hollow sites, similar to that of the Cu { 100}-c(2x2)-Li low temperature overlayer. The first complete Cu layer is buckled, with Cu atoms 5 and 6 at almost the same height, while Cu atoms 4 are buckled outwards by 0.06 A. A significant contraction of the first interlayer spacing between the Cu quartets and the average position of the second layer to 1.67 A, (-7.5%) occurs. This agrees well with the corresponding distance between the outermost and second Cu layers in the Cu{ 100}-p(2xl) phase (1.69 A). The proposed model corresponds to a Li absolute coverage of 0.56 ML.

329 Table 3 Summary of the optimal structural parameters for the Cu{100}-(3x3)-5Li and the Cu{100}(4x4)-10Li surface alloys. Lateral displacements are from four-fold hollow sites in directions indicated by arrows in figures 9 and 10 respectively. Heights are measured from the plane A-A' in figures 9 and 10. Cu { 100 }-p(4x4)- 10Li

Cu { 100 }-p(3x3)-5Li

Atom No.

Lateral displacement A

1

Height A

Interlayer spacing

Atom

Lateral displacement

Height

A

No.

A

A

5.38+0.07 1.50

2

0.30-a:0.26

3.88+0.08

3

0.02+0.11

3.47+0.04

4

0.04+0.11

1.83+0.06

5

0.08+0.08

1.77+0.05

0.41 1.64

0.06

1 2 3 4 5 6 7 8 9

0.19-a:0.28 5.40-a:0.06 3.92+0.16 0.23+0.33 3.82+0.12 3.70-a:0.11 0 . 0 3 + 0 . 2 2 3.47• 0 . 1 3 + 0 . 2 0 3.39+0.07 0.00-a:0.19 1.70a:0.09 0 . 0 3 + 0 . 1 5 1.77+0.06 0.03+0.25 1.74+0.08

0.01 6

1.76+0.09

Further increasing the Li coverage leads to a transition to a p(4x4) structure which has also been probed by TLEED [97]. Figure 10 illustrates the favoured structure, corresponding to a Li surface coverage of 0.625 ML. The (4x4) structure has four Li atoms decorating Cu islands consisting of 9 atoms each with the Cu islands arranged in a p(4x4) array. The missing Cu rows between the Cu islands contain three Li atoms with Li(3) displaced from 4-fold hollow sites leading to a Li interatomic separation of 2.78+0.33 A. In a similar fashion, the Li(1) atoms decorating the Cu islands displace from 4-fold hollow sites leading to a Li-Li separation of 2.81 A. The height of the Li(1) atoms above the Cu island is 1.88/~, compared to 1.91 A in the (3x3) structure. In contrast, the height of the Cu islands above the first complete underlying copper monolayer is larger at 1.78 A.

330

Figure 10. (a) Top and (b) side views of the Cu{100}-(4x4)-10Li structure. Filled circles indicate Li atoms, open circles represent Cu atoms in the mixed CuLi layer. Atom number 1 represents the four symmetrically equivalent Li atoms in the outermost layer while arrows represent directions of displacement from four-fold hollow sites. The line A-A' represents the base line from which atom heights are measured in table 3.; (c) top view of the first complete Cu layer; (d) the mixed CuLi layer. Symmetrically equivalent atoms are numbered [97]. The driving force for the transition from the Cu{100}-p(2xl) to the p(3x3) and p(4x4) alloys appears to be formation of structures that retain a large proportion of Li atoms in preferred substitutional sites. Calculations indicate that adatoms are energetically less favourable than the substitutional sites despite the energy cost to the substrate of creating a missing row restructured surface [99,100]. Thus, as all substitutional sites are occupied in the p(2xl) phase, additional Li atoms are forced to occupy four-fold hollow overlayer sites. The (3x3) and (4x4) structures allow the number of substitutional Li atoms to remain approximately constant: the substitutional Li coverage varies from 0.40 to 0.44 to 0.375 ML as we progress from p(2xl) ~ p(3x3)----~ p(4x4).

331

4.2. The Cu{100}-c(2x2)-Mg Surface Alloy Surface alloy formation has recently been reported for Mg adsorption on Cu{100} by Chen et al. [101]. A c(2x2) LEED pattern is observed, first appearing at a coverage of approximately 0.2 ML with half order beam intensities increasing up-to completion of the c(2x2) phase at 0.5 ML. This behaviour is highly reminiscent of the transition metals Pd, Au and Mn (see section 2). The structure of the c(2x2) has been investigated by TLEED and first principles total energy calculations using the full-potential augmented plane wave (FLAPW) method within the local density approximation. A range of overlayer structures with Mg in hollow, bridge and atop sites were tested along with a sub-surface alloy consisting of a c(2x2) CuMg two-dimensional alloy capped by a copper monolayer. These structures were ruled out in favour of a top layer two-dimensional c(2x2) CuMg alloy. The geometric parameters extracted from the TLEED analysis are given in table 2. A significant buckling exists in the outermost mixed layer of 0.55 A, with Mg tippled outwards. In agreement with the transition metal c(2x2) surface alloys, the Mg-induced tippling in the third copper layer is negligible. Formation of the surface alloy leads to a contraction in the first Cu interlayer spacing to 1.72 A (-4.8%) with all deeper Cu interlayer spacings similar to Cu {100}. Calculations confirm that the top layer substitutional site is most stable by 0.26eV per Mg adsorbate [ 101,102]. The major aspects of the geometric structure determined theoretically are in quantitative agreement with the TLEED analysis: including a first layer tippling of 0.60 A, with Mg outermost and a decrease in the first copper interlayer spacing to 1.66 A (-5.5%). Chen et al. have broken down the total energy into that of formation of a c(2x2) array of vacancies in the outermost layer at an energy cost of 0.93 eV per vacancy and the chemisorption energy of Mg. The greatly increased binding energy in substitutional top layer sites is sufficient to offset the energy cost of vacancy formation. Figure 11 illustrates the result of total energy calculations for a number of possible adsorption sites. The calculated charge re-distribution upon alloy formation for the favoured surface alloy model indicates chemical bond formation with both first and second layer Cu atoms as illustrated in figure 12. Loss of charge density above and below Mg atoms, is re-distributed in polar covalent bonds between first and to a lesser degree second layer Cu atoms. This is consistent with bond lengths determined by TLEED between Mg and top layer (2.61 A,) and Mg and second layer (2.90 A) Cu atoms which, when compared to the sum of metallic radii of Cu and Mg (2.88 A), indicates bond formation.

332

0.93eV 0.69eV

c(2x2)v - (2x 1)v ................

(Ixl)

OeV ;>

r r Ox

c,i -1.0 eV

-2.25eV -2.52eV -2.78eV

(2xl)s

c(2x2)s II

-2.0 eV

,~

__

-3.0 eV Figure 11. Energy diagram for Mg adsorption on Cu{100} in overlayer four-fold hollow (c(2x2)h) and (2xl) and c(2x2) substitutional top layer sites. The energy per atomic vacancy for top layer vacancies in (2xl) and c(2x2) arrays are also indicated relative to a (lxl) surface [ 101].

The large difference in structure of the surface alloys formed by Li and Mg occurs despite similarity in metallic radii and surface energies (see table 1). The grossly different structures formed appears to be due to differences in the valence electron configurations of Li and Mg due to the increased sp-electron density of Mg. In contrast to Cu{ 100}-c(2x2)-Mg, total energy calculations indicate that in the case of Li, c(2x2) overlayer models with Li occupying hollow sites and p(2xl) surface alloys are similar in energy, both being considerably more stable than a c(2x2) surface alloy [99,100].

333

.... -" .

-

(a) ... .... ..~

/ -'" :::-'.',,,'~ /I, ~.~) ~r i)

-

IW

/ - - -..

...... .... --~

, '

.-

f,

~r

..---- ---..,,,~ ~--~

--

tf,

(b) '

!

"-

\

lilI 'W r

/

~.\

I

lll 'WIKJlll(( I

Figure 12. The change in charge distribution for the Cu{100}-c(2x2)-Mg system: (a) a cut along the [011] plane showing Mg ( O ) and Cu (11) atoms in the outermost and third layers; (b) a cut in the [001] plane containing Mg and Cu atoms in the top and second layers. Solid lines indicate increases and broken lines decreases in charge density respectively [ 101 ].

5. D E - A L L O Y I N G T R A N S I T I O N S : IVA and VA METALS

ADSORPTION

OF GROUP

IliA,

The metals/semi-metals located at the foot of groups IliA, IVA and VA have several features in common: large metallic radii and low surface energies (see table 1). Several of these adsorbates, while forming surface alloys at low coverage on Cu{100} exhibit a tendancy to undergo "de-alloying" transitions. Dealloying refers to a reversal of the surface alloy formation process and consists of either a gradual or abrupt reduction of the concentration of adsorbate located in substitutional sites in the first substrate layer and formation of an overlayer structure above an unreconstructed substrate. Two systems have been studied in some detail: Cu { 100}/Pb and Cu { 100}/Bi.

334

5.1. The Cu{100}/Pb System Lead and copper, a bimetallic combination which are bulk immiscible has been extensively studied [103-105]. Auger spectroscopy indicates that lead grows in a Stranski-Krastanov mode with crystallites forming upon completion of the first Pb layer. Three ordered structures form at 300 K : c(4x4), c(2x2) and a (5 ~ x ~)R45 ~ structures at Pb coverages of 0.375, 0.50 and 0.60 ML respectively. The Cu {100}/Pb system has also been studied extensively by atom scattering [106-109]. The two higher coverage structures have been studied by LEED I(V) analysis and have been identified as overlayers [110,111,112]. The c(2x2), which corresponds to a Pb coverage of 0.50 ML, consists of a Pb overlayer with the Pb adsorbate occupying four-fold hollow sites with respect to the unreconstructed substrate [111]. The c(4x4) lower coverage structure was originally suggested by Sepulveda and Rhead to consist of chains of Pb atoms above the Cu{100} surface [104]. The transition from the c(4x4) to c(2x2) phases occurs via a first order transition with nucleation of domains of the more dense c(2x2) phase within the c(4x4) overlayer. Gauthier and co-workers have examined the low coverage c(4x4) structure by LEED I(V) analysis. Auger spectroscopy was used to determine the Pb coverage (0.375 ML), indicating that the c(4x4) unit cell contains 6 Pb atoms. Overlayer models, including the Sepulveda and Rhead chain model could be ruled out in favour of a surface alloy in which Pb chains were embedded in the outermost Cu layer which itself adopts a missing row geometry with a 100% decrease in atomic density [ 113]. Figure 13 illustrates the favoured structure. Table 4 summarises the geometric parameters obtained from the LEED analysis of the c(4x4)-Pb structure [113] along with the higher coverage dealloyed c(2x2) phase [111 ]. Chains of Pb atoms occupy Cu missing rows with four Pb atoms occupying the space originally taken up by five Cu atoms, leading to a Pb-Pb distance of 3.4• A (2.8% smaller than in bulk Pb). The Pb atoms occupy two distinct adsorption sites with one in three occupying four-fold hollow sites with respect to the underlying copper layer. These atoms are located in bridge sites with respect to top layer Cu atoms. An unexpectedly small corrugation amplitude of 0.05 A along the Pb chains was obtained, with the Pb atoms located in bridge sites buckled upwards. The bimetallic layer itself is strongly corrugated with the Pb atoms buckled outwards with respect to the Cu chains by 0.66 A. The strain induced by substitution of Pb into the outermost Cu layer leads to a large distortion within the Cu top layer chains: half of the Cu atoms are shifted laterally by 0.23+0.15 A away from second layer four-fold hollow sites. The Cu interlayer spacings remain rather close to bulk Cu values if the centre-of-mass of the buckled Cu layers is considered.

335

A I.........................

++

dz12 dz23

9Pb

(~

9Culayer I

O

. Culayers2 &3

~

9pb

Figure 13. Top and side views of the Cu{ 100}-c(4x4)-Pb surface alloy. The top view (above) indicates the two in-equivalent Cu atoms in the outermost layer (1 and 2) and two inequivalent Pb sites (1' and 2' ). The side view (below) defines the major geometric parameters. Re-drawn from [113]. A1 represents the Pb-Cu top layer buckling quoted with respect to the centre of mass of the Pb and Cu rows. Nagl et al. [ 114] have probed the detailed mechanism of formation of the c(4x4) surface alloy. STM studies indicate that at extreemly low coverages (0.03 ML), Pb embeddes in the Cu{ 100} surface near step edges on both the upper and lower terraces. As the coverage is raised, copper islands are identified with mono-atomic step heights due to condensation of ejected Cu atoms in-to two-dimensional islands. An enhanced mobility of Pb is observed as disorder in room temperature STM images prior to formation of the c(4x4) LEED superstructure where atomically resolved images could be obtained. Figure 14 illustrates an STM image of the c(4x4) along with a schematic model of the structure originally suggested based on STM observations [ 114]. Nagl et al. were able to define the Pb adsorption sites via comparison with STM images of domains of the higher coverage c(2x2) (where Pb atoms where

W

W Q\

Table 4 Summary of structural parameters of alloyed and de-alloyed phases for Cu{ lOO}/Bi and Cu{ 100)Pb. The geometric parameters for

the low coverage Pb and Bi overlayers are defined in figures 13 and 15, respectively. For the high coverage ~(2x2)structures dCu-Bi(Pb) represents the perpendicular spacing between Bi(Pb) and the outer Cu monolayer with layer spacing being quoted with respect to the centre of mass of Cu layers.

dCu-Bi(Pb)

System

&I2

p(2x2)-Bi (8~,=0.25ML)

SATLEED

NIA

1.77*0.05 (-2.W2.8%)

low coverage-Bi (8~i=0.3ML)

SXRD

N/A

1.87k0.06 (+3.5*3.3%)

c(2x2)-Bi (8si=0.50ML)

SATLEED

2.1 7k0.03

1.82+0.03 (+O. 75 1.7%)

c(2x2)-Bi (8~,=0.50ML)

SXRD

2.18k0.08

1.78 (-1.5%)

~(4~4)-Pb (8pb=0.375ML)

LEED

NIA

1.8W0.09

~(2~2)-Pb (8pb=O.SOML)

LEED

2.2w0.04

1.81

dZ23 -

dz34

AI -

A2 -

A'

Ref

1.84+0.05 (+1.852.8%)

1.83k0.05 (+1.3k2.8%)

0.56 k0.05

N/A

0.05 50.05

[126]

0.61 50. 10

N/A

-

[125]

N/A

0.02 k0.03

N/A

[I261

N/A

[125]

-

[I131

N/A

[lll]

1.8W0.03 (-0.451.7%)

1.84k0.03 (+1.8* 1.7%)

NIA

1.81+0.09

1.81k0.06

0.66 50.12 N/A

N/A

337

know via LEED I(V) analysis to occupy four-fold hollow sites) by imaging at coverages between 0.375 and 0.50 where domains of c(4x4) and c(2x2) co-exist. The STM based model is highly similar to that deduced subsequently by LEED I(V) analysis with the exception that the comer atoms of the c(4x4) unit cell within the LEED analysis correspond to Pb atoms in hollow sites with respect to second layer Cu atoms rather than the bridge sites incorrectly favoured by STM. Robert et al. imaged low coverages of Pb (0.06-0.08 ML) deposited both at room temperature and low temperature (160 K) finding the morphology drastically different. The low temperature deposit nucleates in dense phase at step edges and can be imaged due to the larger metallic radius of Pb, while a lattice gas is imaged for the room temperature deposit [ 115]. The mobility of the low temperature deposit was considerably higher than that adsorbed at room temperature, an observation which could only be rationalised if Pb occupies differing sites when deposited at differing temperature. Effective diffusion activation energies of 0.2-0.4 eV and 0.68 eV were determined for low and room temperature layers respectively. Robert et al. also provided evidence for ejection of Cu in and out of the substrate surface layer: addition of 0.375 ML of Pb at low temperature followed by annealing to room temperature led to formation of a high density of steps of height 1.8 !k. A c(4x4) periodicity was imaged both on upper and lower terraces, the upper terraces coveting 52+5% of the surface, due to half the surface being covered by expelled Cu islands which themselves mix with Pb to form c(4x4) alloy. Tan and co-workers have studied the stabilty of the c(4x4) and c(2x2) phases using Monte-Carlo simulations with Lennard-Jones potentials confirming that a de-alloying transition occurs between 0.375 and 0.50 ML [117]. Within the surface alloy model the outermost mixed layer was found to be strongly buckled with Pb atoms outermost by about 0.8 .A. compared with the LEED value of 0.66 A. A modulation of the top layer Cu chains was also detected in agreement with experiment. The distance between neighbouring Pb atoms was found to be bi-modal with values of 3.08 and 3.22 A compared to the experimental value of 3.4+0.15 A by LEED [113] and 3.3+0.15 A by STM [115]. The c(4x4) surface alloy at 0pb=0.375 ML was the most stable structure with a binding energy of 1.57 eV per Pb atom compared to 1.00 eV for the c(2x2) overlayer (0ab=0.50 ML) at room temperature. In the c(2x2) phase an interplanar separation of 2.3 A between the outermost Pb c(2x2) layer the the underlying Cu substrate was found, highly similar to the value of 2.29A favoured by LEED I(V) analysis [ 111 ].

338

(b)

Q

Pb

Cu (second layer)

Figure 14. (a) A 10xl0 a m 2 atomic resolution STM image of the Cu{100}-c(4x4)-Pb structure. The c(4x4) unit cell is indicated; (b) schematic model (redrawn from [ 114]) of the c(4x4) superstructure suggesting that the comer atoms which appear brighter in the STM image are Pb atoms in bridge sites (the Cu atoms within the outermost layer are not shown for clarity). Recent low energy electron microscopy (LEEM) studies, while confirming that the Cu coverage in the c(4x4) surface alloy is 0.5 ML, have reported that during the proposed de-alloying transition , the amount of Cu

339

displaced is only 0.22 ML (about half that expected), implying that some Cu may remain mixed in the outermost c(2x2) Pb layer [ 118,119]. A more detailed discussion of the LEEM work on the Cu{ 100}/Pb system can be found in the accompanying chapter by G.Kellogg.

5.2 De-Alloying in the Cu{100}/Bi System The adsorption of Bi on Cu { 100 } was initially studied in pioneering work of Rhead and co-workers [120,121]. Delamare and Rhead [120] reported a number of ordered LEED superstructures including a p(2x2) around 0.25 ML and c(2x2) at 0.50 ML along with two phases at higher coverage which were initially identified as c(9~x~-2)R45 ~ and a (~-41xV~l), the latter being also described as c(10xl0). The p(2x2) could not be reproduced in a later investigation by the same group [ 121 ]. The Cu { 100}-c(2x2)-Bi phase has been reported to restructure in-to a Cu{210} p ( l x l ) Bi if defects in the form of vacancies are present in the substrate [122]. Bismuth adsorption leads to a work function decrease of <0.4eV, ruling out significant charge transfer and is consistent with small shifts in the Bi 4f7/2 XPS core lines [123,124]. Detailed structural characterisation was carried out by Meyerheim et al. using surface Xray diffraction (SXRD). A de-alloying transition was reported at Bi coverages >0.33 ML [125]. Mayerheim studied a range of Bi coverages, including 0.10,0.26,0.33,0.43,0.49 and 0.56 ML. At coverages below 0.35 ML, Bi atoms were found to substitute into the outermost Cu layer to form a CuxBit.x monolayer with Bi atoms avoiding occupation of nearest neighbour substitutional sites due to large elastic strain in such configuration. The criteria for avoidance of nearest neighbour sites naturally leads to introduction of some short range order as the Bi coverage is raised. Measurement of the integrated (1/2 1/2 0) X-ray reflection indicated that while weak order begins to develop at Bi coverages between 0.1 and 0.33 M L , an extreemly rapid increase is seen between 0.33 and 0.50 ML as de-alloying proceeds and a c(2x2) overlayer begins to form. SXRD indicates the Cu{ 100}-c(2x2)-Bi structure to be a dealloyed overlayer [ 125]. The surface geometry obtained by SXRD for both the alloyed and de-alloyed phases are summarised in table 4. The same system has been recently studied by A1Shamaileh and Barnes using symmetrised automated tensor LEED (SATLEED) [126]. In this study, a very weak diffiase p(2x2) LEED pattern was observed at a coverage of around 0.25 ML. It was discovered that slight increase in temperature led to dissapearance of the p(2x2) LEED beams, indicating the order-disorder transition temperature of this phase to be rather close to 300 K, perhaps explaining the difficulty in observing this phase in some studies. At room temperature the superlattice spots were exteemly weak and diffuse indicating the phase to possess very poor long range order. Figure 15 illustrates a schematic model of the Cu { 100}-p(2x2)-Bi surface alloy.

340 m

[011]

[001]

[010] p(2x2) unit cell

[011]

A1 dZl2

A3

dz23 --

Key: ~ = Bi ; ( ~ = top layer Cu ; (....i~ second layer Cu ; ~ , - third layer Cu.

Figure 15. Schematic model of the Cu{100}-p(2x2)-Bi surface alloy. The top view (above) indicates the two symmetrically inequivalent Cu atoms (1 and 2) in the p(2x2) unit cell. The side view defines the major geometric parameters.

LEED I(V) spectra were collected from both the partially ordered p(2x2) phase and the Cu{100}-c(2x2)-Bi at a coverage of 0.50 ML. The geometric structures obtained are summarised in table 4. In full agreement with the SXRD study, a surface alloy was the clearly favoured geometry at 0.25 ML transforming to a Cu {100}-c(2x2)-Bi overlayer at 0.50 ML, confirming the dealloying transition proposed by Mayerheim. As LEED has enhanced sensitivity to the substrate geometry in terms of interlayer spacings and Bi induced buckling within the substrate, it provides complimentary crystallographic

341

information to SXRD. Agreement in the major aspects of the geometry including the Bi-Cu nearest neighbour distance and perpendicular height of Bi above the copper surface is excellent.

5.3. Surface Alloy Formation in the Cu{100}/In and Sn Systems Adsorption of Sn has been extensively charactcriscd by AES and LEED [128-130]. The adsorption at Sn at 300 K follows a Stranski-Krastanov growth mode with four ordered phases being detected by LEED with increasing Sn coverage. The Sn phases include a split p(2x2) , p(2x6) , p(3~2xV2)R45~ and finally a p(2V2x2~f2)R45~ monolaycr phase. Rhcad and co-workers proposed ovcrlaycr models for the three higher coverage phases for which surface coverages of 0.42, 0.50 and 0.625 ML were assigned. However, Argilc andRhcad [129] and Abel ct al. [131] observed that adsorption at substratc temperature below 200 K inhibited the formation of the ordered phases. Annealing to 250-350 K led to their irreversible formation. Brccman and Bocrma have detected significant surface mobility for In on Cu {100} already at temperatures as low as 80 K for In deposited at low temperatures [132]. This observation suggests that surface alloy formation/de-alloying plays a significant role in the formation of the observed ordered phases in the Cu {100}/Sn system. Further evidence was provided by Abel ct al. [131] who have reported Sninduced increases in the Cu surface RBS peak at submonolaycr coverages, a result inconsistent with simple overlaycr growth. The authors suggest that Cu atoms arc displaced from their regular lattice sites with each Sn atom displacing one copper. More recently, McLoughlin ct al. have suggested alternative models for the first three ordered phases based on double scattering LEED simulations [133]. For the lowest coverage ordered phase, a model consisting of a p(2x2) ovcrlaycr with a periodic array of light anti-phase domain boundaries yielded good agreement with LEED observations. While models with Sn atoms in both ovcrlaycr and surface alloy configurations were considered, it proved difficult to differentiate between these possibilities via the double scattering pattern simulation approach adopted. Ongoing SATLEED/STM investigations of the Cu{ 100}/Sn are required to clarify the detailed structural transitions occuring [134]. However, the observed temperature dependence of the ordered phase formation makes it extrcemly likely that one or more of the Cu {100}/Sn phases involve surface alloy formation. In the case of Cu{ 100}/In, much work has been performed in the low coverage limit. Brccman and Boerma have used time-of-flight LEISS with 6 kcV Nc + ions as probes to investigate the In adsorption site as a function of temperature of a stepped Cu sample (Cu {17,1,1 }=8.5 {100} x {100 }) at an In coverage of 0.013 ML. At such low coverages, all In atoms deposited are alloyed with 92% on terrace sites and the remaining 8% embedded at step edges

342

[132]. The In atoms within the {100} terraces are buckled outwards by 0.45+0.05 A relative to top layer Cu atoms, while In atoms incorporated in step edges are also buckled outwards by 0.40+0.05 A with only a small movement away from Cu atoms within the step in the surface plane of <0.05 A. Experimental work by Klas and co-workers using perturbed angular correlation (PAC) measurements from radio-active In atoms at low coverage (0.07 ML) also favoured surface alloy formation [135]. These results contrast with calculations of Li et al. in which overlayer sites are found to be energetically preffered to substitutional inclusion [ 136]. Whether surface alloy formation persists up-to higher coverages is less well defined. Nakagawa et al. recently reported a range of complex coverage and temperature dependent structural transitions [137]. A reversible phase transformation from a c(2x2) high temperature (HT) phase to a low temperature (LT) (9~x2~)R45 ~ phase at 350-400 K occurs at a reported In coverage of 1.0 ML. Raising the coverage to 1.3 ML leads to transformation to a c(4x4) phase prior to completion of the first atomic layer, after which In crystallites form. The c(2x2)-HT phase has been suggested to correspond to a surface alloy consisting of two adjacent mixed c(2x2) Culn alloyed layers [137]. The large strain energy associated with substitution of 0.5ML of In in-to two adjacent Cu layers does not favour formation of such a double-layer alloy. Recent work by McEvoy et al. using STM has provided evidence that at intermediate In coverages (0.2 ML) surface alloying occurs without the development of long range order [ 138]. STM images demonstrate the appearance of approximately circular islands with height equal to the Cu { 100} layer spacing of 1.8 A upon In adsorption at room temperature, an observation typical of systems undergoing surface alloy formation due to island formation by ejected Cu atoms. The AES/LEED calibration of McEvoy et al. differs in interpretation of the absolute surface coverages at which the c(2x2)/(9~x~/2)R45 ~ and c(4x4) phases are observed. McEvoy et al have assigned absolute coverages of 0.5 and 0.6 ML to the c(2x2)HT and c(4x4) phases respectively. The question as to whether the c(2x2)-HT phase consists of an ordered overlayer or a surface alloy is answered by SATLEED analysis which indicates that both the c(2x2)-HT and the LT structures are overlayers [138]. Thus, a de-alloying transition occurs between 0.2 and 0.5 ML. Adsorption of Sb on Cu { 111 } and Ag { 111 } and Pb on Ni {111 } leads to formation of (~--3x ~3)R30 ~ surface alloys at adsorbate coverage of 0.33 ML [41,42]. In these systems, the nearest neighbour substrate-adsorbate bond lengths are contracted by between 0.14 and 0.43 A relative to the sum of metallic radii. The nearest neighbour bond lengths are much closer to the sum of the covalent radii of substrate and adsorbate, suggesting a significant degree of covalency in the bonding. In the case of semi-metal adsorption on Cu {100}, the experimentally determined nearest neighbour distances are 2.9 A in the

343

Cu{100}-c(4x4)-Pb structure and 2.62 A in the Cu{100}-p(2x2)-Bi surface alloy. These values compare with the sums of metallic radii of 3.03 and 2.98 A respectively. In the case of In, the nearest neighbour In-Cu distance is 2.59 A which compares with the sum of metallic radii of 2.94 A. Nearest neighbour bond lengths are thus contracted by 0.13, 0.35 and 0.36 A for Pb, In and Bi respectively. Thus, the trend observed on f.c.c.{111 } surfaces is repeated for semi-metal adsorption on Cu {100}. 5.4. De-Alloying Transitions For Transition Metal Adsorbates De-alloying has also been reported for transition metal adsorbates possessing considerably larger metallic radii than copper. Two cases have been documented to date: Au and Ag. In the Cu {100}/Au system, as the Au coverage is raised beyond 0.5 ML a de-alloying transition has been reported in which Au atoms, originally located within the outermost copper monolayer in a c(2x2) array (see section 2) dealloy. This was originally reported by Hansen and co-workers [52] and later confirmed by angle-scanned photo-electron diffraction by Naumovic et al. [139,140] and by Hansen et al. [54] using electron spectroscopy including ARUPS and surface core level shift spectroscopy (SCLSS). As the Au coverage is increased from 0.5 to 1 ML, the Au embedded in the outermost Cu layer is gradually displaced above the surface combining with additional Au to form a close-packed hexagonal monolayer. The transition appears to occur via coexistence of areas of c(2x2) CuAu surface alloy and de-alloyed close-packed Au overlayer islands. The de-alloying is fingerprinted by a transition from Au dstates in ARUPS which exhibit little dispersion as a function of kll for the c(2x2) CuAu surface alloy to more strongly dispersing bands at 1 ML coverage as Au nearest neighbour bonds form within the two-dimensional layer [52]. A significant shift in the shallow Au 4f7/2 core level also occurs upon the transition from c(2x2) to the hexagonal overlayer [52,54]. Shen et al. confirmed the dealloying transition: at 1 ML Au coverage, the Cu LEISS intensity is reduced to 10-15% of that of clean Cu{100} indicative of a surface almost completely covered by Au [141]. Shadowing dips in azimuthal scans of scattered ion intensity from Au yield a symmetry of a {111 } type h.c.p overlayer for the dealloyed layer [141]. Reports of formation of a minimum of three layers of Cu3Au type alloy have also been reported upon adsorption of 1ML of Au on Cu(100} [142]. Spunger et al. have used variable temperature STM to examine the growth of Ag on Cu{100} in the submonolayer coverage regime. A substitutional surface alloy is formed upon deposition at room temperature up-to a critical coverage of 0.13 ML, after which de-alloying occurs via condensation into c(10x2) patches consisting of Ag islands embedded within the Cu {100} surface [143]. Figure 16 illustrates STM images for a low (0.07 ML) coverage of Ag

344

Figure 16. STM images of Cu {100} with a low coverage (0.07 ML) of Ag adsorbed at 440 K and imaged upon cooling to 180 K: (a) a 125x125 A2 image showing Ag atoms imaged as protrusions; (b) a 60x60 A2 image with a Cu(lxl) grid illustrating that Ag substitutes in-to top layer Cu lattice sites [ 143]. deposited and annealed to 440 K. As shown by the grid in figure 16(b), Ag is imaged as protrusions in ( l x l ) lattice sites indicating that Ag forms a disordered substitutional surface alloy. The imaged height of the protrusions is approximately 0.25 A. A pair correlation function constructed from STM images indicates that there are significantly less Ag atoms occupying nearest neighbour sites than expected based on a purely random distribution, indicative of a short range repulsion between Ag atoms along the [011 ] and [011 ] directions due to the overlapping local strain fields. However, there is no enhanced second nearest-neighbour Ag-Ag site occupation expected for c(2x2) alloy formation such as that observed in the Cu{ 100}-c(2x2)-Pd surface alloy [44,45]. At higher Ag coverages, a fraction of the Ag segregates into a c(10x2) phase embedded within the Cu surface co-existing with areas of dilute random substitutional alloy with local Ag coverage of 0.13 ML as shown in figure 17.

Figure 17. STM images of Cu {100} with 0.4 ML Ag adsorbed at 425 K and imaged at 170 K: (a) a 540x540/It 2 image illustrating a mono-atomic step edge running diagonally across the top comer and anisotropically shaped patches oriented along high symmetry directions; (b) 120x120/~2 image showing local c(10x2) Ag patches co-existing with p(lxl) alloy regions ; (c) a 58x58 A2 image showing the hexagonal c(10x2) superstructure (left) with a Ag{111} grid and an area of Ag/Cu alloy on the right [143].

345

Effective medium calculations have confirmed that in the low coverage limit a random substitutional alloy is the lowest energy configuration by 0.19 eV per atom compared to islands of hexagonal Ag overlayer [143]. Calculations also indicate that a surface alloy has a lower energy than Ag substituted into second or deeper layers, hence Cu {100}/Ag is an example of a surface confined alloy. It is suggested that compressive stress and effective repulsions between neighbouring Ag atoms due to overlaping strain fields drives the de-alloying transition at a critical coverage of 0.13 ML. 6. UNDERLAYER 2D ALLOYS AND OVERLAYER TO UNDERLAYER TRANSITIONS A few cases have recently been discovered of ordered two-dimensional alloy underlayers consisting of a mixed alloy second layer capped by a copper monolayer.

6.1 The Cu{100}-c(2x2)-Pd Overlayer to Underlayer Transition The best documented case of underlayer alloy formation is the Cu {100}c(2x2)-Pd surface alloy which undergoes an irreversible transition to a structure consisting of a mixed c(2x2) CuPd second layer capped by a Cu monolayer illustrated in figure 18. This transition was first suggested by Graham et al. [47], who noted a rapid disappearance of the Pd He + LEISS signal at upon annealing at 440 K while a well ordered c(2x2) LEED pattern was retained. Subsequently Andersen et al. [144] using Auger electron spectroscopy (AES), LEED, work function changes, positron anhililation measurements and desorption of a probe molecule (CO) confirmed the temperature induced overlayer to underlayer transition and followed the kinetics of the transition at 353 K, finding the transition to be complete over a time scale of approximately 1 hour. Koyman et al. [ 145] using positron annihilation induced Auger spectroscopy studied the transition at the higher temperature of 423 K. The positron induced Pd Auger features disappeared very rapidly, implying rapid dissapearance of Pd to sub-surface sites at 423 K over a time scale of minutes. In the most recent study by Barnes et al., the kinetics of the transition was studied at a range of temperatures from 340 to 380 K [146]. These temperat~are dependent measurements allowed an approximate activation energy for the overlayer to underlayer transition of 1.13+0.12 eV to be deduced. The derived activation energy is higher than that determined for alloying of Pd into the outermost layer (0.88 eV), yet considerably lower than that for bulk interdiffusion determined by Goupper et al. of 2.1 eV by X-ray diffraction from bulk CuPd alloys [ 148].

346 m

[011]

[011]

A2

dz12 dz23 dz34

A4

Figure 18. Schematic model of a Cu {100}-c(2x2) underlayer ordered alloy including top and side views along the [011] azimuth. The side view defines the major geometric parameters. Open circles indicate top, third and deeper layer Cu atoms, filled circles indicate second layer adsorbate and hatched atoms are second layer Cu.

Recent study of the kinetics for a top layer to underlayer transition for 0.5 ML of Pd on a vicinal Cu { 1,1,11 } surface consisting of { 100} terraces of five unit cells width using He scattering, determined an activation energy barrier of 1.0• eV in good agreement with the value determined for the fiat surface [148]. The mechanism by which the transition occurs is presently unknown. However, as a sharp c(2x2) LEED pattern was retained at all times, random exchange of top layer Pd atoms with second layer Cu atoms may be ruled out. It would appear likely that the exchange occurs via a nucleation and growth mechanism. An STM study would throw further light on the microscopic mechanism by which this transition occurs.

347

The geometric structure of the Cu{ 100}-c(2x2)-Pd underlayer alloy has been determined by SATLEED [ 146]. A top layer surface alloy may be clearly ruled out as it results in significantly worse theory-experiment agreement compared to the underlayer alloy model. The surface geometric structure is summarised in table 5. Substitution of approximately 0.5 ML of Pd in subsurface sites leads to sizeable Pd-induced expansions in the outermost two interlayer spacings. However, the Cu(2)-Cu(4) separation of 3.58 A, represents only a 0.08% contraction relative to the sum of Pd and Cu metallic radii. It should be stressed that only idealised models : namely a second layer with ideal CuPd stoichiometry and all remaining layers being pure copper for the underlayer model have been rigorously tested. It is likely that the real situation is more complex, with small amounts of Pd being located in the outermost and third and deeper layers with consequent imperfections due to Pd defficiency in layer 2.

6.2 Cu{100}/Pt: The Cu{100}-c(2x2)-Pt Underlayer Alloy Despite the similarities between Pt and Pd which both adopt f.r structures with almost identical metallic radii (see table 1), the room temperature growth mechanism of the two metals is quite different. While Pd forms a well ordered r top layer alloy as outlined in section 2, Pt growth involves cluster formation (along with some alloying) [47]. Annealing Pt films of coverage 0.5 ML to 550 K and re-cooling leads to transformation from a p(lxl) LEED pattern with high background with weak diffuse c(2x2) beams to a c(2x2) with sharp half order reflexes. Shen et al. have determined that a 0.5 ML Pt film annealed to 453 K for 10 minutes has 38 at% Pt in the outermost layer and 10 at% Pt in layer 2 [149]. In contrast, Graham and co-workers have reported an essentially Cu-terminated surface at this coverage using He + LEISS after annealing to 525 K [47]. The different top layer compositions is presumably due to the differing thermal activation temperatures utilised. Reilly et al. [150], have studied the formation kinetics of the c(2x2) by deposition of Pt at room temperature and monitoring the intensity and full-width-at-halfmaximum of (1,0) and (1/2,1/2) LEED beams as a function of temperature and time, concluding that annealing to 550 K for 30 seconds produces the optimal ordering of the c(2x2) phase. Using CO as a probe of surface Pt concentration, Reilly et al. reported a drastic reduction in saturation CO uptake relative to room temperature deposited films after annealing to 550 K. This was interpretated as due to transfer of the majority of Pt atoms to sub-surface sites. Reilly et al. suggested the c(2x2) structure formed is due to formation of a Cu {100}-c(2x2)Pt underlayer alloy with an almost pure copper layer outermost. This was subsequently confirmed by A1Shamaileh et al. using SATLEED [151]. Two surface alloy models were considered with the mixed CuPt layer located in either layer 1 (surface alloy) or layer 2 (underlayer alloy). Significantly better

348 agreement was obtained for the sub-surface alloy compared to the surface alloy. The geometric parameters for the Cu{100}-c(2x2)-Pt underlayer alloy are summarised in table 5. It should again be remembered that the geometry of the Cu { 100}-c(2x2)Pt structure does not correspond to a complete structural determination which requires that the layerwise Pt concentration is varied. While it is unlikely that the major aspects of the surface compositional profile and geometry will change, allowing the layerwise Pt concentration to vary within the outermost few atomic layers may lead to small changes in interlayer spacings and bucklings to those presented in table 5. Similar to the Cu{100}-c(2x2)-Pd underlayer alloy, substitution of the larger Pt atom sub-surface leads to significant expansion of the outermost two copper interlayer spacings and buckling in both the mixed CuPt second layer and in layer 4. It is likely that the high surface energy of Pt drives the transition to an ordered sub-surface c(2x2) alloy despite the larger elastic strain for Pt atoms located sub-surface. Li § and He + LEISS studies on a Cu3Pt{ 100} bulk alloy consisting of alternate Cu and c(2x2) CuPt layers also indicate that a pure Cu termination above an ordered c(2x2) CuPt second layer is the preferred geometry [ 152]. Table 5 Summary of geometry of Cu{100}-c(2x2)-Pd(Pt) underlayer alloys determined by SATLEED. Positive values for buckling in layer 2 corresponds to adsorbate atoms relaxed towards the vacuum interface. Positive values for buckling in layer 4 corresponds to Cu atoms below Pd (or Pt) relaxed towards the adsorbate. The values in brackets below the interlayer spacings indicate the percentage contraction (negative) or expansion (positive) relative to the bulk value of Cu{ 100} of 1.807 A. The final entry corresponds to a Pt multilayer alloy with a minimum of two ordered c(2x2) CuPt layers within the LEED probing depth.

Adsorbate

dZl2 A

A2 A

dz.23

A4

dz34

A

A

A

Ref..

Pd

1.86• (+2.9+2.2%)

+0.07•

1.92• (+6.3+2.8%)

+0.20•

-

[ 146]

Pt

1.90• (+5.1+ 1.7%)

-0.08+0.06

1.87• (+3.5• 1.7%)

+0.03•

1.81• (+0.2+2.2%)

[151]

Pt

1.86+0.03 (+2.9•

+0.02•

2.03+0.05 (+12.3+2.8%)

+0.14+0.09 1.83• (+1.3•

[ 160]

349

A third adsorbate which may form a Cu { 100}-c(2x2) underlayer alloy is Rh. An irreversible p ( l x l ) - - ~ c(2x2) transition occurs upon annealing a Rh film of coverage around 0.75 ML to 525 K [47]. The degree of order was significantly lower than the Pd or Pt cases with feint diffuse c(2x2) superstructure spots being visible in LEED with streaks along the [011] directions. LEISS measurements demonstrated that a temperature program of 1 Ks ~ to 525 K reduces the Rh LEISS signal to 29% of that from the room temperature deposited film, indicating the majority of the Rh to penetrate subsurface [47]. In contrast to Pd and Pt, Rh is only moderately soluble in copper. The highly exothermic heats of alloy formation of-14kJ mole ~ for Pd and-10kJ mole l for CuPt compares to a small positive heat of formation for CuRh alloys. The moderate solubility combined with the tendancy not to form ordered bulk compounds in the case of Rh may lead to the poorer degree of ordering observed. Table 1 indicates that the high surface energy of Rh relative to Cu will again favour copper capping upon thermal activation. Further work is required to investigate whether the poor ordering is due to incorrect Rh loading and thermal processing conditions or is an inherent feature of the Cu{100}/Rh surface alloy.

6.3 Cu{100}/Ir: The Unusual Case of p(2xl) Underlayer Formation As the Cu{100}/Ir system has been discussed in detail in the accompanying chapter by H.Nichus, only a brief summary of the most relevant details will be given here. The adsorption of Ir on Cu { 100} at 200 K has been extensively studied by Gilarowski and Nichus using LEISS, LEED and STM [155-157]. LEISS indicates that >80% of adsorbed atoms occupy sub-surface sites up-to an Ir coverage of 1 ML [155]. Figure 19 illustrates a series of STM images with increasing Ir coverage. Defect formation occurs at extremely low Ir coverages (0.01 ML) in the form of rectangular islands of height 1.8 A and chain-like structures with apparent height of 1 A as well as point defects associated with single Ir atoms alloyed into the outermost copper monolaycr. At 0.05 ML, the number of islands increases with the rectangular islands growing in size while the chains remain of an average length of 30 A, preferentially oriented in the [011 ] directions. At this coverage around 15% of the surface is covered by islands. It was thus concluded that the imaged islands consist mainly of copper.

350

Figure 19. A series of STM images for Ir adsorption at 200 K with images recorded at 300 K: (a) clean Cu{100}; (b) 0.01 ML Ir. The insert illustrates two different island types: (A) rectangular or square islands of apparent height 1.8/l, and (B) linear chain like structures of apparent height 1 A; (c) 0.05 ML Ir. The inset illustrates rectangular/square and linear islands; (d) 0.15 ML Ir; (e) 0.3 ML Ir; (f) 1.5 ML Ir. Images (a)-(d): 1000xl000 A 2 , (e) and (f) 1250x1250 A 2. (The insets correspond to 250x250 A2scans) [ 156]. The rectangular islands with height 1.8 A are proposed to be Cu islands with Cu underneath, while the rectangular islands of smaller apparent height were proposed to correspond to chains of sub-surface Ir atoms capped by copper. As the coverage is increased, step roughening occurs along with pit formation , the pit depth being 1.8 /I,. At this coverage the rectangular and elongated islands begin to co-elesce.

351

Annealing submonolayer and monolayer films to 650 K leads to further reduction of the Ir LEISS signal with Ir concentrations in the outermost layer reduced to <1%, while XPS indicates Ir remains in the near surface region. A p(2xl) rotated domain ordered phase is observed in LEED, best formed for a Ir coverage around 0.5 to 0.6 ML for films grown at 620 K. STM images two orthogonal rotated domains of p(2xl) of average width 50 A, with an apparent height corrugation of 0.3 A, assigned to electronic differences between second layer Ir and Cu atoms [155,156]. Gilarowski and Niehus suggest formation of an ordered p(2xl) underlayer alloy consisting of chains of Ir atoms aligned in the [011 ] in one domain and [011 ] in the other with every second copper row being replaced by Ir. Iridium follows the trend for formation of underlayer alloys for elements with similar metallic radii/crystal structure to copper yet having considerably higher surface energy. However, Ir differs from the tendency for c(2x2) underlayer ordering shown by Pd and Pt. In this case, formation of onedimensionally close packed Ir rows may occur due to the higher Ir-Ir bond energy combined with the slightly smaller f.c.c, mismatch of 6.2%.

7. FORMATION OF ORDERED MULTILAYER ALLOYS An ordered multilayer alloy is a surface alloy containing two (or more) ordered bimetallic layers within the selvedge. Two well characterised examples of ordered multilayer alloy formation have been reported to date.

7.1.The Cu{100}-p(2x2)-I ML Pd Surface Alloy As the Pd coverage is raised from 0.5 ML towards 1 ML the Cu{ 100}c(2x2) surface alloy (see section 2) transforms to a more complex phase with p(2x2) periodicity with (0, n+1/2) and (m§ LEED spots missing under normal incidence conditions of the incident beam (n and m are integers). The only 2D space groups consistent with this observation are p4g and pgg. The optimal coverage at which the LEED structure obtains optimal intensity has been reported to be 1.10+0.06 ML [29] and 1.2+0.1 ML [158,159] hence we term it the "monolayer" alloy phase. While three groups have studied the surface geometry of this phase with conflicting results, one aspect of the geometry is uncontroversial: that the second layer consists of a mixed c(2x2) CuPd alloy with third and deeper layers pure copper. Figure 20 illustrates the models previously proposed. Pope et al. were the first to study this phase by MEISS combined with LEED I(V) analysis, concluding that the outermost layer was heterogeneous consisting of 80% domains of pure Pd which was clock-rotated with a p(2x2)p4g symmetry (model 20(a)) co-exising with 20% of double layer c(2x2) CuPd

352 alloy (model 20(b)) [29]. Murray et al. probed the surface at a Pd coverage of 1.3 ML with STM and failed to image a clock-rotation. A p(2x2) periodicity was revealed without significant heterogeneity, although a high concentration of defects in the form of anti-phase domain boundaries were imaged [44,45]. In order to explain these observations, Murray et al. suggested that p(2x2)-pgg periodicity/symmetry originated in layer 2 via a clock rotation of the CuPd underlayer with an outermost layer of CuPd stoichiometry but a 100% atomic density reduction compared to Cu { 100} (figure 20(c)). Shen and co-workers reexamined the structure with LEISS using both He + and Li § ions with pure Cu { 100} and Pd { 100} calibrants and were able to rule out the model of Murray et al., favouring a top layer of close to CuPd stoichiometry and atomic density equal to that of Cu{ 100}. Simulations of angle dependent LEISS experiments suggested a heterogeneous model consisting of 30% clock rotated p(2x2)-p4g { 100} Pd (model 20(a)) and 70% unreconstructed p(2x2) Cu3Pd alloy (model 20(d)) [ 159].

Figure 20. Schematic models of suggested structures for the Cu {100}/Pd monolayer alloy: (a) a p(2x2) clock rotated Pd monolayer above a c(2x2) CuPd second layer; (b) a double layer c(2x2) CuPd alloy; (c) a p(2x2) CuPd overlayer with 100% decrease in atomic density above a clock rotated CuPd underlayer; (d) a p(2x2) Cu3Pd outermost layer above a c(2x2) CuPd underlayer. The surface unit cell is shown in each case by dotted lines [ 160].

353

Pussi et al. applied SATLEED in an attempt to confirm which of these models corrsponded to the correct description of the surface geometry [160]. The model of Murray et al. [45] yielded poor theory-experiment agreement. The model of Pope et al. gave best agreement for a mixed domain model of 10% p4g Pd and 90% double layer c(2x2) corresponding to a Pd coverage of 1.05 ML and a top layer Pd coverage of 0.55 ML. The model of Shen et al. yielded better agreement for a domain mix of 20% p4g Pd and 80% Cu3Pd, correponding to a Pd coverage of 0.9 ML, while a model in which the percentage of Cu3Pd was reduced to 60% gives a higher reliability-factor. The large area of p(2x2) surface would necessarily scatter significant intensity into the "systematic absences", contradicting experimental observation of very small or zero intensities. A new model consisting of a c(2x2) CuPd double layer with the outermost layer clock rotated to form p(2x2) periodicity and a p2gg space group is illustrated in figure 21 [160]. Quartets of atoms consisting of two Cu and two Pd atoms clock rotate about second layer Pd sites as shown in figure 21 leading to an enhanced LEED theory-experiment agreement yielding the structural parameters listed in table 6 and corresponding to a Pendry R-factor of 0.24.

Figure 21. The Cu{100}-p(2x2)-p2gg monolayer alloy: (a) top view illustrating the two orthogonal mirror planes (m) and glide lines (g) with the structure consisting of a double layer c(2x2) CuPd alloy with the p(2x2) periodicity introduced by clock rotation of the outermost CuPd monolayer; (b) side view illustrating the major geometric parameters; (c) motion of top layer Pd and Cu atoms within the "clock rotation" [ 160].

354 The large expansions of the first two Cu interlayer spacings indicated in table 6 are due to high Pd concentration in both layers 1 and 2 to relieve the lattice strain due to the 7.7% larger metallic radius of Pd. A surprisingly good agreement for the degree of displacement of top layer atoms from four-fold hollow sites is obtained in all structural studies, despite the differing nature of the proposed models. The double layer c(2x2) CuPd alloy combined with a top layer clock rotation suggested by the SATLEED analysis of Pussi et al is consistent both with the favoured top layer CuPd composition determined by LEISS and the overall Pd loading. However, it should be noted that LEISS studies also indicate the presence of a portion of the adsorbed Pd atoms in nearest neighbour sites, inconsistent with the model of Pussi et al. Furthermore, STM indicates that a Pd coverage of 1.3ML, a homogeneous p(2x2) top layer, without indication of top layer clock rotation is imaged after gentle annealing to 350K. In contrast, at a Pd coverage of 1.1ML, large areas of c(2x2) periodicity exist in the outermost layer with a heterogeneous surface being imaged [44,45]. Clearly the detailed morphology and structure of this system is highly dependent on Pd coverage and Table 6. Summary of the structural parameters extracted by SATLEED for the Cu{100}p(2x2)-pgg-Pd monolayer alloy. Structural parameters are defined in figure 21. The lateral shift of top layer Pd (and Cu) atoms from four-fold hollow sites is defined as ol. Separation between first and second layer Cu and Pd atoms are also included. The interlayer spacings dz12 and dz23 for the work of Pope et al. correspond to a model with pure Pd outermost above a c(2x2) CuPd alloy, hence the distances are not strictly comparable to the SATLEED counterparts.. Structural

Optimal value

Shenet al. [159] Pope et al. [29]

Atom pair

Spacing A

Parameter

A

A

A

~1

0.25+0.12

0.25+0.07

0.28

A-A

3.11/3.61

dz12

1.934-0.02

2.08

B-B

4.11/3.61

dz23

1.90-a:0.06

1.84

A-B

2.58

dz34

1.80-a:0.03

B-C

2.67

A1

0.06+0.04

B-D

2.51

A2(l)

0.02+0.03

Pd{100}

2.75

A2(2)

0.03+0.03

0.14

355

thermal treatment and vividly illustrates the difficulty in determining the structure and compositional profile of surface alloy systems such as Cu { 100}/Pd which exhibit a high degree of solubility of adsorbate within the bulk of the substrate. In these cases, partition of adsorbate between many substrate layers is possible with the exact surface compositional profile/morphology critically dependent on the thermal treatment undergone by the sample.

7.2. The Cu{100}-c(2x2)-Pt Multilayer Alloy A number of authors have reported that Pt adsorption in the coverage range 1 to 3 ML followed by thermal activation between 440 and 600 K leads to formation of well ordered Cu{100}-c(2x2)-Pt surface alloys [47,149,153]. Graham et al. using LEISS concluded that as the Pt loading is raised, an increasing concentration of Pt remains in the outermost layer after thermal activation to 525 K. For example, at a Pt coverages of 1.5 ML approximately 50% of the Cu{100} surface is Pt after deposition at 300K, which upon annealing reduces to around 25%. At a higher Pt coverage of just over 3 ML, approximately 50% of the surface remains Pt after annealing, interpreted as a mixed c(2x2) CuPt layer outermost. This interpretation has been supported by Belkou et al. using SCLS spectroscopy from the shallow Pt 4f7/2 core line who reported that a film of 3 ML annealed at a higher temperature of 573 K and for 30 minutes produced a sharp low background c(2x2) LEED pattem on cooling [153]. A fitting to the SCLS spectra utilising Pt SCLS data from a Cu3Pt {100} bulk alloy as reference led the authors to conclude a multilayer Cu3Pt{ 100} alloy had been formed with a mixed CuPt termination. A thermally equilibriated Cu3Pt{100} bulk alloy is terminated by a pure Cu rather than a mixed CuPt layer [152]. It is possible that the deconvolution analysis of the Pt 4f7/2 core line based on the Cu3Pt { 100} reference is incorrect and it remains a possibility that this sample is terminated by a pure or almost pure Cu outer layer. The Cu{100}/Pt system does not appear to exhibit clock-type restructuring. A quantitative study of the layerwise composition after thermal activation was performed by Shen and co-workers using both 1 keV He + ions and 1 keV Li + ions to probe the outermost and second layer compositions respectively [149]. Thermal activation involved annealing to 450 K for 10 minutes. At a Pt coverage of 1 ML, the composition of both the outermost and second layers is close to CuPt: Pt concentrations of 46 at% and 40 at% in layer 1 and 2 respectively suggesting an ordered bilayer alloy consisting of two adjacent c(2x2) CuPt monolayers. The LEISS angular scans from both Cu and Pt are practically identical, confirming the formation of an ordered c(2x2) CuPt alloy. Annealing to higher temperature (680 K) leads to surfaces with an almost pure copper layer outermost indicating the mixed CuPt layer outermost structure to be meta-stable.

356

Thermally activated multilayer Cu{100}-c(2x2)-Pt surfaces have been probed by SATLEED [161 ]. These alloys were formed by thermal activation of room temperature deposited Cu{ 100}/Pt interfaces (0pt=l.0-1.5ML) to 550 K. A geometry consisting of alternate layers of c(2x2) CuPt alloy sandwitched between layers of pure Cu with a Cu terminated surface layer was clearly favoured. The stacking pattern favoured was similar to that found for a Cu3Pt {100} LI2 bulk alloy surface [ 152,154]. A structure consisting of a mixed CuPt c(2x2) bilayer outermost as determined by Shen et al. for Pt films of 1ML coverage could be ruled out. It appears that the milder thermal treatment of Shen (450 K) compared to that employed by Pussi et al. (550 K) may lead to a low temperature bilayer CuPt alloy with a mixed CuPt termination with higher temperature thermal processing switching the termination to a Cu-capped Cu3Pt LI2 type structure. A1Shamaileh and Barnes have noted that thermal activation of a 1ML Pt film to a temperatures corresponding to conditions utilised by Shen et al. lead to a c(2x2) surface alloy with significantly differing I(V) characteristics to those of Cu capped Cu3Pt{ 100} formed by the more aggressive thermal processing. This supports the assertion that the surface geometry and layerwise composition may be radically altered depending on the thermal processing to which the system is subjected [162]. The Cu{100}-c(2x2)-Pt multilayer surface alloy formed by thermal processing to 550K exhibits a range of interesting catalytic properties [162]. Cu{100}-c(2x2)-Pt multilayer alloys have been probed by temperature programmed reaction spectroscopy (TPRS) of a formate (HCOO) catalytic intermediate. Adsorption of formic acid on copper surfaces leads to formation of a bidentate formate intermediate bonded through two chemically equivalent oxygen atoms to two nearest neighbour copper sites with the plane of the molecular fragment close to uptight. The formate decomposes stoichiometrically to CO2 and H2 (and a small amount of molecular formic acid) in a decomposition limited kinetics around 450 K [ 150]. In contrast, on Pt surfaces the formate already decomposes below room temperature to CO. Despite the Cu {100}-c(2x2)-Pt-lML surface being Cu-terminated, significant changes occur in the formate adsorption and decomposition kinetics. The presence of Pt increases drastically the rate of formate uptake, indicative of removal of a barrier to dehyrogenation of HCOOH. The formate intermediate is de-stabilised by the presence of Pt, the decomposition kinetics becoming explosive with a reduction of the full-width-at-half maximum of the co-incident CO2 and H2 desorption from around 30 K on clean Cu {100 } to ~5 K on the Cu {100}-c(2x2)Pt multilayer alloy [ 162]. 7.3. The Cu{100}-(4x2)-p2gg-Mn Structure Deposition of approximately 1 ML of Mn at 370 K on Cu{100} leads to formation of a LEED pattern initially identified as (4x2)-p2mg [163] in which

357

the (0, n/4) and (rn/4,0) reflections are absent for odd values of n and m at normal incidence. Subsequent LEED measurements on a stepped Cu{100} surface, allowing preferential growth of only one of the two possible degenerate (4x2) rectangular rotated domains also exhibits a LEED pattern with two perpendicular glide lines in the [011 ] and [011 ] directions and suggested a reassignement of the two-dimensional space group to either cmm2 or p2gg. STM studies of the surface, illustrated in figure 22, were unable to image any mirror planes, hence van der Kraan and van Kempen deduced the appropriate periodicity and symmetry to be (4x2)-p2gg [57]. Structural models were suggested based on STM images, the Mn coverage and the observed p2gg symmetry, consisting of two adjacent alloyed layers with the underlayer consisting of a c(2x2) CuMn alloy. Two possibilities were suggested for the outermost layer: (a) a top layer of composition Mn2Cu and a net atomic density of 0.75 ML with Mn atoms atop second layer Cu atoms and top layer Cu atoms located in four-fold hollow sites; (b) a top layer of composition CuMn with an atomic density of 1.0 ML with top layer Mn atoms close to atop second layer Cu atoms and top layer Cu close to atop second layer Mn [57]. In both cases top layer atoms are located on or close to atop sites, a highly unusual structure for transition metal adsorption on low index metal surfaces. These structural models are not fully consistent with valence band angleresolved photo-emmision studies [43]. In particular, the binding energy of the Mn-3d spin majority peak is not changed relative to the lower coverage Cu{100}-c(2x2)-Mn structure, suggestive of a high-spin ground state of Mn atoms in the bilayer alloy. The increased co-ordination of the sub-surface Mn led to an expectation of a collapse of the magnetic exchange splitting, which itself should lead to a decrease of the majority-minority band splitting or a strong enhancement of the Mn 3d contribution close to the Fermi energy. These effects were absent, with the surface electronic structure as measured by ARUPS being rather similar to the c(2x2) CuMn top layer alloy. Clearly further quantitative structural work on the Cu{100}-(4x2) monolayer is required to confirm the geometry of this interesting surface alloy phase.

Figure 22. STM images of the Cu{ 100}-(4x2)-p2gg monolayer Mn alloy: (a) a 5.6x5.6 nm2 image illustrating the (4x2) unit cell as white rectangles; (b) the same area atter a tip change with the sample rotated by 15~

358

8. CONCLUSIONS This chapter provides a picture of the widespread nature and complexity of surface alloy formation on Cu {100}. A number of patterns have emerged : (a) Surface alloy formation is not limited to bimetallic combinations which are bulk soluble: many examples of surface alloy formation with combinations (Cu/Pb, Cu/Ir etc) which are bulk immiscible occur; (b) Atoms with metallic radii significantly (> 10%) larger than Cu which form surface alloys at low coverage, undergo a de-alloying transition as the coverage is raised. In this class of system, the low coverage alloys are often truly surface localised due to the large elastic strain occuring upon substitution into second or deeper layers; (c) In the case of f.c.c adsorbates which are highly miscible with Cu and form ordered bulk compounds, the tendancy is for ordered surface alloy formation with c(2x2) periodicity at lower coverage transforming to ordered bi or multilayer alloys as the coverage is raised ; (d) Metals with intermediate mismatch in metallic radii and higher surface energy than Cu (Pd,Pt,Rh,Ir) form ordered underlayer alloys in certain cases undergoing overlayer to underlayer transitions, the driving force being lowering of surface energy. Many of the Cu {100} based surface alloys discussed in this chapter are relatively well characterised in terms of their layerwise compositional profile, geometric structure and thermal stability. However, it is clear that the majority of structural studies performed to date have made the (often necessary) assumption that a single homogeneous structural phase with a somewhat idealised compositional profile is present. In many cases, particularly for adsorbates which exhibit considerable bulk solubility in copper this may be a oversimplification. Future work to investigate the sensitivity of quantitative probes of surface structure and composition such as LEED, ion scattering spectroscopies and photo-electron diffraction to structural heterogeneity will be invaluable. Undoubtably a greater focus will be given to the role played by steps and kinks, both in the mechanism of surface alloy formation and in tailoring bimetallic surface alloys of well defined surface morphology on the nano-metre scale. Control of surface geometry and composition on the nano-scale will add considerably to flexibility in future catalyst and material design provided the engineered surfaces have sufficient thermal stability. Another major challenge for the future would be to move towards utilising Cu {100} based surface alloys more fully for reactivity studies with respect to both gas adsorption and the

359 study of kinetics and dynamics of surface reactions using the powerful methodologies available such as molecular beam scattering and high pressure reactor studies.

ACKNOWLEDGEMENTS I gratefully acknowledge the invaluable help provided by Mr Ehab A1Shamaileh during the preparation of this chapter.

REFERENCES [ 1] E.Bauer in: The Chemical Physics of Solid Surfaces, Volume 3B. Editors: D.A.King and D.P.Woodruff. Elsevier, 1984, p 1. [2] E.Bauer, Appl.Surf.Sci 11/12 (1982) 479. [3] U.Bardi, Rep.Prog.Phys.57 (1994) 939. [4] C.T.Campbell, Annu.Rev.Phys.Chem 41 (1990) 775. [5] N.V.Smith, in Photoemission in Solids I, edited by M.Cardonna and L.Ley (Springer, Berlin, 1978) p237. [6] C.G.Chinchen, P.J.Denny, J.R.Jennings, M.S.Spencer and K.C.Waugh, Appl.Catal. 36 (1988) 1. [7] S.Walter, V.Blum, L.Hammer, S.Muller, K.Heinz and M.Giesen, Surf.Sci.458 (2000) 155. [8] M.Wuttig, R.Franchy and H.Ibach, Z.Phys.B 65 (1986) 71. [9] P.Stoltze, J.Phys.:Condensed.Matter 6 (1994) 9495. [ 10] M.Breeman and D.O.Boerma, Surf.Sci.269/270 (1992) 224 [ 11] M.Breeman and D.O.Boerma, Surf.Sci. 287/288 (1993) 881. [12 ] L.Vitos, A.V.Ruban, H.LSkriver and J.Kollar, Surf.Sci 411 (1998) 186. [13] E.Teatum, K.Gschneider and J.Waber in : Compilation of calculated data useful for predicting metallurgical behaviour of elements in binary alloy systems, Los Alamos Scientific Laboratory (1960). [ 14] P.W.Palmberg and T.N.Rhodin, J.Chem.Phys. 49 (1986) 134. [15] G.W.Graham, Bull.Am.Phys.Soc.29 (1984) 222; G.W.Graham, Surf.Sci. 184 (1987) 137. [ 16] Z.Q.Wang, Y.S.Li, C.K.C.Lok, J.Quinn, F.Jona and P.Marcus, Solid State Commun. 62 (1987) 181. [ 17] J.G.Tobin, J.C.Hansen and M.K.Wagner, J.Vac.Sci.Technol. A 8 (1990) 2494. [18] D.Brown, T.C.Q.Noakes, D.P.Woodruff, P.Bailey and Y.Le Goaziou, J.Phys.: Condensed. Matter 11 (1999) 1889. [19] J.Yao, Y.G.Shen, D.J.O'Connor and B.V.King, Surf.Sci 359 (1996) 65. [20] S.M.Foiles, Surf.Sci. 191 (1987) 329. [21] D.D.Chambliss and S.Chiang, Surf.Sci.Lett 264 (1992) L187. [22] D.D.Chambliss, R.J.Wilson and S.Chiang, J.Vac.Sci.Technol. A 10 (1993) 1993. [23] J.Tersoff, Phys.Rev.Lett 74 (1995) 434. [24] G.C.Smith, C.Norris and C.Binns, Vacuum 31 (1981) 523. [25] G.W.Graham, Surf.Sci.171 (1986) L432. [26] S.C.Lu, Z.Q.Wang, S.C.Wu, C.K.C.Lok, J.Quinn, Y.S.Li, D.Tian, F.Jona and P.M.Marcus, Phys.Rev.B 37 (1988) 4296.

360 [27] S.C.Wu, S.H.Lu, Z.Q.Wang, C.K.C.Lok, J.Quinn, Y.S.Li, D.Tian, F.Jona and P.M.Marcus, Phys.Rev.B 38 (1988) 5363. [28] P.Kaukasoina, M.Lindroos, E.AIShamaileh and C.J.Barnes, work in progress. [29] T.D.Pope, M.Vos, H.T.Tang, K.Griffiths, I.V.Mitchell, P.R.Norton, W.Liu, Y.S.Li, K.A.R.Mitchell, Z.-J.Tian and J.E.Black, Surf.Sci.337 (1995) 79. [30] J.E.Black, Phys.Rev.B 46 (1992) 4292. [31] J.E.Black and Z.J.Tian, Comments Condensed Matter Phys 16 (1993) 281. [32] J.Kudmovsky, S.K.Bose and V.Drchal, Phys.Rev.Lett. 69 (1992) 308. [33] R.Tetot, J.Kudmovsky, A.Pasturel,V.Drchal and P.Weinberger, Phys.Rev.B 51 (1995) 17910. [34] G.Bozzolo, R.Ibanez-Meier and J.Ferrante, Phys.Rev.B 51 (1995) 7207. [35] J.E.Garces, H.O.Mosca and G.H.Bozzolo, Surf.Sci.459 (2000) 365 and references therein. [36] C.Binns and C.Norris, Surf.Sci.116 (1982) 338. [37] M.Wuttig, Y.Gauthier and S.Blugel, Phys.Rev.Lett 70 (1993) 3619. [38] M.Wuttig, C.C.Knight, T.Flores and Y.Gauthier, Surf.Sci. 292 (1993) 189. [39] R.Toomes, A.Theobald, R.Lindsay, T.Gieble, O.Schaff, R.Didszhun, D.P.Woodruff, A.M.Bradshaw and V.Fritzsche, J.Phys.:Condens.Matter 8 (1996) 10231. [40] S.D'Addato and P.Finetti, Surf.Sci.471 (2001) 203. [41] D.P.Woodruff, D.Brown, P.D.Quinn, T.C.Q.Noakes and P.Bailey, Nuclear Instruments and Methods in Physics Research B 183 (2001) 128. [42] D.Brown, P.D.Quinn, D.P.Woodruff, P.Bailey and T.C.Q.Noakes, Phys.Rev B 61 (2000) 7706. [43] F.Schiller, S.Danzenbacher and C.Laubschat, Surf.Sci. 482-485 (2001) 442; Y.Huttel, S.Gallego, M.C.Munoz and M.C.Asensio, Surf.Sci.482-485 (2001) 540 and references therein. [44] P.W.Murray, I.Stensgaard, E.Laegsgaard and F.Besenbacher, Phys.Rev.B 52 (1995) R14 404. [45] P.W.Murray, I.Stensgaard, E.Laegsgaard and F.Besenbacher, Surf.Sci. 365 (1996) 591. [46] C.J.Bames, E.AIShamaileh, T.Pitkanen and M.Lindroos, Surf.Sci. 482-485 (2001) 1425. [47] G.W.Graham, P.J.Schmitz and P.A.Thiel, Phys.Rev.B 41 (1990) 3353. [48] T.D.Pope, K.Griffiths and P.R.Norton, Surf.Sci. 306 (1994) 294. [49] M.Valden, J.Aaltonen, M.Pessa, M.Gleeson and C.J.Bames, Chem.Phys.Lett. 228 (1994) 519. [50] T.D.Pope, G.W.Andersen, K.Griffiths, P.R.Norton and G.W.Graham, Phys.Rev.B 44 (1991) 11918. [51] P.Heimann, J.Hermanson, H.Miosga and H.Neddermeyer, Phys.Rev.Lett 26 (1976) 1782. [52] J.C.Hansen, J.A.Benson, W.D.Clendening, M.T.McEllistrem and J.G.Tobin, Phys.Rev.B 36 (1987) 6186; B.J.Knapp, J.C.Hansen, J.A.Benson and J.G.Tobin, Surf.Sci. 188 (1987) L675. [53] J.C.Hansen and J.G.Tobin, J.Vac.Sci.Technol A7 (1989) 2475. [54] J.C.Hansen, M.K.Wagner and J.G.Tobin, Solid State Commun 72 (1989) 319. [55] F.Thielmann, R.Matzdorf and A.Goldmann, Surf.Sci. 387 (1997) 127. [56] H.P.Noh, T.Hashizume, D.Jeon, Y.Kuk, H.W.Pickering and T.Sakurai, Phys.Rev.B 50 (1994) 2735. [57] R.G.P.van der Kraan and H.van Kempen, Surf.Sci. 338 (1995) 19. [58] M.Wuttig, S.Junghans, T.Flores and S.Blugel, Phys.Rev.B 53 (1996) 7551. [59] T.Flores, S.Junghans and M.Wuttig, Surf.Sci. 371 (1997) 1.

361 [60] T.Flores, S.Junghans and M.Wuttig, Surf.Sci 371 (1997) 14. [61 ] H.Li and B.P.Tonner, Surf.Sci. 237 (1990) 141. [62] M.T.Kief and W.F.Egelhoff.Jr, Phys.Rev.B 47 (1993) 10785. [63] A.K.Schmid and J.Kirschner, Ultramicroscopy 4-44 (1992) 483. [64] J.R.Cerda, P.L.de Andres, A. Cebollada, R.Miranda, E.Navas, P.Schuster, C.M.Schneider and J.Kirschner, J.Phys.Condens.Matter 5 (1993) 2055. [65] R.Pentcheva and M.Scheffler, Phys.Rev.B 61 (2000) 2211. [66] J.Fal3bender, R.Allenspach and U.Durig, Surf.Sci. 383 (1997) L742. [67] F.Nouvertne, U.May, A.Rampe, M.Gruyters, U.Korte, R.Bemdt and G.Guntherodt, Surf.Sci. 436 (1999) L653. [68] F.Nouvertne, U.May, M.Bamming, A.Rampe, U.Korte, G.Guntherodt, R.Pentcheva and M.Scheffler, Phys.Rev.B 60 (1999) 14 382. [69] N.A.Levanov, V.S.Stepanyuk, W.Hergert, D.I.Bazhanov, P.H.Dederichs, A.Katsnelson and C.Massobrio, Phys.Rev.B 61 (2000) 2230. [70] C.G.Zimmermann, M.Yeadon, K.Nordlund, J.M.Gibson, R.S.Averback, U.Herr and K.Samwer, Phys.Rev.Lett.83 (1999) 1163. [71] S.-K.Kim, J.-S.Kim, J.Y.Han, J.M.Seo, C.K.Lee and S.C.Hong, Surf.Sci. 453 (2000) 47. [72] J.Sainio, J.Lahtinen, E.A1Shamaileh and C.J.Bames, to be published. [73] S.Blugel, Appl.Phys. A 63 (1996) 595. [74] W.F.Eglehoff.Jr., J.Vac.Sci.Technol. A7 (1989) 2060. [75] B.Hemnas, M.Karolewski, H.Tillborg, A.Nilsson and N.Martensson, Surf.Sci. 302 (1994) 64. [76] S.H. ICdm,K. S. Lee, H. G. Min, J. Seo, S.C. Hong, T.H. Rho and J. S. Kim, Phys.Rev.B 55 (1997) 7904. [77] W.Platow, U.Bovensiepen, P.Poulopoulous, M.Farle, K.Baberschke, L.Hammer, S.Walter, S.Muller and K.Heinz, Phys.Rev.B 59 (1999) 12641. [78] J.Lindner, P.Poulopoulos, F.Wilhelm, M.Farle and K.Baberschke, Phys.Rev.B 62 (2000) 10431; J.Shen, J.Giergiel and J.Kirschner, Phys.Rev.B 52 (1995) 8454. [79] M.Wuttig, B.Feldmann and T.Flores, Surf.Sci. 331-333 (1995) 659. [80] D.A.Steigerwald, I.Jacob and W.F.Egelhoff Jr., Surf.Sci. 202 (1988)472. [81 ] S.A.Chambers, T.J.Wagner and J.H.Weaver, Phys.Rev.B 36 (1987) 8992. [82] D.A.Steigerwald and W.F.Egelhoff Jr., Surf.Sci. 192 (1987) L887. [83] Th.Detzel and M.Memmel, Phys.Rev.B 49 (1994) 5599. [84] K.Kalki, D.D.Chambliss, K.E.Johnson, R.J.Wilson and S.Chiang, Phys.Rev.B 48 (1993) 18344. [85] D.D.Chambliss, R.J.Wilson and S.Chiang, J.Vac.Sci.Technol. A 10(4) (1992) 1993. [86] D.D. Chambliss, K.E.Johnson, R.J.Wilson and S.Chiang, J.Magn.Mater 121 (1993) 1. [87] K.E.Johnson, D.D.Chambliss, R.J.Wilson and S.Chiang, Surf.Sci. 313 (1994) L811. [88] J.Shen, J.Giergel, A.K.Schmid and J.Kirschner, Surf.Sci. 328 (1995) 32. [89] J.H.Kim, K.H.Lee, G.Yang, A.R.Koymen and A.H.Weiss, Appl.Surf.Sci. 173 (2001) 203. [90] R.D.Diehl and R.McGrath, Surf.Sci.Rep 23 (1996) 49. [91] H.Tochihara and S.Mizuno, Prog.Surf.Sci. 58 (1998) 1. [92] H.Tochihara and S.Mizuno, Surf.Sci. 279 (1992) 89. [93] H.Tochihara and S.Mizuno, Surf.Sci. 287/288 (1993) 423. [94] S.Mizuno, H.Tochihara and T.Kawamura, Surf.Sci. 293 (1993) 239. [95] S.Mizuno, H.Tochihara and T.Kawamura, Surf.Sci. 292 (1993) L811. [96] S.Mizuno, H.Tochihara, A.Barbieri and M.A.Van Hove, Phys.Rev.B 51 (1995) 1969. [97] S.Mizuno, H.Tochihara, A.Barbieri and M.A.Van Hove, Phys.Rev.B 52 (1995) R11658.

362 [98] S. Mizuno, H.Tochihara, Y.Matsumoto, K.Yanaka, A.Barbieri and M.A.Van Hove, to be published. [99] I.Ohsaki and T.Oguchi, J.Phys.Soc.Jpn 67 (1998) 978. [ 100] S. Quassowski and K.Hermann, Surf.Rev.Lett 4 (1997) 1209. [101] M.-S.Chen, D.Terasaki, S.Mizuno, H.Tochihara, I.Ohsaki and T.Oguchi, Surf.Sci. 470 (2000) 53. [ 102] S.Mizuno, H.Tochihara, Y.Matsumoto, K.Tanaka, A.Barbieri and M.A.Van Hove, Proc. 9'th Int.Conf. on Solid Surfaces, Yokohama, 25-29'th September 1995. [103] J.Henrion and G.E.Rhead, Surf.Sci. 29 (1972) 20. [ 104] A.Sepulveda and G.E.Rhead, Surf.Sci. 66 (1977) 436. [ 105] M.-G.Barthes and G.E.Rhead, Surf.Sci. 80 (1979) 421. [ 106] A.Sanchez, J.Ibanez, R.Miranda and S.Ferrer, Surf.Sci. 178 (1986) 917. [ 107] A.Sanchez, J.Ibanez, R.Miranda and S.Ferrer, Vacuum 37 (1987) 487. [ 108] A.Sanchez, J.lbanez, R.Miranda and S.Ferrer, J.Appl.Phys.61 (1987) 1239. [ 109] A.Sanchez and S.Ferrer, Phys.Rev.B 39 (1989) 5778. [ 110] W.Hosler and W.Moritz, Surf.Sci. 117 (1982) 196. [ 111 ] W.Hosler, W.Moritz, E.Tamura and R.Feder, Surf.Sci. 171 (1986) 55. [112] W.Hosler and W.Moritz, Surf.Sci.175 (1986) 63. [113] Y.Gauthier, W.Moritz and W.Hosler, Surf.Sci. 345 (1996) 53. [114] C.Nagl, E.Platzgummer, O.Haller, M.Schmid and P.Varga, Surf.Sci. 331-333 (1995) 831. [115] S.Robert, S.Gauthier, F.Bouquet, S.Rousset, J.L.Duvault and J.Klein, Surf.Sci. 350 (1996) 136. [116] S.Robert, C.Cohen, A.L.L'Hoir, J.Moulin, D.Schmaus and M.-G.Barthes-Labrousse, Surf.Sci. 365 (1996) 285. [ 117] S.Tan, A.Ghazali and J-C.S.Levy, Surf.Sci. 377-379 (1997) 15. [ 118] R.Plass and G.L.Kellogg, Surf.Sci.470 (2000) 106. [ 119] G.L.Kellogg and R.Plass, Surf.Sci. 465 (2000) L777. [120] F.Delamare and G.E.Rhead, Surf.Sci. 35 (1973) 172. [121] C.Argile and G.E.Rhead, Surf.Sci.78 (1978) 115. [122] B.Blum, E.W.Plummer, H.L.Davis and D.M.Zehner, J.Vac.Sci.Technol. A9 (1991) 1703. [123] B.Blum, E.W.Plummer and D.M.Zehner, J.Phys: Condensed Matter 5 (1993) A163. [124] W.Li and G.Vidali, Surf.Sci. 287/288 (1993) 336. [125] H.L.Meyerheim, H.Zajonz, W.Moritz and I.K.Robinson, Surf.Sci. 381 (1997) L551. [ 126] E.A1Shamaileh and C.J.Bames, submitted to J.Phys.Chem. [ 127] B.Blum and H.Ascolani, Surf.Sci.482-485 (2001) 946. [128] G.E.Rhead, C.Argile and M-G.Barthes, Surf.Interf.Anal 3 (1981) 165. [129] C.Argile and G.E.Rhead, Thin Solid Films 87 (1982) 265. [130] C.Argile and G.E.Rhead, Surf.Sci 135 (1983) 18. [ 131 ] F.Abel, C.Cohen, J.A.Davies, J.Moulin and D.Schmaus, Appl.Surf.Sci. 44 (1990) 17. [ 132] M.Breeman and D.O.Boerma, Phys.Rev.B 46 (1992) 1703 [133] E.McLoughlin, A.A.Cafolla, E.A1Shamaileh and C.J.Barnes, Surf.Sci. 482-485 (2001) 1431. [ 134] C.J.Barnes, E.A1Shamaileh, T.Pitkanen and M.Lindroos, work in progress. [135] T.Klas, J.Voigt, W.Keppner, R.Wesche and G.Schatz, Phys.Rev.Lett 57 (1986) 1068; T.Klas, R.Fink, G.Krausch, R.Platzer, J.Voigt, R.Wesche and G.Schatz, Europhys.Lett 7 (1988) 151. [ 136] Y.Li, M.R.Press, S.N.Khanna, J.Jena and M.Yussouff, Phys.Rev.B. 41 (1990) 4930.

363 [137] T.Nakagawa, G.I.Boishin, H.Fukioka, H.W.Yeom, I.Matsuda, N.Takagi, M.Nishijima and T.Aruga, Phys.Rev.Lett 86 (1992) 1703. [ 138] T.McEvoy, E.AIShamaileh, A.A.Cafolla and C.J.Barnes, work in progress. [139] D.Naumovic, A.Stuck, T.Greber, J.Osterwalder and L.Schlapback, Surf.Sci. 269/270 (1992) 719. [ 140] D.Naumovic, P.Aebi, A.Stuck, P.Schwaller, J.Osterwalder and L.Schlapbach, Surf.Sci. 307-309 (1994) 483. [ 141 ]Y.G.Shen, J.Yao, D.J.O'Connor, B.V.King and R.J.MacDonald, J.Phys.:Condens.Matter 8 (1996) 4903. [ 142] S.Nakanishi, K.Kawamoto and K. Umezawa, Surf.Sci. 287/288 (1993) 974. [ 143] P.T.Spunger, E.Laegsgaard and F.Besenbacher, Phys.Rev.B 54 (1996) 8163. [144] G.W.Andersen, K.O.Jensen, T.D.Pope, K.Griffiths, P.R.Norton and P.J.Schultz, Phys.Rev.B 46 (1992) 12880; G.W.Andersen, T.D.Pope, K.O.Jensen, K.Griffiths P.R.Norton and P.J.Schultz, Phys.Rev.B 48 (1993) 15283. [145] A.R.Koymen, K.H.Lee, G.Yang, K.O.Jensen and A.H.Weiss, Phys.Rev..B 48 (1993) 2020. [146] C.J.Barnes, E.AIShamaileh, T.Pitkanen, P.Kaukasoina and M.Lindroos, Surf.Sci. 492 (2001) 55. [147] T.D.Pope, K.Griffiths, V.P.Zhadanov and P.R.Norton, Phys.Rev.B 50 (1994) 18553. [ 148] S.Goupper, L.Barbier and B.Salanon, Surf.Sci. 409 (1998) 81. [149] Y.G.Shen, J.Yao, D.J.O'Connor, B.V.King and R.J.MacDonald, Solid State Commun, 100 (1996) 21. [ 150] J.P.Reilly, D.O'Connell and C.J.Barnes, J.Phys:Condensed Matter 11 (1999) 8417. [ 151 ] E.A1Shamaileh, A.O'Hagan and C.J.Barnes, to be published. [ 152] Y.G.Shen, D.J.O'Connor and K.Wandelt, Surf.Sci.406 (1998) 23. [ 153] R.Belkhou, J.Thiele and C.Guillot, Surf.Sci. 377-379 (1997) 948. [ 154] J.Radnik, B.D.Wagner, K.Oster and K.Wandelt, Surf.Sci. 357/358 (1996) 943. [155] G.Gilarowski and H.Niehus, Phys.Stat.Sol.(a) 173 (1999) 159. [ 156] G.Gilarowski and H.Niehus, Surf.Sci 436 (1999) 107. [ 157] S.Heinze, R.Abt, S.Blugel, G.Gilarowski and H.Niehus, Phys.Rev.Lett 83 (1999) 4808. [158] J.Yao, Y.G.Shen, D.J.O'Connnor and B.V.King, Surf.Sci.359 (1996) 65. [ 1.59] Y.G.Shen, A.Bilic, D.J.O'Connor and B.V.King, Surf.Sci. 394 (1997) L 131. [ 160] K.Pussi, M.Lindroos, E.A1Shamaileh and C.J.Bames, submitted to Surface Science. [ 161 ] E.AlShamaileh, C.J.Barnes, K.Pussi and M.Lindroos, submitted to Surface Science. [ 162] H.Younis, E.A1Shamaileh and C.J.Barnes, to be published. [ 163] T.Flores, M.Hansen and M.Wuttig, Surf.Sci 279 (1992) 25.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces 364

D.P. Woodruff, (Editor)

Chapterl 0

Surface and sub-surface alloy formation connected with ordered superstructures Horst Niehus

Humboldt-Universit~it zu Berlin, Institut for Physik, Oberfl~ichenphysik und Atomsto6prozesse, Invalidenstra6e 110, D- 10115 Berlin, Germany 1. I N T R O D U C T I O N The properties of alloy and intermetallic compound surfaces play an important role for the development of new materials. Attention has been stimulated from various topics in microelectronics, magnetism, heterogeneous catalysis and corrosion research. The investigation of binary alloys serves also as a first step in the direction to explore multi-component systems and is of particular regard in material science as a consequence of their widespread use in technical applications. The distribution of two elements in the bulk and at the surface probably results in new characteristics of the alloy or compound as compared to a simple superposition of properties known from the pure constituents. Consequently, surfaces of bulk- and surface- alloys have to be investigated like completely new substances by means of appropriate material research techniques and surface science tools. [ 1-6]. Characteristic features of alloy surfaces such as surface composition, segregation, structural aspects like reconstruction, relaxation and ordering are of great significance for mechanical, chemical and electronic properties of the material. In the past, novel structures, either ordered or disordered, have been created at the surface area depending on the segregation characteristics of the corresponding bulk alloys [7]. Chemical treatment (e.g. oxygen exposure) stimulates sometimes segregation as well [8]. In an alternative approach, comparable alloy layers have been generated by thin film techniques via metal on metal growth. In special situations, even intermixing between different metals is observed which are known to be immiscible in the bulk. Quite a few of those surfaces have been already investigated in the past, because it is well known, that composition and surface structure of a compound influences strongly its connection to the outside world. Moreover, chemical reactions

365

might be effectively influenced by the surface crystallography [9]. In a future vision, the compilation of characteristic data on structure and composition of lots of alloys is aspired to find the way for an understanding of the energetics of the formation of the alloy surface itself. In the following, surface and sub-surface alloy formation of ordered systems in ultra high vacuum will be discussed as an option to generate different surface configurations with dissimilar properties from the same set of material composition. Certainly, alloys develop also at the liquid-solid interface [10-13], yet the topic will not be covered in this chapter being of special devotion to ordering effects. Surfaces of ordered bulk alloys shall be reviewed in a first part. A second subdivision includes the formation of surface and subsurface alloys, whereas in a third section applications are discussed to grow ordered superstructures on top of alloy surfaces. The first example will reference the possibility to change the surface composition of a binary alloy just by varying the sample preparation. NiA1 was depicted as a model of a strongly ordered alloy where different surface compositions at the (111) and (100) surfaces could be prepared. Thereafter, the influence of the surface structure of a bulk alloy on the chemical reactivity will be illustrated for Cu3Au (110) and Cu3Au (100), both containing the same surface composition. Further on, the preparation of completely new materials surface alloys composed of bulk immiscible constituents - will be followed up. Especially, the influence of intermixing and phase separation on the development of the surface arrangement and on the formation of ordered subsurface alloys will be exemplified for the metals of iridium and copper. Finally, two applications with alloys serving itself as substrates for the growth of epitaxial heterostructures will be discussed and demonstrated for vanadium on Cu3Au(100). Depending on the initial surface conditions, supposition for intermixing or phase separation can be established. In order to examine experimentally the relevant surface characteristics, several techniques are applied in the investigations performed in ultra high vacuum (UHV, at a pressure of about lx 10~~ mbar), i.e. scanning tunneling microscopy (STM), spot analysis of low energy electron diffraction (SPALEED [ 14]), low energy ion scattering spectroscopy (ISS), X-ray photoelectron spectroscopy (XPS) and Auger electron spectroscopy (AES). During the last years, microscopy with atomic resolution obtained by STM [15-19], considerably improved the understanding of the surface structure. Since, in spite of this, the information achieved with the STM depends on electronic surface properties, the deduction of the surface topography is not always straightforward. Moreover, a correlation of measured features in STM with the elemental distribution at the surface appears often difficult or impossible (for

366

chemical analysis with STM [20] see also the chapter by Peter Varga). Accordingly, an element specific technique which reacts sensitively to the position of the atom cores in the first few layers like e.g. low energy ion scattering spectroscopy [21, 22] is well suited for a combination with STM for complementary surface structure analysis. STM and ion scattering, in particular low-energy noble gas impact collision ion scattering spectroscopy with detection of neutrals (NICISS) [23, 24] have been used together in the past and offer a suitable instrumentation to locate element specifically the position of surface atoms in real space [25]. 2. SURFACES OF ORDERED BULK ALLOYS 2.1. Preparation dependent surface composition" NiAI Ordered intermetallic compounds of B2 crystallography belong to the significantly examined systems. Aside of their unusual mechanical properties, the catalytic behavior of NiA1 has stimulated recent investigations [26-28]. NiA1 serves as a prototype of strongly ordered binary alloys. The chemical reactivity of these alloy surfaces varies with their composition and structure. At elevated temperatures epitaxial thin films may well occur on NiA1, a process that is usually accompanied by rearrangements of surface atoms through A1 segregation towards the surface. Based on latest studies, the 'rippled relaxation' structure of NiAI(110) is well established [29, 30]. In contrast, the (100) and (111) surfaces happen to be more complex and might form different surface layers. Since NiA1 crystallizes in a CsC1 configuration, the solid consists of alternating Ni and A1 layers in an-A-B-A-B- stacking and may end up in the truncated bulk structure either with a Ni or A1 terminated surface for both, the (111) and (100) plane. All NiA1 surfaces show strong chemical reactivity against oxygen exposure. Thin epitaxial A1203 films can be produced easily at the (110), (100) and (111 ) surfaces [28, 31 ]. The composition of the topmost surface layer was determined straightforward by taking advantage of the high surface sensitivity of low energy ion scattering [21]. Before discussing the NiA1 surface structures in detail, the special technique of ion and neutral particle backscattering-NICISSshall be reviewed briefly below. A pulsed low intensity He + ion beam irradiates the sample surface in the UHV chamber and the energy spectra of backscattered neutralized He projectiles are determined via a time-of-flight (TOF) technique and finally analyzed in terms of surface composition by comparison with the model of elastic single scattering [21, 22]. The signal of backscattered projectiles is recorded as a function of the scattering geometry at fixed scattering angle. A set of TOF spectra of He projectiles, 180 ~ backscattered at

367

the NiAI(100)-(lxl) surface is presented in Fig. 1. The intensity of scattered He is plotted as a function of the azimuthal rotation q~ with a fixed grazing angle of incidence of ~ = 8 ~ (called (p-scan). Two peaks in individual time-of-flight spectra could be easily resolved and are attributed to He scattering at Ni and A1 atoms, respectively.

Fig. 1- Set of experimental NICISS flight time spectra at NiAl(100)-(lxl). q~-scan: the intensity of 180~ scattered He is plotted as function of the azimuth rotation q~(0 ~ - 100~ with a constant grazing angle of incidence ~ = 8~ Energy of the primary He+ particles E0 = 3 keV. (from ref. [37]). Furthermore, structure information can be obtained, to be realized in a first run just by application of a simple triangulation scheme [32]. The angular pattern reflects shadowing and blocking effects for He particles scattered by the atoms in the single crystal arrangement. Whenever target atoms are not shadowed for the incoming ion beam by their next neighbors, backscattering may occur and an intensity rise is measured, in fact mostly enhanced by trajectory focusing close to the shadow and blocking cone edges. By comparison with ion scattering simulations it has been found useful to identify the angle of incidence as a critical angle ~c when the intensity as a function of the polar angle has reached 80% of its maximum [33]. This scheme can be used to associate the critical angles to the intra- and inter-layer distances of atoms via known contours of the shadowing and blocking cones, i.e. by the shape of the interaction potential.

368

A more detailed analysis including the determination of surface relaxation relies on the comparison with computer-simulated data. The temperature dependent motion of the atoms as well as the angular resolution of the instruments mostly limit the precision achieved so far to ascertain atom positions. An accuracy of about +5 pm has been reported [21]. As is known from Rutherford backscattering, the cross section for backscattered particles is extremely low as compared with forward scattering. Hence, ISS simulation codes in which just all incoming particle trajectories are followed, have been proved to be very useful for forward scattering experiments (e.g. MARLOWE [34], SABRE [33], SARIC [35]) but, in fact they are generally not best suited for the backscattering geometry. An alternative simulation procedure (FAN [24]) brings up an another solution: instead of keeping track of all trajectories, just the ones where particles hit centrally the probed atom in the surface near area will be included in the calculation. Only particles of these trajectories are traced back throughout the solid on their way into the detector or to the ion source, respectively. All possible angles of incidence are simulated at once in a fan-like structure and the two overall probability functions for the way in and the way out are connected for the proper scattering geometry in a final step. Thus, just the relevant trajectories for backscattering are followed and the computer time to simulate an entire data set can be shortened down significantly to several minutes on a personal computer (FAN simulation package available [36]). In most cases the FAN simulation code has been employed using Moli6re interaction potentials [38-40] often with a Firsov screening factor reduced by c = 0.7. The preparation dependent formation of different NiA1 surfaces was followed up for the (111) and (100) surfaces. NiAI(111) exposes a rather open surface being composed either of Ni or A1 atoms. From investigations at NiAI(111) annealed below 1000 K it is known that the surface remains often slightly oxygen contaminated. At those surfaces, domain mixing of both, Ni and A1 terminated terraces has been found. Terraces separated by mono-atomic steps might explain this surface configuration. Domain mixing at NiAI(111) for similarly prepared surfaces has been deduced independently from ion scattering [41, 42] and a detailed LEED analysis [43]. In a second attempt, the NiAI(111) surface was re-examined by NICISS and STM where a different surface preparation has been used by annealing the sample finally at 1300 K (instead of 1000 K utilized before [41-43]) in order to dispose of the oxygen contamination [44]. The 180 ~ backscattering patterns for the Ni and A1 signals of the corresponding clean surface are shown in Fig. 2. The intensity of scattered He particles is plotted as function of the grazing angle of incidence qtin (called ~-scan). An easy way to extract from such q~-scans

369

relevant surface crystallographic information is illustrated in the sketched side view of the atomic arrangement (inset in Fig. 2b). Arrows indicate the possible head-on collision geometries. The angular position marked by (A) in Fig. 2a resembles the position of Ni in the first layer.

300

01

200

b) 300-

He"---~ NiA[ (111) Eo = 2000 eV ,9 = 180~ k0inl[ [ 1211 TS= 150K

ol

o.. ~a

Z

on top

ti2 t

200 ~ ~

o o| o

-o|174174

|174o|

.,,.,. (./3

O')

._---,

side

<

100

i

!

20 ~

!

I

400

t

i,

600

t

I

800

~

I

100 -

IB

i

100~ ~in

200

t*O~

600

80~

100~ qJin

Fig. 2: NICISS pattern for He backscattering at NiAI(111). He scattering at a) Ni and b) A1 atoms. W-scan: the intensity of scattered He is plotted as function of the grazing angle of incidence /IJin ( 0 ~ - 120~ with a constant azimuth rotation (q0 11[112] for ll/in -- 0 ~ The inset shows a side view and an on top view of the scattering geometry. Open circles: Ni, hatched circles: A1 atoms. Head-on scattering situation from first layer atoms (A) and second layer atoms (B) is indicated. (from ref. [44]).

The most striking difference with respect to the earlier investigation [41] can be recognized at the A1 signal in Fig. 2b: there is no corresponding A1 first layer peak (A) (which is expected to turn up at grazing angle of incidence). The onset of the Al~s signal occurs at Wi, > 25~ with peak (B), the second layer signal. As a result, for NiAI(111) annealed at 1300 K it has been demonstrated that the first layer consists completely of Ni atoms. Small amounts of oxygen seem to have a large effect on the surface properties of NiA1 and lead in particular to segregation of A1 to the surface. In corresponding STM investigations it has been demonstrated, that oxygen contamination induces the formation of small A1 islands on Ni terminated terraces [44]. Comparable influence has the sample treatment on the surface termination and structure of NiAI(100). Indeed, different experimental findings have been reported earlier depending mainly on various surface preparation techniques. Early LEED investigations [30] indicate that the NiAI(100) surface gives a C(~/2 x 3~/2)R45 ~ LEED pattern when annealed at 750 K. Small amounts of impurities [45] do influence the structure. Upon high temperature annealing at about 1300 K the development of a ( l x l ) LEED pattern has been reported. The surface consists probably of a mixture of Ni and A1 atoms including some

370

surface vacancies as well. Moreover, a surface termination with A1 and a small number of Ni antiside defects incorporated in the surface layer has been suggested [29]. According to the LEED and AES measurements by Roux and Grabke [46], the NiAI(100)-(lxl) surface appears to be A1 enriched after sputtering followed by subsequent annealing at 850 K. Further sputtering at room temperature removes most of the excessive A1 leading to a diffuse LEED ( l x l ) structure. Subsequent annealing at 1100 K restored the LEED pattern.

Fig. 3. SPA-LEED results for NiAI(100): Intensity variation of the normalized (00) spot (full circles) and the background intensity (open squares) with temperature recorded for inphase conditions (E0 = 95 eV). The full lines are guidelines for the eye; the dashed line indicates a result for the situation of surface roughening of the Kosterlitz-Thouless type. (from ref. [37]).

To sum up, basically three dissimilar surface configurations of clean NiAI(100) have been reproduced by the different groups. The resulting surface arrangement depends strongly on the final annealing temperature after the Ar + sputter cleaning procedure of the sample. The surface structures are characterized by the following sequence of LEED patterns which passes from a ( l x l ) to a c(~/2 x 3~/2)R45 ~ and finally again to a ( l x l ) superstructure as a function of increasing annealing temperatures. The correlated occurrence of surface roughening and smoothing during the superstructure development can be followed quantitatively in a SPA-LEED pattern (Fig. 3) by comparison of the intensities of the (00) spot and the background.

371

Based on combined SPA-LEED and NICISS experiments [37] the three different surface phases could be identified as follows. A low temperature ( l x l ) phase (annealing at 500 K) was found and described by an A1 terminated surface showing an inward relaxation of 14 pro: - NiAl(lOO)-(lxl)A t. Annealing at 800 K leads again to an A1 terminated surface exposing additionally ordered vacancy rows to result in a c(~/2 x 3~/2)R45 ~ LEED superstructure: - NiAl(lO0)c(3/2 x 3 ~)R45~ The structure is also in accordance with a configuration proposed by Mullins and Overbury [30]. The surface model is sketched in Fig. 4b and can be described basically by an A1 terminated layer, including missing A1 rows in combination with an inward relaxation of the top A1 layer by 14 pro. In addition a decrease of the A1 intralayer distances of 20 pm has been deduced. The decrease establishes an equal bond length between three A1 atoms within an A1 cluster as sketched in Fig 4b. An alternative model where the vacancies are replaced by Ni atoms which has been recommended [30] could be ruled out. In a high resolution STM investigation the corresponding surface phase has been imaged as can be seen in Fig. 4a.

Fig. 4: a) STM image of the NiAl(100)-c(~/2 x 3~/2)R45 ~ surface. Two domains A and B could be seen on the upper terrace, which is separated by a double step from the lower terrace on the right side (from ref. [47]). b) Hard sphere model of the corresponding surface. (from ref. [37]).

Finally, after annealing at 1300 K another ( l x l ) structure with improved long-range order as compared with the low temperature ( l x l ) phase shows up. Different from the low temperature phase the surface is here Ni terminated and no measurable relaxation could be found: - NiAl(100)-(lXl)ui.

372

The appearance of the different surface phases is closely related to surface segregation and roughening at high temperatures. The creation of vacancies might be basically caused by anharmonicity in the mean square vibrational amplitude of A1 atoms in the surface. A similar increase in vibration amplitudes has been reported for Ni(110), Ni(100) [48-50] and AI(110) surfaces [51 ]. The characteristic range of temperatures for anharmonicities can be found between 500 K and 800 K, which agrees well with the measured start of enhanced vacancy development at the NiAI(100) surface (cf. Fig. 3). The generation and annihilation of vacancies on NiAI(100) seems to be a reversible process over a large range of temperatures. After all, for NiA1 samples with the same elemental bulk composition, different surface compositions can be created due to surface segregation. Either, A1 or Ni terraces develop depending on the final annealing temperature in the range from 600 K to 1400 K.

2.2. Surface properties of alloys with identical surface composition After that, the occurrence of different structures at surfaces exposing the same composition will be ascertained. Several aspects of ordering, structure and composition of alloy surfaces with L12 crystallography are already well characterized in the past [7, 33, 52-56]. Among others, the Cu3Au system was analyzed essentially because it behaves as a classical ordering alloy. The orderdisorder transition at a bulk temperature of T t = 663K has been investigated extensively as a model system. Experimentally, the critical behavior close to T t was studied by many techniques, namely by low energy electron diffraction (LEED) [57, 58], X-ray scattering [59-62], X-ray photoelectron scattering (XPD) [63], medium energy ion scattering (MEIS) [64] and thermal energy atom scattering (TEAS) [65]. Moreover, theoretical predictions are available to describe the transition from the well-ordered alloy to the substitutional disordered phase [66]. In addition, band structure calculations in combination with photoelectron spectroscopy data can be found in literature [67-70]. In the following no attempt will be made to add any supplementary information to temperature dependent structure changes or the order-disorder discussion. Instead an investigation of properties of well ordered Cu3Au- (110) and (100) surfaces at room temperature will be presented. By reason of the L 1 2 crystallography, for bulktruncated surfaces, both samples, Cu3Au(100) and Cu3Au(110), might be terminated by a pure Cu or a mixed Cu-Au topmost layer, respectively. As will be demonstrated below, both single crystals terminate in the mixed layer configuration. Nevertheless, their surface reactivity appears to be genuinely different.

373 Furthermore, Cu3Au surfaces served already in the passed as substrates for epitaxial growth of metal layers especially in view of new magnetic materials (Fe [71-73], Ni [74]). Due to the possibility to develop different ordered CuxAuy alloys with the inclusion of the pure Cu and pure Au metals, an opportunity arises to vary experimentally the lattice parameters over some range. These surfaces might work for different materials as a substrate in use to change the lattice mismatch and consequently the surface stress in epitaxially grown metal layers. It will be confirmed in section 4.2 that Cu3Au may as well act as a good substrate for metal oxide growth.

2.2.1 Cu3Au(110) The Cu3Au(ll0 ) surface has been prepared in UHV as follows, after cleaning by appropriate Ar + sputtering, the crystal was held for 10 hours at about 500 K to allow bulk and surface ordering getting complete. The structure of the bulk-truncated surface is expected to show up in a Cu3Au(110)-(2xl) superstructure to consist of alternating single Cu and Au rows along the [001 ] direction. Indeed, after rapid sample cooling a LEED (2xl) superstructure is measured at room temperature. On the other hand, a clear (4xl) superstructure occurs in the LEED pattern, when the crystal is cooled extremely slowly (about 3 hours) from T t down to room temperature [75, 76]. In fact, a second 'surface phase transition temperature' of T s around 400 K has been proposed by Huang and Cowley [77] for the separation of the (2xl) and (4xl) phases.

Fig. fi: a) STM of the Cu3Au(110)-(4xl) surface at room temperature. High resolution STM (3 nm x 3 nm) with atomic resolution. The main structural element consists of double rows of Au and Cu in the top layer, b) The atomic coordinates for the best-fit model from the LEED IV data. (from ref. [78]).

374

Apparently the (2xl) phase cannot be attributed to the equilibrium structure at room temperature, but may be described as a quenched phase probably due a limited surface diffusion parameter. Aside of finding the ordered surface by LEED, the (4xl) structure has been imaged for the first time in real space by scanning tunneling microscopy [76] which is presented in Fig. 5a. With the help of ion scattering measurements the topmost layer was determined to consist of 50% Au and 50% Cu (mixed layer termination) [7981]. Based on STM and NICISS data, a row-pairing model for the (4xl) superstructure has been inferred by reason of the formation of Au-Au and CuCu rows in the surface [76]. A consecutive performed LEED I-V analysis led to the proposal of a refined surface structure being presented in Fig. 5b [78]. The row-pairing model could be confirmed and in addition precise bond length numbers have been calculated. In the bulk region the Cu-Cu and Au-Au distances are 0.279 nm and 0.285 nm, respectively, whereas in the topmost layer, the Cu-Cu bond length becomes 7 pm shorter, meanwhile the Au-Au separation increases being a sign of the larger distance of Au in an Au bulk crystal. Such an expansion drives the Cu atoms to come closer in the rows. It might turn out to be plausible that a formation of three or more Au atoms in a row is less favorable, because of a pile up of the lattice mismatch as referred to the bulk Cu3Au distances. Accordingly, the process stops after two steps (row pairing). By that, a surface rippling between the Cu-Cu and the Au-Au rows of 9 pm appears in the top layer.

Fig. 6" High resolution STM of Cu3Au(110)-(2xl)-O at room temperature. Area: 1 nm x 1 nm. The added -Cu-O-Cu- rows running along the [001 ] direction are imaged with atomic resolution; unit cell dimensions 0.53 nml0~ j x 0.37 nmf00~l. U,~p = 0.6 V, i - 1 nA.

In addition, the best agreement (Pendry R-factor [82] of 0.30) between calculation and LEED intensity curves is obtained by an additional Au segregation to the second layer by at least 25%. Such enrichment with Au in the surface layers is furthermore consistent with the lower surface energy of Au

375 (1550 mJ/cm 2) as compared to Cu (1850 mJ/cm 2) [83]. The composition of all subsequent deeper layers is identical to that of bulk Cu3Au starting with a pure Cu plane as layer number three. The chemical reactivity of the topmost mixed layer was tested by its reaction against oxygen exposure. Oxygen adsorption at room temperature readily occurs and the formation of-O-Cu- rows has been proved by STM and NICISS measurements [81]. The Cu3Au (110)-(2xl)-O structure presented in Fig. 6 was obtained after 30 Langmuir oxygen exposure at 300 K and subsequent annealing at 700 K. The surface looks very comparable to the well known (2xl) added row structure at Cu(ll0)-O [84]. Yet, the unit cell of Cu3Au (110)-(2xl)-O is slightly larger as compared with Cu(110)-(2xl)-O. Probably, a reduced stress in the overlayer might be in charge for the fact that a self organization of the -O-Cu- stripes into the recognized piano-structure of oxygen on Cu(110) [25, 85] does not occur at the Cu3Au (110) substrate. Anyhow, it is worth mentioning that the configuration of Cu rows at the clean surface Cu3Au(110) seems to be sufficient to drive oxygen dissociation at room temperature. During the adsorption process the strings of Cu atoms act already quite similar to a complete Cu(110) surface! Perhaps, the dissociation is a consequence of the altered electronic structure. Without a doubt, electronic differences at the Cu3Au (110)-(4xl) surface are well expressed in the STM data obtained at the Cu and Au double rows. Indeed, an electronically enhanced apparent corrugation of 30 pm is measured by STM [76] as compared with the 9 pm rippling determined as the topographic height variation by LEED. To put it briefly, the overall chemical activity of the Cu3Au(ll0) surface differs remarkably from bare Au and seems to be at least in part comparable to reactions happening at pure Cu(110). 2.2.2 Cu3Au(100 ) The mixed layer termination with the same surface composition of 50% Au and 50% Cu as explained above for Cu3Au(110) has been advised too for CuaAu(100 ) by experimental and theoretical investigations [69, 86, 87]. In order to determine the actual surface structure in detail, ion scattering was performed and the entire patterns of measured NICISS- and simulated FANdata is presented in figure 7. The intensities of He particles back scattered by 180 ~ at Au atoms are plotted in a contour line graph. The experimental data set for the plot of Fig. 7.1 is obtained from 30 individual ~g-scans for an angular range of incidence gt from 0 ~ to 90 ~ and azimuth q) from 0 ~ to 90 ~ with angular increments of 3 ~ The corresponding FAN simulations of the Cu3Au(100 ) surface, terminated by the mixed layer have been performed by simulation of 30 ~-scans, either for the model of two dimensional crystal slices (2D-FAN)

376 including the first three (Fig. 7.2) and the first five (Fig. 7.4) surface layers or for a full 3D model (Fig. 7.3). Although the prominent features from the 2DF A N are in accordance with the experiment, the best agreement with the calculation is obtained for the three dimensional F A N simulation, which is evidently recognizable by comparison of Fig.'s 7.1 and 7.3. In fact by comparison of the entire set of measured ISS data with the F A N simulation a complete description of the Cu3Au(100) surface has been developed.

Fig. 7: NICISS at Cu3Au(100)-c(2x2): contour plot of He 180~ back scattered at Au. He intensity (white - high; black - low) is plotted as a function of the angle of incidence ~ (0 ~ 90 ~ and azimuth angle q0 (0 ~ - 90 ~ in a linear scale. Eo = 2 keV. The low index directions [011], [001], [101] are indicated. (from ref. [88]). Houssiau and Bertrand [89] refined the model by introduction of a small surface rippling by reason of an outward shift of the Au atoms by 12 pm, which is also in good comparison with the outward shift of Au by 9 pm for the (110)

377

surface as shown above. The mixed layer termination determined by various groups follows the general trend of forming a surface with low surface energy. Regardless of being composed by a 50% Au- 50% Cu- top layer, the atomic arrangements of the two (110) and (100) Cu3Au surfaces are obviously different. In case of the (110) surface, double rows develop consisting either of Au or Cu atoms, whereas the (100) topmost layer consists of [011] chains composed of Au and Cu neighbors. Moreover this Au-Cu alternation brings about next neighbors being never of the same atomic species. In such a surface every Cu atom finds itself surrounded by Au atoms and v i c e v e r s a . As an effect, in variance to Cu3Au(110), the (100) surface behaves completely inert against oxygen exposure at 300 K and seems to be undoubtedly related to properties of a pure Au surface. In particular, one may note, that single Cu atoms surrounded by Au atoms are not able to initiate dissociation of the impinging O2 gas molecules, which is in contrast to the situation of Cu strings at Cu3Au(110) in an otherwise equally terminated surface by 50% Au and 50% Cu.

Fig. 8: a) STM survey of Cu3Au(100)-c(2x2)-O at room temperature. The surface has been sputtered by 1 keV O § and is subsequently annealed at 650 K. b) High resolution STM: unit cell dimensions 0.53 nm x 0.53 nm. Utip = 0.6 V, i = 1 nA.

Bombardment with energetic O + ions forces oxygen to get incorporated into the Cu3Au sample at 300 K. As a result, initially a thin and rough oxygen containing selvedge layer is formed due to sputtering effects in combination with oxygen ion implantation. Subsequent annealing at 650 K drives oxygen out of the bulk towards the surface. The resistance of the CuaAu bulk material against oxidation together with the high mobility of the Cu and Au atoms

378

finally leads to the repair of surface defects upon annealing (Fig. 8). It has been established experimentally by STM and NICISS, that the topmost layer consists of an ordered Cu-O structure [86]. The annealed Cu3Au-O surface heals out completely all sputter induced surface defects and becomes atomically flat (Fig. 8a). Such a well-ordered (Fig. 8b) smooth surface turns up as a perfect substrate for epitaxial film growth (see section 4.2). As a minor contribution, in the high resolution STM image (Fig. 8b) slight background variations may be recognized in the entire image. Possibly, they are caused by electronic differences connected with areas containing still some subsurface oxygen, which remained after annealing in the surface near area. Accordingly, the Cu3Au crystal may act in contrast to a pure Au sample as an oxygen reservoir with oxygen atoms stored in the sub-surface region. Prolonged heating at 650 K, or more effectively by annealing at 700 K will empty this reservoir as can be monitored experimentally by AES. Such an oxygen reservoir capacity is furthermore due to the inert character of Cu3Au against oxidation and thus of real advantage over a pure Cu substrate. Accordingly, just the combination of gold and copper in the alloy substrate gives rise to the novel features of the Cu3Au(100) surface. 3. SURFACE ALLOYS OF BULK I M M I S C I B L E C O N S T I T U E N T S 3.1 Sub-surface alloy formation: Iridium on Cu(100) Novel ordered alloy systems could be created as well via metal on metal growth just by the development of thin mixed metal films. Segregation and intermixing may induce alloying as well as ordering processes in the surface near region. Occasionally, intermixing has been found even between metals that are immiscible in the bulk [4, 17]. As an example, the iridium-copper system will be presented below (for the discussion of additional alloy configurations at Cu(100), please refer to the chapter by Colin Barnes). The phase diagram of Ir and Cu is characterized by a substantial miscibility gap at 300 K. Even at temperatures up to around 1000 K only 3 atomic percent (at.%) of Ir is soluble in a Cu matrix and in the reverse case only 1 at.% Cu in Ir [90]. Since now, no ordered bulk alloy phase has been reported for Cu(100) and Ir. Obviously, the growth mode of Ir on Cu will be directly influenced by the surface free energies of Ir and Cu with 3000 mJ/cm 2 and 1830 mJ/cm 2, respectively [91 ]. Hence, no intermixing is anticipated at least at temperatures below 1000 K and a VolmerWeber growth mode [92] for Ir on Cu seems to be the plausible result. In variance to these expectations, after depositing Ir on Cu(100) at a sample temperature of 200 K, the corresponding ISS measurements display immediately an apparent lack of evaporated material at the surface (Fig. 9). The

379

amount of Ir in the topmost layer appears to be considerably smaller as compared with the quantity of added material. Although an almost linear increase up to 1 ML is found, apparently in the entire range of coverage only about 20% of the total amount of deposited Ir remains in the first layer. For comparison, the dotted line in Fig. 9 indicates the expected dependency for a two dimensional (layer-by-layer) growth where all Ir atoms remain in the topmost layer. The surplus of Cu atoms in the first layer has been rationalized by the proposition of intermixing and direct segregation of Ir atoms into the copper bulk [93, 94]. Even in the case of evaporation on the substrate at 200 K a substantial mass transport is recognized at the surface. 0.20

,

'

O. 18

'growth

~. o.14 r

.~ U)

,

layer-by-layer

0.16

M

i

experimental

'

data

0.12 0.10

L..

0.08

-> 9

0.06

(].) n,'

0.04

, '

0.02 0.00

0.0

,

I

0.2

,

I

0.4

,

I

0.6

,

I

0.8

,

1.0

Nominal Ir coverage (ML)

Fig. 9: ISS: He+ ion scattering spectra obtained for Ir on Cu(100) at 200K substrate temperature. E0 = 1 keV. Amount of Ir in the first layer as a function of the nominal Ir coverage (one monolayer (ML) refers to the layer density of the substrate). Squares: As deduced from measured IS spectra taken after Ir deposition. Dotted line: Expected dependency for 2D (layer-by-layer) growth. (from ref. [93]).

In the light of STM (Fig. 10), the Ir-Cu surface looks like being covered by small islands (apparent height of 0.18 nm due to the surface topography) and chains of lower contrast (about 0.1 nm by reason of a chemical contrast in STM). It was proved that the islands cannot be composed of evaporated iridium, but instead have been identified to consist of Cu that resides on top of Ir islands that are located in Cu terraces. Whereas the displayed chains indicate Ir atoms built in the Cu terrace, which are not yet covered by copper. It becomes energetically more favorable for Ir atoms to be embedded in the Cu substrate via place exchange, rather then staying atop. With increasing

380

coverage, Ir clusters grow in the configuration of implanted Ir chains. These chains successfully act as traps for the Cu atoms stemming from the Cu lattice gas on the terraces. Obviously, the coverage of Ir islands by copper consumes Cu atoms which usually may evaporate onto the terrace from the near by step edges. Therefore, step boundaries start to move back initially (step roughening). With increasing demand of Cu atoms upon larger Ir coverage the step edge movement is eventually hindered by Ir islands in the upper terrace to fix the steps at these points. After all, the evaporation rate of Cu from the steps is no longer sufficient to yield enough atoms to cover up the Ir islands and surface etching sets in which shows up in STM by the occurrence of mono-atomic deep holes on Cu terraces [93].

Fig. 10: STM of Cu(100)-Ir, measured at room temperature after 0.15 ML Ir deposition at 200 K. Image size: 100 nm x 100 nm. (from ref. [93]).

Similar effects of intermixing associated with step roughening and surface etching have been reported additionally for other systems like Co-Cu(111) [95, 96], Fe-Cu(100) [97, 98] and Rh-Ag(100) [99, 100]. As a common feature of all these systems it can be noticed that the surface free energy of the deposited metal exhibits always a higher value with respect to the substrate. Obviously, the driving force for intermixing can be found in the difference of the surface free energy, being furthermore in agreement with theoretic work to propose for

381

Ir on Cu a lowering of the total surface free energy by segregation of Ir into the Cu surface [ 101 ]. Surface segregation depends strongly on the mobility of the atoms in the solid and annealing of the Ir-Cu sample has indeed dramatic influence on composition and surface structure. As by a good choice, the composition has been determined with a combination of ISS and XPS measurements. The different surface sensitivities of the two methods (topmost layer for ISS versus an average over some surface layers for XPS determined by the mean escape depth of the photo-electrons) can be used to find the actual composition of the surface layers (Fig. 11). An amount of 0.28 ML Ir located in the first layer was measured by ISS immediately after the deposition of 1.5 ML Ir at 200 K. With increasing annealing temperature the surface depletion with Ir atoms can be seen, marked by a linear decrease in the ISS signal. At T > 750 K the outermost surface layer is free of Ir atoms. On the other hand, Ir has not yet segregated deep into the bulk, visible by the XPS data to display only a slight decrease for annealing up to 750 K. Finally, at higher temperature Ir starts to segregate completely into the bulk. In the particular situation of annealing at 650 K, the surface contains less then 3% of a monolayer Ir, that is to say the topmost layer consists almost exclusively of Cu atoms while the Ir atoms still remained in the surface near region. 0.40

'

0.35

I

'

I

'

I

'

I

'

I

'

I

'

J

'

I

m

m

m - - - " ~ ' ~ - i

-

.

~

0.30

2.0 1.8

,

=

0.25 -~ (" 0.20 ._~

'

\-

:

9

1.4

o ~-

1.2

._> ~9

iv'

0.15

1.0

o.lo

C)

E

0.8

0.05

0.6 0.00

100

,

i

200

,

,,

I

300

i

I

400

m

I

500

,

I

600

i

l

700

i

I I

800

i 0

1

900

1000

Temperature(K) Fig. 11: ISS - XPS of Cu(100)-Ir: Amount of Ir as function of the annealing temperature: in the topmost layer (left axis; circles) as determined from ISS in comparison with the amount of Ir in the first few layers as deduced from XPS (ratio of Ir 4f and Cu 3p levels, right axis; squares). Initial nominal Ir coverage: 1.5 ML. (from ref. [93]).

382

Simultaneously, massive structural changes directly show up in the STM data after sample annealing. On a larger scale, the complete disappearance of the small ad-islands (seen in Fig. 10) can be recognized in Fig. 12. Obviously, the intermixing process flattens out the entire Ir-Cu surface. The surface visible on individual terraces gives the impression in STM of being almost structureless, except for special tunneling parameters. By tunneling into occupied states of the sample, weak shady depression features become visible on the entire area (Fig. 12). The onset of ordering might be recognized already in small areas of limited size. Considerably better ordering of these shady features have been obtained by evaporation of Ir directly onto the heated sample rather than by subsequent annealing.

Fig. 12: STM of Cu(100)-Ir: 1.5 ML Ir evaporated at 200 K followed by 30s annealing at 650 K. (Image size: 250 nm x 250 nm, Ut~p= 0.2 V). STM image taken at 300 K.

A careful analysis of the novel depression structures can be performed best, by first studying these features for just a few Ir atoms at the Cu surface. An STM image with atomic resolution is presented in Fig. 13a as measured after 0.05 ML iridium deposition at 200 K followed by post-annealing at 650 K. Two basic features can be emphasized, the appearance of an ordered array of white dots and additional star-like depressions which are irregularly spread over the displayed surface area. The array of dots in Fig. 13a has been identified as the location of first layer Cu atoms. Further on, it was verified by corresponding ab initio calculations by Heinze at al. [102] that no bias voltage dependent corrugation reversal as e.g. predicted for W(110) [ 103] occurs on the Cu(100)

383

surface. The apparent height of the star-like depressions is measured to about 0.03 nm depending strongly on the applied tunneling gap voltage. A closer inspection of Fig. 13a reveals, that the centers of gravity of the 'stars' are not located on regular lattice sites of copper atoms in the first layer, but instead in fourfold-hollow sites. Excluding interstitial positions for Ir, the star-like characteristics have to be caused by iridium atoms located at regular lattice sites below or on top of the surface. The latter has been excluded by the ISS measurements reported above. Another distinct property can be recognized in Fig. 13a; sometimes two iridium atoms in the second layer are coming close in a next neighbor configuration, as a result the imaged overlap of this situation manifests itself in the onset of a stripe formation (visible along the [011] direction e.g. in the lower right comer of Fig. 13a).

Fig. 13: a) STM of Cu(100)-Ir: 0.05 ML deposited at 200 K and subsequently annealed at 650 K (Image size: 5 nm x 5 nm, Ut~p- -0.02 V) b) STM image of the ordered surface alloy: 0.6 ML Ir deposited at 620 K (Image size: 10 nm x 10 nm, UTip- 0.3 V). STM images taken at 300 K. (from ref. [93]). After direct deposition of higher doses of Ir (0.6 ML in Fig. 13b) at elevated temperature, long range ordering occurs which can be seen as well in LEED by exposure of a (2xl) superstructure with two domains. The corresponding STM image (Fig. 13b) exhibits a distinct chain like structure with chains running in the [110] directions. The distance between adjacent chains is measured to 0.5 nm, which is about twice the distance of nearest neighbor Cu atoms and thus in good agreement with the (2xl) LEED superstructure. Successful imaging of the striped structure by STM was just possible in a limited range of tunneling gap

384

voltages indicating electronic effects for the origin to measure the stripes. The best ordering of the chains with domain sizes of about 5 nm (calculated from the full width at half maximum of the LEED superstructure spots) have been reported for 0.5 - 0.6 ML Ir deposition on the 620 K hot Cu(100) surface. By following the discussed experiments, a model for the Ir-Cu(100) surface has been suggested and is presented in Fig. 14. As a matter of fact, a two dimensional epitaxial sub-surface alloy has developed and consists of adjacent chains of Ir and Cu atoms along the [011 ] directions to form an ordered (2xl) periodicity. The Ir-Cu sub-surface layer happens to be buried under a monolayer of copper. Remarkably enough, although the surface crystallography of Cu(100) expresses four-fold symmetry, a two fold symmetric pattern is showing up for the chains of subsurface Ir to resemble the (2xl) superstructure.

Fig. 14: Structure model of the ordered Cu(100)-(2xl)-Ir sub-surface alloy. (from ref.

[93]). Hence, on the first sight the proposed model of an ordered sub-surface alloy might appear somewhat surprising because of three facts: firstly, the large miscibility gap in the bulk phase diagram [90], secondly the formation of a (2xl) periodicity on a quadratic surface lattice and finally the assumed possibility to image the buried layer itself by STM. Stimulated by the experimental findings, the Cu-Ir system has been investigated theoretically by Heinze et al. [102] with the help of ab initio calculations. In a first step, the existence of a Cu-Ir sub-surface alloy has been verified via electronic structure-, total energy- and force- calculations by a full potential augmented plane wave method (FLAPW) in bulk and film geometry [104]. For the determination of the alloy structure the surface near region was modeled by nine layers of Cu and

385

one layer containing Cu and Ir atoms placed on both sides of the film either at the surface, sub-surface or deeper layers. For the low coverage Ir situation, impurities were introduced in a p(2x2) surface cell, whereas for the ordered alloy structure a p(2xl) superstructure was assumed. These surface structures are compared with results from a c(2x2) model [4], which often occurs on fcc (100) surfaces (see also the chapter by Colin Barnes). The theoretical outcome explains well that Ir located at the surface is the most unfavorable configuration. This result is again consistent with the fact that the surface free energy for Ir is higher than for Cu and so the overall energy would be lowered by a Cu surface termination. Additionally, the computation revealed in accordance with the presented STM analysis that Ir located in the sub-surface layer presents indeed the energetically most stable configuration. This has been interpreted in terms of the bonding situation: the bond strength of Cu-Ir is expected to increase with the reduction of nearest neighbors in the CuCu environment. Accordingly, among all Cu atoms, the Cu atoms at the surface form the strongest bonds to Ir atoms and the equilibrium position of Ir is found in the sub-surface layer and thus prevents Ir to segregate into deeper layers. From the calculations it turned out as well, that the p(2xl) chain structure at the sub-surface location is 86 meV per Ir atom more favorable as compared with the c(2x2) array of Ir and Cu atoms, which is basically due to directional forces of the straight d-d hybridization between Ir atoms along the chains. These forces are obviously absent in a c(2x2) situation. For the p(2xl) Cu-Ir structure, an energy increase of 49 meV has been determined before segregation of Ir into deeper layers sets in. Such an energy barrier can evidently be overcome by temperature augmentation. Therefore, the experimentally observed diffusion of Ir at T > 650 K into the bulk (cf. Fig. 11) becomes plausible too. In order to estimate the topographic influence of the Ir-Cu structure on the STM data, additional force calculations have been performed by minimizing the total energy. As a result a buckling Az of the Ir vs. Cu atoms of Az/d = 2.9% of the interlayer distance d has been found, which should give rise to a corrugation amplitude in STM topography of less then 5 pm (protrusions for sub-surface Ir atoms). Evidently, pure topography marked by this small height variation (additionally of wrong direction) cannot explain the measured depressions of 30 pm (Fig. 13). In a next step the possibility to image sub-surface impurities in metal surfaces by STM has been investigated [ 102]. The STM images were calculated for room temperature in the Tersoff and Hamann [105] approximation to determine the tunneling current I(r, U) for a gap voltage U. The local density of states (LDOS) of the sample is expressed in n(rll, z, ~F + ~) [ 106] at the position

386

lateral (rl~) and normal (z) to the surface with the Fermi energy ~F. g U,T (~) represents the difference of the Fermi functions f r at (OF- eU + c) and (eF + ~) [106]. 1(1"11,z, U) oc f g ~v (~) n(rls, z, ~'F + ~) d~" In order to describe the wave functions decaying from a single crystal surface into the vacuum, the FLAPW method has been applied which gave the justification to expand the wave functions into 2D basis functions as symmetrized plane waves parallel to the surface (so called 'star' functions ~b,.) with their corresponding z-dependent 'star' coefficients n~:

n(?'ll, z, ~ = Zrti(z, s ~i(?'ll) By this procedure the corrugation amplitudes Az(z, U, n~, n2) as a function of the tip location above the sample surface has been determined [102, 103]. The corresponding 'STM images' consist basically of the information expressed in the first two coefficients n~ and n2, where ~bl is a constant and does not contribute to the STM corrugation pattern. The height modulation of the probe as a function of the tip position is basically determined by ~b2with the sign and strength being settled by the positive or negative n2 coefficient. Fig. 15 represents the calculated STM images for the impurity (p(2x2)) and chain (p(2xl)) structures for Ir buried by the Cu(100) monolayer surface.

Fig. 15: Calculated STM images at UT~p= 0.6 V, z = 0.5 nm for Ir impurity a) and chain b) located at sub-surface locations. Open (full) circles represent Cu (Ir) atoms, big (small)

circles represent atoms at surface (sub-surface) (from ref. [102]). In Fig. 15, obviously the formation of a star-like structure for the impurity case (Fig. 15a) of a single Ir sub-surface atom and the onset of a chain structure

387 for the (2xt) superstructure (Fig. 15b) can be recognizes as depressions (dark). These theoretical predictions nicely reproduce and rationalize the experimental findings for tunneling in the filled states of the sample. The influence on the sampled LDOS profile by changing the bias voltage U is calculated and the obtained corrugation amplitudes are compiled in Fig. 16. The condition of buried Ir impurities and chains can be recognized in the lower part of Fig. 16. A measurable height variation is expected at bias voltages a r o u n d - 0 . 5 V (occupied states in the sample), the corrugation amplitude of about 0.03 nm manifests itself as a depression and is in excellent agreement with the measured data. On the other hand, from the upper part in Fig. 16 it turns out, that Ir atoms or chains being located in the first surface layer would be imaged as protrusions of comparable amplitude. The experimental STM data obtained at Ir impurities (Fig. 13a) undoubtedly excludes the latter occurrence. As a remainder, all calculations resemble only the influence of the electronic effect on the STM data because it has been established above, that ~the topography is not much altered by the substitution of Cu atoms by Ir.

Fig. 16: Calculated corrugation amplitudes of a tip at z = 0.53 nm, as a function of the applied bias-voltage U for the Ir impurity and the Ir chain. In the insets at the upper and lower right corners, filled (open) circles denote Cu (Ir) atoms. Positive (negative) corrugation amplitudes are defined as imaging the Ir site as a protrusion (depression). (from ref. [102]).

Moreover, the charge density distribution above the surface of the buried Ir atoms could be calculated, and by that inferring a correlation with the actual bond situation. The measured STM corrugation was correlated with the variation of the n2 coefficient in terms of the theory. A charge density contour plot based on the calculation of n2 for the Ir chain in the second layer at an energy of 0.6eV below the Fermi energy of the alloy is given in Fig. 17.

388

The hybridization of the Ir d- states with the Cu sp- states yields in tilted pd- orbitals located at the neighbored Cu atoms. Because of the tilt, the charge density maximum, which is for pure Cu(100) right above the Cu surface atom, shifts to the position above the Cu sub-surface atom. As a consequence, in case of a fcc(100) surface, the charge density depletion above buried Ir atoms in combination with a higher intensity aside of the first layer Cu atoms results in the star-like pattern obtained in the experiment (Fig. 13a) and theory (Fig. 15)

Fig. 17: Cross section along the [100] direction through the charge density of a typical state in the 2D Brillouin zone at E = EF- 0.6 eV for Ir chains at the sub-surface location. White (black) color denotes a high (low) charge density. (from ref. [102]).

In addition, the theoretical investigation of the special situation of Ir on a Cu surface was used to compare with possible sub-surface conditions at other transition metal cases. On the basis of the calculations, the hybridization of the Cu sp- and Ir d-states are expected to be of rather general quality and therefore the prospect to detect buried transition-metal atoms by STM should be valid for other couples too. A number of possible candidates have been suggested [102]. For sure, all of the propositions rely on the assumption that buried layers might be created indeed experimentally, which probably will be difficult because of decent miscibility behavior for some elemental pairs recommended below. With decreasing number of d electrons (Ir, Os, Re, W, Ta) it is expected by theory that the d band energy increases with respect to the Fermi energy and therefore the tunneling barrier of the state seen in STM becomes lower. Therefore the corrugation amplitudes are supposed to increase up to 0.05 nm and the subsurface location of these impurities in Cu should turn up even clearer in STM. Also Rh as an iso-electronic pendant of Ir is expected to yield measurable corrugation. Conversely, larger numbers of d- electrons (Ir, Pt, Au) lead to smaller height differences of less then 0.01 nm. This might be the reason why

389 the STM investigation failed to image the sub-surface growth of Pd (being isoelectronic to Pt) in Cu(ll0) [107]. On the other hand, sub-surface alloy formation has been reported for vanadium on Pd(111) and the position of the sub-surface V atoms forming a (~/3 x ~/3)R30 ~ arrangement could be observed in the STM data as depressions appearing in the Pd layer [ 108].

3.2 Intermixing versus phase separation: Copper on Ir(100)-(5xl) In the preceding section it has been confirmed that intermixing occurs for certain bulk immiscible constituents (A) and (B). The question may arise, whether this phenomenon depends on the preparation sequence to evaporate material (A) on substrate (B) in comparison with (B) on substrate (A). With the materials (A) = Ir and (B) = Cu, the mixing properties have been verified for the first situation and were tested afterwards for the reversed order. Of course intermixing is just one option of the system to react. As an alternative way, a clear-cut (perhaps two dimensional) phase separation between the two elements might happen. Another complication may arise because of the more complex structure of the substrate, to be exact, Ir(100)-(5xl) as compared with Cu(100)(Ix1). Apparently, the aspect of a possible lifting of the surface reconstruction upon Cu deposition has to be considered, also since it is known that even small energy variations in the surface e.g. by temperature increase or gas adsorption already might induce a lifting of the surface reconstruction [ 109].

Fig. 18: a) STM image of the clean Ir(100)-(5xl) surface taken at 300 K (Inset: corresponding LEED pattern, E = 180 eV). Image size: 62.5 nm x 62.5 nm. b) STM of clean Ir(100)-(5x 1) with atomic resolution. Image size: 6.5 nm x 6.5 nm. (from ref. [110]). The characteristic (5xl) reconstruction of the clean surface expresses after careful cleaning [ 110] and has been explained in a model structure by coverage of an fcc(100) surface with a quasi-hexagonal close packed monolayer of the Ir

390 atoms ontop [111, 112]. Due to the quasi-hexagonal packing of the first layer, the density of the top layer has to be 20% higher as compared with the fcc(100)( l x l ) surface. Accordingly, the surface layer is marked by a characteristic height modulation leading to the (5xl) periodicity visible in LEED experiments. In the STM image of Fig. 18a the typical corrugation appears as a stripe pattern with lines running parallel to the [011] directions ('reconstruction lines'). Two quasi-hexagonal domains with an orientation rotated by 90 ~ show up and have been found on terraces as well as separated by step edges at adjacent terraces. In Fig. 18b the stripes are measured with atomic resolution. The typical 'double row' height modulation has been attributed to the two-bridge configuration [110] in agreement with LEED I-V investigations [111, 112] and theoretical predictions [ 113 ].

Fig. 19: a) STM constant-current image after deposition of 0.2 ML Cu on Ir(100)-(5• at 300 K. Image size: 100 nm x 100 nm. b) Side view model of the Ir(100) surface before (upper panel) and after lifting of the (5x l) reconstruction due to deposited Cu atoms. The formation of Ir chains embedded in the Cu layer after lifting of the surface reconstruction is illustrated. (from ref. [ 110]).

Deposition of Cu on the Ir surface leads in ISS immediately to an increase of the peak for He scattering at Cu atoms. It could be concluded, that all of the Cu atoms stay ontop and strict 2D layer growth was found up to a coverage of 0.7 ML [110]. At higher coverage, 3D islands start to grow and can be seen with STM. By knowledge of the surface composition, the growth mechanism was followed up in STM measurements. The initial growth of Cu at room temperature on the reconstructed (5x l) Ir(100) surface appears to be strongly

391

influenced by the reconstruction lines of the (5xl) structure on the individual terraces (Fig. 19a). Cu starts to grow by formation of small chains and islands (marked in Fig. 19a) with a preference along the reconstruction lines. In the STM image of Fig. 19a, essentially three different height levels exhibit, connected with surface areas of Cu islands- (bright) and Ir chains- (bright lines), Ir(100)-(5xl)- (medium gray) and Ir(100)-(lxl)- (dark), respectively. In a more specific examination of the STM topographic images it appears that that the intense chains indeed continue straight on across the bright islands as faint dim lines. Similar depression structures have been already successfully identified above as Ir atoms embedded in the Cu surface (cf. section 3.1). Therefore, the bright lines on the Ir terraces as well as their continuation as darker lines in the Cu islands have been recognized as Ir chains resulting from surplus Ir atoms caused by a lifting of the Ir-(5xl) reconstruction. In a cross section image, a model of the evolution to grow the structure is explained in Fig. 19b. The decomposition proceeds as follows" In a side view the Ir atoms of the reconstructed surface (upper panel) residing in the upper corrugation sites (B, D, H, J, N, P) are marked by gray shading. Indeed among all Ir atoms of the first layer, these marked atoms are expected to have the highest chance to be shifted up in the decomposition mechanism. Consequently, the lifting of the (5xl) reconstruction will lead to the formation of one Ir chain at each (5xl) super cell exposing on a larger scale three distinct distances between the chains of 3, 5 and 7 lattice units (cf. Fig 19b). Actually, Cu deposition on Ir(100)-(5xl) at 300 K induces right away the lifting of the reconstruction. The reconstruction is not only lifted in the Cu islands but the action proceeds onto pure Ir terraces. The 20% Ir surplus atoms pop up onto Ir terraces and into the top Cu layer to remain there as embedded Ir chains. This structure might be viewed as a kind of dilute 2D ordered surface alloy. It turns out that the long Ir atom strings in the 2D Cu matrix behave rather fragile against temperature increase and they mark just a metastable situation of this special alloy layer. Surface annealing at about 1000 K influences not much the surface composition as can be monitored by ISS, AES or XPS. However, the surface structure changes dramatically. Indeed, the different surface energies of Cu and Ir play at this point again a dominant role and may explain the effect: Ir strings in the first layer expose long border lines, hence in order to minimize the length of the rims, the Ir chains start to transform themselves into compact round shaped 2D disks via mass transport within the surface layer. In case of the low coverage deposition of Cu, the conversion can be seen directly in Fig. 20a. Four characteristic surface features (A, B, C, D) show up in Fig. 20a. Upon higher pre-coverage at 300 K some 3D islands of Cu have already developed (cf. fig.9

392 in ref. [110]) and finally after annealing an additional gray level (E) can be noticed in the corresponding STM survey topograph (Fig. 20b).

Fig. 20: STM images of Ir(100)-(5• 1) after deposition of Cu at 300 K and subsequent annealing at 1000 K: a) Cu deposition: 0.3 ML. Image size: 250 nm x 250 nm. b) Cu deposition: 0.9 ML. Image size: 250 nm x 250 nm c) barrier height image, d) corresponding STM topography - Cu deposition: 0.9 ML, image size: 100 nm x 100 nm. (from ref. [110]).

After all, five different surface species have been identified: namely A) as the clean ( 5 x l ) Ir surface layer, B) as the clean ( l x l ) Ir surface layer, C) as the pseudomorph C u ( 1 0 0 ) - ( l x l ) overlayer on unreconstructed Ir(100), D) as embedded Ir islands in a Cu(100)-(1 x 1) matrix and E) as Cu ad-islands on top of such embedded Ir areas. In fact, surface annealing enhanced the effect of phase separation and the weak alloy formation found before in the occurrence of Ir chains in a Cu monolayer at 300 K is completely overruled by the formation of compact separated areas of Cu and Ir content, respectively. In addition, depending on the initial coverage, the bare surfaces of Ir islands

393 introduce the tendency to cover themselves up by a monolayer of Cu, in order to minimize the surface free energy. By performing local barrier height measurements in the usual way [114], evidently a chemical contrast of the Ir and Cu areas has been achieved (Fig. 20c). The island types A) and D) -Ir, C) and E) -Cu are indicated in the barrier height image and can be recognized in the simultaneously recorded STM topograph given in Fig. 20d. A correlation between Ir- (A, D) and Cu- (C, E) areas with the corresponding barrier height image has been established by comparing the parts of high surface barrier (white areas, (A, D)- Ir) and low barrier (dark gray, (C, E)- Cu). Moreover, this association is in good agreement with the trend, that barrier heights of not too small surface areas correlate with the values of the macroscopic work function ~b, (~b CuaOO)= 4.6 eV, ~bI r ( l O O ) - ( l x l ) - 5.5 eV and ~bIraOO)-(5~)= 5.4 eV [91, 115]). Furthermore, from the barrier height image in Fig. 20c it becomes evident, that all Cu islands E) are surrounded by white rings, indicating that the Ir island beneath is not completely covered by the Cu atoms. A possible Smoluchowsky effect at the step edges [116] was excluded by direct comparison with barrier height measurements at Cu islands on Cu(100). The incomplete coverage of Ir islands with copper can be explained by the requirement of energy to generate Cu steps and by surface strain which builds up due to the different lattice parameters of Cu and Ir. Probably, the gain of energy by covering the Ir areas does not completely outweigh the energy expense due to Cu island formation on the strained area in combination with the island border line. Similar depletion rings due to substrate strain have been rePorted for oxygen adsorption on Ru(0001) [ 117]. To recapitulate this part, the effect of intermixing for immiscible constituents depends indeed sensitively on the preparation order. For the described system of Cu on Ir, at room temperature a kind of dilute mixture of Cu and Ir can be assigned, basically as a result from the lifting of the surface reconstruction of the Ir substrate. The related release of 20% Ir surplus atoms is being incorporated as atomic strings in the Cu islands. This surface configuration turns out to be metastable and is completely transformed after sample annealing into phase-separated areas of compact Ir islands in a 2D Cu matrix. Whereas intermixing occurs for Ir on Cu, strict phase separation develops for the reversed system of Cu on Ir. 4. A L L O Y SURFACES AS SUBSTRATES FOR ORDERED SUPERSTRUCTURES

The effect of intermixing and phase separation has been applied in a different approach for the creation of additional ordered heterostructures by

394

adding a third component to the system. Hence, either alloy formation of two materials on a single metal substrate [118] or alternatively, alloy formation ontop of a binary alloy system can be discussed. An easy way to use phase separation has been followed up by oxidation of special alloy surfaces. In particular the NiA1 surfaces are of great popularity for creation of thin A1203 surfaces. These structures serve as substrates for catalytic reactions in form of nanostructured oxide arrays [37, 47, 119] or as thin films for catalyst support [114, 120-127]. Recently Franchy published a comprehensive report on the formation of thin oxide structures on several alloy surfaces [28]. In a further step, the alloy surface itself was taken as a substrate for epitaxial film growth of another metal-film material. Among others, the Cu-Au alloy system turned out to yield promising results. This is primarily because several ordered alloy phases Cu• as function of the composition are known to exist [90] and which can be employed to vary the substrate lattice parameter in the range from 0.3614 nm (pure Cu) to 0.4078 nm (pure Au). Such substrates have been proposed for lattice mismatch tailoring in epitaxial metal film growth to make use of the sequence Cu --~ Cu3Au --~ CuAu --~ CuAu 3 ~ Au. In particular the investigation of magnetism for dimensionally reduced systems has proven that the magnetic properties depend strongly on the film stress and morphology. For example, both, ferromagnetic and antiferromagnetic characteristics have been considered for thin fcc iron films depending on the lattice constant in the film. As a consequence, in order to vary the substrate lattice starting with Cu, also Cu3Au(100 ) surfaces have been utilized for epitaxial film growth of Fe or Ni [71-74]. For Fe deposition, a miscibility gap for Cu and Au occurs for bulk material, whereas Ni seems to be miscible with Cu and Au. As an example for the growth of thin films in the two limits of intermixing or phase separation, the case of vanadium on Cu3Au (100) will be shown below. 4.1 Vanadium on Cu3Au(100 ) For the binary system of bulk vanadium and gold, several ordered alloys are known to be present from the corresponding phase diagram [90]. Obviously, for V on Cu3Au the situation for intermixing is fulfilled. After vanadium evaporation on the Cu3Au (100)-c(2x2) surface at room temperature, both, direct clustering and surface embedding of the vanadium atoms shows up in STM. Simultaneously, the fractional spots in LEED vanish at low coverage and after further exposure the remaining ( l x l ) spots become gradually weaker. In the end, no clear LEED pattern is obtained beyond a vanadium coverage of three monolayers. At this stage a highly disordered and rough V surface can be recognized by STM and ion scattering. On the other hand, annealing initiates

395 directly an ordering process and after appropriate heating at a temperature of about 500 K, the c(2x2) superstructure recovers completely. The induced intermixing and surface segregation has been investigated by ion scattering spectroscopy. The corresponding TOF measurements are presented in Fig. 21. A three-monolayer thick vanadium film was deposited at 190 K. From the ISS data the complete coverage of the substrate by V can be deduced, visible by the lack of the ISS peaks for Cu and Au (upper TOF spectrum in Fig. 21). He~V A _

190K

He ~ Cu I [ ~ I He ~ A u I I I I

220K 270K 320K t-

"*-'

fie

370K

0")

._ '-

420K

0 v.

470K

I

I

I

I

I

I

I I I

I I I

F--t,

..Q

520K 570K

4.0

4.5

I I I 5.0

5.5

6.0

6.5

Time-of-flight [ps]

Fig. 21: Ion scattering of He at a 3 ML thick vanadium film on Cu3Au(100). E0 = 3 keV. V was deposited at 190 K, the film was subsequently annealed for 30s at the indicated temperature. ISS data acquired at 190 K. The positions for single scattering flight times of He at V, Cu and Au are indicated. Scattering angle 180~

After annealing at about 450 K first surface segregation of Au can be detected. Upon annealing around 550 K an alloy of V3Au stoichiometry has formed by vanadium atoms to substitute Cu positions. The related LEED pattern reveals the c(2x2) superstructure indeed demonstrating the development of an ordered V3Au (100)-c(2x2) surface alloy. Annealing at higher temperature, leads successively to the formation of a ternary Cu•215 alloy, with x running

396

from 0 at low temperatures to 3 at a temperature of about 800 K. Annealing the ternary alloy at 800 K for longer periods initiates segregation of the entire vanadium layer into the bulk, by that leaving the bare Cu3Au(100)-c(2x2) surface behind. As a result, V deposition on Cu3Au triggers the formation of a substitutional ordered surface alloy by reason of intermixing and surface segregation which, on the other hand can be stopped completely by precovering the Cu3Au substrate with oxygen as will be shown next. 4.2. Vanadium oxide o n C u 3 A u ( 1 0 0 ) - O An appropriate oxygen treatment of Cu3Au(100)-c(2x2 ) has been already described in section 2.2.2. After O + implantation, a flat C u 3 A u ( 1 0 0 ) - c ( 2 x 2 ) - O surface has been established upon annealing at 650 K. The smooth oxygencopper surface layer acts positively in two ways" firstly, it prevents completely intermixing. Secondly, in contrast to a pure Au or Cu crystal, the C u 3 A u sample may proceed as an oxygen reservoir with sub-surface oxygen stored close to the surface, which might be released in a controlled way via temperature treatment of the sample.

Fig. 22: a) high resolution STM image after deposition of 0.1 ML V on Cu3Au (100)-O at 300 K. Positions of individual V atoms are seen as depressions by chemical contrast. Image size: 10 nm x 10 nm; (UT, = -0.35 V, i = 1.0 nA). b) STM survey for higher coverage of 0.6 ML vanadium. Image size: 100 nm x 100 nm; surface wetting of the film can be seen by strict 2D growth.

Evaporation of small quantities of V onto of the Cu3Au (100)-O surface can be monitored directly in the STM image. As a consequence of a strong chemical contrast, tunneling into the empty states marks the position of V atoms at the surface by dark spots, i.e. strong depressions of an apparent depth of about 0.04 nm are visible in Fig. 22a at a tip voltage o f - 0 . 3 5 V. Evaporation of higher

397

quantities of V leads to the formation of a 2D film resulting in no ordering effect at all, neither visible in STM (Fig. 22b) nor in LEED. Ordering of the V film can be just achieved upon annealing and by that oxidizing the vanadium layer in a controlled way. After the preparation of three different initial Cu3Au-O substrates, distinct VO• layers have been generated. All three initial Cu3Au-O substrates provide the same c(2x2) LEED superstructure, but the modification has been obtained by generating dissimilar contents of sub-surface oxygen (low, medium, large). Depending on the preoxygen contents, three different ordered layers of vanadium oxides have been prepared on Cu3Au-O by oxidizing the room temperature deposited V films through annealing in the suggested manner. Indeed, it was possible to produce flat epitaxial and uniform VOx films [88].

Fig. 23: Vanadium oxide layer obtained by vanadium oxidation at a CuaAu(100)-O substrate with medium oxygen content, a) LEED pattern b) survey STM c) Schematic model of the V203 oxide structure. Large white spheres: oxygen; small dark gray spheres: lower half part of the vanadium double layer; small light gray spheres: upper half part of the vanadium double layer, d) High resolution STM. (from ref. [88]). Starting with the sample of low oxygen content, the vanadium oxide structure that is obtained after vanadium-oxidation by substrate annealing

398 consisted of a quadratic unit cell. A homogeneous film covers the entire surface. By means of SPA-LEED measurements the lattice parameter of the oxide was determined to result with 0.28 nm in a slightly larger distance as compared with the 0.26 nm of the Cu3Au substrate. From the knowledge of the crystallographic structure (STM, SPA-LEED) and the rather low oxygen content (AES), this oxide configuration has been correlated with a layer of vanadium monoxide VO(100) carrying vanadium in the V 2+oxidation state. Another oxidation state (V 3+) of vanadium in the VOx film was produced by employing the Cu3Au-O substrate with medium oxygen content. In variance to the quadratic lattice structure of vanadium monoxide, the LEED superstructure of this specific VOx phase give rise to a ring type diffraction pattern (Fig. 23a), which can be best explained by the occurrence of two domains of a hexagonal lattice structure. Indeed the high resolution STM image of Fig. 23d displays a single type of the domains measured at one of the large flat oxide terraces (Fig. 23b). On the basis of the hexagonal structure in combination with the measured lattice constant of 0.522nm, a good structure fit was obtained under the assumption that a V203 type oxide has developed. As a result of the epitaxial relationship, it was concluded that the oxide film forms a V203(0001) surface plane, indeed quite comparable with epitaxial Cr203(0001 ) [128] films grown on Cr(110).

Fig. 24" Vanadium oxide layer obtained by vanadium oxidation at a Cu3Au(100)-O substrate with high oxygen content, a) STM survey showing the different domains, b) Schematic model of the VO2 oxide structure. Large white spheres: oxygen, small dark spheres: vanadium. Unit cell is indicated in b) and c). High resolution STM in c) and d) with different tip configurations. (from ref. [88]).

399 Vanadium in the V 4+ oxidation state has been created too in a two dimensional film configuration. Here, vanadium has been oxidized on the substrate with high oxygen content. This type of vanadium oxide manifests itself in a smooth thin epitaxial layer where several domains can be recognized in a striped oxide pattern (Fig. 24a). The related LEED pattern had changed from a hexagon to a superstructure consisting of 90 ~ angles between the unit cell vectors. Indeed the high-resolution STM data resemble well the rectangular geometry (Fig. 24c and d). The unit cell with the dimensions of 0.264 nm x 0.529 nm is indicated in the images. A model of the vanadium oxide layer is displayed in Fig. 24b and would be in good agreement with an oxide of VO2 stoichiometry. As can be seen in the STM image, vanadium and oxygen atoms can be imaged with slightly different contrast in the gray scale image, the apparent corrugation varies in fact with different temporary tip configurations (Fig. 24c and d). In conclusion, it can be noted that Cu3Au(100)-0 is preferably suited as a metal alloy substrate for growing metal oxides. Phase separation of the substrate and the grown ordered oxide layer is complete. Depending on the preoxygen content at CuaAu(100), the amount of vanadium deposition and annealing temperature, three different epitaxial layers of vanadium oxides have been prepared on the oxygen treated Cu3Au substrate. Following the order of the oxidation states of vanadium, the production of two dimensionally ordered oxide phases ofVO(100), V203(0001) and g o 2 stoichiometry was reported. 5. S U M M A R Y

Ordered metal alloy systems might expose profound different surface characteristics even though consisting of the same elemental composition in the bulk. Intermixing or phase separation is correlated with the surface composition and structure. Differences appear associated by the influence of the free surface energy with segregation and surface ordering. Some prospects have been illustrated at specific metal alloy surfaces. A number of dissimilar surface compositions and structures develop at the NiA1 ordered bulk alloy by preparation dependent effects. Completely different chemical behavior against oxygen adsorption and dissociation has been found for two Cu3Au surfaces, the (100) and (110) plane, consisting of the same surface composition. The (100) surface with Cu atoms surrounded by Au atoms turns up non-reactive alike Au. On the other hand, the (110) surface with Cu chains in the first layer acts similar to a Cu(110) surface. Intermixing or phase separation can be manipulated at Cu3Au too. Upon vanadium deposition on the bare alloy surface, strong intermixing and alloy formation towards a V3Au

400 surface occurs. On the other hand, oxygen at Cu3Au prevents completely intermixing and ordered VOx layers can be grown on top of the Cu3Au (100)-O surface. Finally intermixing has been demonstrated as well for a bulk immiscible system like Ir deposited at Cu(100). After heat treatment an ordered two-dimensional sub-surface alloy has been produced. The position of subsurface Ir atoms could be imaged by STM via electronic effects. Whereas intermixing occurs for Ir on Cu, strict phase separation takes place for the reversed system of Cu on Ir showing neither intermixing nor segregation. As a matter of fact, opposed to solids with single elemental composition, the class of binary alloys obviously resembles an additional freedom to produce from the same bulk material various new surface configurations marked by different chemical reactive states and even completely new surface alloys.

ACKNOWLEDGEMENTS It is a pleasure to acknowledge the excellent cooperation and helpful discussions with Rail-Peter Blum, Dirk Ahlbehrendt, Gerhard Gilarowski and Ralf Spitzl. The work was financed in part by the German Council of Research DFG through the SFB 290 and SFB 546.

REFERENCES H. Brune, Surf.Sci.Rep. 31 (1998) 121. [1] G.L. Kellogg, Surf.Sci.Rep. 21 (1994) 1. [2] J.A. Rodriges, Surf.Sci.Rep. 24 (1996) 223. [3] U. Bardi, Rep.Progr.Phys. 57 (1994) 939. [4] B. Sch6nfeld, Progr.Mat.Sci. 44 (1999) 435. [5] C. Laubschat, K. Mi~ller, K. Wandelt, Physics and Chemistry of Alloy Surfaces, [6]

[7] [8] [9] [10] [11] [12]

[13] [14] [15] [16] [17]

[18] [19]

Surface Review and Letters, Vol. 3, World Scientific, Singapore 1996. M. Polak, L. Rubinovich, Surf.Sci.Rep. 38 (2000) 127. J. A. Rodriguez, Surf.Sci.Rep. 24 (1996) 223. C. T. Campell, Annu.Rev.Phys.Chem. 41 (1990) 775. Interfacial Electrochemistry, Marcel Dekker, Inc., New York 1999. W. Polewska, M.R. Voigt, O.M. Magnussen, R. J. Behm, J. Phys. Chem. B 103 (1999) 10440. M. Wilms, M. Ki'uft, G. Bermes, K. Wandelt, Rev. Sci. Instrum. 70 (1999) 3641. M. Lennartz, M. Arenz, C. Stuhlmann, K. Wandelt, Surf. Sci. 461 (2000) 98. M. Henzler, in H. Ibach (Ed.): Topics in Current Physics, Vol. 4, Springer, Berlin 1977. G. Binnig, H. Rohrer, Ch. Gerber, E. Weibe, Phys.Rev.Lett. 49 (1982) 57. R.J. Behm, N. Garcia, H. Rohrer, NATO ASI SERIES E: Applied Science, Vol. Vol.184, Kluwer Academic Publishers, Dordrecht NL 1990. F. Besenbacher, I. Stensgaard, in D.A.King, D.P.Woodruff (Eds.): The Chemical Physics of Solid Surfaces, Vol. 7, Elsevier 1994. F. Besenbacher, Rep.Prog.Phys 59 (1996) 1737. H. Neddermeyer, Rep.Prog.Phys. 59 (1996) 701.

401

[20] [21] [22] [23] [24] [25] [26] [27] [28] [29]

[30] [31] [32] [33] [34] [35] [36] [37]

[38] [39]

[40]

[41] [42]

[43] [44]

[45] [46] [47] [48] [49]

[50] [51] [52]

[53] [54]

[55] [56] [57]

[58] [59] [60]

P. Varga, M. Schmid, Appl. Surf. Sci 141 (1999) 287. H. Niehus, W. Heiland, E. Taglauer, Surface Science Reports 17 (1993) 213. J.W. Rabalais, Low energy ion-surface interactions, Wiley 1994. H. Niehus, Surf.Sci. 166 (1986). H. Niehus, R. Spitzl, Surf.Interf.Anal. 17 (1991) 287. H. Niehus, Appl.Phys.A53 388 (1991) 388. A.T. Hanbicki, A.P. Baddorf, E.W. Plummer, B. Hammer, M. Scheffier, Surf.Sci. 331 (1995) 811. R.M. Jaeger, H. Kuhlenbeck, H.-J. Freund, M. Wuttig, W. Hoffmann, R. Franchy, H. Ibach, Surf.Sci. 259 (1991) 235. R. Franchy, Surf.Sci.Report 38 (2000) 195. H.L. Davis, J. R. Noonan, Phys.Rev.Lett. 54 (1985) 566. D.R. Mullins, S. H. Overbury, Surf.Sci. 199 (1988) 141. H.-J. Freund, H. Kuhlenbeck, V. Staemmler, Rep.Prog.Phys. 59 (1996) 283. H. Niehus, J.Vac.Sci.Technol. A5 (1987) 751. Y. G. Shen, D. J. O'Connor, K. Wandelt, Surf.Sci 410 (1998) 1. M. T. Robinson, I. M. Torrens, Phys. Rev. B9 (1974) 5008. V. Bykov, C. Kim, M.M. Sung, K.J. Boyd, S.S. Toderov, J. W. Rabalais, Nucl. Instr. Meth, Phys. Res. B 114 (1996) 371. http'//asp 2.p hysi k. hu-ber l in. de/. R.P. Blum, D. Ahlbehrendt, H. Niehus, Surf.Sci. 366 (1996) 107. J.P. Biersack, L. G. Haggemark, Nucl. Instr. Meth. 174 (1980) 257. D. J. O'Connor, J. P. Biersack, Nucl. Instr. Meth. B 15 (1986) 14. J.F. Ziegler, J.P. Biersack, U. Littmark, The Stopping and Range in Solids, Vol. Vol. 1, Pergamon, New York 1985. H. Niehus, Nucl.Instr.Meth. B 33 (1988) 876. S.H. Overbury, D.R. Mullins, J. F. Wendelken, Surf.Sci. 236 (1990) 122. J.R. Noonan, H. L. Davis, J.Vac.Sci.Technol. A6 (1988) 722. H. Niehus, W. Raunau, K. Besocke, R. Spitzl, G. Comsa, Surf.Sci. 225 (1990) L8. P. Gassman, R. Franchy, H. Ibach, Surf.Sci. 319 (1994) 95. J.P. Roux, H. J. Grabke, Appl.Surf.Sci. 68 (1993) 49. R.P. Blum, H. Niehus, Appl.Phys. A66 (1998) 529. Y. Cao, E. H. Conrad, Phys.Rev.Lett. 65 (1990) 2808. L. Yang, T. S. Rahmann, Phys.Rev.Lett. 67 (1991) 2327. Y. Beaudet, L.J. Lewis, M. Perrson, Phys.Rev. (1993) 4127. P. von Blankenhagen, W. Schommers, V. Voegel, J.Vac.Sci.Technol. A5 (1987) 649. Y. G. Shen, D. J. O'Connor, K. Wandelt, Surf.Sci 406 (1998) 23. U. Schneider, G.R. Castro, K. Wandelt, Surf.Sci 287/288 (1993) 146. S. Schomann, E. Taglauer, Surf.Rev.Lett. 3 (1996) 1823. M. Aschoff, S. Speller, J. Kunze, W. Heiland, E. Platzgummer, M. Schmed, P. Varga, B. Baretzky, Surf.Sci.Lett. 415 (1998) L 1051. Y.G. Chen, D.J. O'Connor, K. Wandelt, R. J. MacDonald, Surf.Sci. 328 (1995) 21. V.S.Sundaram, R.S.Alben, W.D.Robertson, Surf.Sci. 46 (1974) 653. S.F. Alverado, M. Campagna, A. Fattah, W. Uelhoff, Z.Phys. B66 (1987) 103. H. Dosch, L. Mail~inder, H. Reichelt, J. Peisl, R.L. Johnson, Phys.Rev. B43 (1991) 13172. H. Dosch, Appl.Phys. A61 (1995) 475.

402 [61] [62] [63] [64] [65] [66] [67]

[68] [69] [70] [711 [72] [73] [74] [75] [76] [77]

[78] [79]

[80] [81] [82]

[83] [84] [85] [86] [87]

[88] [89] [90] [91] [92] [93] [94] [95] [96] [97]

[98]

H. Reichert, H. Dosch, P.J. Eng, I. K. Robinson, Europhys.Lett. 53 (2001) 570. K.F. Ludwig, G.B. Stephenson, J.L. Jordan-Sweet, Y. J. Mainville, S. Yang, M. Sutton, Phys.Rev.Lett. 61 (1988) 1859. A. Stuck, J. Osterwalder, L. Schlapbach, H.C. Poon, Surf.Sci. 251/252 (1991) 670. D.H. Oh, H.J. Kang, K.H. Chae, C.N. Whang, B.V. King, D.J. O'Connor, D.W. Moon, Surf.Sci. 477 (2001) L289. B. Gans, P.A. Knipp, D.D. Koleske, S.J. Sibener, Surf.Sci. 264 (1992) 81. W.E. Wallace, G. J. Ackland, Surf.Sci.Lett. 275 (1992) L685. R. Courths, H. Wern, U. Hau, B. Cord, V. Bachelier, S. Htifner, J.Phys.:Met.Phys 14 (184) 1559. S.V. Halilov, H.Gollisch, E. Tamura, R. Feder, J.Phys: Condens.Matter 5 (1993) 4711. Ch.E. Lekka, N.I. Papanicolaou, G. A. Evangelakis, Surf.Sci. 479 (2001) 287. J.A. Con Foo, A.P.J. Stampfl, B. Mattem, A. Ziegler, M. Hollering, L. Ley, J.D. Riley, R. C. G. Lecky, Sol.State Comm. 107 (1998) 385. M.T. Lin, J. Shen, W. Kuch, H. Jenniches, M. Klaua, C.M. Schneider, J. Kirschner, Surf.Sci. 410 (1998) 290. W.A.A. Macedo, F. Sirotti, A. Schatz, D. Guarisco, G. Panaccione, G. Rossi, J.Magn.Mat. 177 (1998) 1262. F. Bruno, D. Cvetko, L. Floreano, R. Gotter, C. Mannori, L. Mattera, R. Moroni, S. Prandi, S. Terreni, A. Verdini, M. Canepa, Appl.Surf.Sci. 162 (2000) 340. A. Braun, B. Feldmann, M. Wuttig, J.Magn.Mat. 171 (1997) 16. E.G. McRae, T.M. Buck, R.A. Malic, W. E. Wallace, Surf.Sci.Lett. 238 (1990) L481. H. Niehus, Th. Baumann, M. Voetz, K. Morgenstem, Surf.Rev.Lett. 3 (1996) 1899. Y. Huang, J.M. Cowley, Surf.Sci. 285 (1993) 42. H. Over, G. Gilarowski, H. Niehus, Surf.Sci. 381 (1997) L619. T.M. Buck, G.H. Wheatley, L. Marchut, Phys.Rev.Lett. 51 (1983) 43. T.M. Buck, G.H. Wheatley, D. P. Jackson, Nucl.Instr.Phys. 218 (1983) 257. K. Morgenstern, H. Niehus, G. Comsa, Surf.Sci. 338 (1995) 1. J. B. Pendry, J.Phys. C 17 (1980) 937. A. R. Miedema, Z.Metallk. 69 (1979) 455. D.J. Coulman, J. Wintterlin, R.J. Behm, G. Ertl, Phys.Rev.Lett. 64 (1990) 1761. K. Kern, H. Niehus, A. Schatz, P. Zeppenfeld, J. Goerge, G. Comsa, Phys.Rev.Lett. 67 (1991) 855. H. Niehus, C. Achete, Surf. Sci. 289 (1993) 19. Ch.E. Lekka, N. I. Papanicolaou, G. A. Evangelakis, Surf.Sci. 488 (2001) 269. H. Niehus, R.-P. Blum, D. Ahlbehrendt, Phys. Stat. Sol. a 187 (2001) 151. L. Houssiau, P. Bertrand, Surf. Sci. 352 (1996) 978. T. B. Massalski,, ASM International, Ohio 1990. H.L. Skriver, N. M. Rosengaard, Phys. Rev. B 46 (1992) 7157. E. Bauer, Appl. Surf. Sci. 11/12 (1982) 479. G. Gilarowski, H. Niehus, Surf. Sci. 436 (1999) 107. G. Gilarowski, H. Niehus, Phys. Stat. Sol. a 173 (1999) 159. J. E. P. J. de la Figuera, R. Miranda, Surf. Sci. 307-309 (1994) 538. A. Rabe, N. Memmel, A. Steltenpohl, T. Fauster, Phys. Rev. Lett. 73 (1994) 2728. A. Brodde, H. Neddermeyer, Surf. Sci. 287/288 (1993) 988. Th. Detzel, N. Memmel, T. Fauster, Surf. Sci. 293 (1993) 227.

403 [99] [ 100] [101 ] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [ 114] [115] [116] [117] [118] [119] [120] [121] [122] [123] [124] [125] [126] [127] [128]

S.-L. Chang, J.-M. Wen, P.A. Thiel, S. Gfinther, J.A. Meyer, R. J. Behm, Phys. Rev. B 53 (1996) 13747. P.J. Schmitz, W.-Y. Leung, G.W. Graham, P. A. Thiel, Phys. Rev. B 40 (1989) 11477. A. Christensen, A.V. Ruban, P. Stoltze, K.W. Jacobson, H.L. Skriver, J.K. Norskov, F. Besenbacher, Phys. Rev. B 56 (1997) 5822. S. Heinze, R. Abt, S. Blt~gel, G. Gilarowski, H. Niehus, Phys.Rev.Lett. 83 (1999) 4808. S. Heinze, S. Blfigel, R. Pascal, M. Bode, R. Wiesendanger, Phys.Rev. B 58 (1998) 16432. E. Wimmer, H. Krakauer, M. Weinert, A.J. Freeman, Phys. Rev. B24 (1981) 864. J. Tersoff, D. R. Hamann, Phys. Rev. Lett. 50 (1983) 1998. S. Heinze, M. Bode, A. Kubetzka, O. Pietzsch, X. Nie, S. Blfigel, R. Wiesendanger, Science 288 (2000) 1805. P.W. Murray, S. Thorshaug, I. Stensgaard, F. Besenbacher, E. Laegsaard, A.V. Ruban, K.W. Jacobsen, G. Kopodakis, H.L. Skriver, Phys.Rev. B 55 (1997) 1380. Ch. Konvicka, Y. Jeanvoine, E. Lundgren, G. Kresse, M. Schmid, J. Hafner, P. Varga, Surf.Sci. 463 (2000) 199. T. Ali, A.V. Walker, B. K16tzer, D. A. King, Surf. Sci. 414 (1998) 204. G. Gilarowski, J. Mendez, H. Niehus, Surf. Sci. 448 (2000) 290. E. Lang, K. Mfiller, K. Heinz, Surf. Sci. 127 (1983) 347. N. Bickel, K. Heinz, Surf. Sci. 163 (1985) 435. Q. Ge, D.A. King, N. Marzari, M. C. Payne, Surf. Sci. 418 (1998)529. J. Mendez, H. Niehus, Appl.Surf.Sci. 142 (1999) 152. T.N. Rhodin, G. Brod6n, Surf. Sci. 60 (1976) 466. J.F. Jia, K. Inoue, Y. Hasegawa, W.S. Yang, T. Sakurai, Phys. Rev. B 58 (1998) 1193. M. Gsell, P. Jakob, D. Menzel, Science 280 (1998) 717. M. Schick, G. Ceballos, Th. Pelzer, J. Sch~fer, G. Rangelov, J. Stober, K. Wandelt, J.Vac.Sci.Technol. A12 (1994) 1795. R.-P. Blum, D. Ahlbehrendt, H. Niehus, Surf.Sci. 396 (1998) 176. M. Che, C. O. Benett, Adv.Catal. 20 (1998) 153. C.R. Henry, Surf.Sci.Rep. 31 (1998) 231. D.W. Goodmann, Surf.Sci. 299/300 (1994) 837. C.T. Campell, Surf.Sci.Rep. 27 (1997) 1. M. B~iumer, H.-J. Freund, Prog.Surf.Sci. 61 (1999) 127. D.A. King, D. P. Woodruff, The Chemical Physics of Solid Surfaces and Heterogeneous Catalysis, Vol. 1 - 4, Elsevier, Amsterdam 1990. J. Libuda, F. Winkelmann, M. Bgumer, H.-J. Freund, Th. Bertrams, H. Neddermeyer, K. Mt~ller, Surf.Sci. 318 (1994) 61. A. Stierle, V. Formoso, F. Comin, G. Schmitz, R. Franchy, ESRF Newslett. 31 (1998) 25. F. Rohr, M. B~iumer, H.-J. Freund, J.A. Mejias, V. Staemmler, S. Mt~ller, L. Hammer, K. Heinz, Surf. Sci. 372 (1997) L291.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces 404

D.P. Woodruff, (Editor)

Chapter 11

Surface structure and catalytic reactivity of palladium overlayers for 1,3-butadiene hydrogenation J.C. Bertolini and Y. Jugnet Institut de Recherches sur la Catalyse - C N R S 2 avenue Albert Einstein, F-69626 Villeurbanne Cedex, France

1. I N T R O D U C T I O N Transition metals are known to have good catalytic properties for many reactions. Their chemical properties and consequently their catalytic properties (activity, selectivity and stability) can be strongly modified when alloyed to a second element [1, 2]. The behaviour of binary alloys with respect to catalysis is generally interpreted in terms of either the geometric <<ensemble effecb>, associated with the number of nearest neighbours of a given element for the catalytic reaction to occur [3-5] or to the electronic <
405

-

-

-

the difference of surface energy of the two elements: the system tends to minimize its free energy by pushing at the surface the element of lower surface tension, the elastic strain relaxation which tends to expel the minority element mainly when it is the biggest one, the mixing enthalpy which tends to amplify the influence of the two previously invoked effects when it is positive, or on the contrary, tends to moderate this effect when it is negative ; in such a case the layer by layer composition has a tendancy to oscillate, as shown firstly from a LEED intensity analysis on Pt-Ni(111) [8].

The ability to get well characterised thin layers (from a fraction of monolayer up to several atomic layers) of a given metal on another metal by atomic beam deposition has developed strongly over the past few years. Depending on the the relative values of interface energies and of surface free energies, different growth modes can appear once the thermodynamic equilibrium is reached [9, 10] : a layer by layer Franck van der Merwe growth, a three-dimensional Volmer Weber growth or a layer by layer growth followed by island formation (Stranski Krastanov growth). In fact, the growth depends on a number of kinetic parameters, such as deposit rate, temperature . . . . Therefore, the system can be far from equilibrium. For both kinds of materials (single crystal alloys and metallic overlayers on a different metal) strained oveflayers are formed when the lattice constant of the surface element (segregated or deposited) is different from that of the bulk ; the overlayer undergoes compressive (or tensile) strains when the atomic radius of the element constituting the oveflayer is larger (or smaller) than that of the bulk one. The surface adlayers can then either adopt a pseudomorphic geometry or reconstruct in order to relax this stress [ 11-14]. At this point, it is important to keep in mind that surface stress can also be induced by adsorbed molecules [12]. The concept of surface and interface stress has been widely used for several years to generate thin layers of specific magnetic, electronic [15], ... properties. It starts now to be also used as a tool for tuning the chemical reactivity of surfaces. For example, stressed Pd deposited on Mo(110), Ta(110) or Nb(110) [16] has electronic and chemical properties very different from those of pure Pd; it is then difficult to separate the respective contribution of the strain itself from that of the direct chemical interaction, which is important when the two elements are far away in the periodic table. In the case of Pd deposited on Re(0001), where both metals have the same atomic radius, a pseudomorphic overlayer is obtained, showing again chemical properties different from Pd(111) [17]. A way to limit the direct ligand chemical effect is to associate metals located close together in the periodic table, i.e. having similar electronic

406

properties. Moreover, investigation of multilayers can further limit such an effect. Experimentally, direct observation of the influence of surface strain on the chemical reactivity is not easy. Gsell et al. [18] succeeded in measuring, by scanning tunneling microscopy (STM) on Ru(001), the influence of a local strain on the bonding strength of adsorbed oxygen and carbon monoxide. The strain was created by subsurface argon bubbles, generated during ion sputtering, which deform locally the surface layer. The effect of tensile and compressive strain on the chemisorptive properties of metal surfaces has been the subject of a few recent theoretical studies [19-21 ]. A relationship between the lattice expansion, the up-shift of the metal d-states and the CO adsorption energy, wich results in an easier CO dissociation, on Ru(0001) has been established [19]. A lattice compression induces opposite effects. Similarly, Sautet has shown that ethene chemisorption is reduced on compressed Pd(111) [20]. In a recent paper, Pallassama and Neurock studied, by DFT-GCA periodic slab calculations, the chemisorption, hydrogenation and dehydrogenation of ethylene on pseudomorphic Pd layers deposited on various close-packed substrates Re(0001), Ru(0001), Au(111), whose atomic radii are different from that of pure Pd(111) [21]. They correlate the trends in ethylene chemisorption energy and the activation barrier for hydrogenation (and dehydrogenation) to the intrinsic electronic properties of the Pd adlayer, and more precisely to the position of the d-band with respect to the Fermi level, which varies with the strain imposed on the Pd adatoms. A compressive strain would induce a diminution of the ethylene binding energy, and an increase in the hydrogenation rate, as a consequence of an increase of the mean d-band energy. In conclusion, the chemical properties of alloy surfaces or surface alloys differ from those of pure metals. This effect will depend on the way the surface atoms behave relatively to the surface strain. Interestingly this makes possible the use of strain to manipulate the reactivity of a given metal. In this paper we present results related to the atomic structure and catalytic properties of Pd overlayers on various substrates. A reaction has been chosen to test the catalytic properties of these systems, it is the 1,3-butadiene hydrogenation, a reaction for which Pd is known to be the best catalyst. In the following, after a short description of the experimental approach, the 1,3butadiene hydrogenation reaction and the specific properties of Pd for this reaction will be presented. Then the reactivity of several Pd overlayers obtained either by surface segregation in Pd-based alloys or by atomic beam deposition on a metal will be investigated and discussed in terms of structure, composition related to surface segregation and surface stress. The influence of the surface orientation of the substrate will be discussed.

407

2. EXPERIMENTAL APPROACH

In this part, the surface science methodology used in fundamental catalytic research on metal single crystals will be briefly reviewed. This methodology relies firstly on a high degree of characterisation of the studied surfaces under ultra high vacuum (UHV) conditions (with the combined use of many complementary techniques for elemental, structural and electronic properties determination) and secondly to the possibility of carrying out catalytic reactions on samples of very small active area under realistic conditions, i.e. approaching those of technical applications. Indeed, the relevance of single-crystal studies for modeling high surface area supported catalysts has been demonstrated for many catalytic reactions. A schematic representation of a typical set-up for such studies is given in Fig. 1. It gathers several chambers devoted to sample preparation, sample characterisation and reaction. Auger electron spectroscopy (AES) and/or X-ray photoelectron spectrosocpy (XPS) are the most used techniques for elemental quantification, since they measure low/medium kinetic energy electrons which are characteristic of the chromatography yr ep.tum (syringe)

LEED, STM li!ii:.i:.il XPS, AES [~ii:i[ L E I S Evaporation cell Annealing

alve II II .......

-iui I

"~ Gas

inlet

iI

======================= 'e

Ion gun transferprobe

9o

QMS []~Q HREELS ~ . ~

Reactor [

Sample characterisation chamber

Fast entry lock

Sample preparation chamber

Fig. 1. Scheme of a typical combined UHV system with an atmospheric pressure cell for catalytic measurements.

408 elements present in an in-depth limited surface region. The surface sensitivity of the electronic spectroscopies is due to the small mean free path (0.5 - 4 nm) of low/medium energy electrons in solids. In fact, electron spectroscopies probe a depth corresponding to a few atomic layers. A good understanding of the active sites of alloy surfaces or surface alloys needs a good knowledge of the composition and geometric structure of the outer layer itself and of the 3-4 sublayers which may influence the electronic, and hence the chemical properties of the considered material. Low energy ion scattering spectroscopy (LEIS) is a very suitable technique to determine the surface composition of the topmost layer [22-23]. Nevertheless, the absolute quantification of elements needs the use of standards to get the relative sensitivity factors. With medium energy ion scattering spectroscopy (MEIS) complementary informations on composition and structure of the surface and subsurface layers are obtained [25, 26]. Low energy electron diffraction (LEED) and scanning tunneling microscopy (STM) are to date the most frequently used for surface structure investigation. Only a qualitative analysis of the LEED diffraction spots is the most frequently done, but a more sophisticated I,V-LEED analysis is needed to have a 3-4 layers indepth structure determination [8, 27]. Tools for studying temperature dependent chemisorptive properties and surface reactions on metallic single crystal surfaces are frequently added in the UHV chambers. Among them thermal desorption spectroscopy (TDS), vibrational high resolution electron energy loss and infrared reflexion absorption spectroscopies (HREELS and IRRAS), nearedge X-ray absorption spectrocopy (XANES) using synchrotron radiation facilities, ... are the most commonly used. Besides experimental techniques, a lot of calculations are done now through very sophisticated ab initio methods, like density functional theory (DFT). After elaboration and/or cleaning, samples have to be transferred under UHV either to the characterisation chamber to be analysed or into the reactor for catalytic studies. In our design the reactor can be isolated with the help of an all metal UHV valve in order to do reactions under static conditions. Reactants can be introduced at a chosen pressure and the products of the reaction are analysed by mass spectrometry (MS) through a leak valve as a function of time. In addition, during the reaction, a small volume of the reactants and products can be isolated and sampled with a syringe (1 ml) through a nipple to be analysed by gas chromatography in a separate set up. The cleaning of metal and alloy single crystals under UHV is obtained by repeated cycles of ion bombardments, with argon ions of a few keV energy, and annealings. The annealing temperature depends on the metal or alloy under investigation. For deposits, one has to keep in mind that the surface morphology is dictated by the difffusion of adatoms and by the density of defects on the substrate surface.

409 The resuts described in the following correspond to Pd evaporated at a rate of about 2.1014 at.cm-2.min -1 on a sample maintained at room temperature. Whatever the system, the monolayer will be defined with respect to the number of metal atoms in the outer plane of the metallic substrate. The calibration of the evaporation rate of the source and/or the quantification of adlayers following metal on metal deposition is not straightforward. A good way to do it is to combine quantification results from in-situ techniques such as AES or XPS and a further absolute quantification by Rutherford Backscattering Spectroscopy (RBS). For absolute quantification, it is better to use polycrystalline samples in order to avoid any channeling effects during RBS measurements.

3. T H E 1,3 - B U T A D I E N E

HYDROGENATION

REACTION

This reaction is an example of consecutive reactions following a rateau scheme: Butadiene (gas)

Butenes (gas)

~ bdiene

~~ bbutenes Butenes (ads)

Butadiene (ads) ~,'Y///////////////////A

kdiene

~/////////////////////h

utenes

Butane (gas)

The selectivity into butenes, defined by the ratio [butenes] / [butenes + butane] is controlled by two parameters. One is kinetic and measures the butene hydrogenation rate relative to the butene desorption rate. The other one is thermodynamic and depends on the relative adsorption coefficients of butadiene and butenes. It will fix the relative coverages in butadiene and butenes on the surface as a function of their relative pressures. In practice the second parameter, the thermodynamical one, turns out to be preponderant and then fixes the selectivity into butenes as a function of the conversion percentage. A better selectivity will be observed if butadiene is more strongly adsorbed than butene, or in other words, if the ratio between adsorption coefficients bdiene/bbutenesis high. As an example, we show in Fig. 2 a comparison between palladium, a catalyst very selective into butenes even at large conversion, and platinum which is a second rate catalyst for this reaction since it is ten times less active and its selectivity into butenes is only slightly more than 50% even at low conversion [28, 29]. Measurements of competitive hydrogenation of butadiene into butenes on palladium and platinum catalysts supported on silica [30] allowed the determination of the relative adsorption coefficients of dienes and olefins.

410

100

9 ~,3-butadlene

i x

100

,

80

80

I-.. 60

60

E~ a 4O

0n"

13. 20 0

.

butenes _

I

,

,'

,'

X

I

\

0

i

I

20

i

\ l

40

l" 9

\ /

I

butane

60

9

%

.," \

~.

.

/

bute~e~,~ ,.~..

40

"

i

1,3-butadiene butane /.,'

l

80

CONVERSION %

, ,

20 ~ 100

0

t

t

9 j

I

. . s %

.::..,

L.-f".

0

~

S

20

,

\

,

40

,

,

60

, 80

100

CONVERSION %

Fig.2. Butenes and butane production as a function of butadiene conversion on Pd(111) (left side) and on Pt(111) (right side) at 300K and in a large excess of hydrogen [28,29].

Insofar, as the direct measurement of the competition between butadiene and butenes is not possible, butene being a reaction product, it is necessary to run two independent competitive hydrogenation experiments such as butadiene / propene and propene / butene from which it becomes possible to determine adsorption coefficient ratio bdiene/bbutene s. These values are reported in Table 1. From the measured intrinsic relative reaction rates Vdiene/Wbutenes, the rate constant ratio kdiene/kbutene can be calculated. The analysis of the adsorption coefficients shows that under reactive mixture, butadiene will be the main product on the Pd surface while both butadiene and butene will compete on the Pt surface. These observations explain the high selectivity of Pd as compared to Pt.

Table 1 Adsorption coefficients, reaction rate and rate constants measured on Pd/SiO2 and Pt/SiO2 during 1,3-butadiene hydrogenation reaction [30]. Catalyst

bdiene

bpropene

bpropene

bbutenes

bdiene bbutenes

Vdiene Vbutenes

kdiene kbutenes

Pd/SiO2

12

1

12

3.9

0.33

Pt/SiO2

1.9

0.5

0.95

0.75

0.8

411

Chemisorption studies of butadiene and butene on Pt(111) and Pd(111) [31] confirm also this assertion. In Fig. 3 are reported the XANES spectra measured under grazing incidence at the C K-edge following adsorption of butadiene at 300K on Pt(111) and Pd(111). The multilayer spectrum measured at 95K is reported for ease of comparison. It shows one main structure at 285 eV corresponding to the C ls ~ re1* transition (Fig. 3a). The splitting observed on this structure (about 0.66 eV) mirrors the chemical shift, measured by XPS on C ls, between the inequivalent carbons in the molecule. A second transition shifted by 2.7 eV corresponds to a C l s --~ ~2" transition, the ~1"-~2" splitting being due to a degeneracy lifting associated to n conjugation. The other structures appearing at around 289, 296 and 302 eV correspond to transitions from C ls to (a+TZ)*cn, a*C_C and Cy*c=c levels respectively. Marked differences are observed between Pd(111) (Fig. 3b) and Pt(111) (Fig. 3c). Although smaller, the ~Z*l - ~z*2 splitting is still observed in the case of Pd(111) indicating that n conjugation is conserved in the chemisorbed state identified as a di-~z state. At the opposite, on Pt(111) the C ls --~ 7t* splitting is no longer observed, instead a broad band dominates the spectrum centered at 285 eV. The chemisorbed state is associated, in this case, to a di-a state.

di-a L.

v

]

-

-. . . .

r

Z

on Pt(111)

O

di-rc

n,' O o9 t3 <

(~'+") c.

j2

a* on Pd(111)

280

290

300

310

320

E N E R G Y (eV) Fig. 3. C K-edge XANES spectra obtained at grazing incidence of 1,3-butadiene condensed at 95 K (a) and chemisorbed (ML) at 300K on Pd(111) (b) and on Pt(111) (c) from [31 ].

412 At 300K, and in the absence of hydrogen, the characterisation of butene chemisorbed phase is made impossible since it undergoes transformations such as dehydrogenation on P d ( l l l ) and probably formation of butylidyne on Pt(111). By extrapolation of the data to lower temperature, i.e under associative chemisorption conditions, we can however speculate that butene, with a single C=C bond, is n adsorbed on Pd(111) and di-a adsorbed on Pt(111). These results are corroborated by theoretical calculations [32] whose mains results are reported in Table 2. These theoretical results, even if they do not allow to discriminate between n or di-cr states in the case of an olefin on Pd(111) or between di-n and di-c~ states for diene on Pt(111), show that the adsorption energy of the diene is about twice that of the olefin on Pd(111), but it is of the same order on Pt(111). As a conclusion, the adsorption competition between butadiene and butene is actually in favour of butadiene on Pd(111), making this catalyst highly selective in butenes for this hydrogenation reaction. This is not true for Pt(111) which is poorly selective. Moreover, one can remark that the more open (110) faces of fcc metals are more active for the butadiene conversion into butenes than the close packed (111) faces [29, 33]. Some striking results are given in Table 3. Table 2 Adsorption energy (kcal mo1-1)calculated for an olefin (ethylene) and a diene (1,3-butadiene) on Pd(111) and Pt( 111) [32]. Pd(lll)

Pt(lll)

Chemisorbed state

olefin

diene

olefin

diene

n (di-n)

17

(36)

8

(18)

di-cr

19

20

15

15

Table 3 Catalytic activity for the first reaction (butadiene to butenes) on (111) and (110) faces of Pd, Pt and Ni (at 293 K, 10 torr Hz and an hydrogen/hydrocarbon ratio of 5-10 [29, 33 ]. Sample Catalytic activity (x 1015mol.cm2.s-1) Selectivity into butenes (at 50 % conversion)

Pd(lll)

Pd(110)

Pt(lll)

Pt(110)

Ni(lll)

Ni(ll0)

0.42

2.15

0.03

0.1

0.1

0.2

100 %

100 %

58 %

100 %

100 %

413

4. SURFACE AND R E A C T I V I T Y OF Pd BASED A L L O Y SURFACES

4.1. General points The main factors which govern the surface composition of thermally equilibrated binary alloys are the difference of surface energy of the two components, the mixing enthalpy and the elastic strain relaxation [6, 7]. These factors will also govern the tendency to order and/or to reconstruct, a way to relax the stress induced by the difference of atomic radii of the partners, i.e. the surface structure. In Table 4 are reported some comparative values of these parameters for Pd neighbouring elements of the VIII (Ni and Pt) and IB (Cu, Ag and Au noble metals) columns of the periodic table. A pseudomorphic Pd overlayer will have to sustain a large stress in compression on Ni and Cu, and in tension on Ag and Au. On the contrary one expects a quasi no stressed Pd overlayer on Pt. While less important, a similar effect will appear on the surface of alloys where Pd segregates in the top layer. This would be more pronounced on alloys of low Pd concentration exhibiting a very large surface segregation of Pd. In Fig. 4 are reported the values of Pd surface composition as a function of the nominal bulk composition for (111) faces of fcc Pd-X alloys (X = Ni, Pt, Cu, Ag and Au). These values have been calculated with the Equivalent Medium Approximation (EMA)model [7].

1O0

,

,

,

,

,

,

~

' ~

//

8O o~ 60

...jr/

g ~:: 2040 "13 n

."

.-

ooO~ o~

.

--I--

PdNi

- - m - - PdCu ~ ~ o

0

20

40 60 Pd (bulk) at.%

8

' i

PdAg

mO~

P d Pt

~n~

PdAu

Fig. 4. Predicted values for the composition of the first atomic layers of (111) Pd-X alloys (X = Ni, Pt, Cu, Ag and Au) from [7 ].

414 Table 4 Comparative values of the lattice parameter, a, of Pd and surrounding elements, the mixing enthalpy at 300K of Pd alloys, the misfit parameter relative to Pd, the possible phases of Pd alloys [34, 35] and the surface energies [361. Ni

0.352

Cu

~0 +9.3% 2.48

X

a (nm) z~T"Im(kcal. mol 1) misfit (%) surface energy (J. m2) solid solutions

Pd

0.388

0.361 -2.56 +6.96% 1.79 t~-PdCu3 13-PdCu Ag

2.0 Pt

0.392

0.408 -1.20 -5.15% 1.25

Au -0.86 -1% 2.5

0.407 -1.86 -4.9% 1.5 t~-PdAu3 ct-Pd3Au

By opposition to PdxAgl_x, PdxAUl_x and PdxCUl.x, the alloys PdxNil_x and PdxPtl_x are expected to exhibit a large segregation of palladium on the surface layer. In fact the element with the lower surface tension is actually expelled towards the surface. Moreover, these segregation effects are generally more pronounced on more open surfaces, (100) for instance [7] : the lower the coordination number of surface atoms, the higher the amplitude of segregation. In addition, a layer by layer oscillating composition is expected in depth for alloys whose mixing enthalpy is negative. In these predictions, surface reconstruction has not been considered. Actually, this is not always justified, mainly in the case of open surfaces, as will be shown below.

4.2. Surface composition and reactivity of PdsNi9s and PdsPt9s polycrystals Let us first consider, as a striking example, the comparison of two alloys for which a very large surface segregation is expected (to be of the same order) but for which the misfit is very different (see Table 4). These are PdsPt95 and PdsNi95 binary alloys. For both samples the Pd surface segregation is clearly shown by electron spectroscopies, AES or XPS. The peaks characteristic of Pd (PdMvv Auger peak

415 near 330 eV kinetic energy and/or Pd3d photoelectron core level) are greatly enhanced once the samples are annealed after ion sputtering used for cleaning. Moreover the intensity ratio Pd3d/Ni2p and P d 3 d / P t 4 f measured respectively for PdsNi95 and PdsPt95 alloys increase markedly when the detection of photoelectrons is made at a small angle relative to the surface. This is a clear indication that Pd is overconcentrated in the surface layers. LEIS experiments confirm these results, but in addition, they allow to determine the Pd concentration in the ultimate surface layer, which is the one in contact with reactants driving catalytic reactions. A few LEIS spectra measured on the PdsNi95 sample stabilised at 1100 K are reported in Fig. 5 for illustration. The outer surface composition can be obtained by measuring the intensity ratio of the peaks characteristic of Pd (at 872 eV) and Ni (at 778 e V ) , corrected for their relative sensitivity factors determined separately on pure Pd and pure Ni samples [37]. The same procedure was applied to PdsPt95 polycrystalline alloy. The sensitivity factor was found to be 2.2 times lower for Ni than for Pd, and 1.08 times lower for Pt than for Pd under our experimental conditions. Experimental results are reported in Table 5.

/q

Ni

c6 >z

iii

z_

650

700

750

800

850

900

950

K I N E T I C E N E R G Y (eV)

Fig. 5. LEIS spectra of a PdsNi95 polycrystalline sample illustrating the evolution of Pd and Ni characteristic peaks versus time. The first spectrum corresponds to that of the clean sample stabilised at 1100 K ; the time interval between two consecutive further spectra is 35 s. Experimental conditions: 1 keV 4He+ ion source, beam current = 10 nA, beam size about 1 m m

2 .

416 Table 5 Pd concentration in the outer layer of clean PdsNi95 and PdsPt95 stabilised at high temperature, as determined from LEIS experiments. Sample

PdsNi95

PdsPt95

Pd at.% at surface

50

54

One measures a very large Pd surface enrichment, as expected from theoretical predictions (see Fig. 4). Moreover, the rapid decrease of the Pd/Ni (Pd/Pt) ratio as a function of sputtering time indicates clearly that the Pd concentration decreases very rapidly with depth. At low ion current density the sputtering rate is low [38], therefore the main change in concentration with respect to the nominal bulk composition mainly arises from the surface layer, i.e. from the vacuum/sample interface. Let us consider now the catalytic performances of such alloys. Typical curves showing the evolution of reactants/products, analysed by mass spectrometry and gas chromatography, as a function of time are reported in Fig. 6, for the PdsPt95 alloy. From the different known experimental parameters such as sample area, volume of the reactor and products/reactants pressure evolution, the catalytic

4.0 3.5 ~

LU IT" 3.0 09 LU 09

2.5

IT' 2.0 El. ~1 1.5

+utan./ f !

< ,...= I-- 1.o rc < Q. o.5

0.0

lOO

it, ~ "I,,

~m)~ lb'3"bnt~eadiene I

2000

REACTION

3000

4000

T I M E (s)

|

I --"--2"cis'b~aene

~1 --V-- 2-trans-butene'

o~ 60 -1-O" 40 20

q

1000

80

I --II-- 1,3-butadiene I--~ |

5000

0

0

1000

2000

REACTION

=.

3000

,

9

u

4000

T I M E (s)

Fig. 6. Distribution of the reaction products versus time in the course of 1,3-butadiene hydrogenation reaction on PdsPt95 (Sample area 0.8 cm2, temperature 293 K, volume of the reactor 84 cc, C4H6 pressure 3.5 torr, H2 pressure 35 torr) followed by mass spectroscopy (left side) and gas chromatography under similar experimental conditions (fight side).

417 Table 6 Catalytic activity and selectivity into butenes for the first reaction (butadiene to butenes) on PdsPt95 et PdsNi95. These values have been extrapolated to 10 torrs hydrogen, assuming a zero order with respect to hydrogen pressure.

Sample Catalytic activity (mol.cm-2.s1) Selectivity into butenes (at 50% conversion)

PdsPt95

PdsNi95

0.8.1015

8.1015

93 %

100 %

activity can be calculated (see Table 6). While quite high, the selectivity into butenes is not strictly equal to one on the PdsPt95 sample (Fig. 6 and Table 6). This can be explained by a simple superposition of the intrinsic properties of Pt and Pd outer atoms, present in quite the same amount on the surface. Pure Pt is one order of magnitude less active than pure Pd, and only slightly more than 50% selective while Pd is 100% selective into butenes (see Fig. 2 and Table 3). In fact, for the first reaction (butadiene --) butenes) the activity measured on polycrystalline PdsPt95 (0.8 x 1015 mol.cm-2.s-1), whose surface is enriched at 54 at.% Pd, ranges between that of pure P d ( l l 1) (0.42 x 1015 mol.cm-2.s 1) and that of pure Pd(110) (2.15 x 1015 mol.cm-2.s-1). The catalytic activity per Pd surface atom on the PdsPt95 alloy is therefore comparable to that of pure Pd surface atoms. This is a priori surprising since it has been proposed that the electronic properties of Pd were modified by alloying with Pt[39], which would induce a drastic decrease of the hydrogen solubility. Coming back to our results, very interesting is the noticeable increase of the reactivity observed on the the PdsNi95 alloy (8 x 1015 mol.cm2.s -1) with respect to pure Pd. It is now clear that the catalytic activity of PdsPt95 and PdsNi95 alloys has to be related to the presence of a large amount of Pd on the surface. Since the surface composition is quite the same (=50--at.%), two parameters can be invoked to explain the differences in catalytic activity of the two systems. One could first consider some different electronic influence of Pt and Ni surrounding atoms. Such an effect is unexpected since Ni, Pd and Pt belong to the same column of the periodic table, i.e. have similar electronic properties. Another explanation could be related to the difference of the strains undergone by the Pd surface atoms. Indeed, no stress is expected for Pd on top of a Pt rich alloy since Pd and Pt have approximately the same atomic radii. On the contrary, Pd atoms will sustain a strong stress in PdNi alloys due to the large size misfit between the Pd atoms and the Ni rich matrix. Such an explanation remains valid only if no reconstruction takes place to relax, at least partially, the stress. As will be shown

418 further on, surface reconstructions may appear depending on the crystal surface orientation. As a last comment, it has to be noted that in general, on Pd based catalysts, around two thirds of the butenes produced during the first reaction are associated to 1-butene. The rest is composed of 2-butenes, whose trans- configuration dominates largely. Just after the complete transformation of butadiene, the decrease of butenes concentration is very fast (see Fig. 6) ; in fact, 1-butene transforms rapidly to form butane on one hand, by hydrogenation, and 2-butenes on the other hand by isomerisation (fight side of Fig. 6). 2-butenes then hydrogenate to butane, but at a rate which is approximately ten times less than that for the 1-butene hydrogenation [40, 41]. The actual behaviour for the transformation of butenes after the complete conversion of butadiene is quite similar for all the systems considered here after (Pd-Ni, Pd-Cu and Pd-Au) : the 1-butene transforms rapidly via hydrogenation to butane and via isomerisation to 2-butenes. Then, 2-butenes hydrogenate but at a largely lower rate. The analysis of this part of the reaction is therefore complex. In the following, the different systems will be compared only with respect to the first part of the reaction: butadiene --) butenes.

4.3. Influence of the surface orientation on reactivity To illustrate the influence of surface orientation on reactivity, two systems will be described. The first one, PdNi alloy, is a solid solution in the whole range of composition. The second one, PdCu shows a tendency to order.

4.3.1. A solid solution in the whole range of composition : Pd8Ni92 (111) and (110) As expected (Fig. 4), Pd8Ni92 monocrystalline samples show a high rate of Pd segregation in the top layer. The Pd surface concentration determined from LEIS is 76 and 81 at.% for the (111) and (110) faces respectively, which corresponds to a quasi complete monolayer (ML) of Pd [42] on a Ni-rich alloy. Furthermore in order to fit quantitatively AES and XPS data, and LEIS profiles as well, even the second layer of the (110) face has also to be enriched in Pd. While the (111) face exhibits a well defined ( l x l ) LEED pattern, the (110) face shows a rather complex LEED pattern with the appearance of supplementary spots along the [1 10] direction [42]. This surface reconstruction is also observed by STM showing alignments parallel to the [001] direction as shown in Fig. 7 [43]. These alignments are separated by about 1 nm with a corrugation of a few hundredth of nm. The catalytic activity for the butadiene hydrogenation into butenes is 2.1015 mol.cm-2.s -1 at 10 torrs hydrogen on the PdsNi92(111) alloy ; it is near by 5 times that of a pure Pd(111) crystal [29]. The reconstructed Pd8Ni92(110) sample

419

Fig. 7. STM image (lnA, 0.5V) of the Pd8Ni92(110) alloy surface [43]. is largely more reactive; its activity reaches 38.1015 mol.cm-2.s 1, a value 15 times higher than that measured on Pd(110). (see Table 7). Additional compressive stress is expecxted on surface Pd atoms of the PdNi alloys, the Pd atomic radius being quite large compared to that of Ni, the main constituent of this alloy. The (110) oriented Pd8Ni92 alloy relax (at least partially) the stress by reconstruction. This is not the case for the close-packed (111) sample, which exhibits a ( l x l ) <~normal ~ surface structure and which has

Table 7 Catalytic activities of (111) and (110) oriented Pd8Ni92 and PdsoCu50 alloys for the 1,3butadiene hydrogenation, in comparison with pure Pd (at RT, 10 torrs hydrogen, Pm/ Pat = 10). Ni is about ten times less reactive than Pd; pure Cu presents no measurable activity for this reaction. Sample

Pd

Pd8Ni92 PdsoCu50

Pd

Pd8Ni92 PdsoCus0

(111) [29] (111) [42] (111) [44] (110) [29] (110) [42] (110)[441 Structure Catalytic activity (x 1015mol.cm-Z.s-1)

(lxl)

(lxl)

(lxl)

(lxl)

(Nxl)

(2xl)

0.43

2

11.3

2.14

38

1.36

Surface Pd at.%

100

76

45

100

81

32

420

consequently to sustain the stress. One parameter which can play an important role to explain such a behaviour is, for a surface atom, its number of nearest neighbours : 9 (6 in plane and 3 with the second layer) for the (111) oriented fcc crystal instead of 7 (2 in plane, 4 with the second plane and 1 with the third one) for the (110) alloy. The low number of in-plane nearest neighbours in the (110) orientation probably makes easier the movement of surface atoms allowing partial relaxation. Following Palassama et al. [21], one can tentatively explain the modified reactivity of the stressed (111) surface assuming that compressed Pd atoms will induce an increase of the d-band width accompanied by a shift away from the Fermi level leading to a decrease of the binding energy of unsaturated hydrocarbons, and to a higher hydrogenation rate. This is actually coherent with the 0.5-0.7 eV higher XPS binding energy measured on Pd3d for PdsNi92 alloys as compared to pure Pd [42]. Such an explanation does not hold for the reconstructed (110) surface for which the stress of the Pd surface atoms is not so large. In that case, we would rather associate the new chemical properties to the new structure generated. The chemisorptive properties of this reconstructed surface with respect to 1,3-butadiene have been investigated by HREELS and compared to those of Pd(110). Experimental results are reported in Fig. 8.

in and out of plane CH and CH~ deformations I

vCH

I

v

>i---

09 Z Iii F-Z

X1

|

-500

I

d

0

~

500

I

,

I

1 0 0 0 1500

~

I

2000

~

I

2500

,

3000

3500

WAVENUMBER (cm ~)

Fig. 8. HREELS spectra of 1,3-butadiene monolayers chemisorbed on Pd(ll0) at 210K (a) [31] and on Pd8Ni92(l10) at 180K (b) [45]. The spectra were recorded along the specular direction with an incident angle of 50~ (60 ~) from the surface normal and an incident beam of 3eV (5eV) respectively on Pd(110) and PdsNi92(110).

421 The main features observed on both samples are very similar and have been associated to a di-rc like adspecies, previously discussed in w [31]. While the adsorption mode is actually quite the same, changes can exist on the adsorption energy and/or on the activation barrier for the H addition, the limiting step for hydrogenation of unsaturated hydrocarbons. This will be discussed in details with structure and chemical properties of Pd deposited on Ni(110) in w 4.3.2. A system with a tendency to ordering : PdsoCuso (111) and (110) Pd and Cu form a continuous series of solid solutions, and the mixing enthalpy for PdCu alloying is moderately negative. However, these alloys make up ordered PdCu and PdCu3 phases (see Table 4). As observed on PdNi alloys, the (111) oriented PdCu face shows a sharp ( l x l ) LEED pattem indicating that no reconstruction occurs; the surface contains about 45 Pd at.%, i.e. the surface composition is the same as the bulk one [46]. It is again in good agreement with the predictions reported in Fig. 4. The activity of the P d C u ( l l l ) crystal (see Table 7) is largely enhanced compared to that of P d ( l l l ) [44]. This is remarkable since the surface concentration into the active element, Pd, is only 45 at.%. In addition, due to dilution effects, the number of pairs of adjacent Pd atoms is reduced. This pair of atoms is required for a favourable activation of 1,3-butadiene and is assumed to be the active site for the hydrogenation reaction (see w Chemical modifications have also been manifested through the decrease of the heat of adsorption of CO and NO on Pd in the alloy with respect to pure Pd [46, 47]. Since Pd concentration is the same in the bulk and in the surface layer of such alloys, the stress sustained by Pd surface atoms is probably low. One has therefore to consider ligand effects to explain the activity enhancement. XPS measurements have shown that, compared to pure Pd, in PdCu alloys the 3d band exhibits a loss of asymmetry [46] probably owing to the interaction via bond formation between the almost full sd- valence band of Pd and the resonant dsp- band of Cu near the Fermi level. A high degree of mixing of Pd and Cu electronic states has also to be invoked to explain that UPS spectra of PdCu alloys are very different from those of both Pd and Cu [48]. On the (110) oriented PdxCUl_x alloys, the tendency to form a PdCu3 like surface alloy has been often mentioned. On Pd15Cu85(l10) [49, 50], and on PdsoCus0(ll0) [44], (2xl) LEED patterns attributed to the alternate position of Pd and Cu atoms along the [110] rows are observed. The surface is then constituted of alternate layers of 100 at.% Cu and 50:50 at.% Pd:Cu similarly to the ideal (110) face of the L12 PdCu3 ordered phase ; it is schematically represented in Fig. 9. The PdCu3 ordered like structure is in fact the consequence of the tendency for Cu to segregate at the surface together with the tendency for this alloy to order. LEIS experiments indicate that PdsoCus0(110)

422 contains about 32 Pd at.% at the surface ; it can be regarded as a succession of (110) orientated terraces constituted respectively of pure Cu and of well ordered PdCu rows, in approximately the same amount (see Fig. 9). The catalytic activity of the PdCu(110) surface is practically at the same level as that of Pd(110) ; it is less active than the (111) oriented PdCu alloy (see Table 7). This is a priori surprising since open surfaces show generally a higher reactivity than close packed surfaces. One can think that the number of active sites is very low. Firstly the Cu terminated terraces which constitute approximately half the surface area can be considered as inactive. Secondly, on the Pd-Cu terminated terraces there is a lack of adjacent Pd pairs in the (2x l) configuration (Fig. 9). The remaining Pd pairs might then be very reactive. Just above, we have discussed the catalytic activity changes in terms of the possible modifications of the active sites with respect to the unsaturated hydrocarbon, which is often supposed to be of the main importance (see w but we have neglected the possible influence of hydrogen, the second partner of the reaction. In a recent theoretical paper, Sousa et al. [51] argue that the coordination of the surface Pd atoms to other Pd atoms in the second layer are necessary for hydrogen to dissociate and to be trapped with a low energy cost. This points out the importance of the atomic composition and arrangement not only in the outer layer but also in the sublayers to understand the chemical reactivity of alloy surfaces. Anyhow, Cu has a noticeable electronic influence on

Fig.9. Proposed surface structure for the ordered (2xl) surface structure of PdCu(110). A PdCu terminated (upper) terrace and a Cu terminated (lower) terrace are represented.

423

the Pd surface atoms. This influence is positive with respect to the considered reaction, the 1,3-butadiene hydrogenation. In conclusion, closed-packed (111) surfaces tend to retain the ( l x l ) ~ normal ~ surface structure, while the less dense (110) faces tend to reconstruct or to relax, at least partly, the surface stress or to form ordered phases if they are favoured. The closed packed (111) faces have modified chemical reactivities associated to stress effects and/or ligand effects. In the case of more open (110) faces, stressed surfaces tend to relax and generate original structures with peculiar sites of new and specific catalytic properties. The tendency for ordering induces a possible isolation of the surface Pd atoms ; this will decrease the number of active sites for a given reaction or produce specific arrangements of interest for specific reactions.

5. SURFACE AND REACTIVITY OF Pd DEPOSITS

Metal on metal growth is an alternative way to obtain original bi-metallic surfaces. Well controlled atomic beam deposition allows to generate thin layers slightly out of equilibrium, if one works at moderate temperatures. In this way new 2D surface alloys can be generated which have no bulk analog. Therefore, they can have structures, and hence chemical properties different from those of the surface of bulky alloys. In the following we shall describe the behaviour of Pd (A) deposited on several metals (B) whose atomic radii are either smaller (Ni, Cu) or larger (Au) than that of Pd. The growth and stability of Pd will be greatly influenced by thermodynamics (see Table 4). Most systems will be fully described for (110) orientated substrates at the exception of Pd-Ni where a comparison between (111) and (110) orientation will be made. 5.1. Pd in compression on Ni and Cu Pd deposited on Ni or on Cu will suffer a compressive stress because of its large atomic radius. In the first system, Pd deposited on Ni, the thermodynamical parameters (Table 4) favour the stability of Pd on the surface. This will be different with Pd deposited on Cu, system for which Cu prefers to be in the outer layer, and for which a tendency to order exists. 5.1.1. Case thermodynamically favouring A on B : Pd on Ni Pd on Ni(111 )

Pd deposition on Ni(111) at room temperature (RT) is accompanied by a rapid decrease of the LEED spot intensities, and a noticeable increase of the background intensity, with increasing Pd coverage. The initial ( l x l ) LEED

424

pattern reappears after annealing at 475K for 15 minutes. In fact, the Pd adatoms spray all over the surface. Once annealed, the surface structurates without any Pd dissolution in the bulk of the substrate below 600K [52]. These results are coherent with the formation of a pseudomorphic Pd adlayer. The catalytic activity in 1,3-butadiene hydrogenation of 0.5 - 3 ML Pd on Ni(111) is more than one order of magnitude higher than that of pure Pd(111). These results are quite comparable to those obtained for the Pd8Ni92(111). For this system, Terada et al. [53] have observed by STM, at RT and near 1 ML, hexagonal moir6 superstructures which would be characteristic of an hexagonal arrangement of Pd adatoms aligned with a slight rotation with respect to the substrate axis and for which Pd-Pd atomic distances tend towards the known value for pure bulk palladium. This result differs from ours, which could indicate a limit of stability of the stressed Pd oveflayer. Pd on Ni(110) The following results relative to Pd growth on N i ( l l 0 ) are the fruit of experiments combining many surface science techniques 9LEED, STM, AES, XPS, LEIS. It has been also the subject of recent investigations by grazing incidence X-ray diffraction, using synchrotron radiation facilities at ESRF (Grenoble), and by DFT theoretical calculations. At very low Pd coverage (< 0.1 ML) and at RT, islands of monoatomic

height elongated along the [ 110] direction of the substrate are clearly observed by STM. This arrangement could result from an easier diffusion along the [ 110] rows of Pd atoms impinging on the terraces, as generally observed for (110) faces of fcc crystals [54]. The areas located near the steps are not so well defined. Heating at 500-600K induces a diffusion of the Pd islands with formation of rounded, one monoatomic height, areas near or attached to the steps [55], with the probable formation a Pd-Ni surface alloy. In the monolayer range, a (Nxl) superstructure (with N = 5-6) has been observed by LEED, and grazing X-ray diffraction [56]. Based on theoretical works, it can be assumed that the Pd adatoms tend to remain close to the epitaxial hollow sites with the formation of a vacancy each N atoms [57]. The observation of a N order superstructure along the [001] direction indicates an interaction between Pd rows and/or vacancies. A schematic representation of the proposed structure, for N = 6, is reported in Fig. 10. This (Nxl) structure is retained up to a coverage of about 2 ML. The behaviour of 3-4 ML Pd deposits is very striking. The rough surface obtained just after Pd deposition at RT (Fig. 11, left part), structurates after a mild annealing near 500-600K with the occurrence of alignments still oriented along the [1 10] direction (Fig. 11, right part) [58]. Such a surface is very smooth : the corrugation is less than 30-40 pm. In fact a well defined (1 lx2)

425

Fig. 10. Top view of the (Nxl) structure for a Pd coverage approaching the monolayer (N=6). superstructure has then been identified by X-ray diffraction. In Fig. 12 a structure based on quantitative analysis of diffraction spots intensity measured by grazing X-ray diffraction [56] and theoretical calculations [59] is proposed. We note i) the presence of Pd alignments along the [110] direction with Pd atoms located near the Ni(110) hollow sites and the presence of a vacancy each

Fig. 11. STM images obtained for 4 ML Pd deposited on Ni(ll0). Left part : as deposited at RT ; fight part : after heating at 500K for 15 minutes.

426

Fig. 12. Schematic representation of the (11x2) superstructure observed for 4 ML Pd deoosited at RT on Ni(110) and annealed at 500K.

11 Ni substrate atoms at the interface, ii) a tendency for the Pd-Pd distances to expand until they reach a value close to that of pure Pd when approaching the outerlayer and iii) a tendency to generate a missing row like structure generating the (-x2) supplementary spots. The catalytic activity for 1,3-butadiene hydrogenation of Pd deposited on Ni(110), well structurated by annealing nearby 500K, differs strongly from those of Pd(110) taken as a reference (Fig. 13). The reactivity is largely enhanced as soon as the amount of Pd deposited is larger than 0.5 ML, with a special mention for 4 ML. In any case, the selectivity into butenes remains unity up to the complete conversion of butadiene in the reactor. Understanding the reasons for this modified chemical reactivity is not trivial. Nevertheless, this has been attempted theoretically in the case of a lighter hydrocarbon, C2H4, chemisorbed and hydrogenated on the reconstructed structure observed for 4 ML Pd deposited on Ni(110) in comparison with pure Pd(110) [59]. The most stable adsorption sites for both ethene and hydrogen are the same on Pd(110) and on the (1 lx2) reconstructed surface: C2H4 is bridge bonded and H is located in threefold sites. However, the binding energy of C2H4 + H bonded in such a configuration is slightly lower on the reconstructed surface, (state 1 in Fig. 14). For the addition of the first hydrogen to form a C2H5

427

Fig. 13. Compared catalytic activities for 1,-3-butadiene hydrogenation of Ni(ll0), Pd(ll0) and Pd overlayers on Ni(ll0). Pd overlayers have been annealed at 500K. Reaction parameters : RT, 10 torrs hydrogen and PH2 / Pdiene = 10. intermediate, the C2H4 molecule would have to rotate to occupy a top position with its C-C axis perpendicular to the direction of rows (transition from state 1 to state 2), while H would shift to a bridge position (transition from state 2 to state 3) on both surfaces The addition of the second hydrogen on this C2H5 intermediate would be easier. In fact, the most striking feature is the decrease of -0.6

4 M L Pd/Ni(1 10)

Pd(110)

-0.8

.A............

(3)

-, ...........

~ " -1.0

cO

(4)

-1.2

(3)

(4) (5)

rr I.IJ Z ILl -1.4

-1.6

oJi

Fig. 14. Energy diagrams for the chemisorption of H + C2H4 and for hydrogenation on Pd(110) and on the reconstructed Pd (4 ML) on Ni(110) surfaces [59].

428

the activation barrier for the first hydrogenation on the reconstructed surface relative to the normal Pd(110) surface. Briefly summarizing, Pd on Ni(ll0) undergoes a noticeable restructuring with a complete reorganisation of the adlayer, generating very peculiar surface sites (defects, surface atoms of unusual coordination, ... ), very reactive with respect to the 1,3-butadiene hydrogenation. Such a reorganization is surely the consequence of the release, at least partial, of strains in the surface layer. A close behaviour has been observed for other systems with size mismatch, and was associated with strain-induced buckling [14]. Similar structural and chemical effects were observed in the case of PdNi alloys. In conclusion, as already discussed in the case of alloy surfaces, Pd is retained in a stressed configuration, in registry with the substrate for closepacked orientations. On more open surfaces, Pd tends to relax the stress and reconstruct generating peculiar surface sites. In both cases, the chemical properties of surfaces are markedly modified. The mastery of the stress could be a way to tailor the catalytic properties of bimetallics. 5.1.2. Case o f A on B unfavourable : Pd on Cu(110) In the submonolayer range, Pd deposited on Cu(110) is completely inactive for 1,3-butadiene hydrogenation reaction. (Fig. 15, left part). This is likely with the abundance of inactive Cu atoms in the outer surface layer, as demonstrated by LEIS results (Fig. 15 fight part) which show that less than 15 at.% Pd remains on the surface after deposition of 1 ML at RT. STM measurements show that for low coverages (< 0.5 ML) Pd growth is bidimensional [60]. One can then state that Pd adatoms enter in the Cu substrate and/or that Cu atoms migrate to cover the Pd adatoms. This last statement seems to be the most probable since some pits are actually present at the surface which can be seen as sources of Cu atoms [61 ]. Moreover, this 2D morphology is accompanied by the formation of a (2x l) structure clearly seen near 0.5 ML Pd by LEED and grazing incidence X-ray diffraction. One can therefore propose the formation of a 2D PdCu3(110) like surface alloy, with a mixed Pd-Cu plane covered by a pure Cu surface layer. This is one of the possible configurations observed at the surface of PdxCUl_x(110) alloys [44, 49-50], as shown in Fig. 9. Formation of ordered Pd-Cu phases by deposition of Pd on Cu has been often observed. It was found firstly by quantitative LEED analysis of surface alloys formed by Pd deposits on Cu, and subsequent annealing [62] that both the c(2x2) structure on Cu(100) and the (2x l) structure on Pd(110) could be regarded as truncations of the PdCu3 L12 ordered phase. Recent theoretical works relative to low Pd deposits on Cu(110) [63] revealed the exchange mechanism between adatoms and substrate atoms, with the formation of Pd-Cu chains and of Cu islands, in good agreement with all the experimental results. F o r - 1 ML Pd films deposited

429

~. ~

or;

100

18 -

P d / C u ( 1 1 O)

16

80

E 14

o ~

E

12

o~ 60

10

r "o9 40 13_

~-

6 -Pd(110)

>

4-

F"

2-

~

0

1

I

2

II 3

Opd (ML)

II

, U

4

20

5

o

0

,

l

1

,

2

l

3

,

4

5

I

6

epd (ML)

Fig. 15. Catalytic activity for the 1,3-butadiene hydrogenation reaction (at RT, 10 Torr hydrogen and PH2/ P d i e n e "- 10) (left part) and surface composition determined by LEIS (fight part) as a function of the Pd amount deposited at RT on Cu(110). at 310K on Cu(110), an ordered (2xl) surface alloy has been also observed by LEED by Bennett et al. [64]. The catalytic activity increases suddenly for Pd deposits higher than about 1 ML, reaches a maximum near 3ML and further decreases towards a value which would be comparable to that of pure Pd(ll0) at large coverage (Fig. 15, left part). 3 ML Pd means only 4 0 - 50 at% Pd atoms in the surface layer (Fig. 15, fight part) but the activity per Pd surface atom, in 1,3 butadiene hydrogenation, is about 10 times higher than that for Pd surface atoms on pure Pd(110). Looking now at the structure, beyond 0.5-1 ML, a 2D-3D transition occurs in the growth mode. 3D islands elongated along the close-packed [110] direction and about 7-8 nm wide are formed at 0.75 ML [60, 64]. These islands increase in size as the Pd coverage is increased, until they cover the whole surface. LEIS results reported in the fight part of Fig. 15, agree well with these STM observations. A mild annealing (near 500K) does not modify strongly the Pd and Cu amounts; only a moderate increase of Cu concentration has been measured by LEIS. Upon annealing at higher temperatures (603K-723K) Bennett et al. [64] have measured by AES a preferential surface segregation of Cu and significant changes in the surface morphology with larger flat terraces that showed ragged step edges aligned bands running across the surface. These bands would be the consequence of strain relaxation between Cu and alloy sublayers. The surface is then similar to those obtained for deposition of Cu on

430

P d ( l l 0 ) at elevated temperatures; it actually moves towards its equilibrium state. The explanation of enhanced activity is not at first sight directly related to the apparition of 3D islands. However, high resolution STM observations, in the 2-4 ML range, revealed a nanostructuration, similar to that observed when 4 ML Pd were deposited on Ni(ll0) ; the period of alignments along the [110] direction is about 5 nm, instead of 2.5 nm for 4 ML Pd on Ni(ll0). As previously discussed for Pd on Ni(ll0) one can suppose that the surface reconstructs to relax the stress, at least partially, with here again creation of a specific structural arrangement, generating very active surface sites. The understanding of the modified chemical reactivity would recquire complementary informations relative to the electronic properties of the considered thin films. Finally, for larger deposits, Pd recovers its ~normal ~ structural and chemical properties, together with its ~ normal ~ reactivity. To summarize, at low coverage Pd deposited on Cu(110) primarily forms a PdCu3 like surface alloy, with tendency for Cu to come out ; such a surface has no catalytic activity for the 1,3-butadiene hydrogenation reaction. When increasing the amount of deposited Pd, highly strained 3D islands, probably constituted by Pd-Cu alloys with a noticeable amount of Pd atoms in the outer layer, are formed. They exhibit a largely modified chemical reactivity as compared to pure Pd(ll0). For larger deposits, Pd surface atoms gradually recover their own structural and chemical properties.

5.2. Pd in tension on A u ( l l 0 ) Gold has a very low fusion temperature and is therefore expected to be very mobile, i.e. to diffuse easily, even at low temperature. Moreover, its surface tension is very low compared to Pd (Table 4), and it is therefore expected to segregate largely to the surface of Pd-Au alloys (Fig. 4). This has been experimentally verified; for example, the topmost layer of Au3Pd(100) and Au3Pd(110) has been found to consist of Au atoms only [65, 66]. Moreover, the (110) surface of the alloy reconstructs in a (lx2) missing row mode as does pure Au(ll0). The growth of Pd on A u ( l l l ) using atomic beam deposition has been studied by Koel et al. [67]. Initial stages of Pd on Au deposited by electrochemistry have also been investigated on Au(11 l) and Au(100) [68, 69]. The analysis of the reactivity of Pd deposits with respect to CO chemisorpfion [70] and to the cyclization of acetylene to benzene [71 ] has been the subject of a few experimental works. A theoretical work has been devoted to the study of electronic factors governing ethylene hydrogenation and dehydrogenation activity of pseudornorphic Pd on Au(111) [21].

431

The results presented here after, relative to the growth of Pd on Au(110), are based on the combined use of LEED, MEED, grazing X-ray diffraction [72], STM [73], and LEIS [74] techniques. The test reaction is, here again, the 1,3butadiene hydrogenation. The LEED pattern of the clean (lx2) reconstructed Au(110) is presented in Fig. 16 [72] together with a schematic representation of the missing row reconstructed surface. The peculiar geometry of the missing row reconstructed surfaces of fcc metals offers a priori a unique way to generate linear structures of adatoms and to measure their specific properties. However, as will be seen below, in the present case, the situation is more complex. For low deposition, the intensity of the LEED 1/2 spots along the [001] direction decreases rapidly as the Pd coverage is increased. It vanishes at 0.5 ML Pd coverage. The intensities of the (0,1/2) spots observed by MEED (Medium Energy Electron Diffraction) and of the (0,1/2) rods in grazing incidence X-ray diffraction experiments are also largely reduced when the Pd coverage increases and approaches 0.5 ML. This indicates that Pd adatoms do not simply fill the missing rows. Actually, Au has migrated on the surface layer, the Pd concentration of the outer layer measured by LEIS being less than about 10 % for 0.5 ML (Fig. 17, fight part). 2D islands, with the formation of pits probably associated to areas for diffusion of Au atoms which cover and/or associate to the Pd deposited atoms to form a surface alloy, have been observed by STM up to nearby 1 ML. This is quite similar to what happens for Pd deposited on Cu(ll0) (w The activity for 1,3-butadiene hydrogenation is then near zero. A moderate annealing does not strongly modify the surface composition. It is necessary to heat above 650 K to induce a complete dissolution of Pd atoms into the bulk.

Fig. 16. LEED diagram observed for the (lx2) reconstructed Au(ll0) surface (130 eV electron energy) and schematic representation of the corresponding missing row reconstruction.

432

7 ~

100

Pd/Au(11 O)

80

6

E _._o 5 0

E 4 v

0 ',-

3

i

2

o <

0

>I>

"

o~

Pd(110)

60

"o 40 12. 20

0

1

2

3

4

5

ere (ML)

6

7

8

O ~lr

0

i

II

1

,

II

2

,

I

3

~

I

4

0

I

5

,

I

6

~

I

7

8

ep~ (ML)

Fig. 17. Catalytic activity for the 1,3-butadiene hydrogenation reaction (at RT, 10 Torr hydrogen and PH2 / Pdiene = 10) (left part) and surface composition determined by LEIS (fight part) as a function of the Pd amount deposited at RT on Au(110).

Between 0.5 and 3.5 ML Pd, at room temperature, the intensity of the (0,0) specular beam oscillates with a rather regular period, showing a minimum each time the number of deposited monolayers is an integer (N) and a maximum each time this number is equal to (N+l/2). STM observations reveal a quite smooth surface in this range of Pd coverage. That characterises a pseudomorphic layer by layer growth. In the mean time, the outer layer Pd concentration increases monotonically, up to about 50 % at 3.5 ML (Fig. 17). A recent quantitative analysis of diffraction beams intensities, measured by grazing incidence X-ray diffraction, would agree with the formation of large 2D islands whose top layer is rich in gold while the second layer is essentially Pd [72]. A layer by layer growth has also been put forward in the case of Pd deposited on Au(111) [67] ; nevertheless, the tendency for Au to come out seems to be less pronounced on the close-packed (111) Au face than on the more open (110) face. Au migration in the outer layer would need a higher temperature for Au(111) than for Au(110) as a substrate. The catalytic reactivity in the range of 1 to 3 ML Pd on A u ( l l 0 ) is surprisingly low taking into account the noticeable amount of Pd measured by LEIS on the surface (Fig. 17). This low reactivity could be due either to a low number of active sites (two adjacent Pd atoms) resulting from chemical order or to the tensile stress applied to Pd atoms in pseudomorphic epitaxy with Au substrate. Such a hypothesis would be in agreement with the theoretical predictions of Pallassama and Naurock [21] who propose that Pd stressed in

433

tension would chemisorb more strongly unsaturated hydrocarbons, with, as a consequence, a lower hydrogenation rate. Lastly, for thicker layers (4-8 ML), islands elongated along the [110] direction are clearly observed by STM and the intensity measured in the specular direction by MEED does not show any structuration. New spots appear in the X-ray diffraction patterns corresponding to an atomic distance characteristic of pure Pd. This structural change is accompanied by a sudden and noticeable catalytic activity (Fig. 17). The activity for 7 ML is even 3 times higher than that of pure Pd(110). One can suppose that this rough Pd-rich surface contains a lot of very active low cordinated surface sites. In summary, gold has a great ability to migrate to the surface. However, uncovered Pd atoms which are in registry with the substrate are in tension which makes them unactive for the 1,3-butadiene hydrogenation. This coud be a positive effect for some other reactions.

6. S U M M A R Y AND C O N C L U S I O N What is clear from all these results, is that structural parameters and catalytic activity are intimately related. Starting from one metal, palladium in this case, we have shown that, by alloying with or deposition on other metals, it was possible to generate a lot of distinct local structures where surface Pd could exhibit largely modified chemical reactivity. Some particular systems can show large amplifications of activity for the 1,3-butadiene hydrogenation reaction Pd exhibits structures and chemical reactivities which are similar in segregated alloy surfaces and in surface alloys (obtained by atomic beam deposition) as long as the Pd concentration in the outer layer and the misfit between overlayer and substrate lattice parameters are the same. Palladium overlayers deposited on a metallic substrate of higher surface energy are quite easy to produce and control; such systems, after annealing, may show a good structuration with Pd atoms staying out. On the contrary, when the surface energy of atoms contituting the substrate is lower than that of Pd, the substrate atoms will aim to come out and cover thin Pd layers. The migration towards the surface can be effective even at room temperature. On close-packed fcc (111) oriented surfaces a pseudomorphic adlayer is often formed and the stress is actually retained by the surface atoms. In the case of the association of two elements having similar electronic properties, i.e. very close in the periodic table, the modified chemical properties can be tentatively associated to this stress. On the contrary, on more open (110) oriented surfaces there is a tendency to reconstruction which relax, at least partially, the stress ; strained surfaces show then surface sites having very peculiar geometries and strongly modified

434 chemical reactivities. For example, the catalytic activity for 1,3-butadiene hydrogenation, of a well ordered nanostructuration formed by 3-4 ML Pd deposited on Ni(110), is nearby two orders of magnitude higher than that of pure Pd(110). This system can be considered as the most reactive catalyst known for this reaction. The structural changes together with the modified chemical reactivity are for sure associated to modifications of electronic properties which have not been largely discussed in this paper. However, it has been the object of several experimental and theoretical papers in relation with the chemical reactivity 9see for example [75-77, 19, 21, 20]. The trends with respect to chemisorption of CO [19] and unsaturated hydrocarbons [21] can be summarized as follows : on compressed pseudomorphic overlayers, large interaction and overlap between surrounding atoms will result in a broadening of the d valence band shifting downwards from the Fermi level if the d-band filling is larger than 0.5; the adsorption energy of CO and hydrocarbons [21] is decreased. On the contrary, tensile strain induces a narrowing of the d band shifting upward accompanied by an increase of adsorption energies, with important consequences for catalytic reactions. One can expect that compressive stress will make easier the addition reactions such as hydrogenation, while tensile stress will facilitate bond cleavage reactions such as hydrogenolisis. In conclusion, strained surfaces can show very original structures and new catalytic properties. In order to associate the modified catalytic properties to the peculiar structures generated, one has to asume that these original structures are still present under the reactive mixture, at high pressure. Measurements under pressure of reactants are then necessary to measure both the surface structure and the surface species as reaction intermediates. Up to now, only very few data are available in that field. Recent developments around techniques such as STM [79-80], grazing X-ray Diffraction [81] ... and optical vibrational spectroscopies such as IRRAS[82-83] using a polarized light and SFG [79] have demonstrated the possibility to realise such observations. Finally, a good knowledge of the processes involved at alloy surfaces and surfaces alloys, and the ability to control the stress would enable us to carry out the synthesis of new catalysts having tailor made surface sites specific for a particular reaction.

ACKNOWLEDGEMENTS Acknowledgment is made to R6gion Rh6ne-Alpes for its financial support through a project "nanotechnologies" (# PR97039).

435 The authors would like to thank all those who contributed to the results presented in this work : M. Abel, L. Porte and Y. Robach for their STM observations, P. Delich~re for LEIS investigations, P. Ruiz for his contribution in doing the catalytic reactions, R. Baudoing-Savois, P. Dolle, M.C. Saint-Lager, and M. De Santis for the X-ray diffraction experiments and quantitative analysis of the results, J.S. Filhol, D. Simon, P. Sautet for their theoretical contributions on the Pd/Ni(110) system. We would like also to acknowledge several students and postdoctoral fellows for their participation to this work, among them L. Constant, A. Franquet, J.M. Guigner, P. Hermann, L. Lianos, and A.C. Michel. X A N E S and X-ray Diffraction experiments have been done in L U R E (Orsay) and ESRF (Grenoble) respectively. We would like to thank beam operators and beam line local contacts. Lastly, we would like to thank N.S. Prakash for his help in proof reading.

REFERENCES [ 1] J.H. Sinfelt, Bimetallic Catalysts, Wiley, New-York, 1983. [2] B.E. Nieuwenhuys, The Chemical Properties of Solid Surfaces, D.A. King and D.P. Woodruff eds., Elsevier, Amsterdam, 1993, pp. 185-224. [3] J.H. Sinfelt, J.L. Carter and J.T. Yates, J. Catal. 24 (1972) 283. [4] G.A. Martin, Catal. Rev. Sci. Eng. 30 (1988) 519. [5]C.T. Campbell, J.M. Campbell, P.J. Dalton, F.C. Henn, J.A. Rodriguez and S.G.Seimanides, J. Chem. Phys 93 (1989) 815. [6] P.W. Wynblatt and R.C. Ku, Surf. Sci. 65 (1974) 511. [7] J.L. Rousset, J.C. Bertolini and P. Miegge, Phys. Rev. B 53 (1996) 4947 and references therein. [8] Y. Gauthier, Y. Joly, R. Baudoing and J. Rundgren, Phys. Rev. B 31 (1985) 6216. [9] E. Bauer, Z. Kryst. 110 (1958) 372. [10] E. Bauer and J.H. Van der Merwe, Phys. Rev. B 33 (1986) 3657. [11] R. Koch, The Chemistry of Solid Surfaces, vol. 8, the growth and properties of ultrathin epitaxial layers, D.A. King and D.P. Woodruff eds., Elsevier, Amsterdam, 1997. [12] H. Ibach, Surf. Sci. Rep. 29 (1997) 193 and references therein. [13] M. Schmid, W. Hofer, P. Varga, P. Stoltze, K.W. Jacobsen and J.K. Norskov, Phys. Rev. B 51 (1995) 10937. [14] F. Besenbacher, L. Pleth Nielsen and T. Sprunger, The Chemical Physics of Solid Surfaces and Heteregeneous Catalysis, Surface alloying in heteroepitaxial growth, D.A. King and D.P. Woodruff eds., Elsevier, Amsterdam, 1997. Vol. 8, Chapter 10 [15] See for example Th. Pearsall ed., Strained-layer Superlattices- Materials Science and Technology in Semiconductors and Semimetals, Vol. 32, Academic Press, Boston, 1991. [16] J.M. Heitzinger, S.C. Gebhardt and B.E. Koel, Surf. Sci. 275 (1992) 209. [17] R.A. Campbell, J.A. Rodriguez and D.W. Goodman, Phys. Rev. B 46 (1992) 7077. [18] M. Gsell, P. Jakob and D. Menzel, Science 280 (1998) 717. [19] M. Mavritakis, B. Hammer and J.K. Norskov, Phys. Rev. Lett. 81 (1998) 2819. [20] P. Sautet, Stress and Strain in Epitaxy : theoretical concepts, measurements and applications, M. Hanbucken and J.P. Deville eds., Elsevier, Amsterdam, 2001.

436 [21] V. Pallassama and M. Neurock, J. Catal. 191 (2000) 301. [22] A.W. Denier Van der Gon, R. Cortenraad, W.P.A. Jansen, M.A. Reijme and H.H. Brongersma, Nucl. Instr. and Meth. in Phys. Res. B 161-163 (2000) 56. [23] D.O. Boerma, Nucl. Instr. and Meth. in Phys. Res. B 183 (2001) 73. [24] D. Briggs and M.P. Seah eds., Practical Surface Analysis, 2nd edition, volume2, John Wiley and sons eds, Chichester (1992). [25] D.P. Woodruff, D. Brown, P.D. Quinn, T.C.Q. Noakes and P. Bailey, Nucl. Instr. and Meth. in Phys. Res. B 183 (2001) 128. [26] P. Bailey, T.C.Q. Noakes, C.J. Baddeley, S.P. Tear and D.P. Woodruff, Nucl. Instr. and Meth. in Phys. Res. B 183 (2001) 62. [27] D.P. Woodruff, R. Davis, N.A. Booth, A.M. Bradshaw, C.J. Hirschmugl, K.M. Schindler, O. Schaff, V. Femandez, A. Theobald, Ph. Hofmann and V. Fritzsche, Surf. Sci. 357-358 (1996) 19. [28] Y. Gauthier, R. Baudoing, Y. Joly, J. Rundgren, J.C. Bertolini and J. Massardier, Surf. Sci. 162 (1985) 342. [29] J. Massardier, J.C. Bertolini and A.J. Renouprez, Proceedings of the 9 th International Congress on Catalysis, M.J. Philips and M. Teman eds., Calgary, Vol. 3 (1988) 1222. [30] T. Ouchaib, J. Massardier and A.J. Renouprez, J. Catal. 119 (1989) 517. [31] J.C. Bertolini, A. Cassuto, Y. Jugnet, J. Massardier, B. Tardy and G. Tourillon, Surf. Sci 349 (1996) 88. [32] P.Sautet and J.F. Paul, Catal. Lett. 9 (1991) 245. [33] J.C. Bertolini, J. Massardier, P. Ruiz and B. Tardy, Surf. Sci. 211-212 (1989) 1053. [34] J.B. Darby Jr and K.M. Mills, Metallurgical Transactions 3 (1972) 653. [35] R. Hultgen, P.D. Desai, D.T. Hawkins, M. Gleiser and K.K. Kelley, Selected Values of Thermodynamic Properties of Binary Alloys, American Society of Metals, Metals Park OH, 1973. [36] W.R. Tyson and W.R. Miller, Surf. Sci. 62 (1977) 267. [37] P. Miegge, J.L. Rousset, B. Tardy, J. Massardier and J.C. Bertolini, J. Catal. 149 (1994) 404. [38] J.C. Bertolini, P. Hermann and P. Delich6re, Surf. Interfaces Anal. 24 (1996) 34. [39] I. Moysan, V. Paul-Boncour, S. Thiebault, E. Sciora, J.M. Foumier, R. Cortes, S. Bourgeois and A. Percheron-Gu6gan, J. of Alloys and Compounds 322 (2001) 14. [40] J. Goetz, Y. Murzin and R. Touroude, Ind. Eng. Chem. Res. 35 (1996) 703. [41 ] J. Goetz, M.A. Volpe and R. Touroude, J. Catal. 164 (1996) 369. [42] A.C. Michel, L. Lianos, P. Delich6re, N.S. Prakash, J. Massardier, Y. Jugnet and J.C. Bertolini, Surf. Sci. 416 (1998) 288. [43] M. Abel, Y. Robach, J-C. Bertolini et L. Porte, Surf. Sci. 454-456 (2000) 1. [44] L. Lianos, Y. Debauge, J. Massardier, Y. Jugnet and J.C. Bertolini, Catat. Lett. 44 (1997) 211. [45] A.C. Michel, thesis, Universit6 Claude Bemard Lyonl, November 2000. [46] A. Rochefort, M. Abon, P. Delich6re and J.C. Bertolini, Surf. Sci. 294 (1993) 43. [47] Y. Debauge, M. Abon, J.C. Bertolini, J. Massardier and A. Rochefort, Appl. Surface Sci. 90 (1995) 15. [48] R.S. Rao, A. Bansil, H. Aasonen, and M. Pessa, Phys. Rev.B 29 (1984) 1713. [49] D.J. Holmes, D.A. King and C.J. Barnes, Surf. Sci. 227 (1190) 179. [50] M.A. Newton, S.M. Francis, Y. Li, D. Law and M. Bowker, Surf. Sci. 259 (1991) 45. [51] C. Sousa,V. Bertin and F. Illas, J. Phys. Chem. B 105 (2001) 1817. [52] P. Hermann, B. Tardy, D. Simon, J.M. Guigner, B. Bigot and J.C. Bertolini, Surf. Sci. 307-309 (1994) 422.

437 [53] S. Terada, T. Yokoyama, N. Saito, Y. Okamoto and T.Ohta, Surf. Sci. 433-435 (1999) 657. [54] X.D. Xiao, X.D. Zhu, W. Daum, Y.R. Shen, Phys. Rev. Lett. 66 (1991) 2352. [55] L. Porte, M. Phaner-Goutorbe, J.M. Guigner and J.C. Bertolini, Surf. Sci. 424 (1999) 262. [56] M.C. Saint-Lager, private communication. [57] J.S. Filhol, D. Simon and P. Sautet, Surf. Sci. Lett. 472 (2001) 139. [58] J.M. Guigner, thesis, Universit6 Claude Bernard, Lyonl, January 1997. [59] J.S. Filhol, thesis, Ecole Normale Sup6rieure, Lyon, July 2001. [60] L. Constant, P. Ruiz, M. Abel, Y. Robach, L. Porte and J.C. Bertolini, Topics in Catalysis 14 (2001) 125. [61] W. Murray, S. Thorshaug, I. Stensgaard, F Besenbacher, E. Laegsgaard, A.V. Ruban, K.W. Jacobsen, G. Kopidakis, and H.L. Dkiver, Phys. Rev. B 55 (1997) 1380. [62] Y. Fujinaga, Surf. Sci. 86 (1979) 581. [63] J.E. Garces, G.H. Bozzolo, P.Abeland H.O. Mosca, Appl. Surface Sci. 167 (2000) 18. [64] R.A. Bennett, S. Poulston, N.J. Price, J.P. Reilly, P Stone, C.J. Barnes and M. Bowker, Surf. Sci. 433-435 (1999) 816. [65] J. Kuntze, S. Speller, W Heiland, A. Atrei, G. Rovida and U. Bardi, Phys. Rev. B 60 (1999) 1535. [66] J. Kuntze, S. Speller, W. Heiland, P. Deurinck, C. Creemers, A. Atrei and U. Bardi, Phys. Rev. B 60 (1999) 9010. [67] B.E. Koel, A Sellidj and M.T. Paffett, Phys. Rev. B 46 (1992) 7846. [68] L.A. Kibler, M. Kleinert, R. Kandler, D.M. Kolb, Surf. Sci. 443 (1999) 19. [69] L.A. Kibler, M. Kleinert, R. Kandler, D.M. Kolb, Surf. Sci. 461 (2000) 155. [70] A Sellidj and B.E. Koel, Phys. Rev. B 49 (1994) 8367. [71] C.J. Baddeley, R.M. Omerod, A.W. Stephenson and R.M. Lambert, J. Phys. Chem. 99 (1995) 146. [72] P. Dolle, M.C. Saint-Lager, M. de Santis, R. Baudoing-Savois, M. Abel, J.C. Bertolini and P. Delich6re, Surf. Sci., submitted. [73] M. Abel, thesis, Ecole Centrale de Lyon, September 2001. [74] P. Delich6re and J.C. Bertolini, Surf. Interface Anal., submitted. [75] J.A. Rodriguez and D.W. Goodman, J. Phys. Chem. 95 (1991) 4196. [76] J.A. Rodriguez, Surf. Sci. Rep. 24 (1996) 233. [77] J.H. Larsen and Ib Chorkendorff, Surface Sci. Rep. 35 (1999) 163. [78] A. Ruban, B. Hammer, P. Stoltze, H.L. Skriver and J.K. Norskov, J. Molec. Catalysis A : Chemical 115 (1997) 421. [79] G.A. Somorjai, Appl. Surf. Sci. 121-122 (1997) 1. [80] L. Osterlund, P.B. Rasmussen, P. Thostrup, E. Laegsgaard, I. Stensgaard and F. Besenbacher, Phys. Rev. Lett. 86 (2001) 460. [81] H. Isem, K. Peters, P. Steadman, O. Robach, J. Alvarez, E. Lundgren and S. Ferrer, Surf. Sci. 482-485 (2001) 101. [82] M. Endo, T. Matsumoto, J. Kubota, K. Domen and C. Hirose, Surf. Sci. Lett. 441 (1999) 931. [83] Y. Jugnet, L.J. Shorthouse, J.C. Bertolini and P. Colin, Silicon for the Chemical IndustryV, H.A. Oye, H.M. Rong, L. Nygaard, G. Schussler and J. Kr. Tuset eds., Tromso, 2000, pp 331-345.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces 438

D.P. Woodruff, (Editor)

Chapter 12

Electronic and chemical properties of palladium in bimetallic systems: How much do we know about heteronuclear metalmetal bonding? Jos~ A. Rodriguez Department of Chemistry, Brookhaven National Laboratory, Upton, New York 11973, USA

1. INTRODUCTION In many industrial applications, bimetallic systems are superior to their single metal-metal counterparts in terms of catalytic activity and/or selectivity [ 1-4]. For a long time it has been known that a bimetallic surface can exhibit chemical and catalytic properties that are very different from those of the surfaces of the individual metals. Systematic research on alloy catalysts started in the late 1940s [5-7] with the purpose of establishing links between the electronic and catalytic properties of a surface, a knowledge necessary for a scientific design of catalysts. However, due to the lack of adequate techniques for the preparation and characterization of the surface alloys, no real progress was made at an experimental level. In the 1960s and 1970s the development of bimetallic catalysts for hydrocarbon reforming in the petrochemical industry increased the need for a fundamental understanding of the behaviour of bimetallic surfaces, and renewed the interest in catalysis by alloys [2,3,8,9]. This effort provided the basis for the concepts of "ensemble" and "ligand" effects [3,9], which are frequently used to rationalize the superior performance of bimetallic catalysts. "Ensemble" effects are defmed in terms of the number of surface atoms needed for a catalytic process to occur. Changes in catalyst composition modify the ensembles of available active sites. "Ligand" effects refer to those modifications in catalytic activity or selectivity that are the product of electronic interactions between the components of the bimetallic system. Over the years, it has become clear that it is difficult to find pure "ensemble" or "electronic" effects [ 10]. In the last two decades, the development of new experimental techniques [11,12] and reliable theoretical methods [13,14] have

439 made it feasible to study in detail electronic and chemical properties of bimetallic surfaces. Thus, many phenomena responsible for the behaviour of bimetallic surfaces have been identified [ 14-16]. Yet, several important issues associated with heteronuclear metal-metal bonding remain mysterious or badly understood [15,17,181. In this chapter, an overview is presented of studies that deal with the electronic and chemical properties of Pd in bimetallic systems. We will focus on palladium for three main reasons. First, bimetallic catalysts that contain Pd or other Group-10 metals have many uses: isomerization of hydrocarbons, olefin hydrogenation, CO oxidation, alcohol synthesis, acetylene trimerization, etc. [8,10,19-21]. Second, palladium is very sensitive to the formation of bimetallic bonds [22-24]. And third, there is a vast number of experimental and theoretical articles in the literature that examine the properties of Pd in bimetallic systems [14,15,19-23,25-44]. From this large volume of work, one can get a general idea of how deep is our knowledge about the basic nature of bimetallic bonding and how it affects the properties of a metal. The chapter is organized as follows. It starts with a description ofphotoemission and thermal desorption experiments for Pd overlayers on different types of metal substrates. General trends in the experimental data are examined and bonding models that explain them are discussed. Then, the validity of the bonding models is tested through ab initio or first-principles quantum mechanical calculations. From the combination of experiments and theory, a complete picture of the nature of bimetallic bonding is beginning to emerge. 2. PHOTOEMISSION STUDIES Pd atoms bonded to surfaces of early-transition metals exhibit large electronic perturbations in their valence and core levels [15]. The valence photoemission spectra shown in Figure 1 for Pd/Nb(ll0) and Pd/W(100) illustrate this phenomenon [26,43,44]. In early studies examining the interaction of Pd with a Nb(100) surface [43], it was found that the supported Pd monolayer (ML) had a relatively narrow 4d band which exhibited a low density of states (DOS) around the Fermi level (EF). In contrast, Pd multilayers and bulk palladium show emission spectra characterized by a large DOS at EF. More recent photoemission studies for a Pd layer in contact with Wa(ll0) [25], W(100) [26], W(ll0) [26] and Mo(ll0) [15,45] also show a narrow Pd(4d) band with a centroid shifted toward higher binding energy. Thus, it appears that the bonding interactions between Pd and earlytransition metals are quite strong. This will be confmned below by results of

440

Pd/W(100)

Pd/Nb(; 10i '

(~

~,z, / I |

/!i I

~

e(10,5/cm2)

P"

---3.4

t

\! 09

/0=1 ~

0 lo;8;i BINDING ENERGY (eV) J,

--0.0

I,

8

6

4

2

0

BINDING ENERGY ( e V )

Fig. 1 (a) Ultraviolet photoelectron spectra for monolayer (dashed curve) and greater than monolayer coverages of Pd on Nb(ll0). (b) UPS spectra of various coverages (0) of Pd on W(100). Reprinted from ref. [44]. thermal desorption mass spectroscopy. Figure 2 displays photoemission spectra for the valence region of Pd/Rh(111) as a function of admetal coverage [32]. The Pdo.9/Rh(lll) system exhibits a band structure that is very similar to that of Rh(111) or Pd multilayers. Difference spectra showed only minor electronic perturbations for supported palladium near the Fermi level [32].

-

Valence

o.9

~

-

1).4

~

1

1

l

1

1

I

12

10

8

6

4

2

0

Binding Energy (eV) Fig. 2 Valence spectra for the bonding of Pd to Rh(111). Reprinted from ref. [32].

441 The magnitude of the binding-energy shift in the Pd 4d band depends on the position of the metal substrate in the Periodic Table. Figure 3 shows the electronic perturbations observed for Pd in surface alloys (PdTi [46] and PdA1 [47,48]) and Pd monolayers supported on several metals (Ta(ll0) [25], M o ( l l 0 ) [15,45], W ( l l 0 ) [26], Re(0001) [27], Ru(0001) [27] and A I ( l l l ) [49]). The experimental results are ordered according to the group in the Periodic Table of the metal bonded to Pd. One finds that the electronic perturbations for the bonding of Pd to s,p metals like A1 [47-49] or Zn [35,50] are as large as those seen for Pd bonded to early-transition metals, and much bigger than those found when Pd is bonded to late-transition metals. In general, the magnitude of the shill in the Pd valence levels increases when the fraction of empty states in the valence band of the metal substrate rises [23,48]. This phenomenon could result from a simple hybridisation of the admetal and substrate valence bands [14,25]. In addition, a substrate induced Pd(4d)--Pd(5s,5p) rehybridization could contribute to it [23,51,52]. It is interesting that the systems with the largest shifts reported for the centroid of the Pd 4d band (Pd/A1, Pd/Zn, Pd/Ti) also undergo alloy formation [46-50]. Indeed, results to be shown below show a correlation between the strength of the bimetallic bond and the size of the electronic perturbations in Pd. The core levels of Pd are also very sensitive to the formation of bimetallic bonds. Figure 4 shows Pd 3d XPS spectra for different coverages of palladium on

,~__%2.7 "I=t 2.4 ,-~ 2.1 1.8 1.5

a~ 1.2 0.9

.~

o.6

~ o.3 r~

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

I

I

I

I

\\\ \\\ \\\ \\\ \\\

-0.0

Pd/A1 PdTi Pd/Ta Pd/Mo Pd/W Pd/Re Pd/RuPd(100) Fig. 3 Effects of bimetallic bonding on the properties of Pd: Shift in the first peak of the Pd 4d band, the one closer to the Fermi level, as a function of metal substrate. Reprinted from ref. [15].

442

XPS

Pd

3d3/2

03 I-Z

izi re

,a: i

z I,LI I.Z

X8.6

x5.1

3.5

xl

.41

344

342

340

BINDING

338 ENERGY,

336

334

eV

Fig. 4 Palladium3d core-level spectra for the Pd/W(110) system. Reprinted from ref. [53]. W ( l l 0 ) [53]. There is no observable change with coverage in the separation of the palladium 3d core levels. In the top panel of Figure 5, one can observe that the supported palladium monolayer has a Pd 3d~/2 binding energy substantially larger than that seen for palladium multilayers. Photoemission studies indicate that the Pd 3d5/2 binding energy of the surface atoms of Pd(100) is - 0.4 eV smaller than that of bulk Pd [54]. When this is taken into consideration [53], one fmds that palladium atoms bonded to W ( l l 0 ) have 3d core levels shitted - 0.85 eV toward higher binding energy with respect to those of the surface atoms of pure palladium. The perturbations induced by tungsten on the palladium core levels affect not only the first layer in direct contact with the substrate but also subsequent layers (see Figure 5) [53]. This phenomenon has been also observed on Re(0001) [27] and Mo(ll0) [44]. It tracks changes in the structural and chemical properties of the palladium adatoms [27,53].

443 ,I/----

.20 .10 .90 .80

Pd/W(110) .

335.70 IJJ Z

Ill

=,o

.10

(oool)

336.00 Z

.so

t'~

.so

Z

.--- 335.70 rn

U3

"o ~

"13

,

,

,

,

,

,

,

~

.20

,o

,

~:;,~______.

1 o)

o

336.00

.90.80

. . . .

,, i " "

335.70

,, 0

1

2

3

4

5

1

6

i

7

t

8

I

9

10

j/-"

40

Pd COVERAGE, ML Fig. 5 Palladium 3d5/2 core-level positions for Pd/W(ll0) , top, Pd/Re(0001), center, and Pd/Mo(110), bottom, as a function of Pd coverage. Reprintedfrom refs. [27,44]. Figure 6 compares 3ds/2 core-level binding-energy shifts for the deposition of palladium on several metal substrates: A I ( l l l ) [48], Ti [15,46], Ta(ll0) [25,27], Mo(ll0) [15], W(ll0) [53], Re(0001) [27] and Ru(0001) [27]. Alloying takes place in the Pd/AI(lll) and Pd/Ti systems with very big core-level shifts. In all cases, bimetallic bonding shifts the Pd core levels towards higher binding energy. The electronic perturbations in palladium are larger when the element is bonded to a s,p metal or to a transition metal with a valence band almost empty. The case of Pd/Re(0001) is particularly interesting because the palladium adlayer is pseudomorphic to the rhenium substrate, with an atomic density and structure that are very similar to those of the surface atoms in Pd(111) [27]. Yet, the admetal atoms in Pd/Re(0001) are electronically and chemically perturbed due to the effects of bimetallic bonding [27]. The trends in Figures 3 and 6 are identical. In fact, one can say that the shifts in the Pd core levels track shifts in the centroid of the Pd

444

1.8 1.6 1.4 1.2 ,~ ~.o

"~ 0.8 "~ 0.6 "~ 0.4 0.2 0.0

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

\

Pd/AI

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

4\ ..\ \ \ ~,,\\ b,.\\ N.\\ ~_ .. x

PdTi

\[\\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

j\\\

\ \ \

x_ \ \ \ \ \ \\\ \ \.\. \ \ \ \ \ \ \ \ \ x.x.x \ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \

\ \ \ \ \ \ \ \ \ \ \ \

Pd/Ta PdRVlo Pd/W

Pd/Re

Pd/RuPd(100)

Fig. 6 Effects of bimetallic bonding on the properties of Pd: Shift in the Pd 3d5/2 core level as a function of metal substrate. Reprinted from ref. [15]. 4d band [15,17], although the magnitude of the shifts in the core and valence levels is different in many cases. The type of perturbations seen for Pd in the bimetallic surfaces are similar in many aspects to those found in bulk alloys, where heteronuclear metal-metal bonding induces an increase in the binding energy of the core levels and valence band of Pd [56-59]. Results of x-ray absorption spectroscopy indicate that this phenomenon is accompanied by a reduction in the d electron population of Pd [51,52]. Core-level shifts have been detected in many bimetallic surfaces [15]. Other admetals also exhibit well defined trends as seen for palladium [15]. For example, Figure 7 displays core-level shitts measured after depositing a monolayer of platinum or nickel on a series of metal substrates [23]. Two clear trends can be observed in these experimental data. First, the magnitude of the core-level shitt for an admetal increases when the fraction of empty states in the valence band of the metal substrate rises: ruthenium < titanium < aluminum. And second, the larger the occupancy of the admetal d band, the bigger the core level shift in the admetal: nickel < platinum < palladium [23]. The largest electronic perturbations are found in systems that combine and admetal with an electron-rich d band and a substrate with an electron-poor valence band. A priori, it is not clear what causes the core level shifts seen in Figures 4-7 and, in particular, if these shifts come from initial state effects [60,61]. In principle,

445 2.0 Ni 1.5 1.0 r~

.~

0.5 0.0

~

/

1

Ni/AI Ni/Ti

/ Ni/Ta

Ni/W

Ni/Mo Ni/Ru

Pt

O

r,,)

"O

Ni(100)

2.0

1.5 1.0

<

0.5 0.0

Pt/AI Pt/Ta

Pt/Zn

Pt/W

Pt/Re

Pt/Ni

Pt(lll)

Fig. 7 Core-level shifts for supported monolayers ofNi and Pt as a function of metal substrate. Reprinted from res [23]. binding-energy shifts in core-level photoemission can be a consequence of initial state effects (i.e. real variations in the position of the core level produced by charge transfer, orbital rehybridization, volume perturbations, changes in chemical environment, etc) and/or final state effects (i.e. "artifacts" produced by changes in the screening of the core hole) [60-62]. The good agreement between the trends in Figures 3 and 6 suggests that the core level shifts, at least in part, originate from initial state effects. Direct support for this idea is also provided by the experimental data presented in the next two sections. And the theoretical results discussed towards the end of this chapter confmn that, indeed, the core level shifts reflect initial state effects. 3.

T H E R M A L DESORPTION STUDIES

In bimetallic systems like Pd/A1, Pd/Ti and Pd/Zn the bonding interactions between the metals are so strong that intermixing and formation of bulk alloys take place [35,46-50]. Alloy formation does not occur in many other bimetallic systems (Pd/Ta [25], Pd/W [26,53], Pd/Re [27], P d ~ u [27] , Pd/Rh [27]) and one can examine the strength of the corresponding bimetallic bonds using thermal

446 desorption mass spectroscopy [22]. Typical results are shown in Figure 8. After depositing palladium on Rh(111) and taking TDS spectra, two desorpfion states of palladium are seen [32]. The high temperature peak (at N 1390 K) corresponds to desorpfion of the first palladium monolayer (i.e. cleavage of Pd-Rh bonds). The low temperature peak (onset at N 1200 K) is due to multilayer P d desorption [63-67], and its intensity grew continuously with increasing palladium coverage [32]. These results show a clear difference (~ 6 kcal/mol) in the strength of the Pd-Rh and Pd-Pd bonds. Even higher desorption temperatures have been reported for palladium atoms bonded to Ta(ll0) [63,64], Mo(ll0) [65], W(ll0) [66,67], Re(0001) [27] and Ru(0001) [27]. Figure 9 compares desorption temperatures and core-level binding energy shifts observed for a monolayer of palladium on Ta(ll0) [27,63,64], W(110) [53,66,67],

/~

Pd-TDS Pd/Rh(lll)

5.3 ,4..a

~

~

.

1.9

~

.~_____~ 9

.

1100

.

.

_

1.3

~ ~ _

0.6 O.4

~

. I

!

1200

1300

,

i

1400

I

I

1500

,

1600

Temperature (K) Fig. 8 Thermal desorption spectra (m/e= 106) for Pd films on Rh(111). Pd was vapor deposited at a sample temperature of-- 300 K. Heating rate= 7 K/s. Reprinted from ref. [32].

447 M o ( l l 0 ) [15,65], Re(0001) [27], Ru(0001) [27] and R h ( l l l ) [32]. When going from a rhodium to a tantalum substrate, there is an increase of-~ 150 K in the palladium desorption temperature, which indicates an enhancement of 10-12 kcal/mol in the strength of the Pd-substrate bond. At the same time, the Pd 3d5/2 core-level shift increases by more than a factor of four. The stronger the bimetallic bond, the larger the electronic perturbations in the Pd atoms. The strongest metalmetal bonds are seen in systems that combine a metal with an electron-rich d band (Pd) and a substrate with an electron-poor d band (Ta, W, Mo).

g

1

1550

~ 1500 b 1450

1400

1350

m

,~

0.8

m ~0.6 ~ 9'~

0.4

0.2 0.0

Ta(ll0) Mo(ll0)W(110) Re(001) Ru(001) Rh(lll)

Fig. 9 Top: temperatures for the desorption ofa Pd monolayer from Ta(ll0), Mo(ll0), W(110), Re(0001), Ru(0001), and Rh(lll). Bottom: Shifts in the Pd 3ds/z binding energy of a Pd monolayer supported on several metal substrates. The shifts are reported with respect to the Pd 3d5/2 peak position for the surface atoms of Pd(100). Reprinted from ref. [32].

448 The type of correlation seen in Figure 9 has also been observed for Ni [22], Cu [22], Au [68] and Zn [69] overlayers. This suggests that in general the core level shifts do reflect changes in initial state induced by bimetallic bonding. And in most cases the formation of a strong metal-metal bond is associated with substantial perturbations in the electronic properties of the bonded metals [22, 23,68,69]. 4.

CO CHEMISORPTION STUDIES

The electronic perturbations described in section 2 could affect the chemical properties of Pd. Carbon monoxide is an ideal molecule to investigate the chemisorption properties of bimetallic surfaces. There is extensive information about the surface chemistry of this molecule on many monometallic substrates [70], and the bonding mechanism is much better known for CO [ 14,71,72] than for other simple molecule. In addition, CO is involved in many catalytic processes of industrial importance [1,4,70]. Figure 10 displays CO-thermal desorption spectra acquired after adsorbing

CO DESORPTION Pd/Ru(0001)

o3

Pd/W(110)

z n" ,<

pdCra(1 lo)

>.., iN

I rPd/Re(0ool)

.-" i

I

Y !

I

co z uJ z co o~ o9 o9 <

1

200

300 TEMPERATURE

1

1

400

500

-

(K)

Fig. 10 CO thermal desorption spectra for Pd(100), and a monolayer of Pd supported on Re(0001), Ru(0001), W(ll0) and Ta(110). Reprinted from res [27].

449 the molecule on Pd(100), and on a Pd monolayer supported on Ta(ll0) [64], W ( l l 0 ) [67], Re(0001) [27] and Ru(0001) [27]. For CO on Pd(100) a desorption maximum is seen near 480 K, close to the temperature observed on Pd(111) and other fiat Pd surfaces [27,67]. CO interacts strongly with pure Pd with an adsorption energy of N 35 kcal/mol [28,29]. For the Pd/Ta(ll0) and Pd/W(ll0) systems, the large decrease in the CO desorption temperature (180-230 K) indicates that there is a big weakening in the strength of the Pd-CO bond. The isosteric heat of adsorption of CO on the supported Pd monolayers is 15-20 kcal/mol smaller than on Pd(111) [28,29].

The valence photoemission spectra in Figure 11 were acquired after dosing

'1'I

l ! I 1'

I'1"1'I'I'I'I'I 9 (bl

.,m,m.,.

t0 e

9

~$t 4or

t--

%

,4

,

2"

i,,,,, o v )1-O3 Z UJ I-Z

~,",.,,~,t'r'P~,

~(a)

.

t \

Z 0

0 I-0 -r"

v l,i,lllAl -20

-16

-12 BINDING

-8

-4

EF

4

ENERGY (eV)

Fig. 11 Valence photoemission spectra for the adsorption of CO on Ta(110)-supported Pd films: (a) On a thick (> 3ML) Pd(ll 1)-like film, (b) on a pseudomorphic Pd monolayer. Reprinted from ref. [30].

450 CO to a Pd(111)-like thick film and a Pd monolayer supported on Ta(110) [30,31]. The spectrum for a thick palladium film is in very good agreement with that observed for adsorption of CO on a single-crystal Pd(111) surface. The features at ~ 11 and 8 eV correspond to emissions from the 40 and (1~ + 50) levels of CO, respectively [30,31]. In the photoemission spectrum for the Pd monolayer the 4o and (1~ + 50) peaks of CO appear at higher binding energy than in the spectnun for the Pd(lll)-like film, and there is also an extra "shake-up" satellite ('s' peak) around 13.6 eV. The spectrum for CO on the Pd monolayer matches the spectrum seen for CO on C u ( l l l ) [30,31], where the bonding interactions between the admolecule and metal substrate are much weaker than on Pd(111). Evidence for weak bonding between CO and supported palladium monolayers is also seen in XPS experiments (see Figure 12) [27]. The adsorption of CO on Pd(100) produces a binding energy shitt o f ~ 1.2 eV in the 3d core levels of the metal surface (see Figure 13) [32,54]. On the other hand, a palladium 3d shit~ of only 0.6 eV is observed in Figure 12 for CO on Pd/Re(0001) [27]. In fact, palladium monolayers supported on Ta(110), Ru(0001) and Rh(ll 1) also exhibit palladium 3d shifts upon CO adsorption smaller than that seen for the surface atoms of Pd(100) [32], as Figure 13 shows. There is a clear link between the strength of the Pd-CO bond and the CO-induced shitt in the palladium 3d core levels of the bimetallic .L

I-Z

I

I

I---

345

343

Pd(3d) XPS CO/Pd/Re(000~

I

I

/./-x~ /~ !

1

~

I

=

n-' < >!-11.,,,=

O3 Z LU I-Z

347

341

339

337

335

333

BINDING ENERGY (eV) Fig. 12 Pd(3d) XPS spectra for clean and CO-saturated Pd on Re(0001). The Pd was deposited at -~350 K and annealed to 500 K before dosing 10 L of CO at -- 115 K. Reprinted from ref. [27].

451 ~

5OO 0pd

~

1

400

~

300

9

r.) 1.4

ff -

~-~

1.0

=~

o.6

0pd "-" 1

~m 0.2

V-q Ta(ll0)

Re(001)

Ru(001)

Rh(lll)

Pd(100)

Fig. 13 CO desorption temperature and CO-induced shift on the P d 3ds/z binding energy for Pd(100) and Pd monolayers deposited on Ta(ll0), Re(0001), Ru(0001) and Rh(lll). Reprinted from ref. [32]. systems. Pd bonded to Ta(110) is not able to respond in an efficient way to the presence of CO, and essentially behaves as a noble metal. Infrared spectra for the CO/Pd/Ta(ll0) system show that CO is adsorbed linearly on top of the Pd atoms [28]. This is in contrast with the results typically seen for CO adsorbed on singlecrystal Pd surfaces where hollow or bridging CO are the norm, but similar to that found for CO on many copper surfaces [28]. From the experiments described above it is obvious that bimetallic bonding can have a tremendous impact on the chemical properties of a metal. It is important to establish in what kind of bimetallic systems one can expect the largest changes in chemical behaviour. Figure 14 displays the desorption temperature observed for Pdbonded CO on a series of bimetallic surfaces and Pd(100) [15,27,35,64,67,73-77],

452 together with the shift in Pd 3ds/2 binding energy found for each system before the adsorption of CO (i.e. pure metal-metal interactions, Figure 6) [15,25,27,35,46,53]. A clear correlation is seen between the changes in the electronic and chemical properties of Pd. The larger the shift in the Pd 3ds/2 binding energy induced by bimetallic bonding, the lower the CO desorption temperature from Pd. An identical trend is found when using the shifts in the Pd 4d valence band (shown in Figure 3) instead of the shifts in the Pd core levels. The biggest chemical perturbations are observed f o r Pd atoms bonded to early-transition metals or s,p metals, Jn bimetallic systems that essentially revolve the combination o f an element with an electron-rich valence band (Pd) and an element with an electron-poor valence band. In general, for adlayers of the Group-10 metals, one finds positive bindingenergy shifts in the core levels and a decrease in the CO desorption temperature (Figure 15)[22,23]. In contrast, Cu atoms deposited on late-transition metals exhibit negative core-level shifts and an increase in the desorption temperature

77 //

77 // // //

Pd/Ta

Pd/Mo

Pd/W

Pd/Re

Pd/Ru

// // // // // // // // // // // // // // // // // // // // // // // // // // //

Pd(100)

I'--

Pd/Cu

PdZn

I I Shift in Pd(3ds/z) binding energy [T~pd CO desorption t e m p e r a t u r e Fig. 14 Effects of bimetallic bonding on the properties of Pd. The graph displays the CO desorption temperature and the shifts in the Pd 3d~/2binding energy with respect to the surface atoms of Pd(100). Solid bars: shift in Pd 3d5/2binding energy; hatched bars: Pd CO desorption temperature. Reprinted from ref. [35].

453

Pt/Ta, -150 K [ + 1.30 eV

b,\\\\\\\\\\\\\\\\\"~

Pt/Zn, -110 K[ + 0.80 eV

L.\\\\\\\\\"~

Pt/W, -90 K I + 0.60 eV

k',{~~'~

Pt/Mo, -120 K [ Ni/W,-50 K I + 0.35 eV K',\\"~ Ni/Mo,-30 K l + 0.25 eV ~.~"~

] + 50 K, Ni/Ru + 0.05 eV +25 K, Cu/Re + 0.02 eV ]+30 K, Cu/Ru -0.13 eV ~"~

] +70 K, Cu/Pt -0.27 eV

~',,.~\'x,~

[ +70 K, Cu/Rh -0.43 eV

Shift in core-level binding energy Variation in CO desorption temperature Fig. 15 Correlation between shifts in surface core-level binding energy (crossed bars) and the shifts in CO desorption temperature (empty bars). The properties of the platinum, nickel and copper monolayers are compared with the corresponding values of the pure metals. Reprinted from ref. [15]. of CO (Figure 15) [22,65,78,79]. This is very important in the case of the C u ~ t ( 1 1 1 ) and Ct#Rh(100) systems [78,79]. The electronic perturbations induced by bimetallic bonding deactivate Group-10 metals toward CO chemisorption, whereas the same type of phenomena activate copper [65,78,79] and gold [80] adatoms. Bimetallic bonding can also produce interesting changes in the reactivity of a metal towards hydrogen (i.e. H2.os -.2H~s) [15,43,81-83]. Experimental evidence indicates

that hydrogen

dissociatively

chemisorbs

on

Pd/Nb(ll0)

at room

454 temperature when 0pd > 1 ML, but no (or little) hydrogen adsorbs when 0pd= 1 ML [43,81]. A similar behaviour is seen for the interaction of H2 with Pdffa(ll0) [82] and Pd/Mo(100) [83]. Electronic perturbations reduce the adsorption energy of ethylene on a Pd monolayer supported on Mo(100) [84]. Ethylene is weakly chemisorbed on the Pd monolayer (desorption temperature ~ 250 K against-~ 290 K on pure Pd), and the adsorbed species is much less rehybridized from sp2 in the gas phase toward sp3 on this surface compared to C2H4 chemisorbed on the (100) face of pure palladium [84]. 5. MODELS FOR BIMETALLIC BONDING The experimental results in Figures 9, 13, 14 and 15 show strong correlations between the electronic and chemical properties of an dement in a bimetallic surface. In the early 1990s, it became clear that the electronic perturbations induced by bimetallic bonding are associated with the strength of the heteronuclear metal-metal bond [27], and that these perturbations can determine the chemical reactivity of a bimetallic surface [22,44]. To explain the correlations in Figures 9, 13, 14 and 15 a model for bimetallic bonding was proposed [22,27,44]. There were three basic assumptions in the model. First, on the basis of the correlations in Figures 9 and 14, it was assumed that the shifts in the core levels reflected real changes in the initial state of the Pd electrons. Second, since the largest electronic perturbations were found in systems that involved "electron-rich + electron-poor" metal combinations (i.e. Pd/Ta, Pd/W, etc) with an admetal-induced reduction in the work function of the metal substrate, it was thought that bimetallic bonding produced some transfer of electrons (Pd ~) which eventually led to positive shifts in the core and valence levels of palladium. And third, it was proposed that the electronic perturbations in Pd reduced the strength of the Pd-CO bond by weakening z back-bonding. On metal surfaces the CO chemisorption bond is dominated by interactions between the occupied valence levels of the metal and the LUMO (2~ orbital) of the adsorbate (~ back-bonding) [71,72]. For supported Pd the 4d valence band is more stable than in pure Pd, probably weakening ~ back-bonding and leading to smaller CO adsorption energies [44,85]. At the time, this model for metal-metal bonding offered a logical and consistent explanation for the experimental facts [22,27,44]. Its three basic assumptions had to be validated by additional experimental and/or theoretical work. Photoemission studies have shown that in many cases the formation of a bimetallic bond induces positive core-level shills for both metals [17,86,87,88,]. This, obviously, is not consistent with a simple metal--metal charge transfer [60,90]. The phenomenon could be a consequence of combining inter- and intra-atomic charge redistributions (for example, d-.sp rehybridization) induced by bimetallic

455 bonding [23,51,60,90]. Thus, the bond between two different metals can be quite complex [17]. Theoretical studies have been useful for clarifying this issue and other aspects associated with heteronuclear metal-metal bonding. 6. T H E O R E T I C A L STUDIES

6.1 Charge redistribution in bimetallic bonding The nature of the bond between Pd and surfaces of transition or s,p metals has been the subject of a large series of theoretical works [23,33,34,35-42,89-91]. From these studies, it is clear that the Pd-substrate bond is best described as metallic with a small degree of ionic character. The direction of the net charge transfer (i.e. Pd~substrate or substrate-.Pd) varies from one calculation to the other. This discrepancy can be attributed to the lack of charge self-consistency in some of the calculations, and to the intrinsic difficulties associated with determining charge transfer, especially when the net amount of electron density transferred is small. The different schemes used for partitioning the electron population of each atom are more or less arbitrary [90,92,93], and in practice, the results of a given type of analysis can only be justified by comparing against the trends or predictions of experimental measurements. A reasonable approach is to plot the electron density around a metal atom and observe any possible change in the spatial distribution of the electrons [33,34,40,94]. The calculated electron density for a Pd monolayer supported on Ta(ll0) is plotted in Figure 16. These results are from first-principles density-functional calculations with the full-potential linearized augmented plane-wave (FPLAPW) method [33,34]. A strong Pd-Ta bonding interaction can be seen in the charge density difference shown in Figure 16c, where electrons deplete from both the interfacial Ta and Pd sites and accumulate in the region between them [33,34]. The more significant charge redistribution occurs around the Pd atoms, with the average center of electrons shifting away from the plane of Pd nuclei toward the substrate. The complex nature of the bimetallic bond in the Pd/Ta(ll0) system leads to positive core-level shifts for Pd and Ta [27,33,90,95,96]. The Pd-Ta bond cannot be classified as a simple "metallic" or "ionic" bond [33,34]. It involves and important shift of electrons from the Pd atom toward the Ta substrate, as the work function and Pd core levels suggest [48,97], and a simultaneous electron depletion around Ta, as the Ta core-level shifts and a simple Pd-Ta "covalent" interaction imply [87,90,96,97]. The FPLAPW method has also been used to study bimetallic bonding in Pd/W(ll0), Pd/Re(0001) and Pd/Ru(0001) [34]. In general, electron density plots show an important shift of electrons from the Pd layer toward the metal-metal

456

(a)

(b)

(c)

Pd

Ta(I)

Ta(I-1)

Ta(I-2)

Ta(C) Fig. 16 (a) Calculated valence charge density for a Pd monolayer (top) and clean Ta(ll0). (b) Calculated valence charge density for the Pd/Ta(110) system. (c) Charge density difference obtained by subtracting the superposition of the charge densities of the Pd monolayer and Ta(110) from that of Pd/Ta(110). Dashed lines indicate a decrease in the electron density. Reprinted from ref. [33]. interface. A similar result has been found in first-principles density-functional slab calculations for Pd/Mo(110) [40,98]. The larger the movement of electrons from around Pd to the metal-metal interface, the stronger the bimetallic bond [34,98]. The charge redistribution around Pd is in part caused by a Pd(4d)-. Pd(5s,5p) rehybridization that accumulates electrons in the bimetallic bonds [23,98]. Such a rehybridization has been observed in many theoretical studies, using different levels of theory and cluster or slab models [23,37-39,41,48,98]. In general, this redistribution of electrons is more significant than the net charge transfer between the Pd overlayers and metal substrates. From studies of x-ray absorption spectroscopy [51,52], it is known that Pd has a tendency to lose d electrons when forming bulk intermetallic compounds. Figure 17

457 shows the calculated 4d electron population for a Pd atom bonded to clusters that model hollow sites o f A I ( l l l ) , W ( l l 0 ) , R h ( l l l ) and P d ( l l l ) [23,32]. After comparing the results for Pd/Rh9 and Pd/Pdg, o n e can conclude that Rh induces minor changes in the electron distribution around the Pd atoms. This is consistent with the photoemission results in Figures 2 and Figure 9. For a Pd/Rh system the loss in the Pd 4d population, as a consequence o f a d-.s,p rehybridization and a Pd--substrate shift o f electrons, is smaller than for Pd/A1 and Pd/W systems [32]. The qualitative trends in Figures 3, 6 and 17 are identical: as the fraction o f empty states in the valence band o f the substrate rises, there is an increase in the magnitude o f the electronic perturbations in palladium. A similar correlation is observed in DF slab calculations for the bimetallic systems [34,98]. I0.0

Pd/X9 0 ,p,4

"K"K"X'I \ \ \ .,_ .,_ ,_ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ xxx x \ \

0

o

9.5

\ \ \ \ \ \ \ \ \ \ \ \ xxx \ \ \

9.0

X= AI

Cluster I (1:6:8)

W

Rh

,.x,~ ,. - . - . ,

\\\"

~\'?" ~\-,<

Pd

Cluster H (1:4:5)

Fig. 17 Bottom: Pd/X9 dusters used to study the adsorption of a Pd atom (dark sphere) on several metal surfaces. Cluster I models the bonding of Pd to an fcc hollow site of AI(lll), Rh(lll) or Pd(lll). Cluster II models the adsorption of palladium on a hollow site of W(ll0). Top: calculated 4d electron population for the palladium adatom in the Pd/X9 clusters. Reprinted from ref. [32].

458 6.2 Core-level and valence-band shifts

The redistribution of charge and d-.s,p rehybridisation observed in many theoretical calculations [23,33,34,37-39,40,41,48,98] should affect the position of the core levels and valence band of palladium. The Pd(4d) orbitals are more compact than the Pd(5s,5p) orbitals and, therefore, exhibit larger Coulomb interactions with the core electrons of palladium [23,60,99,100]. Thus, a d-.s,p rehybridisation reduces electron-electron repulsion and should shitt the Pd core levels and 4d band toward higher binding energy [51,60]. Early theoretical studies based on a semi-empirical self-consistent tight-binding scheme indicate that the core-level shifts in the Pd/W(110) and Pt/W(110) systems come from initial state effects (d-.s,p rehybridisation, for example) [37]. The calculated shift for the Pd 3d5/2 core level was 0.7 eV versus the value of 0.8 eV measured experimentally [53]. More sophisticated calculations (full-potential linear muffin-tin orbital method with LDF) for the Pd/Mo(110) system also indicate that the Pd 3d core-level shifts reflect initial state effects (substantial polarization of electrons around Pd) [40]. In this case, the calculated Pd 3d5/2 core level (0.9 eV) is identical to the experimental value and most of it (0.77 eV) comes from initial state effects while the rest (0.13 eV) originates in changes in the screening of the core hole [40]. Figure 18 summarizes results of calculations with the FLAPW method for Pd~a(ll0), Pd/W(ll0), Pd/Re(0001) and Pd/Ru(0001) [34]. At the bottom of the figure are shown the calculated and experimental Pd 3dsa core level shifts as a function of the calculated bonding energy for Pd (Eb) on each substrate. For the Pd/Re and Pd/W systems, the agreement between theory and experiment is very good. Discrepancies can be seen for the Pd/Ru and Pd/Ta systems. In general, the theoretical results imply that the core-level shitts measured experimentally mainly derive from initial state effects [34,90]. A strong correlation is obvious between the core-level shifts (for both the calculated and the measured results) and the bonding energy. For the calculated results, there is linearity for Pd/W, Pd/Re, and Pd/Ru, but not for Pd/Ta(110). Experimentally, by contrast, good linearity develops for Pd/Ta, Pd/W and Pd/Re but not for Pd/Ru(0001). The reasons for this discrepancy are not clear [34]. Theoretical studies show that bimetallic bonding increases the stability of the Pd 4d valence band [ 14,23,34,36,40,90,98]. The variation in surface core level shifts for metal overlayers is accompanied by a similar shitt in the center of gravity of the admetal d band [34,40,90]. In the top panel of Figure 18 is shown the calculated density-of-states (DOS) at the Fermi level for a palladium monolayer on four different metal substrates. As one moves from Pd/Ru to Pd/Ta, there is a substantial

459

1.5

'

l

'

'

'

l

'

'

'

l

'

'

'

l

'

'

'

l

'

'

'

l

'

'

'

l

'

'

'

l

1560

'

(a)

.a

Pd/Ru

1540

,, ,, "

~I~

tu

1520 1.0

r.r

ga, 1500 1480

~ ~:I

O

[.r.,.,

o~

0.5

1460

~

1440

~

1420

i~

.

a "

0.0

,

I

Pd/Ta

,

,

,

I

126

,

,

,

128

I

,

J

,

I

130

,

,

~

132

I

,

,

,

134

I

,

,

,

136

I

,

,

,

138

I

,

140

E b (kcal/mol) 1.0

,

i

,

,

,

i

,

,

,

i

,

,

,

i

,

,

,

i

,

0.9

,6 f=

,

,

i

,

,

,

I

,

,

,

I

,

(b) Pd/W .z~ J

0.8

Pd/Ta Pd/Re

0.7 0.6 0.5

Expt.

0.4

- - ~ - - Theo.

r,,,)

0.3 0.2

,

I

126

,

,

,

I

128

,

,

,

I

,

130

,

,

I

132

,

,

,

I

i

,

m

134

I

136

i

i

A

I

138

,

,

.

I

.

140

E b (kcal/mol) Fig. 18 Interaction of a Pd monolayer with Ru(0001), Re(0001), W(110) and Ta(110). Part (a): Calculated value for the density-of-states at the Fermi level and the measured desorption temperature as a function of the theoretical bonding energy. Part (b): Experimental and theoretical palladium 3d5/2 core level shifts versus the calculated bonding energy. Reprinted from ref. [34]. drop in the DOS at EF (i.e. the noble metal character of the bimetallic system increases). This agrees with the experimental results seen in Figure 1. Interestingly, a direct (almost linear) relationship between the bonding energy and the value of the DOS at EF for Pd atoms is observed in Figure 18. Such behaviour can be understood since a substantial redistribution of charge results in both a larger energy shift for the valence states and a larger bonding energy [34]. DF slab calculations have been used to study in a systematic way the effects of bimetallic bonding on the valence band of Pd and many other metals [14,36,101,102]. For metal overlayers, the strain induce by the metal substrate on

460 the structural configuration of the overlayer has a direct influence on the position and width of the admetal valence band [14,102]. The supported Pd monolayers in the bimetallic systems of Figure 18 all adopt a pseudomorphic structure with respect to the metal substrates [27,53,66,67]. In the cases of Pd/W(ll0) and Pd/Ta(ll0), this pseudomorphic configuration leads to a substantial stretching of the Pd-Pd distances with respect to those seen in bulk Pd or the Pd(111) and Pd(100) surfaces [27,53]. The weaker the Pd.-Pd interactions, the stronger the Pd--substrate interactions and the electronic perturbations on Pd. The case of Pd/Re(0001) is particularly interesting, since in this system the metal overlayer has an atomic density that is notvery different from that of Pd(111) [27], and the pure effects of metal-metal bonding shift the Pd 3d core levels and 4d band.

6.3 CO chemisorption The bonding mechanism between CO and a metal involves electron transfer from the CO(5o) orbital into the empty bands of the metal, o-donation, and electron transfer from the occupied bands of the metal into the CO(2~) orbitals, ~backdonation [36,71,72]. From a thermochemical viewpoint, x-backdonation is energetically more important than o-donation [71,72]. In principle, a positive shift in the Pd 4d band (Figures 3 and 18) and a decrease in the Pd 4d population (Figures 16 and 17) should reduce the ability of this metal to donate electrons into the CO(2x) orbital and weaken the Pd-CO bond. The experimental results in Figure 13 are in complete agreement with this idea. When Pd is deposited on metal substrates like Ta or Re, there is a reduction in the CO desorption temperature (weaker Pd-CO bonding) and in the magnitude of the shift induced by CO on the Pd 3d levels (decrease of x-backdonation [27,32]). Several theoretical studies have shown a relationship between a reduction in xbackbonding and the weakening of the Pd-CO on palladium overlayers: CO/Pd/Ti(001) [42,98], CO/Pd/W(ll0) [41,98,103], CO/Pd/Mo(ll0)[40,98], CO/Pd/Ru(0001) [36,42], CO/Pd/Rh(lll) [98], C O / P d / f u ( l l l ) [104], and CO/PdCu [38]. In some surfaces of bulk alloys, P d C u ( l l l ) and Pd3Mn(100), DF calculations show a weakening of the Pd-CO bond without a decrease in xbackdonation [91,105]. In addition to a reduction in ~-backdonation, a weakening of the Pd-CO bond may result from a decrease in 5o donation, variations in Pauli repulsion between adsorbate and surface, and changes in electron correlation [36,91,98,105]. Trends for the adsorption of CO on many bimetallic systems can be reproduced by a simple tight-binding model that includes the interactions between the metal d states and the CO 2x and 5o states, renormalized by the metal s,p continuum [36].

461

>, 0.0 11)

tx0 9

.

Cu:Cu3Pt(111)Ox~ Cu:Ni@Cu(111)/.Z~0u(111)

-0.5

9

Pd/Ru(O001) [ / . , . ~ wY Cu/Pt(111 )

O

-1.0 ~J o

Pd(111)/ / Pt(111) ---i

<

/0 Ni(111)

-1.5 ~:CusPt(111) / 95

-Ni:Ni@Cu(111)

I

I

-1.0

-0.5

i

,

0.0

0.5

Model of the d contribution, Ea_hyb (eV) Fig. 19 Comparison of CO chemisorption energies calculated with a tight-binding model and full DF-GGA calculations. Reprinted from ref. [36].

Figure 19 shows the scaling of the chemisorption energy for CO within the model as compared to results of DF slab calculations [36]. The good agreement indicates that the interactions included in the tight-binding model are responsible for the gross trends in CO chemisorption energies for the monometallic and bimetallic systems examined. The dominant contribution to the Ed-hyb term comes from the hybridisation or mixing of the metal d band and CO 2~ orbitals [36]. The energy released by this hybridisation decreases when going from CO/Pd(lll) to CO/Pd/Ru(0001) [36] or CO/Pd/Cu(111) [104]. The electronic perturbations which reduce the ability of palladium to 7~backdonate electrons to CO also limit electron transfer into the LUMO's of H2, C2H4 and SO2 [98,106]. For these adsorbates, theoretical calculations predict small adsorption energies and no dissociation if palladium is supported on early-transition metals [98,106].

462 7.

CONCLUSION The experimental and theoretical studies described above illustrate the complex

nature of the heteronuclear metal-metal bond. In many cases, bimetallic bonding induces a significant redistribution o f charge around the bonded metals. This redistribution o f charge is usually linked to the strength of the bimetallic bond, affects the position of the core and valence levels o f the metals, and can determine the chemical reactivity of the system under study. New concepts are emerging [22,23,34,36] and eventually the coupling o f experiment and theory can be useful for designing more efficient bimetallic catalysts [98,106,107]. ACKNOWLEDGEMENT Special thanks to W. Goodman for many thought-provoking conversations about the properties of bimetallic surfaces. I am also grateful to C.T. Campbell, J. Hrbek, M. Kuhn, T.K. Sham and M. Strongin for helpful discussions. This work was carried out at Brookhaven National Laboratory under Contract DE-AC0298CH10086 with the US Department of Energy (Division of Chemical Sciences). REFERENCES [1 ] J.M. Thomas and W.J. Thomas, Principles and Practice of Heterogeneous Catalysis, VCH, New York, 1997. [2] J.H. Sinfelt, Ace. Chem. Res. 10 (1977) 15. [3] W.H.M. Sachtler, Faraday Disc. Chem. Soc. 72 (1981) 7. [4] G. Ertl, H. KnSzinger, and J. Weitkamp (eds.), Handbook of Heterogeneous Catalysis, Wiley-VCH, New York, 1997. [5] G.M. Schwab, Disc. Faraday Soc. 8 (1950) 166. [6] A. Couper and D.D. Eley, Disc. Faraday Soc. 8 (1950) 172. [7] D.A. Dowden and P. Reynolds, Disc. Faraday Soc. 8 (1950) 184. [8] J.K.A. Clarke, Chem. Rev. 75 (1975) 291. [9] V. Ponec, Adv. Catal. 32 (1983) 149. [10] J.A. Rodriguez and D.W. Goodman, Surf. Sci. Reports 14 (1991) 1. [11] D.P. Woodruff and T.A. Delchar, Modem Techniques of Surface Science, Cambridge University Press, New York, 1986. [ 12] G. Ertl and J. Kiippers, Low Energy Electrons and Surface Chemistry, VCH, Weinheim, 1985. [13] R.A. van Santen and M. Neurock, Catal. Rev. Sci.- Eng. 37 (1995) 557. [14] B. Hammer and J.K. N~rskov, Adv. Catal. 45 (2000) 71. [15] J.A. Rodriguez, Surf. Sci. Reports, 24 (1996) 223. [16] C.T. Campbell, Annu. Rev. Phys. Chem. 41 (1990) 775. [17] J.A. Rodriguez and D.W. Goodman, Acc. Chem. Res. 28 (1995)477; M.A. Ruckman and M. Strongin, Acc. Chem. Res. 28 (1995) 479. [18] G.K. Wertheim and J.E. Rowe, Science, 260 (1993) 1527; J.A. Rodriguez and D.W. Goodman, Science, 260 (1993) 1528. [19] J. Szanyi, S. Anderson, and M.T. Paffett, J. Catal. 149 (1994) 438.

463 [20] C.J. Baddeley, R.M. Ormerod, A.W. Stephenson, and R.M. Lambert, J. Phys. Chem. 99 (1995) 5146. [21] P. Miegge, J.L. Rousset, B. Tardy, J. Massardier and J.C. Bertolini, J. Catal. 149 (1994) 404. [22] J.A. Rodriguez and D.W. Goodman, Science, 257 (1992) 897. [23] J.A. Rodriguez, Surf. Sci. 345 (1996) 347. [24] Z. Karpinski, Adv. Catal. 37 (1990) 45. [25] M.W. Ruckman, V. Murgai and M. Strongin, Phys. Rev. B, 34 (1986) 6759. [26] G.W. Graham, J. Vac. Sci. Technol. A, 4 (1986) 760. [27] R.A. Campbell, J.A. Rodriguez, and D.W. Goodman, Phys. Rev. B, 46 (1992) 7077. [28] W.K. Kuhn, J. Szanyi and D.W. Goodman, Surf. Sci. 303 (1994) 377. [29] Y.B. Zhao and R. Gomer, Surf. Sci. 239 (1990) 189. [30] M.W. Ruckman, P.D. Johnson and M. Strongin, Phys. Rev. B, 31 (1985) 3405. [31] M.W. Ruckman and M. Strongin, Phys. Rev. B, 29 (1984) 7105. [32] J.A. Rodriguez and M. Kuhn, Surf. Sci. 365 (1996) L669. [33] R. Wu, Chem. Phys. Lett. 238 (1995) 99. [34] R. Wu and A.J. Freeman, Phys. Rev. B, 52 (1995) 12419. [35] J.A. Rodriguez, J. Phys. Cherrt 98 (1994) 5758. [36] B. Hammer, Y. Morikawa and J.K. Norskov, Phys. Rev. Lett. 76 (1996) 2141. [37] S. Pick and P. Miku~ic, Cherrt Phys. Lett, 208 (1993) 97. [38] A. Rochefort and 1L Foumier, J. Phys. Chem. 100 (1996) 13506. [39] M. Femandez-Garcia, J.C. Conesa, A. Clotet, J.M. Ricart, N. Lopez and F. Illas, J. Phys. Chem. B, 102 (1998) 141. [40] D. Hennig, M.V. Ganduglia-Pirovano, and M. Schemer, Phys. Rev. B, 53 (1996) 10344. [41] S. Pick, Chem. Phys. Lett. 239 (1995) 84. [42] J.A. Rodriguez, Surf. Sci. 303 (1994) 366. [43] M. EI-Batanouny, M. Strongin, G.P. Williams and J. Colbert, Phys. Rev. Lett. 46(1981)269. [44] J.A. Rodriguez and D.W. Goodman, J. Phys. Cherrk 95 (1991) 4196. [45] To be published. [46] P. Mikusik and Z. Basil, Phys. Scri. 41 (1990) 130. [47] L.Q. Jiang, M.W. Ruckman and M. Strongin, Phys. Rev. B, 39 (1989) 1564. [48] J.A. Rodriguez, Surf. Sci. 318 (1994) 253; 303 (1994) 366. [49] B. Frick and IC Jacobi, Phys. Rev. B, 37 (1988) 4408. [50] A. Fasana and L. Braicovich, Surf. Sci. 120 (1982) 239. [51] T.K. Sham, Phys. Rev. B, 31 (1985) 1903. [52] Y. Jeon, J. Chen, and M. Croft, Phys. Rev. B, 50 (1994) 6555. [53] R.A. Campbell, J.A. Rodriguez and D.W. Goodman, Surf. Sci. 240 (1990) 71. [54] J.N. Andersen, N. Qvarford, R. Nyholm, S.L. Sorensen, and C. Wigren, Phys. Rev. Lett. 67 (1991) 2822. [55] J.A. Rodriguez and M. Kuhn, Chem. Phys. Lett. 240 (1995) 435. [56] J.C. Fuggle, F.U. Hillebrecht, 1L Zeller, Z. Zolnierek, P.A~ Bennett and C. Freiburg, Phys. Rev. B, 27 (1983) 2145. [57] F.U. Hillebrecht, J.C. Fuggle, P.A. Bennett, Z. Zolnierek and C. Freiburg, Phys. Rev. B, 27 (1983) 2179. [58] G. Wertheim, D. Buchanan and J. Wemick, Phys. Rev. B, 40 (1989) 5319. [59] P.N. Ross, J. Vac. Sci. Technol. A, 10 (1992) 2546. [60] W.F. Egelhoff, Surf. Sci. Reports, 6 (1987) 253. [61] G.A. Benesh and D.A. King, Chem. Phys. Lett. 191 (1992) 315. [62] J.A. Rodriguez, J. Hrbek, Y.-W. Yang, M. Kuhn and T.K. Sham, Surf. Sci. 293 (1993) 260. [63] J.M. Heitzinger, S.C. Gebhard and B.E. Koel, Surf. Sci. 275 (1992) 209.

464 [64] B.E. Koel, ILL Smith and P.J. Bodowitz, Surf. Sci. 231 (1990) 325. [65] J.A. Roddguez, R.A. Campbell and D.W. Goodtmn, J. Phys. Chem. 95 (1991) 5716. [66] W. Schlenk and E. Bauor, Surf. Sci. 93 (1980) 9. [67] P.J. Berlowitz and D.W. Goodman, Langmuir, 4 (1988) 1091. [68] M. Kuhn, J.A. Roddguoz, J. Hrbek, A. Bzowski and T.K. Sham, Surf. Sci. 341 (1995) L1011. [69] J.A. Rodriguez and M. Kuhn, J. Phys. Chem. 100 (1996) 381. [70] G.A. Somorjai, Introduction to Surface Chemistry and Catalysis, Wiley: New York, 1994. [71] K. Herman~ P.S. Bagus and C.J. Nelin, Phys. Rev. B, 35 (1987) 9467. [72] E.1L Davidson, K.L. Kunze, F.B.C. Machado and S.J. Chakravorty, Acc. Chem. Res. 26 (1993)628. [73] A. Sellidj and B.E. Koel, Phys. Rev. B, 49 (1994) 8367. [74] C.J. Baddeley et al, Surf. Sci. 314 (1994) 1. [75] T.D. Pope, K. Griffiths and P.R. Norton, Surf. Sci. 306 (1994) 294. [76] G.W. Graham, Surf. Sci. 171 (1986)IA32. [77] G.W. Graham, P.J. Schmitz, and P.A. Thiel, Phys. Rev. B, 41 (1990) 3353. [78] J.A. Rodriguez, ILA. Campbell and D.W. Goodman, J. Phys. Chem. 95 (1991) 2477. [79] M.T. Paffett, C.T. Campbell, T.N. Taylor and S. Srinivasam, Surf. Sci, 154 (1985) 284. [80] P.J. Schmitz, H.C. Kang, W.-Y. Leung and P.A. Thiel, Surf. Sci. 248 (1991) 287. [81] S.-L. Weng and M. EI-Batanouny, Phys. Rev. Lett. 44 (1980) 612. [82] J. Heitzinger, A. Avoyan and B.E. Koel, Surf. Sci. 294 (1993) 251. [83] J.M. Heitzinger, S.C. Gebhard and B.E. Koel, Chem. Phys. Lett. 200 (1992) 65. [84] J.M. Heitzinger, S.C. G-ebhard and B.E. Koel, J. Phys. Chem. 97 (1993) 5327. [85] J.A. Rodriguez, R.A. Campbell and D.W. Goodman, J. Phys. Chem. 94 (1990) 6936. [86] J.A. Rodriguez and M. Kuhn, Chem. Phys. Lett. 240 (1995) 435. [87] M.W. Ruckman and M. Strongin, Acc. Chem. Res. 27 (1994) 250. [88] J.J. Kolodziej, K. Pelhos, I.M. Abdelreheim, J.W. Keister, J.E. Rowe, and T.E. Madey, Prog. Surf. Sci. 59 (1998) 117. [89] S. Pick and P. Mikusic, Chem. Phys. Lett. 208 (1993) 97; 215 (1993) 319. [90] M. Weinert and R.E. Watson, Phys. Rev. B, 51 (1995) 17168. [91] F. Delbecq and P. Sautet, Phys. Rev. B, 59 (1999) 5142. [92] K.B. Wiberg and P.IL Rablen, J. Comput. Chem. 14 (1993) 1504. [93] J.P. Loew, Quantum Chemistry, Academic Press, New York, 1978. [94] Z. Yang, R. Wu, and J.A. Rodriguez, Phys. Rev. B, submitted. [95] M.W. Ruckman, L.Q. Jiang and M. Strongin, J. Vac. Sci. Technol. A, 10 (1992) 2551; 11 (1993) 466. [96] M. Strongin, M.W. Ruckman, M. Weinert, 1LE. Watson and J.W. Davenport, in Metal Alloys: Experimental and Theoretical Perpectives, Proc. 1993 NATO Advanced Workshop in Alloys (Boca Raton, FL, July 1993). [97] J.A. Rodriguez, R.A. Campbell and D.W. Goodman, Surf. Sci. 307-309 (1994) 377. [98] J.A. Rodriguez and L. Gonzalez, to be published. [99] W.A. Harrison, Electronic Structure and The Properties of Solids, Dover, New York, 1989. [100] H.B. Gray, Electrons and Chemical Bonding, Benjamin, New York, 1965. [101] M. Mavrikakis, B. Hammer, and J.K. Nt~'skov, Phys. Rev. Lett. 81 (1998) 2819. [ 102] A. Ruban, B. Hammer, P. Stoltze, H.L. Skriver and J.K. N~skov, J. Molecular Catal. A: Chemical, 115 (1997)421. [103] R.M. Ferullo and N.J. Castellani, Langmuir, 12 (1996) 70. [104] N. Lopez and J.K. Norskov, Surf. Sci. 477 (2001) 59.

465 [105] F. Illas, N. Lopez, J.M. Ricart, A. Clotet, J.C. Conesa and M. Femandez-Garcia, J. Phys. Chem. B, 102 (1998) 8017. [106] J.A. Rodriguez, T. Jirsak and S. Chaturvedi, J. Chem. Phys. 110 (1999) 3138. [107] F. Besenbacher, I. Chorkendorff, B.S. Clausen, B. Hammer, A.M. Molenbroek, J.K. N~rskov and I. Stensgaard, Science, 279 (1998) 1913.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces 466

D.P. Woodruff, (Editor)

Chapter 13

Interaction of sulphur with bimetallic surfaces: Effects of structural, electronic and chemical properties Jos~ A. Rodriguez and Jan Hrbek Chemistry Department, Brookhaven National Laboratory, Upton, New York 11973, USA

1. INTRODUCTION Sulphur-containing molecules are common impurities in fuels and oil-derived feedstocks [ 1]. Today these impurities constitute a major problem in our industrial society [2-6]. When fuels are burned, the S-containing impurities react with oxygen, fonning sulphur oxides (SOx species). In the atmosphere, the SOx compounds undergo further oxidation and interact with water, producing the acid rain that kills vegetation and corrodes buildings and monuments [2]. Furthermore, the SO2 produced in the engine of automobiles poisons the catalysts utilized for the removal of CO and NO in exhaust catalytic converters [3]. In general, sulphur poisoning can have a very negative impact on the performance of catalysts currently used for the reforming of hydrocarbons in oil refmeries or for the processing of oil-derived feedstocks in the chemical industry [4-6]. Millions of dollars are lost every year in the petroleum and chemical industries as a consequence. The understanding of the interaction of S with bimetallic surfaces is a critical issue in two important areas of heterogeneous catalysis. On one hand, hydrocarbon reforming catalysts that combine noble and late-transition metals are very sensitive to sulphur poisoning [6,7]. For commercial reasons, there is a clear need to increase the lifetime of this type of catalysts. On the other hand, Mo- and W-based bimetallic catalysts are frequently used for hydrodesulphurization (HDS) processes in oil refineries [4,5,7,8]. In order to improve the quality of fuels and oil-derived feedstocks there is a general desire to enhance the activity of HDS catalysts. These facts have motivated many studies investigating the adsorption of S on well-defmed bimetallic surfaces prepared by the deposition of a metal (Co, Ni, Cu, Ag, Au, Zn, A1 or Sn) onto a single-crystal face of another metal (Mo, Ru, Pt, W or Re) [9-29].

467 Depending on the nature of the metal-sulphur and metal-metal interactions several phenomena can occur when sulphur reacts with a bimetallic surface. For some systems [13,16,18,21,23], one can observe the formation of bimetaUie sulphides that exhibit chemical properties very different from those of the pure metals. In another type of systems [11,14,22,30], the interaction between sulphur and one of the metals is repulsive, with sulphur inducing a weakening of the bimetallic bonds and reducing the degree of "mixing" of the metals (i.e. metal--metal segregation). Also, one can have bimetallic systems in which one of the metals increases or promotes the reactivity of the other toward sulphur [1517,19,20,25]. In some situations, this phenomenon accelerates the poisoning of reforming catalysts [15,17,25], while in anothers, the effect can be beneficial enhancing the activity of catalysts for HDS processes [19,30]. And finally, there can be bimetallic systems in which alloy formation decreases the affmity for sulphur of both metals [26-29]. This is a particularly interesting situation, useful for the design of catalysts with a high tolerance toward sulphur poisoning [26-28]. In this chapter we present an overview of all of these different phenomena. 2. REPULSIVE INTERACTIONS BETWEEN GOLD AND SULPHUR ON TRANSITION METAL SURFACES Catalytic reforming is one of the basic petroleum refining processes, yielding a large variety of liquid fuels [1,2]. In reforming, paratTms are reconstructed without changing their carbon number. When an alkane interacts with the surface of a transition-metal catalyst, reactions that lead to the isomerization of the molecule compete with reactions that involve C-C bond breaking and produce species with a lower number of carbon atoms [1,2,7]. By adding an inert noble metal (Au, Ag or Cu) to a transition metal surface, one can reduce the number of active sites that are present in the system, favoring in this way reactions that require a relatively small ensemble of active sites [31-34]. Thus, catalysts that combine gold and latetransition metals exhibit low activity for C-C hydrogenolysis and a high selectivity for the isomerization of hydrocarbons [31,33,35]. In these systems, the "wetting" of the surface of the transition metal by gold is a critical factor for the good performance of the catalyst. A typical Au-thermal desorption spectrum for the pure Au/Ru(0001) system, dotted curve in Figure 1, is characterized by features around 1260 K for desorption of the gold monolayer, with the onset for multilayer desorption appearing at ~ 1100 K [11,36]. Figure 1 shows Au- and S2-TDS spectra acquired after dosing Au to a Ru(0001) surface pre-covered by 0.27 ML of sulphur [ 11]. On clean Ru(0001) this

468

@

Au/S/Ru(0001) Au-TDS

.i.-,, ,u e-

es

o_~

0.27

1.

cO v

t::

t.-

(1)

o')

i5~0

0.27 0.27 0.27

0.29

0.27

0.08

~

..,.%

ooo

~r

=

115

.1/

~

\

~,

...

I

I

I

I

1000

11 O0

1200

1300

Au/S/Ru(0001) S2-TDS 0Au

.mr ~(1)t.. ~

.

/\._..,,/

/

.

~

--=

0.29

s

0.00

(~) 0s=0.27

_

I

I

I

I

1000

1100

1200

1300

Temperature (K) Fig 1 (A) Au-TDS and (B) S2-TDS spectra acquired after dosing Au to a S0m/Ru(0001) surface at 320 K. In (A), the dotted spectrum corresponds to desorption of 1.15 ML of Au from clean Ru(0001). Reprinted from ref. [ 11 ].

sulphur adlayer is stable up to 1300 K [37]. For the {0Au= 0.08, Os= 0.27} and { 0 A ~ 0.29, 0s = 0.27} surfaces, the desorption of Au occurs at ~ 1170 K (Fig. 1A), without desorption of S at temperatures below 1300 K. This Au-desorption temperature is ~ 90 K smaller than that seen for submonolayer coverages of Au on clean Ru(0001) [36]. In the presence of 0.27 ML of S, the adsorption energy of Au has decreased b y - 6 kcal/mol [11]. In Figure 1A, surfaces with Au coverage in excess of 0.3 ML show two Au desorption peaks. The peak at low temperature

469 matches the desorption range for a gold multilayer [36], and its appearance indicates the presence of three-dimensional islands of gold on the surface. The peak at high temperature corresponds to gold atoms in contact with ruthenium. In the {0A,= 1.75, 0s = 0.27} surface, the presence of gold forces the desorption of a fraction of the adsorbed sulphur at-~ 1160 K. Figure 2 illustrates how gold can affect the desorption pattern of sulphur in a drastic way. The Au- and S2-TDS spectra were acquired after dosing several coverages of gold to a surface precovered with 0.5 ML of sulphur. For S0.5/Ru(0001), the signal for desorption of $2 was close to the noise level of the mass spectrometer and covered a temperature range between 1100 and 1600 K [ 11]. The deposition of gold leads to the appearance of a sharp peak for 82 desorpfion from 1150 to 1200 K. No signals for desorption of S, $3 or $4 were detected. In the presence of gold, S atoms that were desorbing or diffusing into the bulk of the sample in clean Ru(0001) [37] are forced to desorb as $2. In the case of Au/S0.5/Ru(0001) systems containing gold coverages in excess of 2.5 ML, XPS measurements indicated that no S was present on the surface after annealing the crystal to 1200 K [11]. A similar experiment for a S0.5/Ru(0001) surface annealed to 1200 K showed a sulphur coverage of-- 0.4 ML [37]. XPS data showed that there was no bonding between S and Au atoms coadsorbed on top of Ru(0001) [ 11]. In fact, when sulphur was deposited on gold multilayers supported of Ru(0001), there was "bailing-up" of the metal overlayers as sulphur migrated to the ruthenium underneath. This is not surprising since the Ru-S bond is much stronger than the Au-S bond [37,38]. From the TDS results, it is clear that the interactions between sulphur and gold on the ruthenium surface are either repulsive or both adsorbates compete for bonding with the metal substrate. The trends in Figures 1 and 2 indicate that the amount of $2 desorbing at 1170 K depends strongly on the coverage of gold present on the surface. This behaviour can be explained in the following way [11]. The deposition of gold on a sulphur-covered Ru(0001) surface leads to the formation of 3D islands of gold which compress the sulphur into small islands of high local coverage. This forces the desorption of a small fraction of the adsorbed sulphur at ~- 1170 K. At this point, the sulphur-free ruthenium sites can be covered by sulphur atoms that still remain on the surface or by gold atoms that diffuse from the three-dimensional gold islands. The first alternative reduces the local coverage of sulphur and stops $2 desorption, whereas the second option maintains the local coverage of S high, favoring $2 desorption. The second option should be the dominant "reaction channel" when the m o u n t of gold deposited on the surface is very large, leading to complete desorption of sulphur at-~ 1170 K.

470

(~

Au/S/Ru(0001)

Au-TDS

I#) t-:::::J

0s=0.5

x::i i,.,. v

%

tao e--

c::

4.65

I",-,. ,%.--

~

~.~, .1.6_2

............ ~ ~ / ...................

~

t._..__..

~

-~ "~

0.81 0.26

~

6;I~ I

'

9( p0

~

_

.

.

I

. . . . . .

.

.

.

I

1100

1000

.

.

.

..............

_ _ . _ ~ ~

I

1200

Au/S/Ru(0001)

1300

~

S2"TDS

|

// t//

0s=05 r'~ i.. v t-(l.)

_.= i

l

1.62

/ ..............

[ 707i,7Y/ill 900

1000

_.,.m

.

.

.

.

.

.

.

1100

.

~

.

.

k._,_

.

.

.

.

.

.

...............

.

.

1200

.

.

.

~

.

.

.

.

.

.

. . . . . . . .

.

.

.

.

.

1300

Temperature (K)

Fig 2 (A) Au-TDS and (B) S2oTDS spectra acquired after dosing Au to a So.JRu(0001) surface at 320 IC Reprintedfrom ref. [11]. Images obtained with scanning tunnelling microscopy (STM) indicate that sulphur and gold have a tendency to segregate into separate domains or regions when coadsorbed on Ru(0001) [39]. On atomically flat Ru(0001), gold grows forming large two dimensional islands that have dendritic shape as shown at the top of Figure 3 [40]. At 300 K, the gold adtoms have a large mobility on the surface and nucleate in these large islands. The bottom of Figure 3 shows an STM

471

Fig 3 STM images (1 [.lm2) for the deposition of Au on clean Ru(0001), top, and a S/Ru(0001) surface with 0.05 ML of S, bottom [39]. image obtained after dosing gold to a Ru(0001) surface pre-covered with 0.05 ML of sulphur [39]. Instead of large islands of gold, one sees small aggregates of the admetal. Repulsive interactions between gold and sulphur impose severe limitations in the mobility of gold. As the sulphur coverage raises (not displayed), the gold deposited on the S/Ru(0001) surface forms three-dimensional clusters or islands instead of "wetting" the Ru(0001) surface [39]. At the same time, the STM images

472 show that gold compresses sulphur into domains of high local coverage that favor S-S coupling and eventually lead to drastic changes in the kinetics of $2 desorption (Figure 2) [ 11]. Figure 4 displays STM images for a Au monolayer supported on Ru(0001) before (left-side panel) and after (fight-side panel) adsorption of sulphur [39]. Due to the mismatch between the lattices of Au and Ru, the gold adlayer forms stripe domains. The image on the lett shows a reconstructed Au layer of individual substrate terraces together with a second layer gold island (lower right comer). Elbows or bends that outline the domain boundary are edge dislocations where the three-fold arrangement of Au atoms is locally distorted to pseudo four-fold sites. Such a distorted sites should have a higher reactivity than sites of the close packed surface. And indeed, sulphur reacts preferentially with this adsorption sites, as shown by the image on the fight-side panel [39]. Notice that at this stage (i.e. low sulphur coverage) only the elbows are decorated by holes created by preferential sulphur adsorption. Upon additional dosing of sulphur, STM shows a drastic change in the morphology of the system, with Au migrating up from the Ru interface and forming three-dimensional clusters [39]. In addition to Ru(0001), the coadsorption of sulphur and gold has been examined on Mo(ll0) [22], Mo(100) [14], Rh(lll) [22] and Pt(lll) [41]. In all of these substrates the interactions between gold and sulphur are repulsive. The case of Att/Mo(110) is particularly interesting since the Au-Mo bonds are very strong with the monolayer of gold desorbing at ~ 1400 K [22]. Nevertheless, the presence of sulphur induces breaking of these bimetallic bonds. In Figure 5, the gold desorption peaks for the {0Au> 2, 0S= 0.74}/Mo(110) surfaces show trends and a zero-order line shape that match those of gold multilayers [22]. No signal is seen for gold attached to molybdenum. A plot of the natural logarithm of the gold desorption rate

Fig 4 STM images (300x300 nm) of a strained Au monolayeron Ru(0001) before (left) and after (fight) coadsorptionwith sulfur [39].

473

@

Au-TDS Au/S/M o(110) 0s=0.74

i

03 . m

t--

_ci t._

>., . m

03 c--

{(33 03 03

0.08 '

I

1000

'

1100

I

'

I

1200

'

I

1300

1400

1500

Temperature (K)

a~

|

Au/S/Mo(110)

n,"

----

.

r

.o {3.. k..

o (D

s

I

0.84

'

i

'

0.86

I

0.88

'

I

0.90

1000/T (K -1)

Fig 5 (A) Au-TDS spectra acquired after dosing Au to a S0.74/Mo(110) surface at 300 K. (13) Activation energy for the desorption of gold. Reprinted from ref. [22]. against 1/T yields a straight line with a desorption activation energy of ~ 80 kcal/mol. This value is close to the heat of vaporization of metallic gold [22,42]. Results of STM for the S/Au/Mo(100) system again show segregation of Au and S into separate areas of the surface [14].

474 Ab initio self-consistent-field calculations and cluster models have been used to

study the bonding of sulphur and gold to Mo(110) [22]. Both adsorbates behave as electron acceptors and, therefore, compete for the electrons of molybdenum. The electronic interactions between sulphur and the metal substrate are considerably stronger than those seen for the adsorption of gold. In addition to withdrawing charge from molybdenum atoms, sulphur substantially reduces the density-of-states (DOS) that the metal atoms exhibited around the Fermi level (or highest-occupied molecular orbital, HOMO). This is illustrated in Figure 6, has been proven by photoemission spectra [43], and mainly arises from hybridisation of the Mo(4d,5s) and S(3s,3p) orbitals. Such a phenomenon considerably weakens the strength of the Mo-Au bonds [22]. From the studies described above, one can expect that sulphur alters (or poisons) the properties of catalysts that combine gold and transition metals by inducing a reduction in the degree of "wetting" of the surface of the transition metal by gold. This effect can explain changes induced by sulphur on the activity and selectivity of bimetallic catalysts used for hydrocarbon reforming [7,22,30]. !

|

........ i

!

S)

(x4)

o -

, -6.0

i HOMOr -5.6

...... ] -5.2

..................................................... ] -6.0

-5.6

-5.2

L

-6.0

1

~

!

-5.6

!

-"L2

Energy (eV) Fig 6 Calculated density-of-states (DOS) for MO13and S/Mo13 dusters. Only occupied states are included and the energies are reported with respect to the vacuum level. The left-side panel shows results for clean Mo13, whereas the center panel shows the corresponding values for S/Mo13. MoB refers to the contribution of a metal atom in a site where sulphur adsorbs. The right-side panel compares the DOS of this metal atom before and after bonding to S. Reprinted from ref. [22].

475 3. INTERACTION OF SULPHUR W I T H Ag/Ru(0001) AND Cu/Ru(0001) Silver and copper are also used as "inert" site blockers when preparing hydrocarbon reforming catalysts [4,7,31,33]. With respect to sulphur, they are more reactive than gold and can form bulk sulphides [42]. Thus, when sulphur is dosed to Ag/Ru(0001) and Cu~u(0001) [13], it weakens Ru-Ag and Ru-Cu interactions at low coverages, but at large sulphur coverages AgSx and CuSx are formed. Figure 7 displays Ag-TDS spectra recorded after depositing silver at 300 K on Ru(0001) surfaces with different coverages of sulphur (0, 0.12, 0.21 and 0.44) [13]. The silver atoms bonded directly to clean Ru(0001) desorb near 1000 K. In the presence of sulphur there is a significant weakening of the Ru-Ag bonds. For {0Ag > 0, 0S > 0.5 } systems, the results of Auger spectroscopy suggest the formation of AgSx on the ruthenium substrate at room temperature [ 13]. The top part of Figure 8 shows Agand S2-TDS spectra acquired during the thermal decomposition of a Ag2S film generated by adsorption of $2 on a Ag/Ru(0001) surface [13]. Desorption of a small amount of $2 is observed between 350 and 450 K, with most of the sulphur evolving into gas phase at temperatures from 750 to 900 K. The $2 desorption peak at high temperatures exhibits a line shape that is characteristic of zero-order desorption kinetics. For this peak, a plot of the desorption rate against 1/T yields a straight line (see Fig 8B), with an apparent activation energy of 48.8 kcal/mol. This value is very close to the enthalpy of decomposition of bulk silver sulphide (2Ag2Ssolid" Ag-TDS: Ag/S/Ru(001 ) 0.44

0.00

o.s8

o_321_

..O v (/) r O E r",.O (/) r

0.62

0.44.~.....~. .'/

k

I

1

800

900

,,

I

1000

1100

Temperature (K)

Fig 7 Ag-TDS spectra acquired after depositing silver at 300 K on clean Ru(0001) and on surfaces precovered with 0.12, 0.21, and 0.44 ML of sulphur. Reprinted from ref. [13].

476

|

TDS: S/Ag/Ru(001 ) ..m

t-:D

0~=5.45

~d

iit

t~

--- - mass 64, S2 ' 9 mass 107, Ag

>, t-

1I

t-

I l

I,,,.

/I /I

E O 1..

o

09

/

r

/

P ,,

I

400

300

,

I

500

1

,,

600

/

.

I

I

700

800

1

.

.

.

.

.

I

.

900

.

.

.

1

1000

1100

T e m p e r a t u r e (K) mass 64, S 2

(~

mass 107, Ag

(~

t~

n- 6

-1

to

~.5 O (D

o

4

E==48.8 Kcal rnol"1

3

1.15

1.20

1.25

1000/T (K")

1.30

I

105

1

1.10

1000/T (K")

Fig 8 (A) S2" and Ag-TDS spectra acquired during the decomposition of a Ag2S film generated by reaction of sulphur with a Ag/Ru(0001) surface at 300 K. At 1100 K, after the thermal desorption experiment, only 0.45 ML of S were lel~ on the Ru(0001) surface. (B and C) Apparent activation energies for the main desorption peaks in part A. From ref. [13]. 4Agsolia + S2,ga~, AH = + 46.4 kcal/mol [13]). Alter decomposition of the silver sulphide at 800-900 K, a substantial amount of sulphur ( - 0.45 ML) remained

477 bonded to the Ru(0001) surface and the Ag adatoms formed three-dimensional clusters or particles. In Figure 8A, the position and shape of the silver desorption peak match those observed for desorption of silver multilayers from Ru(0001) [36]. The graph in Figure 8C indicates that the desorption of silver in Figure 8A follows zero-order kinetics with an apparent activation energy of 63.4 kcal/mol. This desorption activation energy is close to the heat of vaporization of metallic silver and the desorption activation energy for silver multilayers from Ru(0001) or other metal substrates [ 13,36]. Figure 9 shows an STM image recorded after adsorbing - 0.1 ML of sulphur on 0.8 ML of silver supported on Ru(0001) [44]. Initially, a mismatch between the lattice parameters of silver and ruthenium produced misfit dislocations in the structure of the metal overlayer (not shown) [24,44]. The sulphur adatoms attack preferentially these positions. Ag atoms are displaced from the Ru interface and their positions are occupied by sulphur atoms. Within the structure of the metal overlayer a highly ordered triangular lattice of silver vacancy islands forms (Figures

Fig 9 (a) 2000x2000/t~k2 image of a S/Ag/Ru(0001) system. Three ruthenium terraces are shown (stepping down from the bottom left to the upper right comer). The inset shows the Fourier transform of the image. (b) A 700x640/~2 z o o m On the STM image in (a). (c) Size distribution of the vacancy islands induced by sulphur adsorption. (d) Trajectories of the center-of-mass of four neareast-neighbor vacancy islands; the positions were measured every 20 seconds. Reprinted from ref. [44].

478

Fig 10 Atomicallyresolved STM image (115xl 15 ]k) of a large silver vacancy island, about 50 ]k in diameter. The island step edges of this and the smaller islands move much faster than the acquisition rate of the STM images, and thus appear "blurred". The cluster of nearly 50 sulphur adatoms inside the large island exhibits p(2x2) order. Reprinted from ref. [44]. 9a and 9b). The average area of these islands is -- 462 ,/k, with an standard deviation of-- 117 ,/k. Figure 10 displays an STM image for a typical silver vacancy island, where one can see sulphur atoms accommodated in a p(2x2) array. In summary, the results of TDS [13], photoemission [13,45] and scanning tunnelling

microscopy

[24,45]

indicate that at low sulphur coverages the

interactions between S and Ag on Ru(0001) can be classified as repulsive, in the sense that there is weakening of the Ru-Ag bond and no mixing of the adsorbates. Once the ruthenium substrate becomes saturated with sulphur, then attractive interactions between silver and sulphur are possible and AgSx is formed [13,45]. Very similar trends are observed for the coadsorption of sulphur and copper on Ru(0001) [ 13,23]. Figure 11 shows Cu- and S2-TDS spectra for the decomposition of a CuSx film on Ru(0001) [13]. The copper sulphide was formed after the adsorption of sulphur on a supported copper multilayer at 300 K. The initial stoichiometry of the sulphide was CUl.3S. An increase in temperature from 300 to 800 K produced desorption of a significant amount of $2. Photoemission spectra taken after heating the sample to 800 K revealed that at this point a film of Cu2S was present on top of the Ru substrate. This film decomposed at temperatures between 900 and 1100 K, producing evolution of $2 and Cu into gas phase (see Figure 11). After the crystal was heated to 1250 K, only a small amount of sulphur remained on the Ru(0001) surface (-- 0.4 ML) [13]. On Ru(0001), the first copper layer adopts a pseudomorphic structure that reflects the lattice constant of the underlying ruthenium [46]. Because the lattice

479

S/Cu/Ru(O001)

mass 63

fl!

mass 64 JC} v

/

_= (/) r

(D I..

E 0

i-.

/ [ \ J"-"

~

(D {3_ CO (/) (/)

I

[

400

600

800

I

I

1000

1200

Temperature (K) Fig 11 Cu- and S2-TDS spectra acquired during the heating of a Cux.3S film to 600, 800 and 1250 K. The film was prepared by dosing sulphur to a Cu multilayer (0c~= 4.55) at 300 K. Reprinted from ref. [13].

constant of copper is 5.5% smaller than that of ruthenium, the first Cu layer is under tensile strength. Part of the stress is relieved upon addition of more copper to the Cu~.0/Ru(0001) system. A sequence of strain-relieved structures develops for thicker copper films [46,47]. An anisotropically relaxed second Cu layer, consisting of three domains of double stripes is shown in Figure 12 [23]. The bright stripes are misfit dislocations buried at the Cu-Ru interface separating regions of fee and hcp stacking [23]. Figure 13 displays S 2p core level spectra recorded after exposure

Fig 12 STM image for a Cu second layer on Ru(0001). Reprinted from ref.[23].

480

BINDING E N E R G Y (e V) 166

165

164

163

162

161

160

,-p, &

Z;

Z;

.e .....

__._..//.,,

. . . . . .

.,,-. - ~

0.3

/

v

A

x

,

m sulphided o total

-

~

-

. . . . . . -__- i 0 . ~

_

~

0.2

0

0.1 0.0 0.01

'

'

'

'

. . . .

,

0.10

.

.

.

.

.

.

.

,

.

.

.

.

.

.

.

1.00

.

i

'

10.00

EXPOS URE I l A N G M U I R S i

Fig 13 S 2p core-level spectra for the adsorption of sulphur at 300 K on a striped Cu layer (Ocu~ 2 ML) supported on a Ru(0001) surface. Reprintedfrom ref.[23]. of the - 2ML thick copper layer to sulphur. The first spectra in the set display a well-defmed S 2p3/zl/2 doublet with the 2p3/2 component at a binding energy of 161.85 eV, an energy characteristic of adorbed atomic sulphur. The inset at the bottom of the figure shows the sulphur uptake curve based on the curve fitted and integrated experimental data. After the initial adsorption of-~ 0.2 ML of sulphur, a weak shoulder appears in the high binding energy side of the photoemission curve. This new feature is well defined at a sulphur coverage of 0.37 ML. Curve fitting of this spectrum (top of Fig 13) indicates that a sulphide is now present on the surface [23]. The intensity of the sulfide peak grows with increasing sulphur dose, while that of the adsorbed S levels off and even decreases.

481 The long induction period seen in Figure 13 for the formation of the sulphide is unusual. To determine the cause, STM was used to visualize structural changes of the surface [23]. The corresponding images are shown in Figure 14. At very low S coverages (0.001 ML), sulphur adsorbs mainly at the edge dislocations and one sees straight lines that contain 4 to 8 atoms (Figure 14A). As the coverage of sulphur increases, big morphological changes are seen in the Cu overlayer and new dislocations are induced by the adsorbate. At a sulphur coverage of 0.03 ML, Figures 14D and 15, the adsorbate self-organizes into a network of hexagons and close-packed equilateral S-triangles made of 18 atoms that bound the hcp stacking areas (top of Figure 15). This self-organizing network fluctuates in time (bottom of Figure 15). It disappears upon additional dosing of sulphur (not shown), well before the formation of a copper sulphide. The image quality at these higher sulphur coverages degrades and the final details of the conversion cannot be ascertained experimentally with STM [23]. Nevertheless, the results in Figures 14 and 15 illustrate quite clearly the magnitude of the structural perturbations that sulphur can induce in a bimetallic surface. Active sites for catalytic reactions can be completely destroyed in the presence of sulphur.

Fig 14 (A) Early stages of sulphur adsorption on the stripped Cu layer. Individual sulphur adatoms images as black dots are arranged in short rows and are found at the edge dislocations and less frequently on stripes. Estimated sulphur coverage < 0.01 ML. (B-D) Development of sulphur features with increasing sulphur coverage: sulphur adatoms self-organize in rows, hexagons, and equilateral triangles. Sulphur rows can be imaged as dark or bright lines depending on the tip status. Reprinted from ref. [23].

482

Fig 15 An image (7.3 nm x 6.9 nm) of sulphur self-organized in hexagons and equilateral triangles made of 18 sulphur adatoms. At room temperature and fixed S/Cu stoichiometry (0Cu -0.03 ML for this image) the observed structural patterns fluctuate for hours. Lower two time-lapse images (3.5 nm x 3.3 nm) taken 50 s apart show formation of new equilateral triangles. Reprinted from ref. [23].

4. ADMETAL PROMOTED SULPHIDATION OF Pt(111) AND Mo(110) A large number of studies described in this book indicate that the formation of a heteronuclear metal-metal bond can lead to important changes in properties of the bonded elements. large redistribution

of charge

the chemical

In many cases, bimetallic bonding induces a

around

the metals

[48-50].

In principle,

this

redistribution of charge could affect the reactivity of a metal toward sulphur. A very

483

interesting situation is found when silver or copper are added to Pt(111) [ 15,17,41]. Figure 16 compares Pt 4f core-level spectra acquired before and after dosing $2 to Pt(111) and a Ag/Pt(111) system with 2.26 ML of the admetal [ 17]. The exposure of P t ( l l l ) to large amounts of $2 produces only a chemisorbed layer of S, without forming bulk-like sulphides which are thermodynamically very stable (PtS2, AGe= 109 kJ/mol [42]). For the S/Pt(lll) system, two factors make difficult the penetration of S into the bulk of the metal. First, the surface free energy of sulphur (0.08 J m "2 [51]) is much lower than that ofplatinum (2.69 J rn"2 [51]). And second, the cohesive energy of metallic Pt is relatively large (564 kJ/mol [52]). If the influence of these two factors is somehow suppressed, then, the formation of platinum sulphides should take place. In Figure 16, new Pt 4f features for platinum sulphide are detected atter dosing $2 to a Ag/Pt(111) surface. The relative large intensity of these features indicates that a big amount (> 1 ML) of PtSx is formed [17]. Silver has a relatively low surface free energy (1.30 J rn a [51]), and its presence on the Pt surface probably frees sulphur for migration into the bulk of the sample. In the Ag-Pt(lll) bond there is a significant shift of electrons from the admetal toward the metal substrate [50,53] that favors the formation of Pt-.S dative bonds. In addition, silver sulphides could promote the formation of platinum sulphides by inducing changes in the structural geometry that enhance the diffusion of sulphur into the lattice ofmetallic Pt [17]. Figure 17 displays photoemission Pt 4f

.--=

~,

-

S/Pt(111)

.

i._~_~o~_m~.,.----, ~ r--- . . . . .

82

80

78

J

/

76

i.

"-, i >.,,.j

74

i

72

i t \ o.oo

',._x,~ . o.oo

70

68

Binding Energy (eV)

Fig 16 Pt 4fXPS spectra acquired after doing S2 to Pt(111), bottom, and Ag/Pt(111) surfaces, top. Reprinted from ref. [17].

484 Valence: S/Ag/Pt(111 )

,~

"12 (I) .N

i .s (1)

12

....

I

10

i

8

I

6

I

4

I

2

I

0

Binding Energy (eV)

Fig 17 Valence photoemission data for Pt(lll), Ag/Pt(lll) and S/Ag~t(lll) surfaces. Initially, 0.21 ML of silver were vapor-deposited at ~ 300 K, and the Ag/Pt(111) surface was annealed to 550 K before dosing S2 at this temperature. Reprinted from ref. [41]. data for the valence region of Pt(111), Ag/Pt(111) and S/Ag/Pt(111) [41 ]. Pt(111) and Ag/Pt(111) exhibit a substantial DOS near the Fermi level and are chemically and catalytically active. The silver-induced formation of PtSx in S/Ag/Pt(111) leads to a very large drop in the DOS around the Fermi level, hindering the ability of the system to respond to the presence of adsorbates. Thus silver, ideally added as an inert site blocker to reduce C-C hydrogenolysis on Pt reforming catalysts [31-33], can actually accelerate the rate of sulphur poisoning. Copper also promotes the rate of sulphidation of platinum [15], but not all the admetals used as site blockers (Zn, AI, Sn) in Pt-based reforming catalysts behave in this way [15,25,26,29]. Figure 18 shows Pt 4f core-level spectra acquired after adsorbing sulphur on Pt(111) and several bimetallic systems. Strong peaks are seen for PtS~ in S/Ag/Pt(111) [17] and S/Cu/Pt(111) [15]. No platinum sulphide formation is observed for S/Zn/Pt(lll) [15] and S/AIIPt(lll) [25]. In Zn-Pt and A1-Pt bonds there is a net charge transfer toward platinum [54-56] that should facilitate the formation of Pt-* S dative bonds. In addition, zinc and aluminium (like silver and copper) have a smaller surface free energy than platinum [51]. However, the Zn-Pt (or AI-Pt) bonds break apart in the presence of sulphur and the Pt.*ZnSx (or Pt.-A1Sx) interactions are weak. After analysing the results in Figure 18, one can conclude that an admetal~Pt charge transfer and a low surface-free energy for the

485 admetal may be necessary, but insufficient conditions for seeing a promotional effect of the admetal on the formation of platinum sulphides [25]. On the other hand, the relative stabilities of the admetal sulphides may have a direct impact on whether or not sulphidation of the Pt substrate will occur. In bimetallic system~ where the admetals form sulphides of higher stability than those formexl by platinum (Zn/Pt and A1/Pt) [42], the adsorption of sulphur stops once the admetal is saturated with sulphur and no PtS~ is formed. Not included in Figure 18 are data for the S / S n ~ t ( l l l ) system [26]. In this special system, bimetallic bonding acamlly reduces the reactivity of both metals toward sulphur [26-28]. This can be usefifl for the prevention of sulphur poisoning and will be the subject of section 5. Pt 4f: S 2 at 550 K

PtSx

i

Pt

I

-AHfof admetal sulfide

Ag2.2e/Pt(111 )

Cu-Pt surface alio,

Zn-Pt surface

AI-Pt surface alloy

Pt(111)

'

82

I .............. '

80

1

78

'"

I

76

'

I

74

'

I

72

'

I

70

'

68

Binding Energy (eV)

Fig 18 Pt 4fXPS results comparingthe effect of dosing S2 at 550 K to clean Pt(111) and a series of X/Pt systems {X= Ag, Cu, Zn and A1}. The heat of formation for the sulphides of the admetals increases (more exothermic) when going from the top to the bottom of the figure. Reprinted from rcf. [25].

486 The sulphidation of Mo(ll0) is promoted by a series of admetals (Fe, Co, Ni, Cu, Ag and Zn) [19,20,57,58] that form sulphides that are less stable than those formed by molybdenum [42]. Figure 19 displays Mo 3d XPS spectra acquired upon dosing $2 to Mo(110) and Mo(110) surfaces with similar coverages (-~ 1.5 ML monolayers ) of nickel [19], copper [16], zinc [20] and silver [16]. These and other results [57,58] indicate that the amount of MoSx formed depends strongly on the nature of the admetal. Specifically, nickel and cobalt have a unique ability to promote Mo.*S interactions and the formation of molybdenum sulphide [19,57,58]. Results for the reaction of $2 with a series of X/Mo(ll0) surfaces (X=admetal) indicate that the "promotional effect" of an admetal increases following the sequence: Ag = Zn < Cu < Fe < Co < Ni [19,57,58]. Figure 20 compares trends observed in the activity of a series of XSy/MoS2

Mo 3d XPS S2 on X/Mo(110) T=700 K

Mo Or) ~

r

v

S on pure Mo

tr

0s

Ag, 1.4 ML Zn, 1.4ML Cu, 1.3 ML Ni, 1.5 ML

I

MOSy '

I

234

'

I

232

'

i

230

~ '

i

228

'

] I

226

Binding Energy (eV) Fig 19 Mo 3d XPS spectra acquired after dosing large amounts of $2 to clean Mo(110) and X/Mo(110) surfaces (X= Ag, Zn, Cu or Ni) at 700 K. The spectra correspond to systems in which the rate of $2 adsorption has become zero under UHV conditions. Reprinted from ref. [19].

487

E ID T-

e r0 o O

Ni

oo/ "

4

X

m X ~> 3 ~.O

c0

.>__m ~ 2 '6 o if)

Zrl

o

r" -5 o 9 0

Fe Ir

Cu

Mo

go

I

0.00

0.25

'

I

0.50

'

I

0.75

'

I

1.00

1.25

MoSy / Mo 3d5/2 XPS Area Ratio in S/X/Mo(110)

Fig 20 X axis: relative amount of MoSyformed after exposing X1.5/Mo(110) surfaces (X= Zn, Cu, Fe, Co and Ni, with 0 x 1.5 ML) to $2 at 700 K_ Y axis: activity of MoS2 and XSy/MoS2 catalysts for the hydrodesulphurization of dibenzothiophene (DBT). Reprinted from ref. [61]. catalysts (X = Zn, Cu, Fe, Co or Ni) during the hydrodesulphurizafion of dibenzothiophene [59,60] with trends found for the sulphidation of molybdenum in X/Mo(ll0) surfaces [19,58,61]. In general, a good correlation is observed between the changes in the two chemical properties. The presence of Ni leads to a significant enhancement in the Mo~S interactions and a very large HDS activity. In contrast, the effects of Zn, Cu, and Fe on the Mo*.S interactions and HDS activity are less pronounced. From the correlation in Figure 20, it is clear that the effects of bimetallic bonding can be useful in HDS catalysis. Three factors probably contribute to the large HDS activity of NiMoSx catalysts [19,58,61]: (1) the existence of Ni centers that have S-free sites on which a S-containing molecule can adsorb; (2) the presence of Ni-Mo sites that are very reactive for the desulphurization of the adsorbed molecule; and (3) on the S-free Ni sites hydrogen molecules can dissociate, producing in this way a source of hydrogen atoms that helps to remove sulphur from the surface and keeps a large number of unsaturated Mo and Ni sites. Ag/Mo(110) and Zn/Mo(110) are very useful for the synthesis of MoSx films under UHV conditions [16,20,57,61]. The dosing of S: to Ag/Mo(ll0) and Zn/Mo(110) produces bimetallic sulphides, but upon heating to 1000-1100 K the silver and zinc desorb, leaving films of pure MoSx on top of the Mo(110) substrate. Following this methodology, films that have between 2 and 6 sulphide monolayers

488 can be prepared. The films exhibit Mo 3d and S 2p XPS spectra that are very similar to those of MoS2. They show no reactivity toward CO, 02 or H2 at 100-400 K. But they can be activated aider the creation of S vacancies by reaction with atomic H [62], providing convenient surfaces for examining the chemistry of desulphurization reactions on molybdenum sulphide [63]. 5. B I M E T A L L I C BONDING AND THE PREVENTION OF SULPHUR POISONING In the previous section we have discussed several cases in which bimetallic bonding increases the overall reactivity of a system towards sulphur. If the opposite occurs, such a phenomenon could be useful for the prevention of sulphur poisoning. In practical terms, the idea is to fmd bimetallic systems that have a good catalytic activity and are less sensitive to the presence of sulphur-containing molecules in the feedstream than pure metals. Sn/Pt and Pd/Rh satisfy these requirements [26-29]. Pt-Sn bimetallic catalysts are widely used for hydrocarbon reforming or dehydrogenation reactions [4,5,64-66]. In Sn/Pt alloys, there is a redistribution of charge and both metals accumulate electrons around the Pt-Sn bonds [26,67-69]. The effects of bimetallic bonding on the chemical properties are very dramatic in the case of fin [26,27]. In the presence of $2, tin does not get fully sulphided as other metals (A1, Zn, Cu, Ag) do when they are supported on Pt(111) [15,17,25]. The formation of Sn-Pt bonds reduces the electron density of tin and the metal has difficulties responding to the presence of sulphur-containing molecules [26,27,29]. The bottom of Figure 21 compares the uptake of sulphur and SOx species after dosing SO2 to polycrystalline Sn, P t ( l l l ) , and a (~3x-/'3)R30~ surface alloy [27]. The top of the figure shows the structural geometry of the Sn/Pt alloy. Sn atoms are present only in the top layer and protrude 0.22 ]k out of the plane of Pt atoms [70,71]. Each Pt atom present in the surface has the same number of Pt and Sn neighbours (3 and 3). In the alloy, there are plenty of a-top and bridge Pt sites that can adsorb and dissociate a small molecule like SO2. Figure 21 indicates that pure tin is much more reactive than pure platinum. In fact, photoemission studies indicate that even at temperatures as low as 100 K, tin reacts vigorously with SO2 [27]. Therefore, one could expect that Sn adatoms would enhance the ability of the P t ( l l l ) surface to adsorb and dissociate SO2. However, the (-/'3x~3)R30 ~ Sl~t(111) surface alloy exhibits a reactivity smaller than that of pure Sn or Pt(111). It may be argued that the low reactivity of the alloy with respect to tin is due to the fact that the bimetallic system does not have adsorption sites with two or three adjacent tin atoms ("ensemble effects" [31,32]). But the differences in reactivity

489

OSn=0.33 ML

(4"3x~3)R30~

Sulphur Uptake

0.4

300-310 K polycrystalline Sn

0.3

Pt(111)

,~ 0.2

_____--&

Sn/Pt(111)

.~ 0.1

0.0 0

-

I

I

I

I

I

2

4

6

8

10

802 Exposure (L) Fig 21 Top: Structural geometry for a (-]'3x4"3)R30~ surface alloy. The dark and white circles represent Sn and Pt atoms, respectively. The Sn atoms are present only in the top layer. Bottom: Total sulphur uptake (SOx + S) for the adsorption of SO2 on polycrystalline Sn, Pt(111), and a (~f3xC'3)P,30~ alloy. Reprinted from ref. [27].

490 between P t ( l l l ) and ((3x(3)R30~ can only be explained invoking "electronic effects", since in the surface alloy there are plenty of adsorption sites with two or three adjacent Pt atoms and some Pt atoms are being replaced with Sn atoms which, in principle, should be more reactive. The importance of "electronic effects" has been confirmed by theoretical calculations [27,29]. Ab initio SCF calculations indicate that the Pt atoms in ((3x(3)R30~ interact poorly with the LUMO of SO2, leading to a small adsorption energy and hindering the dissociation of S-O bonds [27]. Density-functional slab calculations for the adsorption of atomic sulphur on a p(2x2)-Sn/Pt(lll) surface give adsorption energies on the pure Pt hollow sites that are 7-9 kcal/mol smaller than on Pt(111) [29]. Thus, "electronic effects" probably play an important role in the low chemical affinity of Sn/Pt alloys for sulphur-containing molecules ($2, H2S, SO2, thiophene, etc) [26-29,72]. This does not imply that "ensemble" [32,72] or "geometrical effects" [73] are negligible. For example, in the case of thiophene on p(2x2)SnfPt(111) and (4"3x(3)R30~ one is dealing with a bulky adsorbate and small ensembles of Pt atoms [32,72] or geometrical blocking of Pt.-adsorbate interactions by tin [72,73] help to prevent the decomposition of the sulphurcontaining molecule. Cu, Ag and Sn are added to Pt catalysts as site blockers to improve their selectivity for the reforming of hydrocarbons [4,31,33,64-66]. In this respect the effects of the admetals are more or less similar. From the trends discussed above, it is clear that tin is a much better choice than Cu or Ag when trying to minimize the sensitivity of Pt reforming catalysts toward sulphur poisoning. Palladium has a high catalytic activity for the selective hydrogenation of olefms, the oxidation of alcohols, cyclotrimerization of acetylene, and the removal of CO and NO from automobile exhaust gases [3-5,7]. One of the major limitations in the use of Pd in industrial catalysis is its extreme sensitive to sulphur poisoning [6,74]. Experimental and theoretical studies indicate that bimetallic bonding can reduce the reactivity of palladium toward sulphur-containing molecules [28,72,75-77]. The interaction of SO2 with Pd in bimetallic systems has been studied in detail using a combination of photoemission and theoretical (ab initio SCF, density functional) calculations [28,72,77]. On pure palladium surfaces, SO2 adsorbs molecularly at 100 K and dissociates (60-70%) at temperatures between 200 and 400 K leaving large coverages (> 0.5 ML) of S and O on the surface [28]. A very different behaviour is found for the adsorption of SO2 on a palladium monolayer supported on R h ( l l l ) [28]. At 100 K, SO2 chemisorbs molecularly on a Pdl.dRh(111) surface and heating to 300 K produces the desorption of almost 80% of the adsorbed SO2, leaving a few S adatoms and no SO• species on the surface. In this respect, the Pdl.0/Rh(lll)

491

t

9 ~176

~ O

6 9

E

0eV

Pd site 4d o r b i t a l - ' l t ' ~ E~

9

SO 2 3b 1 LUMO

~

.

~

".

1'1 1r

Q ~ 132/(ELuMO - E,~ )

Fig 22 Bonding interactions between the LUMO of SO2 and an occupied Pd 4d orbital. Reprinted from ref. [28]. system is less chemically active than polycrystalline Pd, Pd(100), or R h ( l l l ) [28]. The results of theoretical studies clearly indicate that bimetallic bonding weakens the Pd'*SO2 bonding interactions [28,77]. In the bond between SO/and palladium, a transfer of electrons from the metal into the LUMO of SO2 (see Figure 22) plays a dominant role in the bonding energy of the molecule [77,78]. This g back donation leads to a weakening of the S-O bonds, since the LUMO of SO2 is S-O antibonding. On the Pdl.0/Rh(lll) surface, the Pd--Rh interactions reduce the electron donor ability of palladium producing weaker Pd-SO2 bonds and stronger S-O bonds than on Pd(111) [28,77]. Even much weaker adsorption bonds are found when Pd is supported on surfaces of s,p or early transition metals [28,72,77]. For example, in Pdl.0/Mo(ll0) and Pdl.0/Al(lll), bimetallic bonding largely shifts the Pd 4d band toward higher binding energy [48] preventing effective interactions with the LUMO of SO2 (i.e. very large End to ELtn~o separations in the diagram of Figure 22) [77]. A similar principle is useful for reducing the rate of thiophene dissociation on Pd/Mo(ll0) [72,78]. When following this approach one has to fmd a good balance between the decrease in the overall catalytic activity of Pd and its affinity for sulphur [77]. Such a balance has been observed in the case ofPd/Rh, Pd/Mn and Pd/Ni catalysts [75,76,79]. All these results together indicate that bimetallic bonding is a viable route for increasing the sulphur tolerance of metal catalysts.

492 6.

CONCLUSION

In recent years, several new interesting phenomena have been discovered when studying the interaction of sulphur with bimetallic surfaces using the modem techniques of surface science. Very small amounts of sulphur can induce dramatic changes in the morphology of bimetallic surfaces. The electronic perturbations associated with the formation of a heteronuclear metal-metal bond affect the reactivity of the bonded metals toward sulphur. This can be a very important issue to consider when trying to minimize the negative effects of sulphur poisoning or dealing with the design of desulphurization catalysts. ACKNOWLEDGEMENT Many of the studies described above were done in collaboration with M. Kuhn, S. Chaturvedi, T. Jirsak, S.Y. Li, J. Dvorak and R.Q. Hwang. Special thanks to all of them for their superb contributions. This work was carried out at Brookhaven National Laboratory under Contract DE-AC02-98CH10086 with the US Department of Energy (Division of Chemical Sciences).

REFERENCES [1] J.G. Speight, The Chemistry and Technology of Petroleum, 2nd ed, Dekker, New York, 1991. ' [2] A.C. Stern, R.W. Boubel, D.B. Turner, and D.L. Fox, Fundamentals of Air Pollution, 2nd ed, Academic Press, Orlando, 1984. [3] K.C. Taylor, Catal. Rev. Sci. Eng. 35 (1993)457. [4] J.M. Thomas and W.J. Thomas, Principles and Practice of Heterogeneous Catalysis, VCH, New York, 1997. [5] B.C. Gates, Catalytic Chemistry, Wiley, New York, 1992. [6] C.H. Bartholomew, P.K. Agrawal and J.IL Katzer, Adv. Catal. 31 (1982) 135. [7] G. Ertl, H. KnSzinger, and J. Weitkamp (eds.), Handbook of Heterogeneous Catalysis, Wiley-VCH, New York, 1997. [8] R.R. Chianelli, M. Daage, and M.J. Ledoux, Adv. Catal. 40 (1994) 177. [9] C.C. Knight and G.A. Somorjai, Surf. Sci. 240 (1990) 101. [10] D.A. Chela, C.M. Friend, and H. Xu, Langmuir 12 (1996) 1528. [ 11] M. Kuhn and J.A. Rodriguez, Chem. Phys. Lett. 231 (1994) 199. [12] M. Kuhn, J.A. Rodriguez and J. Hrbek, Surf. Sci. 314 (1994) L897. [13] M. Kuhn and J.A. Rodriguez, J. Phys. Chem. 98 (1994) 12059. [14] J.C. Dunphy, C. Chapelier, D.F. Ogletree and M.B. Salmeron, J. Vac. Sci. Technol. B, 12 (1994) 1742. [15] M. Kuhn and J.A. Rodriguez, Catal. Lett. 32 (1995) 345. [16] J.A. Rodriguez and M. Kuhn, J. Phys. Chem. 99 (1995) 9567. [17] M. Kuhn and J.A. Rodriguez, J. Catal. 154 (1995) 355. [18] F.H. Ribeiro, A.L. Bonivardi, C. Kim and G.A. Somorjai, J. Catal. 150 (1994) 186. [19] M. Kuhn and J.A. Rodriguez, Surf. Sei. 355 (1996) 85. [20] M. Kuhn and J.A. Rodriguez, Surf. Sci. 336 (1995) 1.

493 [21] W.K. K-uhn, J.-H. He, and D.W. Goodman, J. Vac. Sci. Technol. A, 10 (1992) 2477. [22] J.A. Rodriguez, M. Kuhn and J. Hrbek, J. Phys. Chem. 100 (1996) 3799. [23] J. Hrbek, J. de la Figuera~ K. Pohl, T. Jirsak, J.A. Rodriguez, A.K. Schmid, N.C. BarteR, and R.Q. Hwang, J. Phys. Chem. B, 103 (1999) 10557. [24] K. Pohl, M.C. Bartelt, J. de la Figuera, N.C. BarteR, J. Hrbek, and R.Q. Hwang, Nature, 397 (1999) 238. [25] J.A. Rodriguez and M. Kuhn, J. Vac. Sci. Technol. A, 15 (1997) 1608. [26] J.A. Rodriguez, S. Chaturvedi, T. Jirsak, and J. Hrbek, J. Chem. Phys. 109 (1998) 4052. [27] J.A. Rodriguez, T. Jirsak, S. Chaturvedi, and J. Hrbek, J. Ant Chem. Soc. 120 (1998) 11149. [28] J.A. Rodriguez, T. Jirsak and S. Chaturvedi, J. Chem_ Phys. 110 (1999) 3138. [29] J.A. Rodriguez, J. Hrbek, M. Kuhn, T. Jirsak, S. Chaturvedi and A. Maiti, J. Chem. Phys. 113 (2000) 11284. [30] J.A. Rodriguez and J. Hrbek, Accounts of Chem. Research, 32 (1999) 719. [31] J.H. Sinfelt, Bimetallic Catalysts, Wiley, New York, 1983. [32] W.H.M. Sachtler, Faraday Disc. Chem. Soc. 72 (1981) 7. [33] V. Ponce, Adv. Catal. 32 (1983) 149. [34] J.A. Rodriguez and D.W. Goodman, Surf. Sci. Reports 14 (1991) 1. [35] S. Galvagno et al, J. Catal. 69 (1981) 283; 61 (1980) 223. [36] J.W. Niemantsverdriet, P. Dolle, K. Markert and K. WandeR, J. Vacuum Sci. Technol. A, 5 (1987) 875. [37] S.R. Kelemen and T.E. Fisher, Surf. Sci. 87 (1979) 53. [38] J.A. Rodriguez, J. Dvorak, T. Jirsak and J. Hrbek, Surf. Sci. 490 (2001) 315. [39] J. Hrbek, J. de la Figuera, K. Pohl, A.K. Schmid, N.C. Barter and 1LQ. Hwang, to be published. [40] R.Q. Hwang, J. Schroder, C. Gunther and R.J. Behm, Phys. Rev. Lett. 67 (1991) 3279. [41] J.A. Rodriguez, M. Kuhn and J. Hrbek, J. Phys. Chem. 100 (1996) 15494. [42] Lange's Handbook of Chemistry, 13th ed, McGraw-Hill, New York, 1985 [43] J.A. Rodriguez, J. Dvorak and T. Jirsak, Surf. Sci. 457 (2000) IA13. [44] K. Pohl, J. de la Figuera, M.C. BarteR, N.C. BarteR, J. Hrbek and R.Q. Hwang, Surf. Sci. 433-435 (1999)506. [45] J. Hrbek, M. Kuhn and J.A. Rodriguez, Surf. Sci. 356 (1996) L423. [46] G.O. Potshke and R.J. Behm, Phys. Rev. B, 44 (1991) 1442. [47] C. Gtmther, J. Vrijmoeth, R.Q. Hwang, and ILL Behm, Phys. Rev. Lett. 74 (1995) 754. [48] J.A. Rodriguez, Surf. Sci. Reports, 24 (1996) 223. [49] R. Wu and A.J. Freeman, Phys. Rev. B, 52 (1995) 12419. [50] P.J. Feibelman, Surf. Sci. 313 (1994) L801. [51] L.Z. Mezey and J. Giber, Jpn. J. Appl. Phys. 21 (1982) 1569. [52] C. Kittel, Introduction to Solid State Physics, 6th ed, Wiley, New York, 1986. [53] J.A. Rodriguez and M. Kuhn, J. Phys. Chem. 98 (1994) 11251. [54] J.A. Rodriguez and M. Kuhn, J. Chem. Phys. 102 (1995) 4279. [55] R.E. Watson and L.H. Bennett, Phys. Rev. B, 15 (1977) 5136. [56] J.A. Rodriguez and M. Kuhn, Chem. Phys. Lett. 240 (1995) 435. [57] J.A. Rodriguez, S.Y. Li, J. Hrbek, H.H. Huang and G.-Q. Xu, J. Phys. Chem. 100 (1996) 14476. [58] J.A. Rodriguez, S.Y. Li, J. Hrbek, H.H. Huang and G.-Q. Xu, Surf. Sci. 370 (1997) 85. [59] S. Harris and R.R. Chianelli, J. Catal. 98 (1986) 17. [60] R.R. Chianelli, T.A. Pecoraro, T.R. Halbert, W.-H. Pan, and E.I. Stiefel, J. Catal. 86 (1984) 226. [61] J.A. Rodriguez, Polyhedron, 16 (1997) 3177.

494 [62] S.Y. Li, J.A. Rodriguez, J. Hrbek, H.H. Huang, and G.Q. Xu, Surf. Sei. 366 (1996) 29. [63] J.A. Rodriguez, J. Dvorak, T. Jirsak, S.Y. Li, J. I-Irbek, A.T. Capitano, A.M. Gabelnick, and J.L. Gland, J. Phys. Chem. B, 103 (1999) 8310. [64] C. Xu, J.W. Peck and B.E. Koel, J. Am. Chem. Soc. 115 (1993) 751. [65] O.A. Barias, A. Holmen, and E.A~ Blekkan, J. Catal. 158 (1996) 1. [66] J. Szanyi and M.T. Paffett, J. Am. Chem. Soe. 117 (1995) 1034. [67] S. Pick, Surf. Sci. 436 (1999) 220. [68] Y. Jeon, J. Chen, and M. Croli, Phys. Rev. B, 50 (1994) 6555. [69] P. Ross, J. Vac. Sci. Technol. A, 10 (1992) 2546. [70] S.H. Overbury, D.R. Mullins, M.T. Paffett and B.E. Koel, Surf. Sci. 254 (1991) 45. [71] S.H. Overbury and Y.-S. Ku, Phys. Rev. B, 46 (1992) 7868. [72] J.A. Rodriguez, J. Dvorak and T. Jirsak, to bo published. [73] C. Xu and B.E. Koel, Surf. Sci. 327 (1995) 38. [74] J.A. Rodriguez, S. Chaturvedi and T. Jirsak, Chem. Phys. Lett. 296 (1998) 421. [75] P.C. L'Argentiere, M.M. Cation, N.S. Figoli and J. Ferron, Appl. Surf. Sci. 68 (1993) 41. [76] P.C. L'Argentiere, M.M. Cation and N.S. Figoli, Appl. Surf. Sci. 89 (1995) 63. [77] J.A. Rodriguez and L. Gonzalez, to be published. [78] H. Sellers and E. Shustorovich, Surf. Sci. 346 (1996) 322. [79] D.M. DiCicco, A.A. Adamczyk, and K.S. Patel, Book of Abstracts for the 210 th American Chemical Society National Meeting (Chicago, August 1995), Fuel-145.

9 2002 Elsevier Science B.V. All rights reserved.

Surface Alloys, and Alloy Surfaces D.P. Woodruff, (Editor)

495

Chapter 14

Adsorbate induced segregation at bimetallic surfaces C.J. Baddeley

School of Chemistry, University of St Andrews, St Andrews, Fife, KY16 9ST, United Kingdom. 1. I N T R O D U C T I O N The surface and interface chemistry of bimetallic surfaces is an important subject for a variety of technological reasons including corrosion resistance and hardening, metal-metal interfaces, microelectronics fabrication, electrochemistry, surface magnetic films and heterogeneous catalysis [1]. Bearing in mind, the huge economical importance of heterogeneous catalysis, it can be argued that this aspect of bimetallic surface chemistry can be regarded as the most important. This chapter concentrates primarily on issues of heterogeneous catalysis. The thermodynamic and kinetic factors are outlined that are important in defining the surface chemistry of bimetallic surfaces. In addition, the various approaches will be introduced that are utilised by surface scientists in an attempt to measure the composition of bimetallic surfaces under the influence of adsorbates. Furthermore, the chapter will investigate the difficulties encountered when attempting to obtain accurate measurements on nanoscale bimetallic particles under environments typically encountered in a catalytic reaction. By way of contrast, the relevance of much more accurate measurements on well-defined surfaces under idealised ultrahigh vacuum (UHV) conditions will be questioned. 1.1. Bimetallic surface chemistry - traditional ideas Bimetallic catalysts have often been shown to outperform their monometallic counterparts in terms of both activity and selectivity [2]. There are now many examples of catalytic reactions which have been studied over bimetallic systems some of which are summarised in Table 1.

496 Table 1: Some reactions catalysed by bimetallic systems (adapted from [ 1]). Reaction CO oxidation

Bimetallic system Cu/Cr Cu/Pd Pt,Pd and Rh alloys

Reference [3] [4] [5]

dehydrogenation

Ni/Cu Ni/W Ni/Sn Pt/Co

[6] [7] [8] [9]

acetylene cyclotrimerisation

Pd/Au Pd/Sn other Pd alloys

[10] [11] [12]

Fischer-Tropsch synthesis

Ru-Group IB alloys Fe/Ru, Fe/Ni, Co alloys CufPd

[13] [5] [141

Exhaust emission conversion

P t ~ h or Pd

[15]

Olefin hydrogenation

Pd/T1 Pd/Cu, Sn or Fe Pd/Fe Pd/Co

[161 [ 17] [18] [19]

Hydrocarbon reactions Reforming

Ni, Pt, Pd, Ru based alloys Pt/Re, Pt/Ir, Pt/Au Pt/Sn, multimetallic systems

[20] [21-24] [5]

CO methanation

W/Ni, W/Co, Ru/Cu Ce based intermetallics

[25] [26]

alkane hydrogenolysis

W/Ni, Pt/Ni, Pt/Re, Ru/Cu Cu~d

[25]

hydrodesulphurisation

Co/Mo, Ni/Mo

[28-29]

hydrodenitrogenation

Ni/Mo

[30]

hydrogenation of edible oils and fats

Ni and Ni based alloys

[31]

[27]

Traditionally, there have been two reasons proposed for the enhanced performance of a bimetallic catalyst over each monometallic counterpart. These are k n o w n as ensemble (structural) effects and electronic (ligand) effects.

497

1.1.1. Ensemble effects The idea of the importance of surface structure in the chemistry of bimetallic surfaces relies largely on the concept of the "active site" for a particular surface chemical reaction. If one reaction requires the presence of three-fold hollow sites on an fcc (111) facet, while a competing reaction requires only single atomic sites, then the random dilution of the active metal by an inert second metal would rapidly deplete the number of active three-fold ensembles. By contrast, the number of single atom sites available would be depleted much less dramatically as a function of composition. Thus, the selectivity of the catalyst would vary with surface composition. An example where this effect is observed dramatically is in the trimerisation of ethyne (C2H2) to benzene (C6H6) over Au/Pd surfaces [32]. Detailed investigations by Lambert and co-workers [33] proposed that, over Pd(111), the trimerisation reaction requires a relatively large ensemble of Pd atoms constituting a central atom surrounded by a hexagonal array of Pd atoms - i.e. a Pd7 cluster [33]. On alloying with Au, the activity of the AuxPdl00_x(lll) surfaces, as measured by TPD experiments, varies in a dramatic way as a function of surface composition as shown in Fig. 1.

0

20

40

60

80

100

Composition/atom % Pd

Figure 1: The yield of benzene from random PdAu alloys on Pd(lll) as a function of composition (white circles). Also shown is the theoretical fit (filled black squares) which is formed by summing the contribution from Pd7 ensembles (open squares) and AuPd6 ensembles (white squares) [32].

498 The pronounced maximum in activity at a composition of-~Au15Pd85 was attributed to the presence of a second active ensemble, AuPd6. The number density of AuPd6 ensembles was considered to vary as 06(1-0) where 0 is the mole fraction of Pd in the surface layer. This function reaches a maximum at 0 = 6/7 and is clearly zero at 0 - 0 and 1. It was also known that the selectivity to benzene formation approaches 100% over PdAu surface alloys on Au(111) [34]. The experimental data show that there is considerable activity at 0 - 1, this activity being that of a clean P d ( l l l ) surface. Using the fact that the trimerisation reaction is only 25% selective over Pd(111) due to competing hydrogenation and decomposition reactions, and assuming that the number density of Pd7 ensembles varies as 07, the measured activity of the surface as a function of composition could be modelled closely as a linear superposition of the statistically derived number of AuPd6 ensembles (100% active for the cyclisation process) and Pd7 ensembles (25% active) assuming the surface to consist of a random arrangement of Au and Pd atoms. This conclusion was supported by measurements on AuPd colloidal catalysts where the selectivity of the catalysts showed a strong correlation with the extent of alloying in the particles [10]. X-ray diffraction data showed that the bimetallic colloids consisted of a (-7.5 nm) Au core with a thin (<1 nm) Pd shell. When the bimetallic colloids were annealed in the temperature range 300600 K, no particle size increase was observed, but a change in the lattice parameter of the bulk phases present was detected indicative of alloy formation. ,,, 9 O o

_

i-Butene lrans-2 Butene cis-2 Butene n-Hexane ~

~5t~ "7'

4-

-02' ._

i

1 -

0 300

v

400 500 600 700 Annealing Temperature / K

v

800

Figure 2: Yield of various products from Au core/Pd shell colloidal catalysts as a function of the pre-annealing temperature. Reactions were carried out in a C2H2/H2 mixture at 300 K [10].

499

Over this same temperature range a tenfold increase was observed in the selectivity to benzene formation (Fig. 2). This was consistent with a change in the composition of the surface of the colloidal particles from pure Pd to a bimetallic surface.

1.1.2. Electronic Effects A second explanation for the improved performance of bimetallic systems is the so-called "electronic factor". Here it is presumed that the electronic properties of one element can be modified by the presence of another element and that this in turn can modify the chemisorption and reaction properties of adsorbates. One of the earliest theoretical approaches to explain such effects was the Rigid Band Model. In this model, it was assumed, for example, that upon dilution of a pure element from Group VIII (e.g. Pd) by an element from Group IB (e.g. Au), the Brillouin Zone structure of the pure element is preserved. The overall effect would be that 1 extra electron is added to the common band structure for every additional atom of IB element. The consequence of this would be to alter the density of occupied electron states at the Fermi Energy as the composition of the alloy varied. In the case of pure Pd, the d-band is approximately 95% filled giving rise to a high density of states at the Fermi Energy. Using this model, on addition of Au, a dramatic effect on the adsorption properties would be anticipated at the composition where the d-band became filled due to the extra electrons being supplied by Au. This theory was supported by, for example, the catalytic conversion of ortho to para hydrogen over PdAu alloys [35]. It was found that there was a distinct change in the activation energy for this process at the approximate composition predicted by the Rigid Band Model. It was proposed that the rate limiting step for the reaction was the dissociative chemisorption of H2 which involved electron transfer to the alloy. The rate should be higher when the density of states at the Fermi Energy is high (e.g. in pure Pd) and lower when the density of states at the Fermi Energy is low, as in the case of pure Au. This theory implies that the individual components of the bimetallic system lose their chemical identity and neglects effects due to the segregation of one element to the surface (see below). Once more accurate theoretical calculations [36-37] became available, it progressively became accepted that the two types of atom largely retain their chemical identity and that the important parameter is the local density of states (LDOS) in the vicinity of a particular atom. In general, differentiation between ensemble and electronic effects is rarely clear and often there is a contribution from both effects. However, a major goal of heterogeneous catalytic investigations using bimetallic catalysts is to establish the surface composition under reaction conditions and to attempt to

500 correlate the available surface ensembles with the observed catalytic activity and selectivity. It has long been known that the composition of the surface of a metal particle should differ from that of the bulk on thermodynamic grounds and that provided that kinetic barriers can be overcome, the surface composition will be affected by the presence of the gas phase. Despite the vast array of analytical techniques at the disposal of the surface scientist, the elucidation of the composition of a bimetallic surface under the influence of the adsorbate has proved difficult to achieve. In addition, it is far from the case that conclusions drawn from experiments carried out on single crystal surfaces can be extrapolated to explain the composition of the surfaces of the types of metal nanoparticles routinely used in heterogeneous catalysis. This chapter examines the attempts that surface scientists have made to characterise such effects.

2. ADSORBATE INDUCED SEGREGATION 2.1. Thermodynamic considerations The chemistry of bimetallic surfaces is controlled largely by the composition of the outermost layer of the surface and to a lesser extent by the layers immediately below the surface. It has been predicted for many years that the composition of the bulk of a bimetallic system will differ strongly from that of the surface and a vast number of publications exists where researchers have attempted to measure or calculate the composition of bimetallic surfaces. In general, for an alloy AB, the phenomenon of segregation is controlled by a range of parameters including the differences in bond strengths A-A, B-B and A-B, atomic size, enthalpies of sublimation and surface energies, temperature, the exposed crystal plane, and metal particle size. The driving force for segregation is the difference in the binding energies between the two metal atoms, A-B and the binding energies in the pure components, A-A and B-B [38]. The same change in chemical bonding gives rise to a change in the surface tension of the binary system compared to the surface tension of the pure constituents. There exists a strong correlation for metals between the surface tension and the enthalpy of sublimation, ziH~b. For atoms at an fcc (111) surface, the coordination number is 9 compared with 12 in the bulk. AH~ub can be thought of being proportional to the heat required to convert a 12-coordinate atom into an isolated (gas phase) atom. Thus the difference in energy of atoms in the surface (111) plane relative to that of atoms in the bulk can be considered to be that required to break 3 of the 12 bonds of the bulk metal (i.e. 0.25 zlH~ub). Experimentally a value of 0.16 AH~ub has been obtained [39]. Somorjai [38] showed that the ratio of the mole fraction of surface A atoms to surface B atoms can be expressed by the following equation:

501 XB~/XA~ - xsb/xa b e x p [ 0 . 1 6 (AHsub A - AHsub 8 ) / R T ]

(])

where x8 s and XBb are the surface and bulk mole fractions of component B respectively. From this relatively simple equation, several features emerge. Firstly, the bulk and surface concentrations of a particular element are generally different. Secondly, the element with the lower value of heat of sublimation is likely to be the dominant species at the surface. Thirdly, the surface concentration will depend exponentially on temperature. In deriving the above equation, the metallic alloy was treated as an ideal solution. In general, however, an alloy has a finite heat of mixing. Once this parameter is introduced into the model, it becomes clear that the surface composition is strongly dependent on the heat of mixing, its sign and its magnitude [40]. In the theories presented so far, a major driving force for segregation has been the fact that the surface is a region of reduced atomic coordination. In solids, there is a further driving force, namely the reduction of strain. Solute atoms that differ in size from the solvent lattice atoms create a strain in the lattice [41]. At a grain boundary, there are open sites where more space is available to the atoms. By migrating to these sites, a solute can reduce the strain energy. This approach is extremely simplistic, but nonetheless important features of binary metal systems are effectively explained. Even in the relatively simple case of estimating the surface segregation of an adsorbate free bimetallic surface, problems are encountered with the calculations [42]. For example, in the case of PtNi alloys, there is an orientation dependent change of the segregating component which is not predicted by most theoretical predictions of the type described above. The PtxNil00_x alloy crystallises in the fcc lattice. It has several ordered phases [43] and also forms a solid solution over the entire composition range. For this system interesting segregation phenomena have been observed, in particular a face dependence of the segregating species [44-46]. After thermal treatment the (111) and (100) surfaces are found to be enriched in Pt whereas Ni segregates on the (110) surface. For example, face related segregation reversal has been observed in the case of PtsoNis0(110) [47]. Mezey [48-50] developed a theory known as the modem thermodynamic calculation of interface properties (MTCIP). Within that theory, the conditions of the thermodynamic equilibrium state of an interface separating two homogeneous bodies are given by a system of coupled, non-linear equations. Good agreement was found with available experimental data in all cases investigated, which included correctly predicting face sensitive behaviour of

502

segregation. The analysis can be refined considerably and a database of surface compositional predictions for bimetallic systems has been established [46]. The face sensitivity of segregation is a pointer to potential problems in transferring results obtained on single crystal surfaces to those obtained on bimetallic nanoparticles. Further complications arise due to the fact that surfaces of metals contain irregularities. For example a typical truncated octahedral nanoparticle contains not only (111) oriented faces, but also (100) faces where the coordination number of the surface atoms is 8 rather than 9. In addition, there are edge and comer sites of still lower coordination and on the facets, step and kink defects. Burton et al. [51] showed that the lower the coordination of a particular site, the greater the tendency for segregation. Thus in alloy microclusters, sites at edges and comers are more enriched in the segregating species than are sites on flat terraces. Studies of small particles by Sinfelt [52] have shown that when the particles' size becomes very small and when virtually every atom is at the surface, alloy systems display phase diagrams very different from those that characterise the bulk systems. Additionally in the case of supported bimetallic nanoclusters, the cluster size and the interaction between the clusters and the support are of importance. This may result in large differences in surface segregation between supported nanometre sized clusters and bulk alloys as was illustrated by Gijzemann using Monte Carlo simulations [53]. The cluster size may influence the surface segregation by limiting the supply of atoms and by the reduction in coordination number of the surface atoms. The metal support interaction may influence surface segregation by preferential anchoring of one of the constituents to the support at the cluster-support interface. Since the influence of the cluster size and the metal support interaction on the surface composition is largely unknown, detailed experimental studies of the outermost layer of small supported bimetallic clusters are necessary. Estimating the composition of clean bimetallic surfaces is clearly a complex task. Understanding the effect on that composition of the adsorbate brings an extra dimension of complexity to the problem. In terms of thermodynamics, one might assume that the overriding factor in determining whether an adsorbate, X, induces segregation at a bimetallic surface, AB, is the relative strength of A-X versus B-X. Tomfinek et al. [54] proposed that a modified version of equation (1) to account for the effect of the adsorbate. xss/XA s -- xsO/xA b

exp [QsegChem/RT]

where Qseg chem -- Qseg 4- (EA-EB)O

(2)

503

Here, Qseg is the work involved in exchanging a surface A atom and a bulk B atom; EI is the chemisorption energy of the adsorbate on element I and 0 is the fractional adsorbate coverage. Taking the Au/Pd/CO system as an example, Qseg for Au/Pd is +34.5 kJmol -~ which points to a tendency for Au to segregate to the surface. However, the chemisorption energy of Pd-CO is much stronger than Au-CO (151 v 38 kJ mol-1). Assuming a complete monolayer coverage of CO on the alloy, the authors calculate a value for Qsegchem of--78.5 kJmo1-1 indicating a tendency for CO to segregate Pd to the surface. This approach gives an indication of the direction of segregation under the influence of the adsorbate. However, it was unclear how the authors defined monolayer coverage and no account was taken of the variation in adsorption energy with coverage. Each of these factors will clearly effect the magnitude of Qsegchem. Undoubtedly, kinetics play a major role in the extent of adsorbate induced segregation at a bimetallic surface. Hammer et al. [55], investigating adsorbate induced variation of the structure and composition of the Mov5Re25(100) surface, stated that adsorption at low temperatures does not modify the surface concentration profile as there is a thermal activation for interdiffusion of surface atoms. They stated also, that analysis of the clean surface might be valuable for an understanding of catalytic behaviour since reactions such as hydrocarbon conversions proceed usually at temperatures in the range 400-600 K where a change of the surface composition by interdiffusion is hardly to be expected. In contrast, reactions like high temperature corrosion or carburization are assumed to change the surface stoichiometry severely due to the presence of the adsorbate. Preferential bonding to one of the alloy components may become the dominant driving mechanism for segregation as soon as diffusion is no more kinetically hindered. This conclusion would seem to be supported by Hodak et al. [56] who investigated interdiffusion in Au/Ag core/shell nanoparticles. The diffusion constant for the Au/Ag system is given by

D = Doexp[-EJkT]

(3)

where Do - 0.04 crn2s-~ and Ea = 1.76 eV [57]. The time, t, taken for an atom to diffuse over a distance equal to the radius, R, of the particle is given by [57-59]

t - (R2/6D)

(4)

Using these values, for a particle of radius 5 nm at a temperature, T=1000 K, t - 7.7 x 10 -4 s; at T - 6 0 0 K, t - 632 s and at T - 300 K, t - 3.8 x 1017 S. Clearly, at elevated temperatures, there is no diffusion limitation to the achievement of the new adsorbate induced equilibrium surface composition.

504 However at 300 K, it would appear that adsorbate induced effects should never be observed on a laboratory timescale. There is, though, a large body of experimental evidence which suggests that it is extremely unsafe to assume that the clean surface is an adequate model of the surface under the influence of a reactive gas phase. For example Shen et al. [60] reported that adsorption of oxygen on the (2x2)p4g structure, formed by a clock rotated (001) Pd top layer above a mixed c(2x2) A1/Pd underlayer on Pd(001), results in a lifting of the reconstruction and a segregation of A1 to the surface at 325 K. The driving force is thought to be the relative strength of the A1-O and Pd-O bond emphasised by the large difference in enthalpy of formation of AlzO3 (-1678.2 kJ mo1-1) compared with PdO (-112.6 kJmo1-1) [61]. Paul and co-workers [62] observe an almost pure first layer of Pt atoms on the clean Pt3Ti(111) surface. After 02 exposure, TiO2 islands are formed on top of the Pt-enriched layer, which remains unaffected. Dauscher et al [63] report formation of Ti407 islands together with some oxidised Pt. Bardi et al. [64] show that on the Pt3Ti(510) stepped surface, the ordered step array of the clean surface changes upon oxygen exposure into large facets of TiO2. On polycrystalline Pt-Sn samples, Asbury and Hoflund [65-66] have examined segregation at clean and 02 exposed surfaces. The clean surface is found to be enriched in Pt to an extent dependent on annealing temperature. Upon 02 exposure a tin oxide overlayer is formed with a maximum thickness at 200 ~ Lad and B lakeley [67] reported that the Ni60Fe40(100) surface consists of only FeO and Fe304 after O exposure at 100 ~ On Ni85Crls(111) different stages of oxide formation are found by Jeng et al [68]. At 200-300 ~ initially a chromium oxide is formed and only at higher 02 exposures is NiO also observed. On Ni-Ru [69], XPS and UPS unambiguously indicated the formation of an oxide overlayer. In recent years, it has also become increasingly clear that the adsorbate induced lifting of surface reconstruction is far more common than previously envisaged and may have a profound effect on material selectivity and catalytic activity [70-72]. Later in this chapter, a considerable number of similar examples of room temperature adsorbate induced segregation will be shown. The role of particle size will be examined and the possibility will be discussed that near surface diffusion dominates over bulk diffusion.

505

3. TECHNIQUES FOR CHARACTERISING ADSORBATE INDUCED SEGREGATION The vast array of available analytical techniques enable the surface scientist to probe the surfaces of materials of ever increasing complexity (e.g. supported nanoparticles rather than single crystals) and under ever more difficult conditions (e.g. the gas or liquid-solid interface rather than UHV). However, it is still extremely difficult to obtain quantitative information on the composition of a bimetallic surface under the influence of the adsorbate for reasons which will be outlined below. In this section, several of the techniques utilised to address this problem will be introduced and their relative merits discussed.

3.1. Photoelectron Spectroscopies

3.1.1. X-ray Photoelectron Spectroscopy (AES)

Spectroscopy

(XPS)

and Auger

Electron

X-ray photoelectron spectroscopy (XPS) and Auger electron spectroscopy (AES) have become standard tools of the surface scientist. Their success as surface analytical techniques is largely derived from the fact that the kinetic energies, E, of the emitted photoelectrons lies in the range 0 < E < 1500 eV. In this range the inelastic mean free path (IMFP) of electrons is less than 2 nm [73]. Indeed, at kinetic energies of around 100 eV, the IMFP falls well below 1 nm. Consequently, the great majority of photoelectrons detected have been emitted from the top few layers of the surface. The IMFP of electrons corresponding to a particular XPS or Auger peak can be determined experimentally and, additionally, empirical equations have been derived to calculate their value [74]. If a surface is covered in an overlayer of thickness, d, the intensity, Ia, of a substrate XPS or Auger peak can be expressed as follows.

I~

=

Io exp [-d/& cos 0]

(5)

where Io is the intensity of the clean surface; A, is the IMFP of the emitted photoelectrons which give rise to the XPS peak and 0 is the angle between the detector and the surface normal. Using this equation, if a metal surface is covered in a uniform monolayer of a second metal (d - 0.2 nm) and a core level of the substrate which has a value of A, of 0.5 nm is being monitored at an angle of 45 ~ to the surface normal; the value of Ia/Io should be 0.567. In other words, the contribution of the top layer to the total peak intensity from a clean metal surface under these conditions is 43.3 %. By a similar argument it can be shown that the contribution of the

506

second, third, fourth and fifth layers to the peak intensity is 24.5 %; 13.9%, 7.9% and 4.5% respectively. In a bimetallic system, it is possible to use this type of argument to construct an equation expressing the intensity of a photoelectron peak in terms of the weighted contributions from each layer. Id = 0.433 C 1 "b 0 . 2 4 5 Ce + 0 . 1 3 9 c3 + 0.079 C4 "t" 0 . 0 0 5

C 5 "at" . . . .

where c, is the fractional composition of the element in layer n. In this way, if other data is available either for a different photoelectron peak of the same element, or a photoelectron peak of the other element, it is possible to obtain a reasonably accurate idea of the layer by layer composition of the bimetallic surface using XPS. However, once the surface is covered by an adsorbate layer, it is difficult to use this method to quantify adsorbate induced segregation effects. If, for example, the adsorbate was to preferentially bond to element X, a relatively large attenuation of the XPS peak of element X may be observed due to the fact that this element is covered in an adsorbate layer while the peak intensity of element Y may be unaffected. This could be mistaken for the observation of segregation of element Y to the surface, which may be the complete opposite of the actual effect. Nonetheless, XPS/AES have been used to investigate adsorbate induced segregation effects on single crystal bimetallic surfaces. Wu et al. [75] investigated the adsorption of C12 on CosoNis0(111) using AES. By taking into account the sensitivity factors of the Co (53 eV) and Ni (61 eV) Auger transitions, they calculated the "relative surface concentrations" of each component X u i S / ( X u i s "1- X c o s) (where Xi is the surface mole fraction of component i). At 300 K, they reported an 18% enrichment of Ni relative to the bulk concentration. Thermal processing up to 913 K reverses the segregation such that a Co enrichment of 16% with respect to the bulk is observed. The heat of formation of MClz(s) is slightly larger for Co (-313 kJmol -~) than Ni (-305 kJmol -~) as is the M-C1 bond energy [61]. Invoking the argument that chemisorption induced segregation is expected if the adsorbate binds more strongly to one of the two alloy components, Wu et al. suggest that Co segregation should be anticipated. However, the AES data following chlorine exposure point to Ni segregation leading the authors to propose that CoCl2(g) desorption from the surface occurs leading to a deficiency in Ni at the surface. However, no thermal desorption data were presented and no consideration was given to the possibility that Cl(a) could preferentially attenuate the Co signal relative to that of underlying Ni. A second example of the use of photoelectron spectroscopy to measure adsorbate induced segregation effects was the work of Rodriguez and Kuhn [76]

507

on the reaction of $2 with Pt-A1 alloys. Rather than quantifying compositional changes, the authors utilised the chemical shifting of core level peaks caused by, for example, oxidation of a metallic species, to show that exposure to $2 at room temperature resulted in the formation of A1Sx species by the segregation of A1 to the Pt/A1 alloy surface. The driving force for this segregation was proposed to be the greater tendency for A1 to produce sulphides than is the case for Pt. Angle resolved XPS has been used to investigate the surface segregation and chemisorption of CO an oxygen on a chemically disordered Pt25Ni75(111) surface [77]. The authors were able to estimate the size of the Pt enrichment of the clean surface as a function of temperature by assuming a monotonic layerdependent model concentration and summing up the expected XPS contributions from each layer. In the calculation IMFP of the photoelectrons was assumed to be 0.8 nm for Ni 2p (Ek~n= 399 eV) and 1.4 nm for Pt 4f (Ek~n= 1184 eV). The variation of IMFP with exit angle was taken into account and the theoretical models compared with the experimentally derived Pt/Ni ratio as a function of polar angle. The authors anticipated that the greater interaction between Ni and O should lead to oxygen induced segregation of Ni, but they were unable to detect any such effect using this approach. XPS is one of the more adaptable surface analytical techniques in terms of the types of sample that may be investigated. For example, XPS is a favourite technique of researchers interested in the composition of supported bimetallic catalysts. However, Sault shows that the extent to which XPS and AES can be used to investigate adsorbate induced segregation is limited once the diameter of the bimetallic particles drops below about 6 nm [78]. Sault states that it is not possible to differentiate between clusters consisting of homogeneous alloy particles and those in which one component has segregated to the surface for cluster sizes smaller than approximately 6 nm. Once the particle diameter becomes comparable with the IMFP of the photo- or Auger electrons, essentially every photoelectron or Auger electron generated in the particle is detected regardless of location within the particle. Furthermore, for very small particles (<2 nm), a majority of the total atoms in the particle reside at the surface, and the concept of surface segregation becomes meaningless. A calculation of the relative XPS signals for a homogeneous particle and a segregated particle is carried out. Plotting this change as a function of particle size allows determination of the minimum particle size below which XPS cannot distinguish between the two. Depending on the specific metal pair in question, the minimum particle size varied from 3 to 8 nm.

508 3.1.2. Photoelectron Microscopy (PEEM, S P E M )

Imbihl and coworkers have used both Photoelectron emission microscopy (PEEM) and Scanning Photoelectron Microscopy (SPEM) to investigate adsorbate induced segregation at Pt/Rh surfaces [79-80]. In these experiments, microstructured 7 nm thick Rh films cover a Pt(100) surface. The microstructures consist of circular Pt(100) domains surrounded by Rh. This enables the overall reaction rate to be correlated with local spatiotemporal patterns exhibited as a change in the local work function with PEEM. SPEM uses a synchrotron beam focussed down to a spot of diameter less than 200 nm. The analyser is set to monitor photoelectrons from specific core levels and the sample is then scanned with respect to the x-ray beam. The SPEM measurements demonstrated vertical intermixing in the 7 nm Rh layers on Pt(100) leading to Rh/Pt alloy formation. The Rh/Pt ratio was found to change reversibly with the ambient gas composition. Oxygen induced enrichment of the R h ~ t film was observed. The microstructured Rh film has a higher tendency for oxide formation close to the boundary of the Pt domains.

3.2. Ion Scattering Spectroscopies Ion scattering techniques involve firing a collimated monochromatic beam of ions (typically H § He +, Li § Ne § at a target material and monitoring the scattered ions as functions of ion energy and scattering angle to obtain structural and compositional information on surfaces. One of the crucial features of these techniques is that the incident ion beam interacts strongly with the ion cores of the atoms in the solid. Treating this interaction as a Coulombic interaction, one can show that the trajectories of the incident ions form a shadow cone behind the target atoms whose radius, rs is given by rs - 2 (Zl Z2e2d/4 ~CeoE)1/2

(7)

where Zl and Z2 are the atomic numbers of the incident and target atoms, d is the distance behind the target atoms and E is the incident beam energy [81]. Low energy ion scattering utilises incident ions whose energy is in the range 0<E<5 keV. In this energy range, the radius of the shadow cone is so large even at small distances behind the target atom that, consequently, only the surface layer is visible to the incident beam. Therefore, the scattered ions measured by the detector are derived from the surface layer alone. In addition, any ions that penetrate the topmost layer are neutralised effectively and will not contribute to the scattered ion signal [82]. These factors make LEIS an extremely surface sensitive tool.

509 MEIS, by contrast, is frequently used to achieve depth profiling information with close to monolayer resolution. The energy of the incident ion beam is in the order of 100 keV. At such energies, the shadow cone radius is relatively small and incident ions are able to channel hundreds of nanometres into the bulk of a crystalline lattice. It is possible, with careful sample alignment, to selectively illuminate a given number of surface layers (see below), in which case one may achieve layer by layer compositional information as a function of depth.

3.2.1. Low Energy Ion Scattering (LEIS) Novacek and Varga [83] outline in detail how LEIS may be used to quantify adsorbate induced segregation effects at bimetallic surfaces using the example of oxygen adsorption on Pt•215 alloys. The decrease of the LEIS signal of the components of the clean surface (in this case Ni and Pt) due to the adsorption of oxygen and also the increase of the LEIS signal of oxygen gives a semiquantitative composition of the first layer. The intensity of backscattered ions depends on several parameters. Firstly, the specific scattering cross sections, which can be calculated relatively easily, need to be taken into account. Secondly, the intensity depends on the concentration of the respective elements, which is the desired parameter to be derived from the calculation. Thirdly, the specific survival probability of the incident ions needs to be calculated. This parameter needs careful consideration with LEIS (in contrast to MEIS (see below)). Novacek and Varga took great care to ensure that there was no element specific change in the survival probability. Firstly, they took measurements at three different primary energies and showed that the results obtained were independent of energy. Secondly, they plotted the intensity of Ni derived LEIS peaks as a function of the equivalent intensity of Pt peaks for several different compositions and showed this to result in a straight line. Further complications can arise if the experimental geometry is not chosen carefully. Specific shadowing of one component may occur. To investigate the effect of adsorbing oxygen onto the bimetallic surface, it is important to carry out additional experiments on each individual clean surface. On adsorption of oxygen, the LEIS signal due to the metal will decrease due to the shadowing of top layer metal atoms by the adsorbate. By comparing how much the Ni and Pt signals decreased upon oxygen adsorption onto the monometallic surface with the corresponding decrease on the bimetallic surface, Novacek and Varga were able to conclude that oxygen adsorption induces Ni segregation to the surface. LEIS has been used to investigate adsorbate induced segregation at the surfaces of bimetallic nanoclusters [84]. van den Oetelaar et al. showed that for Pt/Pd catalysts with low metal dispersions of about 0.3 and 0.8, Pd surface

510

segregation takes place on heating in hydrogen to approximately the same extent as in Pt/Pd bulk alloys. For very small P t ~ d catalyst particles with a dispersion close to 1 (i.e. nearly all atoms are at the surface); surface segregation is completely suppressed during heating in hydrogen and oxygen. This particle size effect is attributed to the limited supply of bulk material in the nanoclusters. Clearly, LEIS outperforms XPS in this type of experiment in the sense that meaningful surface information can be gathered even for the very smallest nanoparticles. However, van den Oetelaar et al. struck a note of caution in their work. They state that great care should be taken when performing LEIS experiments on small clusters because damage by sputtering may be significantly larger than that of bulk metal [85]. Sputtering may also change the surface composition by preferential sputtering and radiation enhanced segregation and diffusion [86-88]. It is suggested that static measurements using a very low ion dose minimises this type of problem.

3.2.2. Medium Energy Ion Scattering (MEIS) One of the main differences between MEIS and LEIS is demonstrated by examination of equation (7). The shadow cone radius for a 100 keV ion beam (typical for MEIS) is much narrower than the low energy ion beam. The medium energy ions effectively sense the lattice atoms as an array of discrete point charges and are able to channel deep into the bulk of the material. Fig. 3 shows that, by careful alignment of the sample with respect to the beam, the presence of the shadow cone may be utilised to selectively illuminate a chosen number of surface layers. In addition, by analysing the scattered ions along particular "blocking" directions (Fig. 3), any "unwanted" contribution to the scattered ion signal from deeper lying layers is minimised. Deckers et al. developed an approach [89] whereby careful alignment of the sample along particular azimuthal directions resulted in the incident beam striking the top layer, the top two layers or the top three layers of the PtsoNis0(111) surface. Taking into account the scattering cross sections of Pt and Ni it is trivial to calculate the mole fraction of Pt in the top layer, the top two layers etc. Hence it is equally straightforward to calculate the layer by layer composition by solution of a series of simple simultaneous equations. Using this approach, Deckers et al. demonstrated that the surface layer was enriched in Pt (Pt75Ni25), the second layer was depleted in Pt (Pt27Ni73) and the third and subsequent layers quickly approached the bulk composition. Upon oxygen treatment (350 L 02 at 400~ the top layer was found to be disordered [90] and contained only Ni and O indicating substantial oxygen induced segregation of Ni to the surface. This type of approach was subsequently adopted by Baddeley and co-workers [91-93] who were interested in adsorbate induced segregation effects at the CusoPds0(110) surface.

511

In this case, the nature of the adsorbate was much more complex than in other similar studies. A range of chloroethene molecules were adsorbed onto the bimetallic surface with the aim of characterising the surface composition as a function of adlayer coverage and temperature. Investigating the fcc(110) surface proved to be considerably more difficult than the (111) investigations of Deckers et al.. Baddeley and co-workers measured the relative number of Cu and Pd atoms contributing to the scattered ion signal for each experimental geometry. They achieved this by normalising the peak intensities to the atomic number dependence of the scattering cross section (z2), the angular dependence of the scattering cross section (sin4(O/2)) and the beam footprint [81]. They found that rather than the total number of contributing Cu and Pd atoms varying in the ratio 1:2:3:4 for the 1, 2, 3 and 4 layer geometries, the ratios were 1:1.37:2.06:2.47. As it is physically impossible for the incident beam to strike less than 4 layers in the 4 layer geometry, these findings indicated that even using the double alignment geometries shown in Fig. 3, the layer illumination in the 1, 2 and 3 layer geometries must be significantly greater than 1.0, 2.0 and 3.0 layers respectively.

Figure 3: Schematic diagram showing a cross section of an fcc (110) surface in the <-112> azimuth. The arrows indicate the experimental geometries used to selectively illuminate

(from left to right) two, one, four and three surface layers [93]. There are several possible factors which may account for the difference between the fcc (111) and (110) surfaces. Firstly, the presence of the surface layer relaxation means that, in the one layer incident geometry (Fig. 3), the shadow cone created by the ions incident on the surface atoms is sufficiently narrow at the second layer atoms that the latter become visible to the incident beam. If the surface layer relaxation is different for the (111) and (110) surfaces of CuPd, this may account for some or all of the extra illumination. The second major factor is potentially the enhanced vibrations of surface atoms. The top layer atoms of the fcc (110) surface are only 7 coordinate compared with 9 coordinate atoms on the (111) surface and 12 for atoms in the bulk. A consequence of this low coordination number is that the top layer atoms vibrate with a considerably larger amplitude than those of the bulk. As a result, even in

512

the one layer double alignment geometry, a substantial fraction of the second layer atoms are visible to the incoming beam since their vibrational amplitude coupled with that of the surface atoms results in them spending a finite time outside the shadow cone. In addition, ions scattered from these atoms are able to reach the detector due to incomplete blocking caused by the vibrations of the surface atoms. Baddeley et al. [93] sought to use Monte Carlo simulations (the VEGAS code [94]) to estimate the contribution of each layer to the scattered ion signal for each experimental geometry. The model structure used in the simulations consisted of an fcc(ll0) lattice with the lattice parameter of CusoPds0. To surmount problems associated with defining a unit cell for a random substitutional alloy, the atoms in the model were considered to have an atomic number midway between Cu and Pd (i.e. Sr). The bulk Debye temperature of CuPd was utilised to account for bulk lattice vibrations. It was found that varying the surface layer relaxation over the range 0 - 10 % resulted in only a small variation in the predicted layer illumination. This variation was far too small to account for the rather high value of the first layer illumination measured experimentally. The surface vibrational enhancement emerged as the most critical parameter. An excellent fit to the experimental data was achieved assuming that there exists a surface to bulk vibrational enhancement of 2.0, a value consistent with literature values for the surface Debye temperature of Cu [93]. Furthermore, it was considered that the vibrational enhancement decays exponentially with depth into the bulk such that the enhancement in the second layer is 1/e of that of the top layer. This approach was similar to that adopted by Fowler and Barth [95] and Jiang et al [96] for their investigations on Cu(100). It seems reasonable to assume that the second layer on the CuPd(110) surface will have some enhanced vibration as the second layer atoms are also coordinatively unsaturated compared to the fcc bulk. Baddeley et al. [93] also investigated the effect of the adsorbate on the number of visible atoms by measuring the total number of visible Cu and Pd atoms as a function of C1 coverage. They concluded that any effect was negligible and could be discounted in the calculations. This is another advantage of the MEIS approach compared to LEIS. The shadow cone radius varies a s Z 1/2, SO the shadowing effect of H(a), C(a) and O(a) are all relatively weak and even the effect of Cl(a) can be neglected since it is very unlikely that C1 will sit in a site which is a continuation of the bulk bimetallic lattice. The combination of experiment and theory enabled Baddeley et al. to establish that for an incident He ion beam, using the geometries indicated in Fig. 3, the number of layers illuminated in the nominally 1, 2, 3 and 4-layer geometries was 1.45, 2.26, 3.15 and 4.00. Experimentally, assuming the sample is correctly aligned, the quality of the data is such that a high degree of

513

confidence in attached to the ratio WcJ(Wcu + Wpd) where Wo, and Wpd are the normalised peak intensities for ions scattered from surface Cu and Pd atoms in the 1 layer illumination geometry. The product 1.45 x Wcu/(Wcu + Wpd) gives the total number of Cu layers illuminated by the beam. This value is equal to 1.00 acu + 0.42 bcu + 0.03 Ccu + 0.00 dcu where acu, bcu, Ccu and dcu are the fractional compositions of Cu in the top, second, third and fourth layers respectively and the coefficients of ncu are derived from the output of the VEGAS code. Similarly, equations may be constructed for the 2.26, 3.15 and 4.00 layer illuminations giving the following series of simultaneous equations: 1.Oacu + 0.42 bo, + 0.03 Ccu + 0 dcu = 1.45 Wcu/(Wcu + Wpd) 1.Oacu + 1.0 bcu + 0.20 co, + 0.06 dcu = 2.26 XcJ(Xcu + xpd) 1.0 acu + 1.0 bcu + 1.0 Ccu + 0.15 dcl, = 3.15 YcJ(Ycu + YPd) 1.0 acu + 1.0 bo, + 1.0 Ccu + 1.0 dcu = 4.00 Zcu/(Zcu + Zed)

and similar equations for Wpd etc. The solution of these equations is trivial and leads to the calculated values of the layer compositions for each of the top four layers. These values are summarised in Table 2. The conclusions of this work were that the clean CusoPds0(110) surface was enriched in Cu (Cu65Pd35) with a corresponding depletion of Cu in the second layer in good agreement with the previous studies of Lobodacackovic and coworkers [97-99] and Bertolini and coworkers [100]. On adsorption of ethene at 325 K, Pd segregation occurred to the surface layer at the expense of second layer Pd giving a surface composition much closer to that of the bulk. Presumably this effect was due to the much stronger interaction of the ethene re-system with Pd in comparison to Cu. By contrast, when C1 was adsorbed onto the surface, and when the hydrocarbon moieties decomposed to leave just C(a), significant Cu segregation was observed to the surface layer, in certain cases resulting in an almost pure Cu top layer and a corresponding Cu depleted second layer. The C1 induced segregation of Cu to the surface can be understood in terms of the relative heats of formation of CuC12(s) (-219.9 kJmo1-1) and PdC12(s) (-198.6 kJmo1-1) [61]. One of the most revealing features of these experiments is that the composition of the top two layers of the bimetallic surface remains almost constant throughout. It appears that any segregation effect merely involves swapping atoms between the top two layers rather than any diffusion of material from the bulk. This may account for the fact that adsorbate induced segregation is observed even at room temperature and that diffusion of atoms from the bulk

514 to the s u r f a c e s i m p l y c a n n o t o c c u r at s u c h l o w t e m p e r a t u r e s on the t i m e s c a l e o f a typical experiment. Table 2 The calculated layer compositions of the Cus0Pds0(110) surface as a function of gas treatment and thermal processing. Experiment

1st layer

2 na layer

3ra layer

4 ra layer

(atom % Cu)

(atom % Cu)

(atom % Cu)

(atom % Cu)

Clean CusoPds0(110)

65

38

71

50

10 L ethene at 323 K

57

53

45

80

10 L ethene at 323 K

55

47

61

66

72

30

69

71

76

28

62

81

71

30

66

78

72

29

62

73

88

7

76

65

flashed 383 K 10 L ethene at 323 K flashed 473 K 10 L ethene at 323 K flashed 573 K 10 L 1,2-dichloroethene at 323 K 10 L 1,2-dichloroethene at 323 K; flashed 383K 10 L 1,2-dichloroethene at 323 K; flashed 473 K 10 L 1,2-dichloroethene

surface

at 323 K; flashed 573 K

restructuring

515 The ability to be able to ignore the presence of the adsorbate atoms in calculating the surface composition is a powerful aspect of the MEIS technique. However, it should be stated that this technique has so far only been used for single crystal substrates. No useful information has yet been reported for more catalytically relevant systems such as supported bimetallic clusters.

3.3. X-ray Absorption Spectroscopies 3.3.1. Extended X-ray Absorption Fine Structure (EXAFS) Essentially, the EXAFS technique involves the incident X-ray energy being scanned across an absorption edge (typically the K-edge, but also L-edges of heavier elements are often probed). Since the K- absorption edges of transition metals are relatively high in energy (e.g. Cu K edge 8.98 keV; Pd K-edge 24.35 keV), the incident X-rays are highly penetrating making this technique easily usable at atmospheric pressure or even at the liquid-solid interface. Hence, EXAFS is a popular technique for the study of supported metal catalysts. The outgoing photoelectron wave is able to interact with reflected waves from neighbouring atoms giving rise to energy dependent constructive and destructive interference - the characteristic EXAFS wiggles. The intensity of the wiggles is dependent on the number of neighbouring scatterers and the strength of vibration of the emitter-scatterer bond. The wavelength of the wiggles is dependent on the distance between emitter and scatterer and the elemental nature of the scatterer (as this defines the phase shift on reflection of the electron wave). The theory of EXAFS is well established [101] and fitting procedures are available which essentially treat the EXAFS wiggles as being a linear sum of all the contributions from all the atoms in the coordination shell of the emitter. If the sample is extremely well defined (e.g. a polycrystalline metal foil) then EXAFS data can be analysed very reliably. The output of the analysis gives the number of scatterers and their distance from the emitting atom. If the sample is less well defined as in the case of a typical supported metal catalyst where a distribution of particle sizes is present, the data analysis is more open to criticism. Essentially, EXAFS gives the average environment around a particular elemental species in the sample. Several workers have attempted to use EXAFS to provide information on adsorbate induced segregation effects on bimetallic particles. Hills et al. [102] studied the effect of a hydrogen atmosphere on bimetallic Pt-Ru nanoparticles. Their interpretation of adsorbate induced segregation effects relied on comparing the coordination environments of Pt and Ru and trying to identify differences from what would be expected for a random distribtion of Pt and Ru within the particles. They concluded that hydrogen treatment at 673 K resulted

516

in segregation of Pt to the surface of hemispherical 1.2 nm diameter nanoparticles. The danger of this type of analysis is not to be underestimated. The coordination numbers derived from EXAFS are strongly correlated with the Debye-Waller factors used to simulate the vibrations of atoms in the lattice. Essentially, in the fitting procedure if one overestimates the Debye-Waller factor, this has a dampening effect on the theoretical prediction of the amplitude of the EXAFS wiggles which can be compensated by increasing the coordination number of the scattering atoms. Secondly, for very small particles (as in the work of Hills et al.) a large fraction of the atoms are to be found at the surface of the particles. Many of the fitting programs for EXAFS analysis assume that the M-M bond vibrates isotropically. For atoms at the surface, the bond vibration is anisotropic. This factor may lead to inaccuracies in the data analysis [103]. Hills et al. [102] use energy dispersive X-ray analysis (EDX) and scanning transmission electron microscopy (STEM) to support the conclusions of the EXAFS work. EDX probes the overall particle composition and STEM gives information on particle size distributions. This corroborating evidence coupled with the fact that data were taken from both Pt and Ru edges add considerable weight to the conclusions of the EXAFS data. In general, with extreme care with EXAFS analysis and with corroborating data from other techniques, EXAFS can give qualitative information on adsorbate induced segregation at bimetallic particles, which for reasons outlined earlier, may be considerably more relevant than quantitative analysis of such effects on single crystal surfaces under UHV conditions. Similar investigations were carried out by Hansen et al. [104], who investigated alloy formation and surface segregation in zeolite supported Pt-Pd bimetallic catalysts and concluded that hydrogen treatment at 573 K results in Pd enrichment at the particles surface. Furthermore, Yang et al. [105] used EXAFS to show that upon decarbonylation of MgO supported PtRh5 clusters, particle aggregation occurred which resulted in Pt segregation to the surface of the aggregates in a hydrogen atmosphere at 548 K.

3.4. Vibrational Spectroscopies

3.4.1. Infra-red Spectroscopy The adsorption of CO followed by infrared (IR) spectroscopy has long been used to characterise supported and unsupported transition metals [106]. CO adsorbed on atop sites gives IR bands in the range 2100-2040 cm -~, while bridging and three-fold CO give bands in the range 1860-1780 cm -~. Extensive studies on monometallic systems have led researchers to assign individual peaks in a bimetallic system to CO adsorbed to each individual component. The

517

relative intensifies of the various CO bands, therefore, are thought to give an indication of the overall composition of the bimetallic surface. However, using CO as a tool to calculate the composition of a bimetallic surface has potential flaws. There is a considerable body of evidence that CO itself causes segregation of individual components at bimetallic surfaces. Bradley and coworkers [ 107] showed that the IR spectrum of CO(a) is time dependent at 300 K over a period of several days and is indicative of Pd enrichment in the surface layers of colloidal PdCu alloys. Anderson et al. [108] showed that an adsorbate induced segregation of Pd occurs following exposure of the bimetallic catalysts to a CO/H2 reaction mixture. They also attribute the fact that a two-fold bridging band is observed indicates that Pd is preferentially segregated on (100) facets of the bimetallic particles. Boellaard et al. [109] report that CO adsorption induces Fe segregation to the surface of alumina supported Fe/Cu catalysts at room temperature. Nerlov and Chorkendorff [110] report that CO induced surface segregation of Ni promotes methanol synthesis from CO, CO2 and H2 over Ni/Cu(100). Essentially, the picture emerging from CO adsorption studies is that IR spectroscopy of species adsorbed on the supported clusters and chemisorption methods give only indirect information about the surface of clusters, and the surface may be modified by the adsorbates used.

3.5. Other Techniques

3.5.1. Scanning Tunnelling Microscopy (STM) Despite the atomic resolution achievable with the STM technique, the lack of chemical specificity which is generally the case makes it difficult to glean much information on the composition of bimetallic surfaces. Varga and coworkers have achieved chemical specificity at PtRh(100) surfaces with the tip clearly imaging the different types of atoms [111 ]. This enables highly detailed information on the extent of local chemical ordering within the bimetallic surface layer and therefore of the range of different adsorption sites available to gas phase molecules. Once the molecular layer has been formed however, the extent of information on adsorbate induced segregation obtainable with STM is currently limited. STM has been used to image rather more dramatic adsorbate induced effects at bimetallic surfaces. For example, Baddeley et al. [112] investigated the adsorption of oxygen on a range of PtCe ordered alloys on Pt(111). This work showed that surface alloys based on the (0001) plane of the hcp bulk alloy PtsCe were formed on P t ( l l l ) . The PtsCe(0001) surface may either be terminated by a pure Pt layer in the form of a Kagom6 net of Pt atoms or by a bimetallic layer of composition PtzCe (Fig. 4a). In all, five different structures

518 were observed. At submonolayer Ce precoverage, produced two distinct structures.

annealing

the sample

0 0

..e"

C)

Ce atoms

9

Pt atom s

Figure 4a: Schematic diagram showing successive (0001) planes of PtsCe. Note the vacancy in each pure Pt layer directly above and below Ce atoms in adjacent layers [ 112].

Figure 4b: STM image showing a 0.5 ML Ce film on Pt(111) annealed to 1000 K. The insert shows an STM image of the clean Pt(111) surface. Note the rotation of the close packed directions [ 112]. The first structure (shown in Fig. 4b) consisted of a presumably pure Pt Kagom6 net terminating layer. The close packing directions of the surface

519 atoms was rotated by 30 ~ from that of the underlying Pt(111). The reason for this rotation is unclear. The consequence of this rotation is that Moir6 patterns are observed with a height variation of surface atoms with a periodicity of about 1.4 nm. Interestingly, when this surface is exposed to background CO or dosed with oxygen, it is remarkably unreactive. However, adsorbates can be seen to preferentially attack the brighter Pt atoms. This was interpreted as being due to the brighter atoms being in lower coordination (e.g. atop) and hence more reactive sites. A second structure was observed where the overlayer was rotated by only 4 ~ with respect to the underlying Pt. The STM image (Fig. 4c) showed a periodicity of--2.7 nm which was interpreted as the distance required for Ce atoms in the bimetallic Pt/Ce layer to return to purely three fold hollow sites on the underlying Pt surface. It appears that the Ce-Ce distance prefers to stay at -~0.54 n m (as in the PtsCe) structure rather than relax to 0.554 nm which would enable each Ce atom to sit in a 3-fold hollow site on P t ( l l l ) in a p(2x2) structure. This surface was also Pt terminated and very unreactive when exposed to CO or 02 even at elevated temperatures. In this case, each surface Pt atom is coordinated to 4 Pt atoms in the (0001) plane and 2 Ce and 2 Pt atoms in the layer beneath. Despite the buckling associated with the second layer Ce atoms being in different adsorption sites, the surface remains unreactive, presumably since each Pt atom is in an approximately similar environment.

Figure 4c: STM image showing a 2.0 ML Ce film on Pt(111) annealed to 1000 K [112].

520 Two of the other structures were interpreted as being thicker films of PtsCe which were rotated by --30 ~ with respect to each other. In these cases, STM picked out the Ce-Ce spacing of the PtsCe alloy. Presumably these films had grown respectively on the two low coverage surface alloys.

Figure 4e: STM image showing a 2 ML Ce film on Pt(111) deposited at 900 K and exposed to 10.9 torr CO for 1 hour [ 112].

521

A fifth structure is shown in Fig. 4d. This has a much longer range periodicity. Interestingly, exposure of this surface to CO at room temperature resulted in corrosion centres being nucleated at brighter features in the STM image. The corrosion centres eventually spread (Fig. 4e) until alternate layers of presumably pure Pt and cerium oxide are observed (upper part of Fig. 4e). On annealing in vacuum, this process was found to be reversible. The likely reason for this selective corrosion is that the surface layer is actually the bimetallic Pt2Ce layer. The brighter (more reactive) atoms are likely to be Ce atoms forced into low coordination sites due to the interface between the surface alloy and the underlying Pt(111).One is able to make several important conclusions from this work. Firstly, adsorbate induced segregation is almost certainly much more effective at low coordination sites such as the edges and comers of bimetallic nanoparticles. Secondly, ordered surface alloys may prefer to maintain a structure parallel to the underlying substrate rather than relaxing to be accommodated in a commensurate structure. The consequence of this is that novel reactive sites are produced on the bimetallic surface due to the change in coordination number of the surface atoms caused by the periodic transformation from hollow sites to bridge or atop sites.

3.5.2. Low Energy Electron Diffraction (LEED) Sporn et al. [113] investigated the oxygen induced p(3xl) reconstruction on PtzsRh75(100) using LEED I(E). Sporn et al. were able to conclude that oxygen adsorption reverses the segregation observed for the clean surface. The LEED I(E) approach involves measuring the intensity of diffraction beams in a LEED pattern as a function of beam energy to achieve I v E curves. A full multiple scattering calculation is carried out to predict the I(E) curves for a theoretical model and this is then compared with the experimental data. The technique relies on an accurate starting model such that only relatively small modifications are required to achieve the final structure. In addition, such studies are more difficult for multiple component systems such as the Pt/Rh/O system as this necessitates an increase in the number of parameters to be varied in the structural calculation. This study required considerable information from other techniques to formulate the starting model. For example, it utilised the STM work of Wouda et al. [114] on oxygen adsorption on PtsoRhs0(100) which concluded that oxygen adsorption induced Rh segregation to the surface layer. Hammer et al. [55] used LEED I(E) to study adsorbate induced variation of the structure and composition of the Mo75Re25(100) surface under the influence of adsorbed H, C and O. They concluded that above 600 ~ O(a) and C(a) induce segregation of Mo to the surface.

522 3.5.3 Nuclear Magnetic Resonance (NMR)

Savargaonkar et al [115] measured the surface compositions of a series of alumina supported Pt/Rh catalyst in the presence of chemisorbed hydrogen by 1HNMR spectroscopy. They found that the surface was slightly enriched in Rh compared with the bulk composition and significantly enriched compared to adsorbate free samples which were enriched in surface Pt. The interpretation of these data is complex however. The derivation of surface composition comes via the following argument. In the absence of diffusion of H atoms, two resonance chemical shifts should be observed ~Pt and 8 ~ consistent with H on pure Pt and pure Rh. The presence of one resonance in the spectra suggests that H is in fast exchange on Pt and Rh adsorption sites relative to the NMR time scale. The observed shift ~Rh/Pt can be expressed by (~Rh/Pt -- (~Pt Spt S + (~Rh SRh S

(8)

where Xpts and XRhs are the surface atomic fraction of Pt and Rh respectively. This approach has the advantage that data are taken from real catalysts at relatively high pressures (compared with UHV!) of, in this case, 7 torr. 4. C O N C L U S I O N S Adsorbate induced segregation is a phenomenon which is underestimated by many researchers interested in the chemical properties of bimetallic surfaces. The existing database points to the driving force for such segregation being the different heat of adsorption of the adsorbate with each alloy component. It is also an effect which is very difficult to quantify. The most successful quantitative methods are MEIS and LEED I(E) especially when the latter technique is coupled with LEIS. However, these techniques rely on single crystal measurements which may have limited relevance to measurements on bimetallic nanoparticles. Nonetheless, the ability to achieve layer by layer compositional analysis under the influence of an adsorbate means that MEIS data helps to explain why segregation is observed even at relatively low temperatures where bulk diffusion is extremely slow. Activation barriers for near-surface to surface diffusion must be significantly lower than bulk diffusion barriers. Ideally, the catalytic chemist would desire a technique capable of analysing the surface composition of bimetallic nanoparticles. Currently, no truly reliable methods exist for this type of analysis. XPS and EXAFS are not sufficiently accurate and CO adsorption may itself induce compositional changes in the surface layer. However, these techniques, perhaps used in tandem with high resolution techniques such as HRTEM, do provide useful qualitative insight into the surface composition of bimetallic particles.

523 In terms of heterogeneous catalysis, it is the study of nanoparticular systems in realistic environments where data are really required to enable a detailed characterisation of surface composition to optimise catalyst preparation conditions. REFERENCES [1] G.R. Rao; Current Science 75 (1998) 901 [2] J.H. Sinfelt; Accounts of Chemical Research 20 (1987) 134 [3] J. Laine, Z. Ferrer, M. Labady, V. Chang and P. Frias; Applied Catalysis 44 (1988) 11 [4] W.M.H. Sachtler and R.A. van Santen; Advances in Catalysis 26 (1977) 69 [5] V. Ponec and G.C. Bond; Studies in Surface Science and Catalysis; 95 (1995) 449 [6] J.H. Sinfelt, J.L. Carter and D.J.C. Yates; Joumal of Catalysis 24 (1972) 283 [7] Y.I. Yermakov and B.N. Kusnetsov; Journal of Molecular Catalysis 8 (1980) 13 [8] M. Masai, K. Moil, H. Muramoto, T. Fujiwara and S. Ohnaka; Journal of Catalysis 38 (1975) 128 [9] G.K. Reddy and P. Kanta Rao; Catalysis Letters 45 (1997) 93 [10] A.F. Lee, C.J. Baddeley, C. Hardacre, G.D. Moggridge, R.M. Ormerod, R.M. Lambert, G. Schmid and H. West; Journal of Physical Chemistry 99 (1995)6096 [ 11] A.F. Lee, C.J. Baddeley, C. Hardacre, G.D. Moggridge, R.M. Ormerod, R.M. Lambert, J.P. Candy and J.M. Basset; Journal of Physical Chemistry B 101 (1997) 2797 [12] I.M. Abdelrehim, K. Pelhos; T.E. Madey, E. Joseph and J.G. Chen; Joumal of Molecular Catalysis A 131 (1998) 107 [13] J.C. Kelzenberg and T.S. King; Journal of Catalysis 126 (1990) 421 [ 14] J.S. Feeley, A.Y. Stakheer, F.A.P. Cavalcanti and W.M.H. Sachtler; Joumal of Catalysis 136 (192) 182 [15] K.C. Taylor Catalysis Review 35 (1993) 457 [16] R. Ohnishi, H. Suzuki and M. Ichikawa; Catalysis Letters 33 (1995) 341 [ 17] A. Benedetti, G. Fagherazzi, F. Pinna, G. Rampazzo, M. Selva and G. Strukul; Catalysis Letters 10 (1991) 215 [18] J.C. Bertolini, J.L. Rousset, P. Miegge, J. Massardier and B. Tardy; Surface Science 287/288 (1993) 346 [19] A. Sarkany, Z. Zsoldos, G. Stefler, J.W. Hightower, L. Guczi; Joumal of Catalysis 157 (1995) 179 [20] V. Ponec; Advances in Catalysis 32 (1983) 149 [21] V.K. Shum, J.B. Butt and W.M.H. Sachtler; Joumal of Catalysis 99 (1986) 126 [22] J.H. Sinfelt and G.H. Via; Joumal of Catalysis 56 (1979) 1 [23] J.W.A. Sachtler and G.A. Somorjai; Joumal of Catalysis 81 (1983) 77 [24] R.C. Yeates and G.A. Somorjai; Journal of Catalysis 103 (1987) 208 [25] J.A. Rodriguez and D.W. Goodman; Surface Science Reports 14 (1991) 1 [26] C.A. Luengo, A.L. Cabrera, HB MacKay and MB Maple; Journal of Catalysis 47 (1977) 1 [27] K.I. Choi and M.A. Vannice; Joumal of Catalysis 131 (1991) 36 [28] P. Dufresne, E. Payne, J. Grimblot and J.P. Bonnelle; Journal of Physical Chemistry 85 (1981) 2344 [29] D.K. Lee, H.T. Lee, I.C. Lee, S. Park, S.Y. Bae, C.H. Kim and S.I. Woo; Journal of Catalysis 159 (1996) 219

524 [30] Y, Park, E. Oh and H. Rhee; Indian Engineering Chemical Research 36 (1997) 5083 [31] J.W. Veldsink, M.J. Bouma, N. Schoon and A.A.C.M. Beenackers; Catalysis Review 39 (1997) 253 [32] C.J. Baddeley, M.S. Tikhov, C. Hardacre, J.R. Lomas and R.M. Lambert; Journal of Physical Chemistry 100 (1996) 2186 [33] R.M. Ormerod and R.M. Lambert; Journal of Physical Chemistry 96 (1992) 8111 [34] C.J. Baddeley, A.W. Stephenson, R.M. Ormerod and R.M. Lambert; Journal of Physical Chemistry 99 (1995) 5146 [35] A. Couper and D.D. Eley; Discussions of the Faraday Society 8 (1950) 172 [36] D.H. Seib and W.E. Spicer; Physical Review B 2 (1970) 1676 [37] P. Soven; Physical Review 178 (1969) 1136 [38] G.A. Somorjai; Introduction to surface chemistry and catalysis; John Wiley and Sons, New York (1994) Chapter 3 [39] S. Overbury, P. Bertrand and G.A. Somorjai; Chemical Reviews 75 (1975) 547 [40] P. Wynblatt and R.C. Ku; Surface Science 65 (1977) 511 [41] D. McLean; Grain Boundaries in Metals (Oxford University Press, London 1957) [42] L.Z. Mezey and W. Hofer; Surface Science 352-354 (1996) 15 [43] A. Christensen, A.V. Ruban, P. Stoltze, K.W. Jacobsen, H.L. Skriver, J.K. Norskr and F. Besenbacher; Physical Review B 56 (1997) 5822 [44] C.E. Dahmani, M.C. Cadeville, J.M. Sanchez and J.L. Moran-Lopez; Physical Review Letters 55 (1985) 1208 [45] P. Weigand, B. Jelinek, W. Hofer and P. Varga; Surface Science 269-270 (1992) 1129 [46] P. Weigand, B. Jelinek, W. Hofer and P. Varga; Surface Science 301 (1994) 306 [47] Y. Gauthier, R. Baudoing, M. Lundberg and J. Rundgren; Physical Review B 35 (1987) 7867 [48] L.Z. Mezey; Surface Science 162 (1985) 510 [49] L.Z. Mezey and J. Giber; Acta Phys. Hung. 66 (1989) 309 [50] L.Z. Mezey and J. Giber; Surface Science 234 (1990) 210 [51] J.J. Burton, E. Hyman and D.A. Fedak; Journal of Catalysis 37 (1975) 106 [52] J.H. Sinfelt; Science 195 (1977) 641 [53] O.L.J. Gijzeman; Applied Surface Science 64 (1993) 9 [54] D. Tomanek; S. Mukherjee, V. Kumar and K.H. Bennemann; Surface Science 114 (1982) 11 [55] L. Hammer, M. Kottcke, K. Heinz; K. Muller and D.M. Zehner; Surface Review and Letters; 3 (1996) 1701 [56] J.H. Hodak, A. Henglein, M. Giersig and G.V. Hartland; Journal of Physical Chemistry B 104 (2000) 11708 [57] K.N. Tu, J.W. Mayer and L.C. Feldman; Electronic Thin Film Science for Electrical Engineers and Materials Scientists; Macmillan, New York, 1992 [58] S. Kakac, Y. Yenev; Heat Conduction, Taylor and Francis, Washington (1993) [59] H.J.V. Tyrrell and K.R. Hams; Diffusion in Liquids; Butterworth, London (1984) [60] Y.G. Shen, D.J. O'Connor and J. Yao; Applied Surface Science 125 (1998) 300 [61] D.R. Lide (Ed), CRC Handbook of Chemistry and Physics (72 nd edition), CRC Press, Ann Arbor, 1991-2. [62] J. Paul, S. Cameron, D. Dwyer and F. Hoffmann; Surface Science 177 (1986) 121 [63] A. Dauscher, L. Hilaire, J. Spirlet, W. Muller and G. Maire; Surface Science 204 (1988) 161 [64] U. Bardi, P. Ross and G. Rovida; Surface Science 205 (1988) L798

525 [65] G. Hoflund and D. Asbury; Surface Science 161 (1985) L583 [66] D. Asbury and G. Hoflund; Surface Science 199 (1988) 552 [67] R. Lad and J. Blakeley Surface Science 179 (1987) 467 [68] S.P. Jeng, P. Holloway, C. Batich and S. Hofmann; Journal of Vacuum Science and Technology A 5 (1987) 650 [69] M. Hegde, T. Sampath Kumar and R. Mallya; Surface Science 188 (1987) 255 [70] G.A. Somorjai and M.A. van Hove; Progress in Surface Science 30 (1989) 201] [71] C.T. Campbell; Annual Review of Physical Chemistry; 41 (1990) 775 [72] U. Bardi; Reports on Progress in Physics; 57 (1994) 939 [73] M.P. Seah and W.A. Dench; Surface and Interface Analysis 1 (1979) 2 [74] M.P. Seah and W.A. Dench, Practical Surface Analysis by Auger and X-ray Photoelectron Spectroscopy (D. Briggs and M.P. Seah (Eds), Wiley, New York, 1993 [75] J.X. Wu, S.X. Zhuang, M.R. Ji, M.S. Ma, W.X. Fang and K. Wandelt; Surface Science 352 (1996) 218 [76] J.A. Rodriguez and M. Kuhn; Journal of Vacuum Science and Technology A 15 (1997) 1608 [77] A. Pantforder, J. Skonieczny, E. Janssen, G. Meister, A. Goldmann and P. Varga; Surface Science 333 (1995) 824 [78] A.G. Sault; Catalysis Letters 29 (1994) 145 [79] R. Imbihl; Journal of Molecular Catalysis A - Chemical; 158 (2000) 101 [80] F. Esch, S. Gunther, E. Schulz, A. Schaak, I.G. Kevrekidis, M. Marsi, M. Kiskinova and R. Imbihl; Surface Science 443 (1999) 245 [81] R.M. Tromp; Practical Surface Analysis- Ion and Neutral Spectroscopy (D. Briggs and M.P. Seah (Eds)), Wiley, New York, 1992 [82] P. Weigand, P. Novacek, G. van Husen, T. Neidhart, L.Z. Mezey, W. Hofer and P. Varga; Nuclear Instruments and Methods B 64 (1992) 93 [83] P. Novacek and P. Varga; Surface Science 248 (1991) 183 [84] L.C.A. van den Oetelaar, O.W. Nooij, S. Oerlemans, A.W. Denier van der Gon, H.H. Brongersma, L. Lefferts, A.G. Roosenbrand and J.A.R. van der Veen; Journal of Physical Chemistry B 102 (1998) 3445 [85] W.K. den Otter, H.H. Brongersma and H. Feil; Surface Science 306 (1994) 215 [86] N.Q. Lam and H. Wiedersich; Nuclear Instruments and Methods B 18 (1987) 471 [87] R. Shimizu; Nuclear Instruments and Methods B 18 (1987) 486 [88] R. Kelly; Nuclear Instruments and Methods B 39 (1989) 43 [89] S. Deckers, F.H.P.M Habraken, W.F. van der Weg, A.W.D. van der Gon, B. Pluis. J.F. van der Veen and R. Baudoing; Physical Review B-Condensed Matter 1990, 42, 3253 [90] S. Deckers, F.H.P.M. Habraken, W.F. van der Weg, A.W. Denier van der Gon, J.F. van der Veen and J.W. Geus; Applied Surface Science 45 (1990) 121 [91] T.C.Q. Noakes, P.Bailey, S.C. Laroze, L.H. Bloxham, R. Raval and C.J. Baddeley; Surface and Interface Analysis 30 (2000) 81 [92] C.J. Baddeley, L.H. Bloxham, S.C. Laroze, R. Raval, T.C.Q. Noakes and P.Bailey; Surface Science 435 (1999) 827 [93] C.J. Baddeley, L.H. Bloxham, S.C. Laroze, R. Raval, T.C.Q. Noakes and P.Bailey; Journal of Physical Chemistry B 105 (2001) 2766 [94] R.M. Tromp; Practical Surface Analysis (2nd Edition), (D. Briggs and M.P. Seah (Eds)), Wiley, New York 1992 Volume 2, 577-612 [94] J.W.M. Frenken, R.M. Tromp and J.F. van der Veen; Nuclear Instruments and Method B 17 (1986) 334

526 [95] D.E. Fowler and J.V. Barth; Physical Review B 52 (1995) 2117 [96] Q.T. Jiang, P. Fenter and T. Gustaffson; Physical Review B 44 (1991) 5773 [97] J. Loboda-Cackovic, M.S. Mousa and J.H. Block; Vacuum 46 (1995) 89 [98] J. Loboda-Cackovic; Vacuum 48 (1997) 1405 [99] J. Loboda-Cackovic; Vacuum 48 (1997) 571 [100] A. Rochefort, M. Abon, P. Delich~re, and J.C. Bertolini,; Surface Science 1993, 294, 43 [101] EA Stem; Physical Review B 10 (1974) 3027 [102] C.W. Hills, M.S. Nashner, A.I. Frenkel, J.R. Shapley and R. Nuzzo; Langmuir 15 (1999) 690 [103] B.S. Clausen, L. Gr~bek, H. TopsCe, P. Stoltze, L.B. Hansen, J.K. NorskCv and O.H. Nielsen; Journal of Catalysis 141 (1993) 368 [104] P.L. Hansen, A.M. Molenbroek and A.V. Ruban, Journal of Physical Chemistry B 101 (1997) 1861 [105] O.B. Yang, M. Shirai, W.A. Weber and B.C. Gates; Journal of Physical Chemistry B 102 (1998) 8771 [106] L.-C. de Menorval, A. Chaqroune, B. Coq and F. Figueras; Journal of the Chemical Society - Faraday Transactions 93 (1997) 3715 [107] J.S. Bradley, E.W. Hill, B. Chaudret and A. Duteil; Langmuir 11 (1995) 693 [108] J.A. Anderson, M. Fernandez-Garcia and G.L. Hailer; Journal of Catalysis 164 (1996) 477 [109] E. Boellaard, F.T. van de Scheur; A.M. van der Kraan and J.W. Geus; Applied Catalysis A 171 (1998) 333 [110] J. Nerlov and I. Chorkendorff; Catalysis Letters 54 (1998) 171 [111] P.T. Wouda, B.E. Nieuwenhuys, M. Schmid and P. Varga; Surface Science 359 (1996) 17 [112] C.J. Baddeley, C. Hardacre, A.W. Stephenson, M. Tikhov and R.M. Lambert; Physical Review B 56 (1997) 12589 [113] M. Sporn, E. Platzgummer, E.L.D. Gruber, M. Schmid, W. Hofer and P. Varga; Surface Science 416 (1998) 384 [114] P.T. Wouda, M. Schmid, W. Hebenstreit, P. Varga; Surface Science 388 (1997) 63 [115] N. Savargaonkar, B.C. Khanra, M. Pruski and T.S. King; Journal of Catalysis 162 (1996) 277

527

Index A ab initio calculations, 2

acetylene, 209 acetylene cyclisation, 430, 439, 496 acid rain, 466 active computational cell, 49 adsorbate-induced segregation, 495 adsorption energy, 412, 449 adsorption on alloys, 144 AES, 108, 154, 155, 186, 202, 326, 334, 342, 365,407, 505 AFM, 188 Ag(100)/Cu, 33, 46 Ag(100)/Mn, 291 Ag(100)/Rh, 380 Ag(111)/CO, 461 Ag(l 11)/Sb, 278, 281 structure, 280 Ag/In, 219 Ag2A1, 111 Ag33Pd67(111), 139 Ag33Pd67(111)/O, 144 Ag3Pd(100), 123 Ag-Au, 503 Ag-Cu, 8 Ag-Pd, 123, 138,413 Ag-Pt, 8 Ag-Sb, 285 AgSx, 478 A1 clean surface structure, 229 A1 alloys, 106 AI(100), 232 AI(100)/alkalis, 245 AI(100)/Li, 247, 250, 268 AI(100)/Na, 246, 248, 261,267 AI(100)/Ni, 93 AI(100)/Yb, 246, 299 AI(110), 372 AI(110)/alkalis, 253 AI(110)/Li, 253 AI(110)/Na, 253,257 AI(111), 230 surface vacancies, 265 Al(111)/Ag, 111

AI(111)/alkali coadsorption, 243 AI( 111 )/alkalis, 233A1(111)/Cs, 234, 240 AI(111)/K, 235, 258, 265 AI(111 )/Li, 237 AI(111)/Na, 226, 235,237, 241,264, 266 AI(111)/Na+K, 244 AI(111)/Ni, 95 AI(111)/Pd, 441,457,491 AI(111)/Rb, 234, 236, 258 Al/alkalis surface phase transitions, 257 survey of adsorption phases, 227 A1/Li, 226 A1203, 366 A13Li, 226, 251,269 A13Ti, 252, 269 A1-Ag, 109 ALISS, 188, 207 alkali ion scattering spectroscopy. See ALISS alkali-aluminium surface alloys, 225 A1-Li, 214 alloy composition, 5 alloy nomenclature, 220 alloy ordering, 23 alloy phase diagrams, 19 alloy surface ordering, 86 alloying trends, 12 AI-Ni, 445 A1Ni(111), 216 A1-Pt, 445,485 A1Sx, 507 aluminide surfaces, 105 aluminium alloys, 106 angle-resolved ultra-violet photoelectron spectroscopy. See ARUPS anisotropic deformations, 42 antiferromagnet, 289 anti-phase domain boundaries, 321 anti-site defects, 131 ARUPS, 315, 320, 343, 357 atomic chain formation, 59 atomic chains, 68 atomic force microscopy. See A F M atomic radius, 287, 293, 294, 413,423 atomistic modelling, 30

528 Au dimer, 54 Au(110), 202, 431 Au(110)/Pd, 430 Au(110)/Pd/butadiene, 432 Au(111), 8, 297 Au(111)/A1, 301 Au(111)/CO, 461 Au( 111)/Pd, 498 Au(111)/Ru, 10 Au/Pd/CO, 503 Au3Pd, 189, 210 Au3Pd(100), 140, 430 Au3Pd(110), 430 Au75Pd25, 215 Au-Ag, 503 AuCu3, 185 Auger electron spectroscopy. See AES Au-Pd, 413 Au-Pd(111)/acetylene, 497 Au-Ru, 8

1,3-butadiene, 404 bcc-fcc transformation, 35 benzene, 209 BFS, 313 BFS chemical energy, 44 calculation, 44 BFS method, 36. See BFS BFS modelling of surface alloys, 47 BFS parameters, 45 BFS reference state, 45 BFS strain defect, 47 BFS strain energy, 39 bimetallic bonding, 454 binary alloy, 5 bi-phase binary alloy segregation, 109 blocking curves, 283 Bozzolo-Ferante-Smith method. See BFS Bragg-Williams theory, 90 bulk alloy types, 296 bulk alloys, 19 bulk chemical order, 130, 132 bulk immiscibility, 270 bulk modulus, 34, 41 bulk-surface exchange, 5 butadiene, 409 butane, 209, 409

butenes, 409 C catalyst poisoning, 466 catalytic activity, 404, 417, 434, 438 catalytic selectivity, 417,438 catalytic site blockers, 490 charge redistribution, 455 chemical reference energy, 32 chemical affinity, 144 chemical energy, 32, 38 calculation, 44 chemical ordering, 118, 130 chemical potential, 6 chemical reactivity, 433 chemical reference energy, 38 chemical shifts, 230 See also CLS

clock reconstruction, 353 CLS, 227, 229, 231,240, 241,250, 260, 263, 320, 343,355,420, 44 1,444, 447, 450, 458,480 cluster variation method, 89 clusters, 141 CO adsorption, 147, 209, 319, 345,430, 448,460, 503, 522 CO oxidation, 439, 496 Co3Pt(111), 104 cohesive energy, 4, 34, 41 compressive stress, 423 configurations, 52 CoNi(111)/CIz, 506 core level binding energies, 13, 231 core level photoemission, 229 intrinsic linewidths, 232 kine shapes, 232 line shapes, 251 core level photoemission spectroscopy. See CLS covalent radius, 294 coverage, 7 coverage dependence, 59 Cowley parameter, 102 Cr(100), 120 Cr(110)/Cr203, 398 Cu monolayers, 453 Cu stepped surfaces, 175 Cu(lO0), 8

529 Cu(100) surface alloys, 305 Cu(100)/Ag, 343 Cu( 100)/Ag-Cu, 10 Cu(100)/Au, 48, 72, 73,288, 296, 298, 308, 311,343 Cu(100)/Bi, 300, 336 de-alloying, 339 structural phases, 336 Cu(100)/Co, 322, 323 Cu(100)/Cu, 31 Cu(100)/CuAu, 122 Cu(100)/Fe, 219, 380 Cu(100)/In, 179, 341 Cu(100)/Ir, 349, 378 Cu(100)/Li, 326 structural phases, 329 Cu(100)/Mg, 309, 331 Cu(100)/Mn, 278, 288, 298, 308, 313, 321, 356 Cu(100)/Ni, 325 Cu(100)/Ni/CO, 517 Cu(100)/Pb, 121,165, 334 structural phases, 336 structure, 166 Cu(100)/Pd, 31, 48, 62, 71, 179, 288, 308, 312, 316, 344, 345,348, 351 Cu(100)/Pt, 68, 71,347, 348, 355 Cu(100)/Rh, 349 Cu(100)/Sn, 341 Cu(100)c(2x2)-Pd, 64 Cu(100/Fe, 322 Cu(110)/Au, 48, 72, 77 Cu(110)/Mn, 278 Cu(110)/Pb, 172 structure, 174 Cu(110)/Pd, 23, 48, 68,428 Cu(110)/Pd/butadiene, 429 Cu(111), 8 Cu(111)/Co, 15, 380 Cu(111)/CO, 461 Cu(111)/Co-Cu, 16 Cu(111)/Cs, 234 Cu(111)/Cu, 153 Cu(111)/Ni, 95 Cu(111)/Ni/CO, 461 Cu(111)/Ni-A1, 95 Cu(lll)/Pb, 121,158 structure, 162 Cu( 111)/Pd, 24

Cu(111)/Pd/CO, 460 Cu(111)/Pt-Cu, 10 Cu(111)/Sb, 278 structure, 280 Cu(111)/Sn, 291 Cu/Co, 153 Cu/Fe/Cu layers, 326 Cu/Pb, 152, 175 survey of surface phases, 177 Cu2Pd, 26 Cu2Sb, 279, 296 Cu3Au, 185, 190, 211, 251,269, 296, 308 Cu3Au(100), 101,140, 212, 314, 320, 365, 372, 375 Cu3Au(100)/Fe, 394 Cu3Au(100)/Ni, 394 Cu3Au(100)/O, 377, 399 Cu3Au(100)/V, 394 Cu3Au(100)/VOx, 396 Cu3Au(110), 365, 372, 373 Cu3Au(110)/O, 375 CuaPd, 23, 25, 32, 314, 352 Cu3Pd(110), 101 Cu3Pt, 210 Cu3Pt(100), 355 Cu3Pt(111), 138 Cu3Pt(111)/CO, 461 CuaPt(111)/Pt/CO, 461 Cu-Ag, 8 Cu-Au, 79, 214, 310 Cu-Mn, 289 CuPd, 312 Cu-Pd, 22, 345, 413 CuPd(110), 510, 513 CuPd(110)/dichloroethene, 514 CuPd(110)/ethene, 513 Cu-Pt, 8, 485 Cu-Sn, 153 CuSx, 478 cyclohexane, 209 D

de-alloying, 152, 160, 167, 333, 336, 343 Debye temperature, 512 decay chains in energy level spectra, 50 dehydrogenation, 496 dendritic islands, 471 density functional theory. See DFT

530 desegregation, 107, 110 desorption energy, 477 de-wetting, 469 DFT, 2, 61,157, 226, 238, 241,246, 264, 269, 284, 323,408, 461,490 diffuse LEED. See DLEED direct methanation fuel cells, 185 discommensurate structure, 174 disordered alloys, 23 DLEED, 317 domain coarsening, 171 Doniach-Sunjic function, 232

EAM, 157, 214, 217, 312 ECT. See Equivalent Crystal Theory ECT parameters, 41 effective atomic radius, 287, 301 effective chemical potential, 6 effective medium theory. See EMT electronic effect, 490, 499 embedded atom method, 89. See EAM EMT, 157, 162, 217 endothermic processes, 88 energetics of surface alloy formation, 178 energy of formation, 36, 53 energy of mixing, 9 Engel-Brewer model, 211 ensemble effect, 119, 145,404, 438,490, 497 enthalpy of segregation, 89 entropy, 6 entropy of segregation, 89 epitaxial growth, 18, 45,207 equiatomic binary alloys, 97 equilibrium properties, 1 equilibrium shape, 164 equilibrium state, 88 equivalent chemical lattice parameter, 44 equivalent crystal, 32 Equivalent Crystal Theory, 33, 39 ethanol, 209 EXAFS, 515 exothermic process, 88 extended X-ray absorption fine structure. See EXAFS

faceting, 199, 202, 203 fcc(100) surfaces, 140 Fe(lO0), 120 Fe3Pt, 210 FeAI(100), 216 FeAl(110), 108 FeAl(111), 107 FeAI2, 108 Fe-Pd, 113 ferromagnetic ordering, 289, 313 field ion microscopy. See FIM FIM, 189 fire gilding, 184 first-principles methods, 1, 32 Fischer-Tropsch synthesis, 496 FLAPW, 268, 331,455,458 floating stacking fault, 279 formate, 356 formation energy, 53 formic acid, 356 free energy, surface, 7 free-energy concentration expansion method, 90 free-energy expansion method, 89 Frenkel pair, 266 Friedel model, 4, 14 fuel cells, 185 full-potential linearised augmented plane wave. See FLAPW G gas chromatography, 416 giant magnetoresistance. See GMR Gibbsian segregation, 129 GMR, 153 gold wetting, 467 grain boundary segregation, 501 growth, 15, 46 I-I hard-sphere model, 278, 290 hard-sphere radii, 239 HAS, 156 helium atom scattering. See HAS high-resolution electron energy loss spectroscopy. See HREELS hill and valley structure, 204, 215

531 HREELS, 407, 420 Hund's rules, 289 hydrocarbon reforming, 466, 496 hydrodesulphurisation, 466, 487, 496 hydrogenation, 404, 409, 434, 439, 496 reaction products, 416 hydrogenolysis, 434, 496

ICISS, 366 immiscible metals, 31 immiscible systems, 270 impact-collision ion scattering spectroscopy. See ICISS, NICISS infrared reflection-absorption spectroscopy. See IRAS infra-red spectroscopy, 516 interatomic distances, 277, 293,295 interatomic potentials, 32 interface stress, 405 interfacial stacking fault, 279 interlayer spacing, 327 interstitial solid solutions, 88 ion scattering spectroscopy. See LEIS, MEIS, ALISS, ICISS, NICISS ionic bond, 455 Ir(100), 392 Ir(100)/Cu, 389 IRAS, 407,434 Ising model, 89 Ising model Hamiltonian, 91 island formation, 15 island growth, 168 islands, 59 isobutane, 209 isomerisation, 418,439 ISS. See LEIS 1UPAC, 220 K

Kagom6 net, 517 kink sites, 266

Langmuir-McLean theory, 88, 127 lattice constant, 161 lattice gas, 156 layer spacing relaxation, 229

layer-by-layer growth, 153,405 LDA, 157, 312 LDOS, 120, 211,325, 387 LEED, 104, 108, 119, 122, 129, 134, 154, 155, 166, 173, 186, 197, 202, 207, 208, 226, 228, 238, 240, 241,246, 257, 259, 264, 284, 292, 309, 312, 318, 323, 325, 327, 331,334, 336, 341,347, 349, 365, 369, 370, 373,407, 431,521 dynamical theory, 228 quantitative structure determination, 228 LEEM, 154, 156, 163, 168, 338 LEIS, 118, 128, 129, 166, 186, 188, 192, 197, 202, 286, 292, 318, 341,343,348, 349, 365, 379, 381,395,407, 415, 431, 508 ligand effect, 119, 147, 404, 438 local density approximation. See L D A local density of states, 120, 123 long-range order, 86, 131 low energy electron diffraction. See LEED low energy electron microscopy. See LEEM low energy ion scattering. See LEIS M

machine bearings, 154 magnetic exchange splitting, 357 magnetic moment, 278, 289, 314 mean-field Bragg-Williams theory, 90 medium energy ion scattering. See MEIS MEIS, 101,129, 134, 282, 284, 286, 289, 292, 309, 408, 509, 511 blocking curves, 283 metallic bond, 455 metallic radius, 287, 290, 307 methanation, 185 methanol, 209 methanol synthesis, 305 misfit dislocation, 129 missing row structure, 202, 330, 431 mixing energy, 9, 25 MnA1, 108 Mo(100)/Au/S, 472 Mo(100)/Pd, 299 Mo(110)lAg/S, 486 Mo(110)/Au/S, 472, 473 Mo(110)/Cu/S, 486

532 Mo(110)/Ni/S, 486 Mo(110)/Pd, 405, 441, 491 Mo(110)/Pd/CO, 460 Mo(110)/S, 482 Mo(110)/Zn/S, 486 Mo/S, 474 Mo75Re25(100), 503 Mo75Re25(100)/C, 521 Mo75Re25(100)/O, 521 Mo-Ni, 445 monolayer growth, 15 Monte-Carlo simulations, 88, 102 MoSy, 487 muiltilayer relaxation, 36 multi-component alloys, 90 multilayer alloys, 217, 351 multilayer growth, 15 multilayer segregation, 90 N

Nb(110)/Pd, 405,440 neutral-detection ICISS. See NICISS NEXAFS, 411 Ni monolayers, 453 Ni(100), 372 Ni(100)/AI, 22 Ni(100)/Cu, 48 Ni(100)/Mn, 278, 288 Ni(100)/Sn, 291 Ni(110), 372 Ni(110)/AI, 20 Ni(110)/Au, 48, 51,217 Ni(110)/butadiene, 412 Ni(110)/Cu, 48, 78 Ni(110)/Cu-Au, 79 Ni(110)/Ni-Al, 21 Ni(110)/Pd, 48, 60, 424 Ni(110)/Pd/butadiene, 426 Ni(110)/Sn, 291 Ni(111 )/Sn, 291 Ni(111)/butadiene, 412 Ni(111)/CO, 461 Ni(111)/K, 234 Ni(111)/Pb, 286, 296 Ni(111)/Pd, 423 Ni(111)/Pt, 405 NiaA1, 20, 33, 210 Ni3Al(100), 212

NiaPt(111), 105 Ni60Fe40(100)/O, 504 NiasCrls( 111)/O, 504 NiAI(100), 212, 366, 367, 370 NiAI(110), 134, 366 NiAI(111), 366, 368 NiA1/AI203, 394 Ni-AI-Cu, 93 near-surface concentrations, 94 NICISS, 366, 367, 369, 376 NiPb, 297 Ni-Pd, 413 Ni-Pt, 445 NIXSW, 238, 241,264 NMR, 522 noble metal adatoms, 467 nomenclature for alloys, 220 non-equiatomic binary alloys, 99 normal incidence X-ray standing waves. See NIXSW nuclear magnetic resonance. See NMR O oil refining, 466 on-top adsorption, 239 open system, 1 order-disorder transition, 114, 170, 262 ordered alloys, 23 ordered alloys, surface segregation, 96 ordered binary alloy, 87 overlayer stability, 17 overlayer to underlayer transition, 345 oxygen adsorption, 369 P

pattern formation, 167 Pd monolayers, 447 Pd thin films, 423 Pd(100), 318 Pd(100)/Al/O, 504 Pd(100)/CO, 448 Pd(100)/Mn, 278, 288, 296, 298 Pd(110)/butadiene, 412, 420 Pd(110)/Cu, 68, 70 Pd(110)/ethylene, 427 Pd(111), 405 Pd(111)/acetylene, 497 Pd(111)/Ag, 216

533 Pd(l 11)/butadiene, 410 Pd(111)/CO, 421,461 Pd(111)/ethylene, 412 Pd(111)/NO, 421 Pd(111)/Pd, 457 Pd/Au/CO, 503 Pd/SiO2, 410 Pd3Mn, 291 Pd3Mn(100), 460 PdsNi95, 414 PdsPt95, 414 PdsNi82(110), 418 PdsNi92(110)/butadiene, 420 PdaNi92(111), 418 Pd-Ag, 413 Pd-A1, 441 Pd-Au, 413 Pd-Cu, 22, 413 PdCu(110), 421 PdCu(110)/butadiene, 419 PdCu(111), 421 PdCu(111)/butadiene, 419 PdCu(111)/CO, 460 Pd-Cu/CO, 517 PdCu3(110), 428 Pd-Ni, 413, 421 Pd-Pt, 413 Pd-Ti, 441 PEEM, 508 phase diagrams, bulk alloys, 19, 32 phase separation, 88, 389 phase transitions, 257, 345 photoelectron diffraction, 78, 187, 285, 289, 309 photoelectron emission microscopy. See PEEM preferential sputtering, 128 pseudomorphic layer, 2, 405,433 Pt alloys, 135, 210, 445 Pt monolayers, 453 Pt(100), 214 Pt(100)/Rh, 508 Pt(100)/Rh/O, 508 Pt(100)/Sn, 209, 291 Pt(110), 202 Pt(110)/butadiene, 412 Pt(lll), 8 Pt(111)/Ag, 10 Pt(111)/Ag/S, 483

Pt(111)/AI/S, 484 Pt(111)/Au/S, 472 Pt(111)/butadiene, 410 Pt(111)/Ce, 517 Pt(111)/Ce/CO, 520 Pt(111)/Co, 218 Pt(111)/CO, 461 Pt(111)/Cu/CO, 453, 461 Pt(111 )/Cu/S, 484 Pt(111)/ethylene, 412 Pt(111)/Fe, 136 Pt(111)/S, 482 Pt(111)/Sn, 200, 207, 218, 291 Pt(111 )/Sn/S, 488 Pt(l 11)/Zn/S, 484 Pt/SiO2, 410 Pt25Co75(100), 124 Pt25Co75(111), 136 Pt25Ni75(lll), 130, 136, 145, 216 Pt25Ni75(111)/CO, 507 Pt25Rh75(100), 142, 145, 521 Pt25RhT5(111), 126, 128 PtasCu65(111), 136 Pt3Mn, 210 Pt3Sn, 185, 190 density of states, 193 oxygen adsorption, 196 surface defects and disorder, 215 surface structure, 212 Pt3Sn(100), 197, 212 Pt3Sn(110), 202 Pt3Sn(111), 104, 191,213, 216 surface composition and morphology, 192 Pt3Ti, 211 Pt3Ti(111), 212 Pt3Ti( 111)/O, 504 Pt3Ti(510), 216 Pt4o_Ni6o(100), 141,143 PtsoFeso(111), 136 PtsoRhso(100), 142 PtsCe(0001), 517 PtsoCo2o(100), 213 PtsoFe20(111), 213 Pt-Ag, 8 Pt-AI/S, 507 PtCo/CO, 147 PtCo3(100), 140 Pt-Cu, 8

534 PtCu(111), 138 Pt-Ni, 501 PtNi(100), 124 PtNi(111), 124, 510 PtNi(111)/O, 510 Pt-Ni/O, 509 Pt-Pd, 413, 516 Pt-Rh, 516 PtRh(100), 123 PtRh(100)/O, 521 Pt-Rh/H, 522 Pt-Ru, 515 PtS2, 483 PtSn, 190 Pt-Sn, 184, 190, 488 phase diagram, 190 Pt-Sn surface alloys, 207 Pt-Sn/O, 504 PtSx, 483 pyramid morphology, 199

Q quaternary alloys, 36 R

Rayleigh phonons, 306 RBS, 317, 341,368, 409 Re(0001)/Pd, 405,406, 441,459 Re(0001)/Pd/CO, 448,450 reference state, 32 reflection high energy electron diffraction. See RHEED Re-Pt, 445 reservoir, 49 Rh(100)/Cu/CO, 453 Rh(111)/Au/S, 472 Rh(111)/Pd, 440, 446, 457 Rh(111)/Pd/CO, 460 Rh( 111)/Pd/SO2, 490 Rh(111)/Sn, 291 Rh3Pt(111), 105 RHEED, 187, 200, 323,325 rigid band model, 499 row-pairing model, 374 Ru(0001)lAg/S, 475 Ru(0001)/Au, 467 Ru(0001)/Au/S, 468 Ru(0001)/CO, 406

Ru(0001)/Cu, 479 Ru(0001)/Cu/S, 475 Ru(0001)/Ni/CO, 461 Ru(0001)/O, 393,406 Ru(0001)/Pd, 406, 441,459 Ru(0001)/Pd/CO, 460 Ru(0001)IS, 468 Ru-Au, 8 Ru-Ni, 445 Rutherford backscattering. See RBS

scanning transmission electron microscopy. See STEM scanning tunnelling microscopy. See STM scanning tunnelling spectroscopy. See STS second harmonic generation. See SHG secondary electron imaging, 111 segregation, 36, 86, 127, 364, 505 Gibbsian, 129 radiation induced, 129 site-specific, 141 segregation energy, 10, 88 segregation energy, general trends, 13 segregation enthalpy, 88, 128 segregation in multi-element alloys, 90 segregation/order factor, 114 selectivity of reaction, 417 self assembly, 164, 167 SEXAFS, 226, 238, 241,246, 264, 313 SHG, 258 short-range order, 25, 86, 131, 137 site blockers, 490 site exchange, 54 site-specific segregation, 141 'skin' alloy surfaces, 219 Smoluchowski effect, 204, 393 Sn, 292 Sn-Pt, 184, 488 SO2 production, 466 solders, 153 solid solutions, 88 SOx species, 466 SPA-I~EED, 154, 155, 158, 161,176, 187, 192, 370 spin glass, 289 spot-profile analysis LEED. See SPALEED

535 sputtering, preferential, 128 stability of overlayers, 17 stacking fault, 277, 279 stacking fault energy, 284 statistical-mechanical approximation, 88 STEM, 516 step edge, 68, 393 step interaction, 164 step roughening, 380 stepped surfaces, 175, 216, 504 steps, 242, 270 STM, 24, 119, 156, 159, 162, 167, 173, 188, 194, 197, 203,238, 241,281,311, 316, 322, 337, 342, 344, 349, 365, 371, 373, 377, 385, 386, 392, 407, 419, 431, 477, 517, 520 chemical contrast, 119 chemical contrast mechanisms, 121 chemical discrimination, 120 electronic effects, 123 tip-surface interaction, 125 topographic effects, 121 strain defect, 47 strain energy, 32, 37 calculation, 39 strained overlayers, 405 Stranski-Krastanov growth, 158, 334, 405 stress relief, 433,479 STS, 120 substitutional solid solutions, 88 subsurface alloy, 384 subsurface composition, 127 subsurface impurities, 385 sulphidation, 487 superalloy, 134 surface alloy configuration, 7 surface alloy formation, 30, 178 surface alloy, generic classes, 8 surface alloys multilayer, 217 surface buckling. See surface rumpling surface chemical ordering, 118, 132 surface composition, 127 surface core level shifts, 13 surface coverage, 7 surface de-alloying, 152, 160, 167 surface defects, 311 surface diffusion, 306, 321,337, 424 surface energies

table, 307 surface energy, 2, 3, 12, 16, 36, 433 surface extended X-ray absorption fine structure. See SEXAFS surface ordering, 86, 130 surface phase transitions, 257 surface reactions, 404 surface reactivity influence of surface orientation, 418 surface reconstruction, 391 surface relaxation, 229, 297, 327, 371 surface rippling. See surface rumpling surface roughening, 370 surface rumpling, 277, 285, 287, 290, 310, 313, 385, 519 surface segregation, 36, 86, 118, 127, 364, 381,413,415, 501,505 bi-phase binary alloy, 109 ordered alloys, 96 surface states, 315 surface steps, 242, 270 surface stress, 163, 204, 297, 306, 405, 419, 479 surface structure, 36, 277 surface structure nomenclature for alloys, 220 surface vacancies, 6, 265 surface X-ray diffraction. See S X R D surface-bulk exchange, 5 surfactant, 153, 179 SXRD, 155,279, 282, 284, 336, 424, 431, 434

Ta(110)/Pd, 405,441,456, 459 Ta(110)/Pd/CO, 448 Tamm surface states, 315, 319 Ta-Ni, 445 Ta-Pt, 445 TEAS, 156, 158, 161,172 tensile stress, 430, 434 ternary alloys, 36 thermal desorption, 445,447, 468 thermal energy atoms scattering. See TEAS thermal vibrations, 512 thermodynamic equilibrium, 127 thermodynamics, 1

536 thiophene dissociation, 491 Ti(001)/Pd/CO, 460 Ti-Ni, 445 total energy, 3 total energy calculations, 1 transition metal alloys, 11,142 transition metals, 4d, 4, 11 transition metals, 5d, 11 transition metals, segregation energies, 14 truncated octahedral particles, 502 U UBER. See universal binding energy relationship underlayer alloy, 345,384 universal binding energy relationship, 34 V V203, 397

V3Au(100), 395 vacancies, 6 vacancy formation energy, 178 VAI, 108 valence photoemission, 439, 449, 484 vibrational amplitudes, 239, 255, 372 vibrational spectroscopy, 516 vicinal surfaces, 175, 216, 346 VO(100), 398 VO2, 399 Volmer-Weber growth, 18, 405 VOx, 396 W W(100)/Cu, 217, 299 W(lOO)/Pd, 440 W(110)/Pd, 441,459 W(110)/Pd/CO, 448,460 Warren-Cowley SRO parameter, 25 wetting, 164, 467,469 Wigner-Seitz radius, 33 W-Ni, 445 Wood notation, 220 W-Pt, 445 X

XANES. See NEXAFS XPD, 187, 200, 208, 241,246, 319

XPS, 13, 110, 186, 325, 365,381,407, 442, 444, 450, 483, 505 X-ray diffraction, 139, 155, 173 X-ray photoelectron diffraction. See XPD X-ray photoelectron spectroscopy. See XPS

Zn-Pt, 445,485

Oxide Surfaces (The Chemical Physics of Solid Surfaces Vol. 9)

Oxide Surfaces (The Chemical Physics of Solid Surfaces Vol. 9)

Solid state physics: surfaces

Solid state physics: surfaces

Solid Surfaces and the Gas-Solid Interface

Solid Surfaces and the Gas-Solid Interface

Physics of Surfaces and Interfaces

Physics of Surfaces and Interfaces

Physics of Surfaces and Interfaces

Physics of Surfaces and Interfaces

Physics of Surfaces and Interfaces

Physics of Surfaces and Interfaces

Methods of Experimental Physics - Solid State Physics - Surfaces

Methods of Experimental Physics - Solid State Physics - Surfaces

Photochemistry on Solid Surfaces (Studies in Surface Science and Catalysis)

Photochemistry on Solid Surfaces (Studies in Surface Science and Catalysis)

Surfaces

Surfaces

Surfaces

Surfaces

Surfaces

The Chemical Physics of Solid Surfaces : Growth and Properties of Ultrathin Epitaxial Layers

The Chemical Physics of Solid Surfaces : Growth and Properties of Ultrathin Epitaxial Layers

Surfaces

Solid surfaces, interfaces and thin films

Solid surfaces, interfaces and thin films

Solid Surfaces, Interfaces And Thin Films

Solid Surfaces, Interfaces And Thin Films

Chemical Bonding at Surfaces and Interfaces

Chemical Bonding at Surfaces and Interfaces

Chemical bonding at surfaces and interfaces

Chemical bonding at surfaces and interfaces

Chemical Bonding at Surfaces and Interfaces

Chemical Bonding at Surfaces and Interfaces

Reactions at Solid Surfaces (Baker Lecture Series)

Reactions at Solid Surfaces (Baker Lecture Series)

Physics and Chemistry at Oxide Surfaces

Physics and Chemistry at Oxide Surfaces

Physics and chemistry at oxide surfaces

Physics and chemistry at oxide surfaces

Equilibria and Dynamics of Gas Adsorption on Heterogeneous Solid Surfaces (Studies in Surface Science and Catalysis)

Equilibria and Dynamics of Gas Adsorption on Heterogeneous Solid Surfaces (Studies in Surface Science and Catalysis)

Surfaces Algebriques

Surfaces Algebriques

Rough Surfaces

Rough Surfaces

Spillover and Mobility of Species on Solid Surfaces ,Volume 138 (Studies in Surface Science and Catalysis)

Spillover and Mobility of Species on Solid Surfaces ,Volume 138 (Studies in Surface Science and Catalysis)

Equilibria and Dynamics of Gas Adsorption on Heterogeneous Solid Surfaces (Studies in Surface Science and Catalysis)

Equilibria and Dynamics of Gas Adsorption on Heterogeneous Solid Surfaces (Studies in Surface Science and Catalysis)

Riemann surfaces

Riemann surfaces

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close