Handbook of Infra-red Technologies

Handbook of Infrared Detection Technologies

Handbook of Infrared Detection Technologies

This Page Intentionally Left Blank

Handbook of Infrared Detection Technologies

Edited by Mohamed Henini and Manijeh Razeghi

ELSEVIER

UK USA JAPAN

Elsevier Science Ltd, The Boulevard, Langford Lane, Kidlington, Oxford OX 5 1GB, UK Elsevier Science Inc, 360 Park Avenue South, New York. NY 10010-1710, USA Elsevier Science Japan, Tsunashima Building Annex, 3-20-12 Yushima, Bunkyo-ku, Tokyo 113. Japan Copyright ~ 2002 Elsevier Science Ltd. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means: electronic. electrostatic, magnetic tape, mechanical, photocopying, recording or otherwise, without prior permission in writing from the publishers.

British Library Cataloguing in Publication Data Handbook of infrared technologies 1 .Infrared technology I.Henini, Mohamed II.Razeghi, M(Manijeh) technologies 621.3'62

III.Infra-red

ISBN 1 85617 388 7

Library of Congress Cataloging-in-Publication Data Handbook of infrared technologies / edited by Mohamed Henini and Mahijeh Razeghi. p. cm Includes bibliographical references and index. ISBN 1-85617-388-7 (hardcover) 1. Infrared technology-Handbooks, manuals, etc. I. Henini, Mohamed. II. Razeghi, M. TA 1 5 7 0 . H 3 5 8 2002 621.36'2-dc21

2002040767

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. Published by Elsevier Advanced Technology, PO Box 15() Kidlington Oxford OX5 1AS. UK Tel.: +44(0) 1865 8 4 3 0 0 0 Fax: +44(0) 1865 843971 Typeset by Variorum Publishing Limited, Lancaster and Rugby Printed and bound in Great Britain by Biddles Limited, Guildford and King's Lynn

Contents

List of contributors

xiii

Chapter 1

Introduction

Chapter 2

Comparison of photon and thermal detector performance 2.1 Introduction 5 2.2 Fundamental limits to infrared detector performance 6 2.2.1 Photon detectors 9 2.2.2 Thermal detectors 18 2.2.3 Comparison of the fundamental limits of photon and thermal detectors 21 2.3 Focal plane array performance 23 2.4 FPAs of photon detectors 26 2.4.1 InSb photodiodes 30 2.4.2 HgCdTe photodiodes 33 2.4.3 Photoemissive PtSi Schottky-barrier detectors 36 2.4.4 Extrinsic photoconductors 36 2.4.5 GaAs/A1GaAs OWIPs 38 2.4.6 OWIP versus HgCdTe in the LWIR spectral region 39 2.5 Dual-band FPAs 52 2.5.1 Dual-band HgCdTe 53 2.5.2 Dual-band QWIPs 54 2.6 FPAs ofthermal detectors 58 2.6.1 Micromachined silicon bolometers 58 2.6.2 Pyroelectric arrays 61 2.6.3 Thermoelectric arrays 64 2.6.4 Status and trends of uncooled arrays 65 2.7 Conclusions 67 Appendix 70 References 75

vi

Handbookof Infrared Detection Technologies

Chapter 3

GaAs/AIGaAs based quantum well infrared photodetector focal plane arrays

3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9

Chapter 4

Introduction Detectivity D* comparison Effect of nonuniformity 640 • 512 pixel long-wavelength portable OWIP camera 640• 486 long-wavelength dual-band imaging camera 640 • 512 pixel broad-band OWlP imaging camera 640 x 512 spatially separated four-band QWIP focal plane array OWIPs for low background and low temperature operation Summary Acknowledgements References

83 89 91 92 99 102

110 113 115 117 117

GalnAs(P) based QWIPs on GaAs, InP and Si substrates for focal plane arrays 121 4.1 Introduction

4.2

4.3

4.4

4.5

4.6

4.1.1 Overview of infrared detector 4.1.2 Quantum well infrared photodetector 4.1.3 State-of-the-art Fundamentals of OWlP 4.2.1 Intersubband absorption 4.2.2 OWIP parameters 4.2.3 Comparison of n-type and p-type QWIPs 4.2.4 Growth. fabrication and device characterization of a single QWIP device Fabrication of infrared FPA 4.3.1 Infrared FPA fabrication steps 4.3.2 Indium solder bump fabrication steps 4.3.3 ROIC for infrared FPA p-type OWIPs 4.4.1 p-type MWIR QWIPs 4.4.2 p-type LWlR QWIPs n-type OWIPs 4.5.1 n-type LWIR QWIPs 4.5.2 n-type VLWIR OWIPs 4.5.3 Multi-color OWIPs Low cost QWIP FPA integrated with Si substrate 4.6.1 Overview of OWIPs on Si 4.6.2 Growth of GaInAs/InP QWIP-on-Si 4.6.3 Detector performance of GaInAs/InP OWIP-on-Si How to fabricate a monolithic integrated FPA 4.6.4 with Si substrate

121 123 125 126 126 130 133 134 135 135 139 140

141 142 144 144 144 146 147 147 147 148 148 151

Contents

4.7 4.8

New approaches of OWIP Conclusions References

vii

151 153 154

Chapter 5

InAs/(Galn)Sb superlattices: a promising material system for infrared detection 5. ] Introduction 159 5.2 Materials properties 159 5.2.1 BandstructureoflnAs/(GaIn)Sbsuperlattices 159 5.2.2 X-ray characterization 163 5.2.3 Interfaces 165 5.2.4 Sample homogeneity 16 7 5.2.5 Residual doping 16 7 5.3 Superlattice photodiodes 169 5.3.1 Diode structure 169 5.3.2 Diode processing 172 5.3.3 Photo response 172 5.3.4 I-V measurements 173 5.3.5 C-V measurements 181 5.3.6 Noise measurements 183 5.4 Summary and outlook 18 7 References 18 7

Chapter 6

GaSb/InAs supperlattices for infrared FPAs 6.1 Type-II heterostructures 6.1.1 Historical review 6.1.2 Definition oftype-II band alignment 6.1.3 Features oftype-II band alignment and their applications 6.2 Type-II infrared detectors 6.2.1 Principle of operation 6.2.2 Band structure of type-II superlattices 6.2.3 Optical absorption in type-II superlattices 6.2.4 Modeling and simulation of type-II superlattices 6.3 Experimental results from type-II photoconductors 6.3.1 Uncooled type-II photoconductors in the X=8-12 l~m range 6.3.2 Cooled type-II photoconductors for )v > 2013m 6.4 Experimental results from type-II photodiodes 6.4.1 Uncooled type-II photodiodes in the )v=8-12 ~m range 6.4.2 Cooled type-II photodiodes in the )v > 14 l.tm range 6.5 Future work References

191 191 191 192 193 193 193 194 19 7 201 201 208 211 211 215 227 231

viii

Handbook of Infrared Detection Technologies

Chapter 7

Chapter 8

MCT properties, growth methods and characterization 7.1 Preface 7.2 Introduction 7.2.1 Brief history 7.3 MCT Characteristics and material properties 7.3.1 Composition and crystal structure 7.3.2 Bandgap 7.3.3 Intrinsic carrier concentration 7.3.4 Doping and impurities 7.3.5 Carrier mobility 7.3.6 Carrier lifetime 7.3.7 Defects 7.4 MCT crystal gowth methods 7.4.1 Phase diagrams 7.4.2 Bulk growth 7.4.3 Epitaxial growth 7.5 Material characterization methods 7.5.1 Material composition 7.5.2 Measurements of carrier concentration and mobility 7.6 Summary References HgCdTe 2D arrays- technology and performance limits 8.1 Introduction 8.1.1 Historical perspective 8.2 Applications and sensor design 8.3 Comparison of HgCdTe with other 2D array materials 8.4 Multiplexers for HgCdTe 2D arrays 8.4.1 Photocurrent injection techniques 8.4.2 Scanning architectures 8.4.3 Future trends 8.5 Theoretical foundations for HgCdTe array technology 8.5.1 Thermal diffusion currents in HgCdTe 8.5.2 Leakage currents 8.5.3 Photocurrent and quantum efficiency 8.6 Technology ofHgCdTe photovoltaic devices 8.6.1 Materials growth technology 8.6.2 Junction forming techniques in homojunction arrays 8.6.3 Device structures 8.7 Measurements and figures of merit for 2D arrays 8.7.1 NETD - theoretical calculation 8.7.2 NETD- experimental measurement 8.7.3 Relationship of NETD with other figures of merit

233 233 234 235 235 236 238 239 240 241 245 245 246 247 248 256 256 260 260 260

269 270 271 274 276 277 278 279 279 280 281 282 282 284 287 288 292 293 294 294

Contents

8.8 8.9

8.10 8.11

8.12

HgCdTe HgCdTe 8.9.1 8.9.2

2D arrays for the 3-5 lum (MW) band 2D arrays for the 8-12 ~tm (LW) band Array design issues Introduction to performance limitations in LW arrays 8.9.3 Cause of defective elements in HgCdTe 2D arrays HgCdTe 2D arrays for the 1-3 ~m (SW) band Towards GEN III detectors 8.11.1 Two-colour array technology 8.11.2 Higher operating temperature (HOT) device structures 8.11.3 Retina level processing Conclusion and future trends Acknowledgements References

ix

296 296 296 297 298 299 301 301 303 304 304 305 305

Chapter 9. Status of HgCdTe MBE technology 9.1 Introduction 9.2 HgCdTe MBE equipment and process sensors 9.2.1 Vacuum equipment and sources 9.2.2 HgCdTe MBE process sensors 9.3 HgCdTe MBE growth process 9.3.1 Substrate preparation 9.3.2 Growth conditions 9.3.3 Defects 9.3.4 Doping 9.4 Device applications 9.4.1 Multispectral HgCdTe infrared detectors 9.4.2 Near-infrared avalanche photodiodes High-performance MWIR detectors 9.4.3 9.4.4 Large-format arrays on silicon substrates Acknowledgements References

309 310 310 312 318 318 320 320 322 324 324 331 33 7 341 348 348

Chapter 10 Silicon infrared focal plane arrays 10.1 Introduction 10.2 CooledFPAs 10.2.1 Schottky-barrier FPAs 10.2.2 Heterojunction internal photoemission FPAs 10.3 UncooledFPAs 10.3.1 Silicon On Insulator (SOI) diode FPAs 10.3.2 Si-based resistance bolometer FPAs 10.3.3 Thermopile FPAs 10.4 Summary References

353 354 354 3 71 374 374 379 383 386 387

x

Handbookof Infrared Detection Technologies

Chapter 11 Infrared silicon/germanium detectors 11.1 Introduction 11.2. Near Infrared detectors 11.2.1 General operation principle 11.2.2 Detector growth and fabrication 11.2.3 Results and discussion 11.3 Mid-and long-wavelength SiGe IR detectors 11.3.1 Introduction 11.3.2 Principle of operation of HIP detectors 11.3.3 Growth and material characterization 11.3.4 Experimental results and discussion 11.3.5 Calculation of optical properties of SiGe HIP detectors 11.3.6 R6sum6 and outlook for SiGe MWIR detectors Acknowledgements References Chapter 12 PolySiGe uncooled microbolometers for thermal IR detection 12.1 Introduction 12.1.1 Uncooled resistive microbolometers 12.1.2 Microbolometer terminology 12.1.3 Microbolometer process options 12.2 Structural, thermal and electrical properties ofpolySiGe 12.2.1 Deposition of polySiGe 12.2.2 Structural properties 12.2.3 Thermal properties 12.2.4 Electrical properties 12.2.5 High-temperature vs. low-temperature polySiGe 12.3 PolySiGe bolometer pixels 12.3.1 Process development 12.3.2 Absorber comparison and trade-offs 12.3.3 Pixel optimization 12.3.4 Vapor HF processing 12.3.5 Stiffness enhancement techniques 12.4 Readout and system development 12.4.1 Introduction 12.4.2 Readout of polySiGe bolometer arrays 12.5 Zero-level vacuum packaging 12.5.1 Introduction 12.5.2 Indent-Reflow Sealing using metal solder 12.5.3 Zero-levelpackagingusingBCB 12.5.4 Hermeticitytestingusingmicrobolometers 12.6 Conclusions and outlook Acknowledgements References

393 398 398 403 408 411 411 412 413 421 438 444 445 445

449 449 451 452 453 453 454 455 455 457 457 457 459 460 462 463 467 467 468 472 472 473 475 475 476 477 477

Contents

xi

Chapter 13 Fundamentals of spin filtering in ferromagnetic metals with application to spin sensors

13.1 13.2

13.3 13.4

13.5

Index

Introduction Theoretical IMFP variation 13.2.1 The simplest m o d e l - mathematical bases of the calculation 13.2.2 Amore complete treatment 13.2.3 An intuitive derivation 13.2.4 Comparison with the Sch6nhense and Siegmann model Experimental study of A c~ Spin precession and spin filters 13.4.1 Density-operator formalism 13.4.2 Electron transmission through ferromagnetic bilayers 13.4.3 The bilayer with collinear magnetizations 13.4.4 The bilayer with perpendicular magnetizations Discussion and conclusion Acknowledgements References

481 483 483 488 492 495 495 498 499 501 502 503 504 506 506 509

This Page Intentionally Left Blank

List of Contributors

Ian M. BAKER, BAE Systems Infrared Ltd., P.O. Box 217, Southampton, Hampshire SO 15 0EG, UK S. V. BANDARA, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91 ] 09, USA L. BURKLE, Fraunhofer Institut f/ir Angewandte Festk6rperphysik, Tullastrasse 72, D-79108 Freiburg, Germany Henri-Jean DROUHIN, Laboratoire de Physique de la Mati~re Condens~e (UMR 7643-CNRS), Ecole Polytechnique, 9 ] 128 Palaiseau cedex, France F. FUCHS, Fraunhofer Institut fiir Angewandte Festk6rperphysik, Tullastrasse 72, D-79108 Freiburg, Germany S.D. GUNAPALA, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA

M. HENINI, Department of Physics and Astronomy, University of Nottingham, Nottingham, UK Chris van HOOF, IMEC Kapeldreef 75, B-30()1 Haverlee, Belgium and ESATINSYS Department, University of Leuven, Belgium J. E. JENSEN, HRL Laboratories, 3011 Malibu Canyon Road, Malibu, CA 90265. USA

J. JIANG, Center for Quantum Devices, Electrical and Computer Engineering Department, Northwestern University, Evanston, Illinois 60208, USA

xiv Handbook of Infrared Detection Technolo,qies Masafumi KIMATA, Senior Technology Department, Advanced Technology, R&D Center, Mitsubishi Electric Corporation, 8-1-1, Tsukaguchi-Honmachi, Amagasaki, Hyogo 661-8661. Japan Randolph E. LONGSHORE, Raytheon Missile Systems, P.O. Box 1137, MS 8 4 0 / 7 Tuscon, AZ 85734, USA Terry de LYON, HRL Laboratories. 3 O l l Malibu Canyon Road, Malibu, CA 90265, USA H. MOHSENI, Center for Quantum Devices. Electrical and Computer Engineering Department, Northwestern University. Evanston. Illinois 60208. USA Piet de MOOR, IMEC, Kapeldreef 7 5. B- 3()() 1 Heverlee, Belgium Hartmut PRESTING, DaimlerChrysler Research (REM/C), Dep. FT2/H, WilhelmRunge Strasse 11, D-89081 ULM, Germany R. D. RAJAVEL, HRL Laboratories, 3011 Malibu Canyon Road. Malibu, CA 90265, USA

Manijeh RAZEGHI, Center for Quantum Devices. Electrical and Computer Engineering Department, Northwestern University. Evanston. Illinois 6()2()8, USA Antoni ROGALSKI, Institute of Applied Physics, Military University of Technology, 2 Kaliskiego Str.. ()()-908 Warsaw. Poland J. A. ROTH, HRL Laboratories, 3011 Malibu Canyon Road, Malibu, CA 90265. USA

Chapter 1

Introduction M. R a z e g h i a n d M. H e n i n i

Nature has provided numerous examples of efficient detection systems. Almost all types of life, from bacteria, to plants, to h u m a n beings, have evolved some type of optoelectronic detection system for perceiving the world around them. These systems have had millions of years to develop, and demonstrate a seamless integration of optoelectronics with biological systems. The jewel beetle (Melanophila acuminata) thrives on the remnants of forest fires. Its larva feed on the dead wood, which gives evolutionary incentive for the beetle to find dead wood before other species. Towards this end, the beetles have developed an infrared detection system which allows them to sense a 10 hectare forest fire from up to 12 km away. As shown in Figure 1.1, a pit organ, called a sensilla, is located on either side of the beetle's thorax, which allows both intensity and directional information to be obtained. Absorption of infrared (2.4-4 lxm wavelength) light triggers a mechanical expansion which triggers nerve impulses. Obviously, this system must be small and easy to use. Further, as

Figure 1.1 The jewel beetle and its infrared sensor.

2

Handbookof Infrared Detection Technologies

a beetle does not have a large built-in power supply or cryogen, the system must be power efficient and be uncooled. Our eyes are also excellent examples. Nature has provided a multi-spectral detection system based on microscopic variation in detector design. These differentiated detector cells add another dimension to the versatility of the eye. With a broadband detector, there is no way to differentiate between the intensity of a source and its emissivity at different wavelengths. This is akin to trying to pick out a matching wardrobe with a black and white camera. Multispectral systems allow separate waveband analysis of objects, which allows faster and more accurate identification to be made. On an evolutionary perspective, this ability allows more efficient target identification, allowing faster response to a potentially hazardous situation. The goal of science is to enhance our senses and better understand the universe around us. Infrared detectors broaden our vision into the realm of heat, allowing remote sensing of an object's temperature. This has had a dramatic impact on how we perceive our environment, and has led to many types of thermal imaging, including night vision, infrared astronomy, medical diagnostics, and failure analysis. These newfound abilities have spurred the development of m a n y new systems, as shown in Figure 1.2. Infrared detectors have seen a remarkable surge in interest over the past several decades. This is thanks in part to the successful development of highperformance devices which have become the core of all the infrared systems listed above. The natural progression of these systems is a multispectral, uncooled, infrared camera, which can, by itself, address most of these applications. As in nature, a good system should be flexible, power efficient, lightweight, and easy to use. While we cannot expect to match the sophistication of natural systems, we can be inspired by them.

Figure 1.2 Examples of mainstream thermal imaging systems.

Introdl~ction

3

One inspiration involves the exploitation of quantum size effects for higher efficiency and added functionality. Most infrared photon detectors have a limited photocarrier lifetime and peak detection wavelength that is fixed by the bandgap of the material. Without changing the chemical composition of the material, patterning on an atomic scale can allow an increase in carrier lifetime and tuning of the peak detection wavelength. This type of effect has already been demonstrated in the form of the type-II InAs/GaSb semiconductor detector. Used in another way, similar to the eye, microscopic alterations can be made to the lateral size of individual detectors to demonstrate multispectral sensitivity in a single focal plane array. The purpose of this book is to present current methods and future directions in infrared detection. By bringing together experts in physics, material science, fabrication technology, and application, we will develop a well-rounded view of how far we have progressed towards the goal of an integrated, versatile, infrared detection system.

This Page Intentionally Left Blank

Chapter 2

Comparison of photon and thermal detector performance A. Rogalski

2.1 Introduction At present, HgCdTe is the most widely used variable gap semiconductor for infrared (IR) photodetectors. Over the last forty years it has successfully fought off major challenges from extrinsic silicon and lead-tin telluride devices, but despite that it has more competitors today than ever before. These include Schottky barriers on silicon, SiGe heterojunctions, A1GaAs multiple quantum wells, GaInSb strain layer superlattices, high temperature superconductors and especially two types of thermal detectors: pyroelectric detectors and silicon bolometers. It is interesting, however, that none of these competitors can compete in terms of fundamental properties. They may promise to be more manufacturable, but never to provide higher performance or, with the exception of thermal detectors, to operate at higher or even comparable temperatures. The main motivations to replace HgCdTe, are technological problems of this material. One of them is a weak Hg-Te bond, which results in bulk, surface and interface instabilities. Uniformity and yield are still issues. The slow progress in the development of large photovoltaic HgCdTe infrared imaging arrays and the rapid achievements of novel semiconductor heterostructure systems have made it more difficult to predict what types of arrays will be readily available for future systems applications. For spaceborne surveillance systems, low background IR seeker/tracker systems, reliable and affordable sensors with long life are needed which can function effectively at temperatures higher than the 2 0 - 3 0 K currently required by bulk photon detectors. The only alternative to HgCdTe that had been available so far was extrinsic Si, which operates at much lower temperatures where a problematic three-stage cryocooler would be required. Improvement in surveillance sensors and interceptor seekers requires large area size, highly uniform and multicolour (or multispectral) IR focal plane arrays (FPAs) involving long wavelength IR (LWIR) and very long wavelength IR

6 Handbookof Infrared Detection Technologies

(VLWIR) regions. Among the competing technologies are the quantum well infrared p h o t o c o n d u c t o r s (QWIPs) based on lattice matched GaAs/A1GaAs and strained layer InGaAs/A1GaAs material systems. In comparison with photon detectors, thermal detectors have been considerably less exploited in commercial and military systems. The reason for this disparity is that thermal detectors were popularly believed to be rather slow and insensitive in comparison with photon detectors. As a result, the world-wide effort to develop thermal detectors has been extremely small relative to that of photon detectors. In the last ten years, however, it has been shown that extremely good imagery can be obtained from large thermal detector arrays operating uncooled at TV frame rates. The speed of thermal detectors is quite adequate for non-scanned imagers with two-dimensional detectors. At present, uncooled, monolithic FPAs fabricated from thermal detectors, revolutionise the development of low cost thermal imagers. In this paper, we discuss the performance of photon detectors as compared to thermal detectors. In comparative studies, more attention is paid to a wide family of photon detectors, especially to HgCdTe photodiodes and QWIPs. The potential performance of different materials used for photon detectors is examined utilizing the o~/G ratio, where cz is the absorption coefficient and G is the thermal generation. Different types of detectors operated as single element devices, are considered. Also such FPA issues as array size, uniformity, operability, multicolour capability and cost of systems, are discussed.

2.2 Fundamental limits to infrared detector performance Spectral detectivity curves for a number of available IR detectors are shown in Figure 2.1. Interest has centered mainly on the wavelengths of the two atmospheric windows 3-5 ~m [middle w a v e l e n g t h IR (MWIR)] and 8 - 1 4 pm (LWIR region) (atmospheric transmission is the highest in these bands and the emissivity maximum of the objects at T,~3()()K is at the wavelength ;.~10 micron), though in recent years there has been increasing interest in longer wavelengths stimulated by space applications. Depending on the detection mechanism, nature of interaction and material properties, the various types of detectors have their own characteristics. These characteristics result in advantages and disadvantages when the detectors are used in field applications 1-4. Table 2.1 shows a comparison of various IR detectors. Progress in IR detector technology is connected with s e m i c o n d u c t o r IR detectors, which are included in the class of photon detectors. In this class of detectors the radiation is absorbed within the material by interaction with electrons either bound to lattice atoms or to impurity atoms or with free electrons. The observed electrical output signal results from the changed electronic energy distribution. The photon detectors show a selective wavelength dependence of response per unit incident radiation power. They exhibit both perfect signal-to-noise performance and a very fast response. But to

Comparison of photon and thermal detectors performance

7

1012

101,

10'~ o

"D 109

108

1

1.5

2

3

4

5 6 7 8 9 10 Wavelength (lum)

15

20

30

40

Figure 2.1 Comparison of the D* of various infrared detectors when operated at tile indicated temperature. Chopping frequency is 1000 Hz.for all detectors except the thermopile (10 Hz), thermocouple (10 Hz), thermistor bolometer (10 Hz), Golay cell (10 Hz) and p!lroelectric detector (10 Hz). Each detector is assumed to view a hemispherical surround at a temperature of 300 K. Theoretical curves for the backgroundlimited D'for ideal photovoltaic and photoconductive detectors and thermal detectors are also shown.

achieve this, the photon detectors require cryogenic cooling. Photon detectors having long-wavelength limits above about 3 pm are generally cooled. This is necessary to prevent the thermal generation of charge carriers. The thermal transitions compete with the optical ones, making non-cooled devices very noisy. Cooling requirements are the main obstacle to the more widespread use of IR systems based on semiconductor photodetectors, making them bulky, heavy, expensive and inconvenient to use. Depending on the nature of the interaction, the class of photon detectors is further sub-divided into different types as shown in Table 2.1. The most important are: intrinsic detectors, extrinsic detectors, photoemissive detectors (PtSi Schottky barriers), and quantum well detectors. Depending on how the electric or magnetic fields are developed, there are various modes such as photoconductive, photovoltaic, photoelectromagnetic (PEM), and photoemissive ones. Each material system can be used for different modes of operation. In this paper we focus on photodiodes. Photodiodes with their very low power dissipation, easy multiplexing on focal plane silicon chip and less stringent noise requirements for the readout devices and circuits, can be assembled in twodimensional (2D) arrays containing a very large number of elements, limited only by existing technologies. Current cooled IR detector systems use material such as HgCdTe, InSb, PtSi, and doped Si. OWIP is a relatively new technology for IR applications. Among these cooled IR detector systems, PtSi FPAs are highly uniform and

3 Table 2.1

2

Comparison o f infrared detectors

0 ?7

Iletector Type

Advantages

Disadvantages

Thermal (thermopile, bolometers, pyroelectric)

Light, rugged, reliable, a n d low cost Room temperature operation

Low detectivity at high frequency Slow response (ms order)

I'hoton

Easier to prepare More stable materials

Very high thermal expansion coefficient Large permittivity

r.,

Easy bandgap tailoring Well developed theory and cxp. Multicolour detectors

Non-uniformity over large area High cost in growth a n d processing Surface instability

*

(;ood material and dopant.s Advanced technology I'ossiblc monolithic inlcgriition

Heteroepitaxy with large I;itticr mis~iiatcli I.ong wavelength cut-offlimited to 7 p m ( a t 77 K )

Very long wavelength operation Kelativcly simple technology

lligh thermal generation 1:xtremely low temperature operation

I,ow-cost, high yields 1.arge a n d close packed 21) arrays

I.ow q u a n t u m el'liciency [,ow temperature operation

'I'ypc I ((;aAs/Al(;aAs. In(;aAs/Al(;aAsJ)

Matured material growth (;ood uniformity over large area Multicolour detectors

High thermal generation Complicated design and growth

'l'y pe I I (InAs/ln(;aSh. InAs/lnAsShJ

1,ow Auger recombination rate Iksy wavelength control

Complicated design and growth Sensitive to the interfaces

Intrinsic

2

IV-VI (I'bs. I'bSc. I'bSn'l'e)

'5 2 2, u -,

-. 9 ? .

Y

5

0 9. s,

Ill-V (In(;:iAs. InAs. IliSb. InAsShJ

1:rec carriers (I'tSi. l't2Si, IrSi) ()uantum wells

- --

~- --

~

-

-~

~

~

~

-~-.-ppp-p

Comparison of photon and thermal detectors performance 9

manufacturable, but have very low quantum efficiency and can only operate in the MWIR range. The InSb FPA technology is mature with very high sensitivity, but it also can operate in the MWIR spectral range. Doped Si has a wide spectral range from 0.8 to 30 pm and it can only operate at very low temperatures. PtSi, InSb, and doped Si detectors do not have wavelength tunability or multicolour capabilities. Both QWIPs and HgCdTe offer high sensitivity with wavelength flexibility in the MWIR, LWIR and VLWIR regions, as well as multicolour capabilities. HgCdTe can also operate in the short wavelength IR (SWIR) region, while QWIP has to go to a direct band-to-band scheme for SWIR operation. The second class of IR detectors is composed of thermal detectors. In a thermal detector the incident radiation is absorbed to change the temperature of the material, and the resultant change in some physical property is used to generate an electrical output. The detector is suspended on lags, which are connected to the heat sink. The signal does not depend upon the photonic nature of the incident radiation. Thus, thermal effects are generally wavelength independent; the signal depends upon the radiant power (or its rate of change) but not upon its spectral content. This assumes that the mechanism responsible for the absorption of the radiation is itself wavelength independent, which is not strictly true in most instances. Attention is directed toward three approaches which have found the greatest utility in infrared technology; namely, bolometers, pyroelectric and thermoelectric effects. In pyroelectric detectors a change in the internal electrical polarization is measured, whereas in the case of thermistor bolometers a change in the electrical resistance is measured. In contrast to photon detectors, the thermal detectors are typically operated at room temperature. They are usually characterized by modest sensitivity and slow response (because heating and cooling of a detector element is a relatively slow process), but they are cheap and easy to use. They have found widespread use in low cost applications, which do not require high performance and speed. Being unselective, they are frequently used in IR spectrometers. Uncooled FPAs fabricated currently from thermal detectors revolutionize the development of thermal imagers s'6. 2.2.1 Photon detectors

The photodetector is a slab of homogeneous semiconductor with the actual 'electrical' area, A e, that is coupled to a beam of infrared radiation by its optical area, Ao (Figure 2.2). Usually, the optical and electrical areas of the device are the same or close. The use of optical concentrators can increase the Ao/Ae ratio. The current responsivity of the photodetector is determined by the quantum efficiency, r/, and by the photoelectric gain, g. The q u a n t u m efficiency value describes how well the detector is coupled to the radiation to be detected. It is usually defined as the number of electron-hole pairs generated per incident photon. The idea of photoconductive gain, g, was put forth by Rose 7 as a simplifying concept for the understanding of photoconductive phenomena and is now widely used in the field. The photoelectric gain is the number of carriers passing contacts per one generated pair. This value shows how well the

10 Handbook of Infrared Detection Technologies

Figure 2.2 Model of a photodetector. generated electron-hole pairs are used to generate the current response of a photodetector. Both values are assumed here as constant over the volume of the device. The spectral c u r r e n t responsivity is equal to (1)

R i - -~ccq g

where 2 is the wavelength, h is Planck's constant, c is the light velocity, and q is the electron charge. Assuming that the current gains for p h o t o c u r r e n t and noise c u r r e n t are the same, the current noise due to generation and recombination processes is 7 12 - 2(G + R)AetAfq2g 2

(2)

where G and R are the generation and recombination rates, Af is the frequency band and t is the thickness of the detector. Detectivity D* is the main p a r a m e t e r characterizing normalized signal-tonoise performance of detectors and can be defined as D* --

R , ( A o A f ) 1/2 I,

(3)

According to eqs ( 1 )-(3)s D* -

-

2(A~ 1/2 hc Aee ~[2(G+ R)t] -1/2

(4)

Comparison of photon and thermal detectorsperformance 11 For a given wavelength and operating temperature, the highest performance can be obtained by maximizing q/[t(G+R)] 1/2 which corresponds to the condition of the highest ratio of the sheet optical generation to the square root of sheet thermal generation-recombination. This means that high q u a n t u m efficiency must be obtained with a thin device. For a given wavelength and operating temperature, the performance can be optimized by reducing the total number of generation and recombination acts, which is (G + R)(Aet). In further considerations we put that Ao/Ae = 1. Assuming a single pass of the radiation and negligible frontside and backside reflection coefficients, the quantum efficiency and detectivity are r/= 1 - exp(-c~t)

(5)

D* - hcc "~ (1 - e-at) [2(G + R)t] - 1/2

(6)

where a is the absorption coefficient. The highest detectivity can be obtained for t = l . 2 6 / a for which (1-eat)t -1/2 achieves a maximum value of 0.62a 1/2. This thickness is the best compromise between the requirements of high quantum efficiency and low thermal generation. In this optimum case 17= O. 716 and detectivity is equal 8

D* - 0.4S

2

) 1/2 ot C+R

(7)

To achieve a high performance, the thermal generation must be suppressed to possibly the lowest level. This is usually done by cryogenic cooling of the detector. For practical purposes, the ideal situation occurs when the thermal generation is reduced below the optical generation. At equilibrium, the generation and recombination rates are equal, and we have

d. (or) 1/2

D* - 0.31~c ~

(8)

Considerations carried out in ref. 8 indicate that, for a double pass of radiation, the detectivity of an optimized photodetector, leads to the following expression

~ (~)1/2 D* -- 0"31~cck

(9)

where 1 ~
12

Handbook of Infrared Detection Technologies

10~

~10 4

300 K 77K "'-..,

--.

10-' l t

E

410~

" ' C ....

PbSe

r

1101"~

.g_

102 &

~1.

InAs~!Te~ L i I

101~

1

I

2

I',

3

I:

4

:

5

i ~// t Hgc,,Cdo,-Te 1

6

!

7

8

J

HgI Cd [ e

9

10 11 12 13 ,14,03t

Wavelength (lxm)

Figure 2.3 Absorption coef~cientfor various photodetector materials in spectral range 1-14 lzm.

Figure 2.3 shows the measured intrinsic absorption coefficients for various narrow gap photodetector materials. The absorption coefficient and corresponding penetration depth vary among the semiconductor materials. It is well known that the absorption curve for direct transitions between parabolic bands at photon energy greater than energy gap, Eq, obeys a square-root law ~(hlJ) - ~ ( h l ) -

Eg) 1/2

(ao)

where/~ is a constant. As can be readily seen in Figure 2.3, in the MWIR spectral region, the absorption edge value changes between 2 x l ( ) 3 c m -1 and 3 x 103 cm- 1. in the LWIR region it is about 10 ~cm- 1. Since cz is a strong function of the wavelength, for a given semiconductor the wavelength range in which an appreciable photocurrent can be generated, is limited. Near the material's band gap, there is tremendous variation causing a third order of magnitude variation in absorption. In the region of the material's maximum usable wavelength, the absorption efficiency drops dramatically. For wavelengths longer than the cut-off wavelength, the values of a are too small to give appreciable absorption. Figure 2.4 shows the infrared absorption spectra for different n-doped, 50 period GaAs/Al• OWIP structures measured at room temperature using a 45 ~ multipass waveguide geometry. The spectra of the bound-to-bound continuum (B-C) QWIP (samples A, B, and C) are much broader than the boundto-bound (B-B) (sample E) or bound-to-quasibound (B-QB) OWIP (sample F). Correspondingly, the value of the absorption coefficient for the B-C OWIP is significantly lower than that of the B-B OWIP, due to conservation of oscillator strength. The values of the absorption coefficient at 77 K, peak wavelength ).p, cut-off wavelength 2c (long wavelength for which c~drops to half C~p),and spectral

Comparison of photon and thermal detectors performance

700

1 /~1

/

E600

I

/B ~

I

r

g

1600

T = 300 K 1200E

Z" 500 c._o E- 400 .~

'

13

g 800

300

g

o 200

400

100

6

8

10 12 Wavelen9th (gin)

14

16

Figure 2.4 Absorption coefficient spectra measured at T= JO0 K for different QWIP samples described in Table 2.2 (after ref. 9).

width A2 (full width at half etp) are given in Table 2.2. It appears that the lowtemperature absorption coefficient ~ p ( 7 7 K)~l.3C~p(300 K) and % ( A 2 / ) . ) / N D is a constant (ND is the well's doping). 9 A typical value of the absorption coefficient in 7 7 K in the LWIR region is between 600 and 800 cm -]. Comparing the last two figures (Figures 2.3 and 2.4) we can notice that the absorption coefficients for direct band-to-band absorption is higher than that for intersub-band transitions. The ot/Gth ratio versus temperature for different types of tunable materials with hypothetical energy gap equal 0.25 eV (2 - 5 ~tm) and 0 . 1 2 4 eV (2 - 10 ~m) is shown in Figures 2.5 and 2.6. Procedures used in calculations of Gth for different materials are given in Appendix A1-A4. It is apparent that the HgCdTe is by far the most efficient detector of IR radiation" it is characterized by high absorption coefficient and relatively low thermal generation rate. We can also notice that OWIP is better material than extrinsic silicon. The above two figures are completed by Figure 2.7, where the ol/Gth ratio of dependence on wavelength is presented for different materials at 77 K. An optimized photodetector should consist of ] (): 9 9

9

A lightly doped active (base) region, which acts as an absorber of IR radiation. Its band gap E,, doping and geometry should be selected. Electric contacts to the base region, which sense optically generated charge carriers and which should not contribute to the dark current of the device. Surfaces of the absorber regions which must be insulated from the ambient by a material, which also does not contribute to the generation of carriers; in addition the carriers, which are optically generated in the absorber,

Table 2.2 Structure parameters for different n-doped. 5 0 period AI,Gal .As QWlP structures (after ref. 9) Samplc

A

I3 C' I< I:

Wcll width

I3arricr width

C'ornposition

(A)

(A)

Y

Iloping density (lOiXcrn '1

Intersub-hand transition

i,, (pm)

40 40 (10 50 45

500 500 500 500 500

0.Lh 0.25 0.15 0.2h 0. 30

1.O 1.6 0.5 0.42 0.5

13-CT 13-C' 13-cT 13-13 13-013

9.0 9.7 5 8.6

- -1.1

I.,

(1.1rn) 10. 3 10.9 14.5 9.0 8.1 5

Ai (~un)

A ('Kt)

u,,( 7 7 K ) (crn ' )

3.0 2.9

33 30 1 (1 9 11

410 hi0 450 1x20 Xi5

2.1 0.75 0.85

Comparison of photon and thermal detectors performance

100

~

10 "-

~

Xc= 5 lam ~ .

d~ 10-8E

r

C0

_

15

\\

.

HgCdTe

.

QWlP

.

Si

\ ".. ".

0_12

".

\\

10-~6..

10_~o_ 1 0 .24

".....

........

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

-- _.

......

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

100 150 Temperature (K)

50

200

Figure 2.5 o~/Gthratio versus temperature for MWIR (,;.r 5/lm) photon detectors.

Xc =

0-11_

-

.

-

.

-

.

-

.

.

101am HgCdTe QWlP

~- 10 '~E o 10 '~-

"'..

~ 9

""'"".........

~ ~ ~ ~

..~ ...~ .~....~

1 0 -23

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

1 0 -27

50

100 '

15 0

Temperature (K)

200 '

Figure 2.6 c~/Gth ratio versus temperature for L WIR ( ~.~.=10 Ilm ) photon detectors.

should be kept away from surfaces where recombination can reduce the quantum efficiency. A backside mirror for double pass of IR radiation.

16

Handbook of Infrared Detection Technologies 10 ~ T-77K 10 3

10 -3

oo

E 0 CO

HgCdTe

QWIP

\

10 -~

"". \ \ '"\i \ \

10 -~5

10

9 ',nS0 .9.

-2~

~ \ \ 9. ~ '-. "~

.~. 9-....

"-~---..... ~..

..............................-.:

1 0 -27

0

5

10 15 Wavelength (~um)

20

25

Figure 2.7 ot/Gth ratio versus cut-off u,aveh'ngth for different t!lpes of photon detectors operated at 77 K.

The above conditions can be fulfilled using heterojunctions like N+-p-p + and P + - n - n + with heavily doped contact regions (symbol '+' denotes strong doping, capital l e t t e r - wider gap). Homojunction devices (like n-p, n+-p, p+-n) suffer from surface problems: excess thermal generation results in increased dark current and recombination, which reduces the photocurrent. To achieve a high performance, the thermal generation must be suppressed to possibly the lowest level. This is usually done by cryogenic cooling of the detector. For practical purposes, the ideal situation occurs when the thermal generation is reduced below the optical generation. The requirements for the thermal generation rate can be highly reduced in heterodyne systems, in which optical excitation by the local oscillator can dominate the generation, even for high thermal generation. The total generation rate is a sum of the optical and thermal generation G - Gth + Got,

( 1 O)

The optical generation may be due to the signal or background radiation. For infrared detectors, usually background radiation is higher than the signal radiation. If the thermal generation is reduced much below the background level, the performance of the device is determined by the background radiation (BLIP conditions for Background Limited Infrared Photodetector). This condition can be described as qOPB T t

> nrl,

(11)

Comparison of photon and thermal detectors performance

17

where nth is the density of thermal carriers at temperature T, and r is the carrier lifetime, and ~R is the total background photon flux density (unit cm -2 s -1) reaching the detector. Rearranging we have, for the BLIP requirements, 17r B

G~

t

=

>

tlth

r

-

G.,

(12)

i.e., the photon generation rate per unit volume needs to be greater t h a n the thermal generation rate per unit volume. The carriers can be either majority or minority in nature. The direct bandgap semiconductor photodiode is a minority carrier device and in thermal equilibrium t/rain

--

tlmaj

(13)

where ni is the intrinsic carrier concentration, and n,,,,,j is the majority carrier concentration. The ultimate limit on carrier lifetime in a direct gap semiconductor is given by band-to-band recombination, by either radiative or Auger processes. Humphreys indicated 11 that the van Roosbroeck and Shockley theory of radiative recombination underestimates the radiative lifetime due to noiseless photon re-absorption. As a result, Auger recombination is the dominant process in narrow gap semiconductors like, e.g., HgCdTe ternary alloy. The extrinsic semiconductor photoconductor is strictly a majority carrier device. The background limited detectivity, or so-called 'photovoltaic' BLIP detectivity, is given by 12'13

DRuP = hc

(14)

D'BLIP for photoconductors is v ~ times lower than for photodiodes. This is attributable to the recombination process in photoconductors, which is uncorrelated with the generation process, which contributes to the detector noise. The background photon flux density received by the detector depends on its angular field of view (FOV) of the background and on its ability to respond to the wavelengths contained in this source. Plots of D*~LIp as a function of wavelength for TR/.ip= 300 K and for full 2x FOV are shown in Figure 2.1 BLIP temperature is defined when the device is operating at a temperature at which the dark current is equal to the background photocurrent, given an FOV, and a background temperature. In Figure 2.8. plots of the calculated temperature required for background limited (BLIP) operation in 30 ~ FOV are shown as a function of cut-off wavelength. We can see that the operating temperature of 'bulk' intrinsic IR detectors (HgCdTe and PbSnTe) is higher than for other types of photon detectors. HgCdTe detectors with background limited performance operate with thermoelectric coolers in the MWIR range. However, the LWIR detectors (8~<)., ~< 12 lam) operate at ~1()()K. HgCdTe photodiodes

18

Handbook of Infrared Detection Technologies

"~\

-

FOV = 30 ~ Scene temperature = 300 K

",~..\ ~ p-on-n HgCdTe Auger limited photodiodes ',,~..\.~ N~= 5x10 TM cm~, t = 10 ~m ',',,-.k--

200

I

\

",,~..~ ~ n*-on-p HgCdTe Auger limited photodiodes ',,~,.",~j/... N. = 5x10 '5 cm 3, t= 10 lum

\

\\

,,z' v

I-

._1 m

"*.X,. ~HgCdTe Auger limited photoconductors x .. X 14 -3 " ' , ~ ~ ....... N , = 3 x l 0 cm , t = 1 0 p m "', "".":b<.~,.. n'-on-p PbSnTe Auger limited photodiodes ', "'~~' N~ = 10" cm3, t = 10 pm ",,, "--f. ......

100 Schott

Q = 10 ~~ [__L

00

i

i

]

L__

5

L

.

~

L

10 Cutoff wavelength (lum)

. . . . . . . .

15

1

......

20

Figure 2.8 Estimation of the temperature required for background limited operation of different types of photon detectors. In the calculations FOV= 3() ~ and TB- 300 K are assumed (after ref. 14).

exhibit higher operating temperatures compared to extrinsic detectors, silicide Schottky barriers and QWIPs. However, the cooling requirements for QWIPs with cut-off wavelengths below 10 l.tm are less stringent in comparison with extrinsic detectors and Schottky barrier devices. 2.2.2 Thermal detectors Thermal detectors operate on a simple principle, that when heated by incoming IR radiation their temperature increases and the temperature changes are measured by any temperature-dependent mechanism, such as thermoelectric voltage, resistance, pyroelectric voltage. The simplest representation of the thermal detector is shown in Figure 2.9. The detector is represented by a thermal capacitance Cth coupled via a thermal conductance Gth to a heat sink at a constant temperature T. In the absence of a radiation input, the average temperature of the detector will also be T, although it will exhibit a fluctuation about this value. When a radiation input is received by the detector, the rise in temperature is found by solving the heat balance equation. Assuming the radiant power to be a periodic function, the change in temperature of any thermal detector due to incident radiative flux is 12.1 s AT--

(15)

Comparison of photon and thermal detectors performance

19

Figure 2.9 Thermal detector mounted via lags to heat sink.

Equation (15) illustrates several features of the thermal detector. Clearly it is advantageous to make AT as large as possible. To do this, the thermal capacity of the detector (Cth) and its thermal coupling to its surroundings (Gth) must be as small as possible. The interaction of the thermal detector with the incident radiation should be optimized while reducing as far as possible all other thermal contacts with its surroundings. This means that a small detector mass and fine connecting wires to the heat sink are desirable. A characteristic thermal response time for the detector can therefore be defined as Cth 75th -- ~Gth -- CthRth

(1 6)

where R t h - - 1 / G t h is the thermal resistance. The value of the thermal time constant is in the millisecond range. This is m u c h longer than the typical time of a photon detector. There is a trade-off between sensitivity, AT, and frequency response. If one wants a high sensitivity, then a low frequency response is forced upon the detector. For further discussion we introduce the coefficient K=AV/AT, which reflects how good the temperature change translates into the electrical output voltage of the detector. The voltage responsivity Rv of the detector is the ratio of the output signal voltage A V to the input radiation power and is given by

20

Handbookof InfraredDetectionTechnologies KsRth

R,, =

(17)

In order to determine the detectivity of the detector, it is necessary to define a noise mechanism. One major noise is Johnson noise. Two other f u n d a m e n t a l noise sources are i m p o r t a n t for assessing the ultimate performance of a detector: thermal fluctuation noise and b a c k g r o u n d fluctuation noise. Thermal fluctuation noise arises from t e m p e r a t u r e fluctuations in the detector. These fluctuations are caused by heat c o n d u c t a n c e variations between the detector and the s u r r o u n d i n g substrate with which the detector element is in thermal contact. The spectral noise voltage due to t e m p e r a t u r e fluctuations is ~2.1 s

4kT 2 A f K2

V~, = 1 4- ,02 r~,

(18)

Rt,,

A third noise source is b a c k g r o u n d noise which results from radiative heat exchange between the detector at t e m p e r a t u r e Td and the s u r r o u n d i n g e n v i r o n m e n t at t e m p e r a t u r e Tl,, that is being observed. It is the ultimate limit of a detector's performance capability and is given for a 2 rr FOV by 12.1 s

V~,

-

8ksc~A(T2 + T~,) K2 , 1 + ofir~,

R~-,,

(19)

where c~ is the Stefan-Boltzmann constant. The f u n d a m e n t a l limit to the sensitivity of any thermal detector is set by t e m p e r a t u r e fluctuation noise, i.e., r a n d o m fluctuations in the t e m p e r a t u r e of the detector element due to fluctuations in the radiant power exchange between the detector and its surroundings. Under this condition at low frequencies (to < < 1/rth) w e have

Dt,, -

4kTdGt,,,j

(20)

It is assumed here that E is independent of wavelength, so that the spectral D*; and blackbody D*(T) values are identical. If radiant power exchange is the d o m i n a n t heat exchange mechanism, then G is the first derivative with respect to temperature of the Stefan-Boltzmann function. In that case, k n o w n as the background fluctuation noise limit, we have

, Db--

I.

8 8ka(T dg -4- Tb~)

] 1,/2

(21)

Comparison of photon and thermal detectors performance

21

Note that D*b is independent of A, as is to be expected. Equations (20) and (21) assume that background radiation falls upon the detector from all directions when the detector and background temperature are equal, and from the forward hemisphere only when the detector is at cryogenic temperatures. The highest possible D* to be expected for a thermal detector operated at room temperature and viewing a backgrounds at room temperature is 1 . 9 8 • l~ cmHz 1/2 W -1. Even if the detector or background, not both, were cooled to absolute zero, the detectivity would improve only by the square-root of two. This is the basic limitation of all thermal detectors. The background-noiselimited photon detectors have higher detectivities as a result of their limited spectral responses. The performance achieved by any real detector will be inferior to that predicted by equation (21 ). The degradation of performance will arise from: 9 encapsulation of detector (reflection and absorption losses at the window), 9 the effects of excess thermal conductance (influence of electrical contacts, conduction through the supports, influence of any gas-conduction and convection), 9 the additionalnoise sources. Typical values of detectivities of thermal detectors at 10 Hz change in the range between ]08 and 109 cmHz ~/2 W -~. 2.2.3 Comparison of the fundamental limits of photon and thermal detectors

The temperature dependence of the fundamental limits of D* of photon and thermal detectors for different levels of background are shown in Figures 2.10 and 2.11. From Figure 2.10 the results show that in the LWIR spectral range, the performance of intrinsic IR detectors (HgCdTe photodiodes) is higher than for other types of photon detectors. HgCdTe photodiodes with background limited performance operate at temperatures below about 80 K. HgCdTe is characterized by a high optical absorption coefficient and quantum efficiency and a relatively low thermal generation rate compared to extrinsic detectors and OWIPs. The extrinsic photon detectors require more cooling than intrinsic photon detectors having the same long wavelength limit. The theoretical detectivity value for the thermal detectors is much less temperature dependent than for the photon detectors. At temperatures below 50 K and zero background, LWIR thermal detectors are characterized by D* values lower than those of LWIR photon detectors. However, at temperatures above 60 K, the limits favour the thermal detectors. At room temperature, the performance of thermal detectors is much better than LWIR photon detectors. The above relations are modified by the influence of background, shown in Figure 2.10 for a background of 1017 photons/cm2s. It is interesting to notice that the theoretical curves of D* for photon and thermal detectors show similar fundamental limits at low temperatures.

22

Handbook of Infrared Detection Technologies

1014~-\\ i

,_~

QB-"

#

.

-

~,\~

-OLO-

\

101'Ph/cm~s

-'----,,

/nermal

~

HgCdTe

~

-1- 1 E

-

10 8 ~-

"~xtrinsic

_

-

14 pm

;Lc =

10 6

,

10

1

30

,

~ 1

.

.

_ .

.

50 70 90 Temperature (K)

110

130

Figure 2.10 Theoretical performance limits of LWIR photon and thermal detectors at zero background and background of 1017 photons/cm2s, as afimction of detector temperature (after ref. 16).

, o1,ii,',,'-..'-., , "7

10,~_ .

",,",, .

.

.

QB- 10'~Ph/cm2s Q~ - 0 ,"-- ......

Thermal

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

otodiode

~N 10 ~~ "1" E

~

10 ~ ~.~trinsic

10 ~

lo

Xc = 28 p m

3o

~;o

~

zo

~;o

Temperature (K)

1~o

l ao

Figure 2.11 Theoretical performance limits of VL WIR photon and thermal detectors at zero a background and background of 10 lz photons/cm2s, as afunction of detector temperature (after ref. 16).

Comparison of photon and thermal detectors performance 2 3

Figure 2.12 Thermal imager system configuration.

Similar experiments have been carried out for very long wavelength infrared (VLWIR) detectors operated in the 1 4 - 5 0 p m spectral range. Results of calculations are presented in Figure 2.11. Detectors operating within this range are cryogenic Si and Ge extrinsic photoconductors and cryogenic thermal detectors, usually bolometers. Nevertheless, theoretical prediction for intrinsic detectors (HgCdTe photodiodes) is also included. Figure 2.11 shows that the theoretical performance limit of a VLWIR thermal detector at zero and high backgrounds in a wide range of temperatures, equals or exceeds that of photon detectors. The comparison of both types of detector indicates that theoretical performance limits for thermal detectors are more favourable as the wavelength of operation moves from the LWIR to the VLWIR. It is due to the influence of a fundamentally different type of noise (generation-recombination noise in photon detectors and temperature fluctuation noise in thermal detectors), these two classes of detectors have different dependencies of detectivities on wavelength and temperature. The photon detectors are favoured at long wavelength infrared and lower operating temperatures. The thermal detectors are favoured at very long wavelength spectral range. The temperature requirements to attain background fluctuation noise performance, in general favour thermal detectors at the higher cryogenic temperatures and photon detectors at the lower cryogenic temperatures.

2.3 Focal plane array performance For FPAs the relevant figure of merit is the noise equivalent temperature difference (NEDT), the temperature change of a scene required to produce a signal equal to the rms noise. The configuration of the basic thermal-imager system is shown in Figure 2.12. The spectral photon incidence for a full hemispheric surround is

24 Handbookof Infrared Detection Technologies

O-

r(;t)O(T, 2)d2

(22)

).

if a zero-emissivity bandpass filter having in-band transmission, r(2), cut-on wavelength kl, and cut-off wavelength k2 is used (zero emissivity is practically obtained by cooling the spectral filter to a temperature where its self-radiation is negligible). The photon flux density incident on the detector focal plane arrays is

OB__=

1

] + 4(f/#) 2

0__

(23)

where f / # is the ratio of the focal length to the diameter of the limiting aperture or lens. Under these conditions the background-induced photocurrent in any photon detector of area Ad is:

Iph

1 + 4(f/#) 2 ;. r(2)0(2)0(r, 2)d2

(24)

where rl()~) is a spectral response per photon (quantum efficiency) and z(k) is an optics (filter) transmission spectrum. The thermal contrast is one of the important parameters for IR imaging devices. It is the ratio of the derivative of spectral photon incidence to the spectral photon incidence

c

=

~176176 o

(2s)

Figure 2.13 is a plot of C for several MWIR sub-bands and the 8 - 1 2 pm LWIR spectral band. We can notice, that the contrast in the MWIR bands at 300 K is 3.5-4 percent compared to 1.6 percent for the LWIR band. The noise equivalent difference temperature of a detector represents the temperature change, for incident radiation, that gives an output signal equal to the rms noise level. While normally thought of as a system parameter, detector NEDT and system NEDT are the same except for system losses. NEDT is defined as NEDT -

Vn(3T/3O) AT (3Vs/3O) = V,, A Vs

(26)

where Vn is the rms noise and AV~ is the signal measured for the temperature difference AT. It can be shown that 17 N E D T - (rC~n1,,pV~w)-'

(27)

Comparison of photon and thermal detectors performance

25

0 . 0 6 0 .. _

...... 9

.9. . . . . . . . ..

0.050

0.040

cO

-. ,,,",.,

.....

3.5-4.1

- .... 3.5-5.0 ........

---~

.........

i

... 9

pm

pm

4 . 5 - 5 . 0 lam 8.0-12.0 pm

.. . . . . . . . . . . . . . . ".iiiiiii

E 0.030 r !--

0.020

0.010 250

~

l 270

l

l L l I 290 310 S c e n e t e m p e r a t u r e (K)

1 330

I

350

Figure 2.13 Spectral photon contrast in the MWIR and L WIR (after ref. 17).

where Nw is the number of photogenerated carriers integrated for one integration time, tint:

Nw -

rlAcltintQB

(2

The percentage of BLIP, FPA noise

rlst,e

rlBLIp,is simply the

~Nvhoto,,+NFi,A2) 9

2

r

o

w

_

_

8)

ratio of photon noise to composite

(29)

It results from the above formulas that the charge handling capacity of the readout, the integration time linked to the flame time, and the dark current of the sensitive material, become the major issues ofIR FPAs. The NEDT is inversely proportional to the square-root of the integrated charge and therefore the greater the charge, the higher the performance. The distinction between the integration time, and the FPA flame time must be noted. At high backgrounds it is often impossible to handle the large amount of carriers generated over flame time compatible with standard video rates. Off-FPA flame integration can be used to attain a level of sensor sensitivity that is commensurate with the detector-limited D* and not the charge-handlinglimited D*. It is of interest to compare the performance of uncooled photon and thermal detectors in the MWIR (2= 5 lam) and LWIR (k= 10 pm) spectral range. The paper recently published by Kinch as follows us in this comparison. Figure 2.14 compares theoretical NEDT of detectors operated at 290 K for f/1 optics and a

26 Handbook of Infrared Detection Technologies I mil pixel size. Parameters typical for micromachined resonant cavity bolometers given in Appendix A5 are assumed in calculations. As a photon detector, N+-1r-P + HgCdTe photodiode is chosen, first proposed by Elliott and Ashley. 19 7r designates an intrinsic region containing a p-type background dopant equal to 5x 1014 cm -3 with carrier lifetime limited by the Auger 7 process. It is also assumed that the detector node capacity can store the integrated charge due to detector dark current. Figure 2.14 shows that the ultimate performance of the uncooled HgCdTe photon detectors is far superior to the thermal detectors at wide flame rates and spectral bands. Also, any other tunable bandgap alloy, such as type-II InAs/ GaInSb superlattices, could be worthwhile with regard to the development of an uncooled photon detector technology. 1'2~ The ultimate performance of HgCdTe photodiodes with optimally doped base region are comparable with that of InAs/ InGaSb strain layer superlattices in the temperature range between 300 K and 77K. 21 Comparing both curves of Figure 2.14 for thermal detectors, we see that for long integration times in the LWIR region, excellent performance is achieved, with NEDT values below 10 mK for frame rates of 30 Hz. However, for snapshot systems with integration time below 2 ms, the available NEDT is above 100 mK even at the LWIR region. For the MWIR band the thermal detector has obvious performance limitations at any frame rate.

2.4 FPAs of photon detectors There are a number of architectures used in the development of IR FPAs. In general, they may be classified as hybrid and monolithic. The central design questions involve performance advantages versus ultimate producibility. Each application may favour a different approach depending on the technical requirements, projected costs, and schedule. In the monolithic approach, some of the multiplexing is done in the detector material itself rather than in an external readout circuit. The basic element of a monolithic array is a metal-insulator-semiconductor (MIS) structure as shown in Figure 2.15(c). An MIS capacitor detects and integrates the IR-generated photocurrent. Although efforts have been made to develop monolithic FPAs using narrow-gap semiconductors, silicon based FPA technology with Schottkybarrier detectors is the only technology, which has matured to a level of practical use. Microbolometer FPAs are also based on silicon technology (see Figure 2.15(d)). Hybrid FPAs detectors and multiplexers are fabricated on different substrates and mated with each other by the flip-chip bonding or loophole interconnection (Figure 2.16). In this case we can optimize the detector material and multiplexer independently. Other advantages of the hybrid FPAs are; near 100% fill factor and increased signal-processing area on the multiplexer chip. In the flip-chip bonding, the detector array is typically connected by contacts via indium bumps to the silicon multiplex pads. The detector array can be illuminated from either

l O 2 ~(a)

102p-

LWlR 300 K

-3

lo4

'

l

L

"

Integration time (s)

lo1' t

I

MWlR 300 K

lo'/

" " i lI

t

10'

8

1

,

'

'

L

'

I

,

' ~ " ' ' ' ~

n

~

"

"

'

~

l

10' Io2 Integration time (s)

lo3

'

I

I C " '

1

1:igrirr 2 . 14 Tllrorctic~crlNE:'L)Ti.o~tlpirrisc,,~ o$lir~cool~d tlrernlul crtlrl HgCd'l'r ~ir~cool~~ilpl~oton 1,WIK ( u )citld MWlK ( 1 1 ) il(~1i~c~tor.s (riftiv' rqf. 18).

28 Handbookof Infrared Detection Technologies

Figure 2.15 Monolithic IR FPAs: (a) all-silicon: (b) heteroepitaxy-on-silicon: (c) non-silicon (e.g.. HgCdTe CCD): (d) microbolometer.

the frontside or backside (with photons passing through the transparent detector array substrate). In general, the latter approach is most advantageous. In HgCdTe hybrid FPAs, photovoltaic detectors are formed on thin HgCdTe epitaxial layer on transparent CdTe or CdZnTe substrates. For HgCdTe flip-chip hybrid technology, the maximum chip size is of the order of 10 mm square. In order to overcome this problem, the technology is being developed with saphire or silicon as the substrate of HgCdTe detectors. When using opaque materials, substrates must be thinned to 1 0 - 2 0 ~tm in order to obtain sufficient quantum efficiencies and reduce the crosstalk. There is a large research activity directed towards 2D staring arrays detectors consisting of more than 106 elements. IR FPAs have nominally the same growth rate as dynamic random access memory (RAM) integrated circuits (ICs) (it is a consequence of Moore's Law, which predicts the ability to double the transistor integration on each IC about every 18 months) but lag behind in size by about

Comparison of photon and thermal detectors performance 29

Figure 2.16 Hybrid IR FPA with independently optimized signal detection and readout: (a) indium bump technique; (b) loophole technique.

5 - - 1 0 years. ROIC's are somewhat analogous to dynamic R A M m o n l y readouts require a m i n i m u m of three transistors per pixel compared to one per memory cell. Consequently, whereas various 6 4 x 6 4 FPAs were available in the early 1980s, several vendors are now producing monolithic FPAs in the TVcompatible 1 0 4 0 x 1040 formats. Figure 2.17 illustrates the trend of array size over the past 25 years and some projections of w h a t will evolve in the coming decade. Rockwell has developed the world's largest HgCdTe SWIR FPA for astronomy and low background applications. The format of the device is a hybrid 2 0 4 8 x 2 0 4 8 with a unit cell size of 1 8 ~ t m x l 8 ~ t m . Table 2.3 contains a description of representative IR FPAs that are commercially available as standard products and/or catalogue items from the major manufacturers. Twenty years ago, high quality single element detectors were often priced over $ 2 0 0 0 , but now, some current IR FPA production costs are less than $0.1 per detector and even greater reductions are e x p e c t e d in the near future. As the commercial market for uncooled imagers expands, the cost of commercial systems will inevitably decrease. At present, the cost of 3 2 0 x 2 4 0 , 50• 50 ~tm2, bolometer arrays for thermal imagers is $15 0 0 0 - 2 0 000. Two types of silicon addressing circuits have been developed: CCDs (CMOS) switches. The and complementary m e t a l - o x i d e - s e m i c o n d u c t o r photogenerated carriers are first integrated in the well, formed by a photogate, and subsequently transferred to slow (vertical) and fast (horizontal) CCD shift registers.

30

Handbook of Infrared Detection Technologies

101~

4 megapixel infrared

109- focal plane arrays .-Q" c- 10" O t._ 10~ Cu oo 106 .-=,_ .Q 0

~

before 2000

_.~

-

I1~ e s i c c D

~

-

DRAM 1 ~ " / PtSi.......[] .....~ K .production/~ ....[] ......... ~ 2x512 ~A480x640

10 5

0 x

il. 10'

_ ......~ r ~

103 4 102

K ,

.~..-'OHiOdTe

~ J

S

QwlP

.........

_

'''Cr''''" ~

L

72 74 z6 78 80 82

,

~

,

88 90 92 94 96 98 2000 &

Year

~

o4 06 68 2010

Figure 2.17 Increase in array format size over the past 2 ~ years and projections for the coming decade. PtSi, InSb, and HgCdTe have been following the pace of d!lnamic RAM. offset by about a decade. O WIP detectors have been recently reported in sizes as large as 640• 480 pixels (after ref. 22).

An attractive alternative to the CCD readout is coordinative addressing with CMOS switches. The advantages of CMOS are that existing foundries, which fabricate application specific integrated circuits, can be readily used by adapting their design rules. Design rules of 0.18 pm are in production with pre-production runs of the 0.12 lum design rules. At present, CMOS with m i n i m u m feature ~<0.5 l.tm is also enabling monolithic visible CMOS imagers.

2.4.7 InSb photodiodes InSb material is more mature t h a n HgCdTe and good quality bulk substrates of larger t h a n 7 cm diameter are commercially available. 23 InSb photodiodes have been available since the late fifties and they are generally fabricated by impurity diffusion (usually Cd) and ion implantation (usually Be + or B+). Epitaxy is not used; instead, the standard m a n u f a c t u r i n g technique begins with bulk n-type single crystal wafers with donor concentration about 101 ~ cm -3. Wimmers et al. have presented the status of InSb photodiode technology for a wide variety of linear and F P A s . 24'2 s A typical InSb photodiode RA product at 77 K is 2 x 106 ~ c m 2 at zero bias and 5 • facm 2 at slight reverse biases of approximately 1 0 0 m V . This characteristic is beneficial when the detector is used in the capacitive discharge mode. As element size decreases below 10 -4 cm -2, some slight degradation in resistance due to surface leakage occurs. InSb photodiodes can also be operated in the temperature range above 77 K. Of course, the RA products degrade in this region. At 120K, R A products of 104 Eacm2 are still achieved with slight reverse bias, making BLIP operation

Table 2.3 Representative IR FPAs offered by some major manufactures Manufacturerjweb site

Pixel size (~m)

Detector material InSb InSb %:As BIB HgCdTe HgCdTe VO,(bolometer) I'yro (BST) HgCd'l'e HgCdTe HgCd'l'e HgCdTe HgCd'l'e VO,(bolorneter)

BAE Systems

HgCdTe HgCdl'e HgCd'l'e HgCdrl'e

Spectral range (~m)

Oper. temp. (K)

( c ~ H z ~ ~ ~ ~ w ) NETD (mK)

W

Table 2.3 (continued) ManufacturerIWeb site

N

Pixel size ( ~ m )

Detector material Amorphous Si(bolometer)

1 IgCd'l'e HgC'd'l'c l'yro ( I'S'I') l'yro ( I'S'l7)

Sensors llnlimited/ www.sensorsinc.com H -- hybrid. M -monolithic

Spectral range ( ~ m )

Oper. temp. (K)

Comparison of photon and thermal detectorsperformance 3 3

possible. The quantum efficiency in InSb photodiodes optimized for this temperature range remains unaffected up to 160 K. InSb photovoltaic detectors are widely used for tactical applications, groundbased infrared astronomy and for applications aboard the Space Infrared Telescope Facility. An array size of 1 0 2 4 • is possible (see Table 2.3) because the InSb detector material is thinned to less than 10 ~tm (after surface passivation and hybridization to a readout chip) which allows it to accommodate the InSb/silicon thermal mismatch. 26 Linear array formats of 64, 128 and 256 elements are also produced with frontside-illuminated detectors for both highbackground and astronomy applications. Element sizes depend on device format and range from 20 x 2 0 - 2 0 0 • 200 Bm. The cryogenically cooled InSb and HgCdTe arrays have comparable array size and pixel yield at MWIR spectral band. However, wavelength tunability and high quantum efficiency have made HgCdTe the preferred material. 2.4.2 HgCdTe photodiodes

During the past four decades, HgCdTe has become the most important semiconductor for the middle and long wavelength (k=3-30~tm) IR photodetectors. The short wavelength region has been dominated by III-V compounds (InGaAs, InAsSb, InGaSb). Spectral cut-off can be tailored by adjusting the HgCdTe alloy composition. Epitaxial techniques are the preferred technique to obtain device-quality HgCdTe material for IR devices. Epitaxial growth of the HgCdTe detector array on a Si substrate, rather than CdZnTe, has emerged as a particularly promising approach to scale up wafer dimensions and achieve cost-effective production. 27 MBE offers unique capabilities in material and device engineering including the lowest growth temperature, superlattice growth and potential for the most sophisticated composition and doping profiles. The growth temperature is less than 200~ for MBE but around 350~ for MOCVD, making it more difficult to control the p-type doping in the MOCVD due to the formation of Hg vacancies. Different HgCdTe photodiode architectures have been fabricated that are compatible with backside and frontside illuminated hybrid FPA technology. 3'28 The fabrication of HgCdTe photodiodes was usually based on the most common n+-p and P+-n double layer heterojunction (DLHJ) structure. In these photodiodes the base p-type layers (or n-type layers) are sandwiched between CdZnTe substrate and high-doped (in n+-p structure) or wider-gap (in P+-n structure) regions. Due to backside illumination (through CdZnTe substrate) and internal electric fields (which are 'blocking' for minority carriers), the influence of surface recombinations on the photodiodes performance is eliminated. Both optical and thermal generations are suppressed in the n+-region due to the Burstein-Moss effect and in the P+-region due to wide gap. The influence of surface recombination is also prevented by the use of suitable passivation. Passivation of HgCdTe has been achieved by several techniques and a comprehensive review was given by Nemirovsky and Bahir. 29 Recently,

36 Handbook of Infrared Detection Technologies Photovoltaic HgCdTe FPAs are based on both p-type and n-type materials. Linear ( 2 4 0 , 2 8 8 , 4 8 0 , and 960 elements), 2D scanning arrays with time delay and integration (TDI), and 2D staring formats from 64 • up to 2048 • have been made. 18 Pixel sizes ranging from 18 ~m square to over 1 mm have been demonstrated. The best results have been obtained using hybrid architecture. 2.4.3 Photoemissive PtSi Schottky-barrier detectors

The most popular Schottky-barrier detector is the PtSi detector which can be used for the detection in the 3-512m spectral range. 3~.~4 Radiation is transmitted through the p-type silicon and is absorbed in the metal PtSi (not in the semiconductor), producing hot holes which are then emitted over the potential barrier into the silicon, leaving the silicide charged negatively. Negative charge of silicide is transferred to a CCD by the direct charge injection method. The effective q u a n t u m efficiency in the 3-_S ~m atmospheric window is very low, of the order of 1%, but useful sensitivity is obtained by means of near full flame integration in area arrays. The q u a n t u m efficiency has been improved by thinning PtSi film and implementation of an 'optical cavity'. Due to very low q u a n t u m efficiency, the operating temperature of Schottky barrier photoemissive detectors is lower than another types of IR photon detectors (see Figure 2.8). Schottky photoemission is independent of such factors as semiconductor doping, minority carrier lifetime, and alloy composition, and, as a result of this, has spatial uniformity characteristics that are far superior to those of other detector technologies. Uniformity is only limited by the geometric definition of the detectors. Most of the reported Schottky-barrier FPAs have the interline transfer CCD architecture. Using a 1.2 txm charge sweep device technology, a large fill factor of 71% was achieved with a 2 6 • 2 pixel in the 5 1 2 • monolithic structure. 34 The NETD was estimated as 0.()33 K with f/1.2 optics at 300 K. The 1040• element CSD FPA has the smallest pixel size (17• 17 l.tm2) among two-dimensional IR FPAs. Current PtSi Schottky barrier FPAs are mainly manufactured in 150 mm wafer process lines with around 1 l~m lithography technologies; the most advanced Si technology offers 200 mm wafers process with 0.25 ~tm design rules. However, the performance of monolithic PtSi Schottky-barrier FPAs has reached a plateau, and slow progress is expected from now on. 2.4.4 Extrinsic photoconductors

Extrinsic photoresistors are used in a wide range of the IR spectrum extending from a few~m to approximately 300pro. They are the principal detectors operating in the range ;k > 20 l~m. 3s Detectors based on silicon and germanium have found the widest application as compared with extrinsic photodetectors on other materials. Si has several advantages over Ge: for example, three orders of

Comparison of photon and thermal detectorsperformance 3 3

possible. The quantum efficiency in InSb photodiodes optimized for this temperature range remains unaffected up to 160 K. InSb photovoltaic detectors are widely used for tactical applications, groundbased infrared astronomy and for applications aboard the Space Infrared Telescope Facility. An array size of 1 0 2 4 • is possible (see Table 2.3) because the InSb detector material is thinned to less than 10 ~tm (after surface passivation and hybridization to a readout chip) which allows it to accommodate the InSb/silicon thermal mismatch. 26 Linear array formats of 64, 128 and 256 elements are also produced with frontside-illuminated detectors for both highbackground and astronomy applications. Element sizes depend on device format and range from 2 0 x 2 0-2 O0 x 2 O0 ~m. The cryogenically cooled InSb and HgCdTe arrays have comparable array size and pixel yield at MWIR spectral band. However, w a v e l e n g t h tunability and high quantum efficiency have made HgCdTe the preferred material. 2.4.2 HgCdTe photodiodes

During the past four decades, HgCdTe has become the most important semiconductor for the middle and long wavelength (~.=3-3()~m) IR photodetectors. The short wavelength region has been dominated by III-V compounds (InGaAs, InAsSb, InGaSb). Spectral cut-off can be tailored by adjusting the HgCdTe alloy composition. Epitaxial techniques are the preferred technique to obtain device-quality HgCdTe material for IR devices. Epitaxial growth of the HgCdTe detector array on a Si substrate, rather than CdZnTe, has emerged as a particularly promising approach to scale up wafer dimensions and achieve cost-effective production. 27 MBE offers unique capabilities in material and device engineering including the lowest growth temperature, superlattice growth and potential for the most sophisticated composition and doping profiles. The growth temperature is less than 200~ for MBE but around 350~ for MOCVD, making it more difficult to control the p-type doping in the MOCVD due to the formation of Hg vacancies. Different HgCdTe photodiode architectures have been fabricated that are compatible with backside and frontside illuminated hybrid FPA technology. 3'2s The fabrication of HgCdTe photodiodes was usually based on the most common n+-p and P+-n double layer heterojunction (DLHJ) structure. In these photodiodes the base p-type layers (or n-type layers) are sandwiched between CdZnTe substrate and high-doped (in n+-p structure) or wider-gap (in P+-n structure) regions. Due to backside illumination (through CdZnTe substrate) and internal electric fields (which are 'blocking' for minority carriers), the influence of surface recombinations on the photodiodes performance is eliminated. Both optical and thermal generations are suppressed in the n+-region due to the Burstein-Moss effect and in the P+-region due to wide gap. The influence of surface recombination is also prevented by the use of suitable passivation. Passivation of HgCdTe has been achieved by several techniques and a comprehensive review was given by Nemirovsky and Bahir. 29 Recently,

34

Handbook of Infrared Detection Technologies

however, most laboratories have been using CdTe or CdZnTe (deposited by MBE, MOCVD, sputtering and e-beam evaporation) for photodiode passivation. 3o For example, the important element of the DLHJ structure is an In-doped (Na.~.lO is cm -3) HgCdTe absorber layer with a wider bandgap cap layer grown by LPE (MBE or MOCVD), arsenic diffusion (or implantation: N . ~ I O is cm-3), a passivation layer, metal contacts to the diode, and indium bumps for mating to the readout integrated circuit multiplexer. It appears that, for the lowest doping levels achievable in a controllable manner in the base regions of photodiodes (N,,= 5 x l 015 cm- 3 for n+-p structure, and Nd=5• 14cm -3 for P+-n structure), the performance of both types of photodiodes is comparable for a given cut-off wavelength and temperature. 31 1081

\

~\

~ k

~

p-on-n HgCdTe N,=5x 10 TM cm -3 t= 10 ~tm

106

A(35K)

--Q

04

9

9

o

_

Q '~o

(40K) (40K)o

10" " ~ 0

2

rr 1 0 -

~o

N(80K/

o"

(40K)e

"~,

o

50K

~176

+ "

9 0

Zx +

~,

~ x

100

60K

(6OK)

,,.

o

9

o

A

7K

x x

+

o (70K) u .~.

(60oK)

1 0 2L-

8

I

12

]

1

16 kc (pm)

~

1

20

I

24

Figure 2.18 Dependence of the RoA product on the long wavelength cut-off for L WIR p-on-n HgCdTe photodiodes at temperatures <<.77 K. The solid lines are cah'ulated assuming that the performance of photodiodes are due to thermal generation governed by the Auger mechanism in the base n-type region of photodiodes with t= 10 #m and Na = 5 • 1014 cm- ~. The experimental values are taken from different papers.

Comparison of photon and thermal detectors performance

37

magnitude higher impurity solubilities are attainable, hence thinner detectors with better spatial resolution can be fabricated from silicon. Si has lower dielectric constant than Ge, and the related device technology of Si has now been more thoroughly developed, including contacting methods, surface passivation and mature MOS and CCD technologies. Moreover, Si detectors are characterized by superior hardness in nuclear radiation environments. Figure 2.20 illustrates the spectral response for several extrinsic detectors. Although the potential of large extrinsic silicon focal plane arrays for terrestrial applications has been examined, interest has declined in favour of HgCdTe and InSb, with their more convenient operating temperatures. Strong interest in doped silicon continues for space applications, particularly in low background flux and for wavelengths from 13 to 20 pm, where compositional control is difficult for HgCdTe. The shallower impurity energies in germanium allow detectors with spectral response up to beyond 100 ~m wavelength and major interest still exists in extrinsic germanium for wavelengths beyond about 20 Hm. To maximize the quantum efficiency and detectivity of extrinsic photoconductors, the doping level should be as high as possible. This idea is realized in blocked impurity band (BIB) devices. The longer spectral response of the BIB Si:As device compared with the bulk Si:As device (see Figure 2.20) is due to the higher doping level in the former that reduces the binding energy of an electron. BIB devices made from either doped silicon or doped germanium are sensitive in the infrared wavelength range of 2 - 2 2 0 ~m. BIB devices in large staring array formats are now becoming commercially available. Hybrid FPAs with Si:As BIB detectors operating in the 4 - 1 0 K temperature range have been optimized for

10 0

(D

r.f) cO c~ oo

CD "-' 1 (I.)

01

-

>

. ~ . , .

c~ CD

:/

n"

10 .2 0

, Si:ln 5

,

,

Si:Ga '.~

.i

10

15

20

25

Wavelength (pm)

30

Figure 2.20 Examples of extrinsic silicon detector spectral response. Shown are Si:In, Si:Ga, and Si:As bulk detectors and a Si:As BIB (after ref. 36).

38 Handbook of Infrared Detection Technologies low, moderate, and high IR backgrounds. 37 The 2 5 6 • format with 30~tm pixels and 2 4 0 • format with 50 ~tm pixels are available for low- and high background applications, respectively. Antimony-doped silicon (Si:Sb) arrays and 128 • 128-pixel Si:Sb hybrid FPAs having response to wavelengths > 40 ~m have been also demonstrated, primarily for use at low and moderate backgrounds. Germanium BIB devices have been developed on an experimental basis, but they have not been reported in large 2D array formats yet. 2.4.5 GaAs/AIGaAs QWlPs

QWIP technology is based on the well-developed A3BS material system, which has a large industrial base with a number of military and commercial applications. OWIP cannot compete with the HgCdTe photodiode as a single device, especially at higher temperature operation ( > 70 K) due to fundamental limitations associated with intersub-band transitions.3S However, the advantage of HgCdTe is less distinct in the temperature range below 50 K due to problems involved in a HgCdTe material (p-type doping, Shockley-Read recombination, trap-assisted tunnelling, surface and interface instabilities). Even though OWIP is a photoconductor, several of its properties such as high impedance, fast response time, long integration time, and low power consumption, comply well with the requirements of large FPAs fabrication. Due to the high material quality at low temperature, OWIP has potential advantages over HgCdTe for VLWIR FPA applications in terms of the array size, uniformity, yield and cost of the systems. A key factor in QWIP FPA performance is the light-coupling scheme. A distinct feature of QWlPs is that the optical absorption strength is proportional to an incident photon's electric-field polarisation component normal to the q u a n t u m wells. For imaging, it is necessary to be able to couple light uniformly to 2D arrays of these detectors, so a diffraction grating or other similar structure is typically fabricated on one side of the detectors to redirect a normally incident photon into propagation angles more favourable for absorption. The pixels of 2D arrays are thinned to about 5 ~tm in thickness. The thinning traps diffracted light inside the illuminated pixels, increasing responsivity and eliminating crosstalk.

Table 2.4

Essential properties of LWIR HgCdTe p h o t o d i o d e s and QWIPs at T=77 K

Parameter

HgCdTe

QWlP (n-type)

IR absorption

Normal incidence

Eoptic, d _Lplane of well required

Quantum efficiency Spectral sensitivity Optical gain Thermal generation lifetime RoA product (2c=10 ~tm) Detectivity (2c= 10 lam,FOV=O)

>~70% Wide-band 1 ~ 1 ~s 300 f2cm2 2 • 1012 cmHz12 W-1

Narrow-band (FWHM~ 1+2 lain) 0.4 ( 30-50 wells) 10 ps 104 ~ c m 2 2 x 1()1o cmHz1,,_,W- 1

Normal incidence: no absorption <~10%

Comparisonof photon and thermaldetectorsperformance 39

The thinning also allows the detector array to stretch and accommodate the thermal expansion mismatch with the Si ROIC. At present QWIP FPAs are into a mainstream IR technology. The OWIP 2000 workshop on 2 7-29 July 2000 in Dana Point, CA, USA (see Infrared Physics and Technology, Vol. 42, No. 3-5, 2001) covered many team efforts in bringing the technology into the commercial market. 2.4.6 QWIP versus HgCdTe in the LWIR spectral region

Table 2.4 compares the essential properties of HgCdTe and QWIP devices at 77 K. Quantum efficiency HgCdTe has large optical absorption and a wide absorption band irrespective of the light polarization, greatly simplifying the detector array design. Quantum efficiency is routinely produced around 70% without anti-reflection (AR) coating and in excess of 90% with AIR coating, and is spectrally constant from less than 1 l~m out to near the cut-off of the detector. The wide-band spectral sensitivity with near-perfect rl enables greater system collection efficiency (smaller aperture) making the FPA useful for imaging, spectral radiometry, and long-range target acquisition. It should be noticed, however, that the current LWIR staring array performance is mostly limited by the charge handling capacity on the IROIC and the background (warm optics). Due to intersub-band transitions in the conduction band, the n-type QWIP detection mechanism requires photons with non-normal angle of incidence to provide proper polarization for photon absorption. The absorption quantum efficiency is relatively small, about 20% using 2D grating. Since OWIP is a photoconductive detector, the responsivity is proportional to the conversion efficiency, which is the product of the absorption quantum efficiency times the optical gain. The optical gain of OWIP structures is typically 0.4. From this it follows that the r/ is typically below 10% at the maximum response, rapidly rolling of both the short and long wavelength sides of the peak. Figure 2.21 compares the spectral r/of the HgCdTe photodiode with a QWIP. Dark current and RoA product Figure 2.22 shows typical current-voltage characteristics of HgCdTe photodiode at temperatures between 40 and 90K for a 12 l~m cut-off detector at 40K. Leakage current is less than 10 -5 A/cm 2 at 77K. The biases-independent leakage current, aids in achieving FPA uniformity as well as reducing detector bias-control requirements during changes in photocurrent. Usually for Hgl_xCdxTe photodiodes with x~().22, in the zero-bias and lowbias region, diffusion current is the dominant current down to 60K. 3 For medium reverse bias, trap-assisted tunnelling produces the dark current, and also dominates the dark current at zero bias below 50 K. For a high reverse bias, bulk band-to-band tunnelling dominates. At low temperatures, such as 40 K, large spreads in RoA product distributions are typically observed due to onset of tunnelling currents associated with localized defects. Moreover, the HgCdTe

40

Handbook of Infrared Detection Technologies

100 800

c

HgCdTe Xo = 9.2 gm

~ ~1 ~

:3 r

m=3

0

60-

,i

//

X~ = 8.5 g m

/ [FWHM - 1.4 .p...m

/

l

6 ~ c

~r 4 E ~

//

20-

r

-

0 o

2

\1

11

40-

0

8

]QWlP ]

. - - .

9 E

10

,,

P "~ "

4

i- . . . . . . . .

6

"i'-

-

""

"

8 10 Wavelength (#m)

I

12

0

14

Figure 2.21 Ouantum efficiency versus wavelength for a HgCdTe photodiode and GaAs/A1GaAs OWIP detector with similar cut-off.

photodiodes often have additional dark current, particularly at low temperature, which is related to the surface. 3 The average value of RoA product at 77 K for a 10 ~tm cut-off for HgCdTe photodiodes at 77K is around 300 ~ c m 2 and drops to 30 ~ c m 2 at 12 l~m. At 40 K, the RoA product varies between 1()~ and 10 s ~ c m 2 with 90% above 10 ~ ~2cm2 at 11.2 ll,m. 28'3~ In comparison with HgCdTe photodiodes, the behaviour of the dark current of QWIPs is better understood. At low temperatures (T < 40 K for 2,,= 10 ~m), the dark current is mostly caused by defect related direct tunnelling. In the medium operating range between 40 and 70 K (for 2,,= 10 pro), the thermally assisted tunnelling dominates. In this case, electrons are thermally excited and tunnel through the barriers with assistance from the defects and the triangle part of the barrier at high bias. At high temperature (> 70K for kc=lOlam), thermally excited electrons are thermionically emitted and transport above the barriers. It is difficult to block this dark current without sacrificing the photoelectrons (transport mechanisms of thermionically emitted current and photocurrent are similar). Minimizing thermionically emitted current is critical to the commercial success of the QWIP, as it allows the highly desirable high-temperature camera operation. Dropping the first excited state to the top, theoretically causes the dark current to drop by a factor of ~ 6 at a temperature of 70 K. This compares well with the four-fold drop experimentally observed for 9 ~m cut-off QWIPs. 4~ The value of the QWIP dark current could be adjusted using different device structures, doping densities, and bias conditions. Figure 2.23 shows the I-V characteristics for a range of temperature between 3 5 and 77 K measured on a device with 9.6 ~m spectral peak. Typical operation at 2 V applied bias in the

Comparison of photon and thermal detectors performance

10

-~

Operation bias/,,',."

/,',.":

/,,'//

-25mY

r

41

E 10-~

o

90 K 77 K

~~r 1 0-5

i/:

/," /:'

-,,1i ,' v; :

"" "~

I

I

II I, I"

~_1 ~

::3

0-7

..........

'''

o

10 .9 -0.5

......

"\.i

40 K -0.4

i

60 K

"----.--v.-: . . . . . . . . .

-0.3 -0.2 Bias voltage (V)

-0.1

"',,

0.0

0.1

Figure 2.22 Current-voltage characteristics at various temperatures for a 12 I~m cut-off HgCdTe photodiode (after ref. 22 ).

slowly-varying region of current with bias between the initial rise in current at low voltage and the later rise at high bias. Typical LWIR OWIP dark current at 7 7 K is about 10 -4 A/cm 2, which is in the n a n o a m p e r e range for 2 4 x 2 4 ~tm2 pixel. 42 Comparing Figures 2.22 and 2.23 we can see that a 9.6 pm OWIP must be cooled to 60 K to achieve leakage current comparable to a 12 pm HgCdTe photodiode operating at 2 5 degrees warmer. QWIP operates at a bias voltage from 1 to 3 V depending on the structure and periods of the devices. Using the voltage divided by the dark current density, the RoA products are usually larger t h a n 107 fZcm 2 and 104 f2cm 2 w h e n operated at 40 K and 77 K, respectively. 42 These values indicate very high impedance.

Detectivity We can distinguish two types of detector noise" radiation noise and intrinsic noise. Radiation noise includes signal fluctuation noise and background fluctuation noise. For infrared detectors, background fluctuation noise is higher compared to the signal fluctuation noise. Usually for photodiodes, shot noise is the major noise. In the case of QWIPs, the major source of noise is the dark current. Due to high dark current, Johnson noise is neglected in most cases, especially at high temperature operation. But at lower temperature and w h e n the array pixel size is smaller, Johnson noise becomes comparable to dark noise. Owing to stable surface properties, there is very little 1 ff noise observed in OWIPs. At FPA level, the pattern noise (which results from local variation of the dark current, photoresponse, and cut-off wavelength) is the major limitation to the

42

Handbook of Infrared Detection Technologies

1O~I

300 K FOV=2x

~_

Eo _

I

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

....

104

77K

t ~

/

/

,~..-'""

'""-.

.-""' 66 K

oB 10-~t "~

10 -1~

f

-5

I

-4

I

-3

1

-2

1

1

I

-'1 0 1 2 Bias voltage (V)

I

3

1

4

5

Figure 2.23 Current-voltage characteristics of a QWIP detector having a peak response of 9.6 I~m at various temperatures, along with the 300 K background window current measured at 30 K with a 180 ~ FOV (after ref. 41).

array performance, especially at low temperature. This type of noise is a nonuniformity appearing across the array, which does not vary with time and reflects the intrinsic properties of a FPA. The fixed pattern noise is smaller for QWIP arrays than for HgCdTe arrays due to their material quality and better controlled cut-off wavelength. Figure 2.24 compares the detectivities of p-on-n HgCdTe photodiodes with GaAs/A1GaAs OWIPs. The theoretical curves for HgCdTe photodiodes are calculated assuming constant cut-off wavelengths of 10 pm and 11 ~tm. The VLWIR results for HgCdTe (14.8 lam at 80 K and 16.2 pm at 40 K) and the QWIP at 16 pm show the intrinsic superiority of the HgCdTe photodiodes. HgCdTe has roughly an order of magnitude higher detectivity, although the advantage decreases as the temperature is reduced. The best example of where the QWIP could have a performance advantage, is at low temperature. As we see from Figure 2.24, the QWIP at 7.7 Bm peak wavelength offers superior performance relative to a ,~ 10. 6 l~m HgCdTe at temperature ~<45 K. Since 1987, 44 rapid progress has been made in the detectivity of long wavelength QWIPs, starting with bound-to-bound QWIPs, which had relatively poor sensitivity, and culminating in high performance bound-to-quasibound OWIPs with random reflectors. All the OWIP data is clustered between 101~ and 1011 cmHzl/a/W at about 77 K operating temperature. The above performance comparison of QWIPs with HgCdTe in the low temperature range is less profitable for photodiodes in the case of n+-p structures due to the same non-fundamental limitations (contacts, surface, SchockleyRead processes). Including the influence of tunnelling, the comparison of

Comparison of photon and thermal detectors performance

43

detectivity is more advantageous for GaAs/A1GaAs OWIPs in the spectral region below 14 ~m and at temperature below 50 K. ]4 Noise equivalent difference temperature The noise in HgCdTe photodiodes at 77 K is due to two sources; the shot noise from the photocurrent and the Johnson noise from the detector resistance. It can be expressed as

(30) where k is Boltzmann's constant and R is the dynamic resistance of the photodiode. Assuming that the integration time, "tint, is such that readout node capacity kept half full, we have 1

Af

(31)

- - - ,

2tint

and then

10.6 ~m LPE HgCdTe (1990)

1015

~

"

10

'". .. ~'_"-~"

01.

1

.>

1012

O

"-.

-,.

E

v

9 ~""".,

y--v.

TM

T-

a "r'

...... ...... 9 1991 AIGaAs QWlP (qCp=2.3%)

- 4 1 - - Est: 10 ~m MBE HgCdTe --O--- Est: 11 ~m MBE HgCdTe _A 15 p.m QWlP (Gunapala) ~ Est: 15.8 ~m MBE HgCdTe I...................10.7 ~m QWlP (Levine) h. . . . . 19 ~m QWlP (Levine) i: - HgCdTe Theory @ 11 ~m i . . . . . . . . . . HgCdTe Theory @ 10 pm various S-o-A 9-10 pm QWlP's

0 "O

(D a_

I

"

101~ 10 ~

30

", ""

",.x

'-v ~::~ .......

[] ",,

_4'0

(~]

"-

~'50

6()

,~

80

100 1 89

150

Temperature (K) Figure 2.24 LWIR detector detectivity versus temperature for GaAs/AIGaAs QWIPs and p-on-n HgCdre photodiodes (after refs. 9 and 43).

44 Handbook of Infrared Detection Technologies

~/(4_~Ta) In --

1

2qlph + ~

2tint"

(32)

At tactical background levels, the Johnson is much smaller than the shot noise from the photocurrent. In the case where the number of electrons collected in a flame is limited by the capacity of the ROIC charge well, which is often true, the signal-to-noise ratio is given by S

qNw/2r

N

(qNw) l

=~

(33)

2 q \ 2r ,] 2--r Assuming that the temperature derivative of the background flux can be written to a good approximation as O__QO=h__f_cQ aT

(34)

A:T2

and using equation (2 6), the NEDT under these conditions is NEDT -

2kT2X

(35)

hc2x/X~w

In the last two equations ~=(21+}.2)/2 is the average spectral band between 21 and 22. For a typical storage capacity of 2 x 107 electrons, 2 - 1 0 from equation (35) we have an NEDT of 19.8 mK. The same estimations can be made for OWIP. In this case negligible compared to the generation-recombination noise,

/

1

wavelength of the pm, and Tu= 300 K, the Johnson noise is and then (36)

where dark current can be approximated by Ia-Ioexp

( Eo) -~

.

(37)

In the above expressions, g is the photoconductive gain, Id is the dark current, Io is a constant that depends on the transport properties and the doping level, and Ea is the thermal activation energy, which is usually slightly less than the energy corresponding to the cut-off wavelength of the spectral response, g, Ivh, and Io should also be bias-dependent parameters.

Comparison of photon and thermal detectorsperformance 45 The signal-to-noise ratio for a storage capacity-limited QWIP is given by

s

qNw/2__ ~4

(~r~)

1

Nw

(38)

1 , ~ ~~'

and the NEDT is NEDT-

2kT~ i~ ~ h---~ ,

(39)

Comparing equations (35) and (39) we notice that the NEDT value for chargelimited OWIP detectors is better than for HgCdTe photodiodes by a factor of (2g) 1/2 since a reasonable value of g is 0.4. Assuming the same operation conditions as for HgCdTe photodiodes, the NEDT is 17.7 mK. So, a low photoconductive gain actually increases the S/N ratio and a QWIP FPA can have a better NEDT than an HgCdTe FPA with similar storage capacity. This deduction was experimentally confirm by Schneider et al. 4 s Using a photovoltaic 'low-noise' OWIP structure in which g is only 0.05, the group achieved a NEDT of 7.4 mK with 20 ms integration time and 5.2 mK with 40 ms. For a system operating in the LWIR band, the scene contrast is about 2%/K of change-in-scene temperature. Therefore, to obtain a pixel-to-pixel variation in apparent temperature of less than, e.g. 20 mK, the nonuniformity in response must be less than 0.04%. This is nearly impossible to obtain in the uncorrected response of the FPA, so a two-point correction is typically used. The nonuniformity can be different depending on the specification of operability; e.g., a higher requirement on the operability usually leads to a lower uniformity and vice versa. Typical uncorrected response nonuniformity in QWIP FPAs is 1-3% with an operability (the fraction of good pixels) greater than 99.9%. For the 1 2 8 x 1 2 8 151am array fabricated by the Jet Propulsion Laboratory (see Table 2.5), the uncorrected standard deviation is 2.4% and the corrected nonuniformity 0.05%. For recently described large 6 4 0 x 4 8 6 9 pm FPA, the uncorrected noise nonuniformity is about 6%, and after two-point correction improves to an impressive 0.04%. For the same format FPA, demonstrated by Lockheed Martin, an operability of greater than 99.98% was described. 5~ It is very hard for HgCdTe to compete with OWIP for high uniformity and operability with large array format, especially at low temperature and VLWIR. The variation of x across the Hgl_xCdxTe wafer causes a much larger spectral nonuniformity (e.g. at 77 K, a variation of Ax=0.2% gives a A2,,=0.064 Bm at 2c=5 ~tm, but A2~=0.51 ~m at 14 ~m), which cannot be fully corrected by the two or three-point corrections. Therefore, the required composition control is much more stringent for VLWIR than for MWIR. High uniformity and high operability, as shown in the above examples, demonstrate the maturity of GaAs growth and processing technology. In this

46 Handbook of Infrared Detection Technologies context, the n o n u n i f o r m i t y and operability h a v e been an issue for HgCdTe, a l t h o u g h recently published values for Sofradir and SBRC a r r a y s are as high as 99%. W h e n we c o m p a r e the p e r f o r m a n c e of both types of FPAs (see Tables 2.5 and 2.6), the a r r a y operability is h i g h e r for OWIPs, above 99.9%. In the case of OWIPs, e x t e n d i n g cut-off w a v e l e n g t h to VLWIR is relatively easier since there is little c h a n g e in material properties, g r o w t h and processing. However, a serious r e q u i r e m e n t for m a i n t a i n i n g the device p e r f o r m a n c e is to lower the o p e r a t i n g t e m p e r a t u r e . Due to lower q u a n t u m well barriers, the dark c u r r e n t of t h e r m i o n i c emission d o m i n a t e s at a lower t e m p e r a t u r e . In order to achieve e q u i v a l e n t p e r f o r m a n c e of a 1 0 ~ m cut-off OWIP at 7 7 K , the t e m p e r a t u r e needs to be cooled d o w n to 55 K for a 15 ~tm cut-off and 35 K for a 19 ~m cut-off (see Table 2.5 and Figure 2.24). Figure 2.25 c o m p a r e s the p e r f o r m a n c e of two Sofradir's HgCdTe staring arrays, sensitive b e t w e e n 7.7 ~m and 9 ~m, and 7.7 ~m and 9.5 ~m. Higher p e r f o r m a n c e with improved t e c h n o l o g y has been obtained using, on the one h a n d , a reduced dark c u r r e n t detector t e c h n o l o g y and, on the o t h e r hand, n e w r e a d o u t circuit a r c h i t e c t u r e w h i c h maximizes both c h a r g e h a n d l i n g capacity and responsivity. 51 For a c o n s t a n t t e m p e r a t u r e , the p e r f o r m a n c e is h i g h e r for 2c-9 ~m due to the fact t h a t dark c u r r e n t is lower and i n t e g r a t i o n time can be increased. In c o n s e q u e n c e , the focal plane o p e r a t i n g t e m p e r a t u r e of improved arrays with 2c=9 ~m can be increased up to 105 K and to 102 K for a r r a y with 2~=9.5 ~tm (for NEDT < 18 mK, for the given example). The next figure (Figure 2.26) shows the m e a s u r e d and estimated NEDT as a function t e m p e r a t u r e for 8.9-~m OWlP FPA. In c o m p a r i s o n with a r e p r e s e n t a t i v e HgCdTe FPA (Figure 2.2:3), this p a r a m e t e r exhibits strong

Table 2.5

Properties of JPL 9 ~tm and 15 ~tm GaAs/AIGaAs QWIP FPAs (after refs. 4 6 - 4 9 ) .

Parameter

2,,=9 Jam

Array size

256x256 (ref. 46) 38x38 28x28 2D periodic grating 8.5 8.9

320x256 (ref. 47) 3()x 30 28x28 2D periodic grating 8.5 8.9

640x486 (ref. 48) 25x25 18x 18 2D periodic grating 8.3 8.8

128x 128 (ref. 49) 50x 50 38x 38 2-D periodic grating 14.2 14.9

99.98 5.4

99.98

99.9 5.6

>99.9 2.4

0.04

0.05

2.3

3 1.6x 10 l~ (55K) 30 (45K)

Pixel pitch (~m) Pixel size (lxm) Optical coupling Peak wavelength (lxm) Cutoffwavelength, 50% (lain) Operability (%) Uncorrected nonuniformity (%) Corrected uniformity. 17-27~ Quantum efficiency (%) D*(cmHz1/2W- 1) NEDTwith f/2 optics (mK)

2,,= 15 ~m

0.03 6.4 2.0x l 0 ll (7OK) 23 {70 K)

6.9

2 . 0 x 1011

33 (70 K)

(7OK) 36 (70 K)

Comparison of photon and thermal detectors performance 47

temperature dependence. At temperatures < 70K, the signal-to-noise (SNR) ratio of the system is limited by multiplexer readout noise, and shot noise of the photocurrent. At temperatures > 70 K, temporal noise due to the QWIP's higher dark current becomes the limitation. As mentioned earlier, this dark higher current is due to thermionic emission and thus causes the charge storage capacitors of the readout circuitry to saturate. Comparing the values of NEDT parameters for both types of FPA (see also Tables 2.5 and 2.6), we can see that the performance of LWIR HgCdTe arrays is better. The well charge capacity is the m a x i m u m a m o u n t of charge that can be stored on the storage capacitor of each cell. The size of the unit cell is limited to the dimensions of the detector element in the array (of large LWIR HgCdTe hybrid array, a mismatch in the coefficient of thermal expansion between detector array and the readout can force the cell pitch to 20 pm or less to minimize lateral displacement). However, the development of heteroepitaxial growth techniques for HgCdTe on Si has opened up the possibility of cost-effectively producing significant quantities of large-area arrays t h r o u g h utilization of large-diameter Si substrates. For a 3 0 x 3 0 l~m2 pixel size, the storage capacities are limited to 1 to 5 • 107 electrons. For example, for a 5 • 107 electron storage capacity, the total current density of a detector with a 30• 30 lam a pixel size has to be smaller t h a n 2 7 pA/ cm 2 with a 33 ms integration time. 42 If the total current density is in the 1 m A / c m 2 range, the integration time has to be reduced to 1 ms. For the LWIR HgCdTe FPAs the integration time is usually below 1 O0 ps. Since the noise power bandwidth Af= 1/2t~nt, a small integration time causes extra noise in integration. Usually, LWIR OWIP FPAs using conventional ROIC typically operated at 6 0 65 K. Due to a smaller q u a n t u m efficiency of OWIP, filling the charge capacitor is not a problem at high background application. OWIP allows a longer integration Table 2.6

Performance specifications for LWIR HgCdTe FPAs (after SOFRADIR and SBRC

d a t a sheets) Parameter

SOFRADIR

Format Cut on--cut off (gm) FPA temperature (K) Detector pitch (gin) Fill factor (%) Charge handling capacity Frame rate (Hz) D* peak RMS/T~,,t/pitch (average) (cmHzX/aW-1) Pixel NETD (average) NEI (photons/cmas)(max) Typical FOV Fixed pattern noise Crosstalk (optical and electrical) (%) Array operability (%)

128 x 128 7.7-10.3 < 85 50 > 70 > 118 106 eto 300 1.1 • 1 ()11

SBRC 3 2 0 x 2 56 7.7-9.0 < 90 30

256x256 8.5-11.0 77 (upto 100) 30

12 or 36 to 4 0 0

8 x 106 e- (min) to 120

l O m K for 275 Hz

18

f/2 7% RMS 2 99

f/2

1.52x1012

99

48

Handbook of Infrared Detection Technologies

50

j/

9.5 pm / 9 Std MC techno. /

40-_

/

/

E,E

Improved

/

//

-

I-

pm

MCT techno. 9.5

,,If /

/

. / o 9 pm

m 20Z

10-

~ !

i

75

I

I

80

85

,

~'1

I

I

*

90 95 100 FPA temperature (K)

105

110

115

Figure 2.25 NEDT of 1 2 8 x 128 HgCdTe FPA (f/2 optics. ~0% well Jill. pitch ~0 l~m) as a fimction of operating temperatures (after ref. f~1 ).

time, which gives a relatively lower NEDT. However, at higher temperatures, the dark current of OWIP is high and fills the charge capacitor very quickly. The current subtraction and switched capacitor noise filtering capabilities of ROICs permit low NEDT at higher operating temperatures. In this case however, the readout circuit is complicated which limits the size of array. A goal of third-generation imagers is to achieve sensitivity improvement corresponding to NEDT of about 1 mK. From equation (2 7) we see that in a 300 K scene in the LWIR region with thermal contrast of 0.04, the required charge storage capacity is above 109 electrons. This high charge-storage density within the small pixel dimensions is probably not possible with standard CMOS capacitors. Norton et al. 52 have suggested using of stacked hybrid structures as at least an interim solution to incorporate the desired charge storage density in detector-readout-capacitor structures. Cost

The cost of a FPA depends strongly on the maturity of the technology and varies with production quantity in different companies. So far, large size LWIR FPAs are developed in R&D laboratories without mass production experience. According to Sofradir, HgCdTe experience, by continuous effort in the domain of industrialization decreased the cost of HgCdTe detectors by a factor of 5-10. 53 The cost of making high performance cooled components can be broken down into three parts of about equal weight: the chip (detector and ROIC): the dewar;

Comparison of photon and tllermal detectors performance

49

50

40~" 30-

a

o 20UJ Z

=

10

Estimated Experiment

_

0

50

I

55

1

60 65 70 Temperature (K)

75

80

Figure 2.26 NEDT of 3 2 0 • QWIP FPA ().c=8.9 I~m. f/2 optics. 50% well fill, pitch 30 I~m) as a function of operating temperature (after ref. 4 7).

integration and tests, s4 In addition, the user must add the cryogenic machine cost that is not negligible compared to those of the component. Even if the detection circuit is free of charge, the total cost would only be reduced by about ] 5-20%. This explains why the cost of PtSi and QWIP detectors is not markedly less than that of q u a n t u m detectors of the same complexity, even though the raw materials (Si or GaAs) are m u c h less than for HgCdTe. Moreover, since PtSi requires a very wide optical aperture to obtain acceptable performance, and since OWIP requires lower operating temperatures than other photon detectors, a possible reduction in the purchase price is counterbalanced by a significant increase in operating costs. HgCdTe detectors have been the centre of a major industry for the last three decades. The technology is relatively mature at MWIR but it does not fold over to LWIR. To make components with more pixels requires reducing the pitch or mastering the thinning operation needed to withstand the thermal cycling (differential thermal expansion between CdZnTe and silicon). In the future, a more advantageous approach would seem to be the use of Si substrates, which offer m a n y well-known advantages relative to bulk CdZnTe substrates (much larger available size at lower cost, a thermal expansion m a t c h to Si readout chips, higher purity, and compatibility with automated wafer processing/ handling methodology due to their superior mechanical strength and flatness). Promising results have been achieved in the SWIR and MWIR spectral region. During last four years the defect density for MWIR layers of HgCdTe grown by MBE on silicon substrates has decreased from 2 0 0 0 cm -2 to below 500 cm-2. s2 Currently MWIR arrays with pixel operability of 98% can be produced from this material. For comparison, CdZnTe material typically has operability of 99% or

50 Handbookof Infrared Detection Technologies better. Defect densities for LWIR material grown on silicon substrates continue to limit performance, but they have been reduced by an order of magnitude in the past decade. ~2.~~ In comparison with HgCdTe FPAs, the industrial experience in OWIP FPAs is lower and improvements can be expected because this technology is at a lower stage of development. The major challenge is changing the device and grating designs to improve the device performance. Because of the maturity of the GaAs growth technology and stability of the material system, no investment is needed for developing OWIP substrates, MBE growth, and processing technology. 42 Development of LWIR and multicolour HgCdTe detectors are extremely difficult, especially for low background applications. It means a lower cost in OWIP technology development and production compared with HgCdTe. Reliability In our discussion the reliability issue has been omitted due to the fact that statistical data on this subject is not available. In several applications, especially military systems high reliability is required to ensure both the success of the mission and minimal risk to the user. Two reliability challenges affect both FPAs; survival in high temperature system storage environments and withstanding repetitive thermal cycles between ambient and cryogenic temperatures. In HgCdTe as well as in OWIP FPAs the indium bumps are used to hybridize both types of detectors with a silicon multiplexer. However, certain problems can be expected in the case of QWIP arrays, since the indium bumps have many known alloys with III-V compounds. Very large FPAs may exceed the limits of hybrid reliability engineered into current cooled structures. Hybrids currently use mechanical constraints to force the contraction of the two components to closely match each other. This approach may have limits, when the stress reaches a point where the chip fractures. Three approaches offer an opportunity to resolve this issue:

9

to eliminate the thick substrate which limits the detector active region from deforming at the slower rate of the silicon readout, 9 to subdivide the array into a plurality of regions, 9 to use silicon as the substrate for growth of the detector material.

A technology for making HgCdTe photodiodes and QWIPs on silicon a substrate would be the ultimate, simply because of the vast existing silicon technology. However, major issues with this approach are: ~2 9 9 9

less area would be available for readout circuitry, microlens arrays would be required to regain the fill factor, material quality may not be adequate for low-leakage detectors, particularly for LWIR HgCdTe photodiodes, 9 silicon integrated circuits are processed on (100) oriented silicon, but e.g. the preferred orientation for HgCdTe growth on silicon is near the (211/ orientation.

Comparison of photon and thermal detectors performance

51

Less demanding approaches to the elimination of the thick detector substrate is the loophole or high density vertically integrated photodetector device structures already practised by GEC Marconi and DRS, respectively. In this approach, HgCdTe material is glued to the readout and contacts are made through the thin layer (1()-20 ~tm) after the substrate is removed (see Figure 2.16(b)). Another approach is to remove the substrate after hybridization with indium bumps. Substrate removal is standard practice with very large hybrid InSb arrays ( 1 0 2 4 • pixels). This approach has been recently adopted for OWIP arrays. 46'47's6 After epoxy backfilling of the gaps between the array and the readout multiplexer, the substrate is thinned to a very thin membrane ( ~ 1 0 0 0 A). This not only eliminates the thermal mismatch problem between the silicon readout and the GaAs based detector array, but also completely eliminates pixel-to-pixel crosstalk, and finally, significantly enhances an optical coupling of IR radiation into OWIP pixels.

Summary

LWIR QWIP cannot compete with the HgCdTe photodiode as a single device, especially at temperatures above 70 K due to fundamental limitations associated with intersub-band transitions. However, the advantage of HgCdTe is less distinct at temperatures below 50 K. Comparing photovo|taic HgCdTe and OWIP technologies, we arrive at the following conclusions"

9 two major issues that impede the performance of OWIPs should be overcome: optical conversion efficiency and dark current, 9 for HgCdTe, improvement of the array uniformity is necessary, 9 QWIP has more potential to realize VLWIR FPA operation (also with multicolour detection). The main drawbacks of LWIR QWIP FPA technology are the performance limitation for low integration time applications and the low operating temperature. Their main advantages are linked to performance uniformity and to availability of large size arrays. Next, the main drawback of LWIR HgCdTe FPA technology is the unavailability of large size arrays necessary for TV and larger formats. Several properties of OWIP such as high impedance, fast response time, long integration time, and low power consumption, comply well with requirements for fabrication of large FPAs. Due to the high material quality at low temperature, OWIP has potential advantages over HgCdTe for VLWIR FPA applications in terms of the array size, uniformity, yield and cost of the system. Three-band and four-band FPAs will be demonstrated in the near future. State of the art QWIP and HgCdTe FPAs provide similar performance figures of merit, because they are predominantly limited by the readout circuits. The performance is, however, achieved with very different integration times. The very short integration time of LWIR HgCdTe devices of typically below 300 l~s is very useful in order to freeze a scene with rapidly moving objects. OWIP

52

Handbookof Infrared Detection Technolo#ies

devices achieve, due to excellent homogeneity, an even better NEDT, however, the integration time must be 1()-1 ()() times longer for that, and typically 5-20 ms. Choice of the best technology is therefore driven by the specific needs of a system. Observation of the global market over the past several years has indicated that even HgCdTe photodiodes intrinsically exhibit higher performance than the OWIP detectors, the market tendencies for the future are: 9 HgCdTe for small formats (e.g. 128 x128), small pitch, high flame rates and low integration times, 9 OWIP for large formats (e.g. 64()• and larger), low flame rates and large integration time. Despite serious competition from alternative technologies and slower progress than expected, HgCdTe is unlikely to be seriously challenged for highperformance applications and applications requiring multispectral capability and fast response.

2.5 Dual-band FPAs Multicolour capabilities are highly desirable for advanced IR systems. Systems that gather data in separate IR spectral bands can discriminate both absolute temperature and unique signatures of objects in the scene. By providing this new dimension of contrast, multiband detection also enables advanced colour processing algorithms to further improve sensitivity above that of single-colour devices. Multispectral detection permits rapid and efficient understanding of the scene in a variety of ways. In particular, two-colour IR FPAs can be especially beneficial for threat-warning applications. By using two IR wavebands, spurious information, such as background clutter and sunglint, may be subtracted from an IR image, leaving only the objects of interest. Multispectral IR FPAs can also play many important roles in Earth and planetary remote sensing, astronomy, etc. Thus, the effective signal-to-noise ratio of two-colour IR FPAs greatly exceeds that of single-colour IR FPAs for specific applications. Currently, multispectral systems rely on cumbersome imaging techniques that either disperse the optical signal across multiple IR FPAs or use a filter wheel to discriminate spectrally the image focused on single FPA. These systems contain beam-splitters, lenses, and bandpass filters in the optical path in order to focus the images onto separate FPAs, responding to different IR bands. Also, complex alignment is required to map the multispectral image pixel for pixel. Consequently, these approaches are expensive in terms of size, complexity, and cooling requirements. At present, considerable efforts are directed to fabricate a single FPA with multicolour capability to eliminate the spatial alignment and temporal registration problems that exist whenever separate arrays are used, to simplify optical design, and reduce size. weight, and power consumption. Considerable progress has been recently demonstrated by research groups at Hughes Research

Comparison of photon and thermal detectors performance

53

Laboratory (Raytheon), 57'58 Lockheed Martin (BAE Systems), s~ DRS Infrared Technology 60 AIM,61 Rockwell ~2 and Leti ~ 3 in multispectral HgCdTe detectors employing mainly MBE (although LPE and MOCVD are also used) for the growth of a variety of devices. Also QWIP's technology demonstrates considerable progress in fabrication of multicolour FPAs. 4 7. s~.-36.64-66 It is, perhaps, the niche in which OWIPs have an intrinsic advantage due to relative ease of growing multi-band structures by MBE with very low defect density. Devices for the sequential and simultaneous detection of two closely spaced sub-bands in the MWIR and LWIR radiation have been demonstrated. 2.5.1 Dual-band HgCdTe In the back-illuminated dual-band detectors, the photodiode with longer cutof wavelength is grown epitaxially on top of the photodiode with the short cut-off wavelength. The shorter cut-off photodiode acts as a long-wavelength pass filter for the longer cut-off photodiode. Both sequential mode and simultaneous mode detectors are fabricated from multi-layer materials. The simplest two-colour HgCdTe detector, and the first to be demonstrated, is the bias-selectable n - p - n back-to-back photodiode shown in Figure 2.2 7(a). The sequential-mode detector has a single indium bump per unit cell that permits sequential bias-selectivity of the spectral bands associated with operating tandem photodiodes. When the polarity of the bias voltage applied to the bump contact is positive, the top (LW) photodiode is reverse biased and the bottom (SW) photodiode is forward biased. The SW photocurrent is shunted by the low impedance of the forward-biased SW photodiode, and the only photocurrent to emerge in the external circuit is the LW photocurrent. When the bias voltage polarity is reversed, the situation reverses: only SW photocurrent is available. Switching times within the detector can be relatively short, on the order of microseconds, so detection of slowly changing targets or imagers can be done by switching rapidly between the MW and LW modes. One bump contact per unit cell. as for single-colour hybrid FPAs, is the big advantage of the bias-selectable detector. It is compatible with existing silicon readout chips. The problems with the bias selectable device are that its construction does not allow independent selection of the optimum bias voltage for each photodiode, and there can be substantial MW crosstalk in the LW detector. Many applications require true simultaneity of detection in the two spectral bands. This has been achieved in a number of ingenious architectures described by Reine. 67 All these simultaneous dual-band detector architectures require an additional electrical contact to an underlying layer in the multijunction structure of both the SW and LW photodiode. The most important distinction is the requirement of a second readout circuit in each unit cell. Integrated two-colour HgCdTe technology has been developed for nearly a decade with a steady progression having a wide variety of pixel size ( 30-6113m), array formats ( 6 4 • up to 32()• and spectral-band sensitivity (MWIR/ MWIR, MWIR/LWIR and LWIR/LWIR). Figure 2.2 8 shows examples of spectral response from different two-colour devices. Note that there is a minimal crosstalk

54 Handbook of Infrared Detection Technologies

Figure 2.27 Cross-section views of unit cells for various l~ack-illuminated dual-band HgCdTe detector approaches: (a) bias-selectable n-p-n structure reported b!t Ra!ltheon: ~7 (b) simultaneous n-p-n design reported b!t Raytheon: ~ (c) simultaneous l ~ n - n - p reported b!t BAE S!lstems: ~9 (d) simultaneous n - l ~ p - p n design reported by Leti: 6 ~ and (e) simultaneous structure based on p-on-n junctions reported b!l Rockwell 62 (after ref. 6 7).

between the bands, since the short wavelength band absorbs nearly 1 ()()% of the shorter wavelengths. Test structure indicates that the separate photodiodes in a two-colour detector perform exactly as single-colour detectors in terms of achievable R o A product variation with wavelength at a given temperature. 2.5.2 Dual-band QWIPs

Sanders was first to fabricate two-colour. 2 56 • 2 56 bound-to-miniband OWIP FPAs in each of four important combinations: LWIR/LWIR. MWlR/LWlR, near IR (NIR)/LWIR and MWIR/MWIR-with simultaneous integration, s(~'~s

Comparison of photon and thermal detectors performance

55

A device capable of simultaneously detecting two separate wavelengths can be fabricated by vertical stacking of the different OWlP layers during epitaxial growth. Separate bias voltages can be applied to each OWIP simultaneously via the doped contact layers that separate the MOW detector heterostructures. Figure 2.29 shows schematically the structure of a two-colour stacked OWIP with contacts to all three ohmic-contact layers. The device epilayers were grown by MBE on a three-inch semi-insulating GaAs substrate. An undoped GaAs layer, called an isolator, was grown between two A1GaAs stop layers, followed by Au/ Ge ohmic contact of a 0.5 ~m thick doped GaAs layer. Next, the two OWIP heterostructures were grown, separated by another ohmic contact. The long wavelength sensitive stack (red OWIP, ;.,= 11.2 pm) is grown above the shorter wavelength sensitive stack (blue QWIP. ,;.,=8.6 lam). Each OWIP is a 20-period GaAs/AlxGal_xAs MQW stack, in which the thickness of the Si-doped GaAs OWs (with typical electron concentration 5 • 1 ()17 cm-~) and the A1 composition of the undoped AlxGal_• barriers (~ 5 5()-6()() A) is adjusted to yield the desired peak position and spectral width. The gaps between FPA detectors and the readout multiplexer were backfilled with epoxy. The epoxy backfilling provides 1.0 0.8 c 0

Q.

4.5 #m utoff

cutoff

14.9 lam\ ~-om0.8-cutoff. l / 0

0.6

Q.

~ 9 0.4

0.2 0.02

0.6-

> 9 0.4-

~ !

m

n"

cutoff

. . . .

3

J

n, 0.2 4

. . . .

Wavelength (pm)

0.0 ' ~ ' ................ 2 3 4 5 6 7 8 910 11121314

5

Wavelength (gm)

1.0-

o t) c 0

0.8

9

~11.4

pm

0.6

=> 9 0.4 rr 0.2

j

0.0 . . . . . . . . . . . . . . . . .

~

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

2 3 4 5 6 7 8 910 11121314

Wavelength (gm)

Figure 2.28 Spectral response curves for two-colour HgCdTe detectors in various dual-band combinations of MWIR and L WIR spectral bands (after ref. 22 ).

56 Handbook of Infrared Detection Technologies the necessary mechanical strength to the detector array and readout hybrid, prior to the array's thinning process. Most OWIP arrays use 2D grating, which is very wavelength dependent so that the efficiency gets lower when the pixel size gets smaller. Lockheed Martin has used rectangular and rotated rectangular 2D gratings for their two-colour LW/ LW FPAs. Although random reflectors have achieved relatively high quantum efficiencies with a large test device structure, it is not possible to achieve the similar high quantum efficiencies with random reflectors on small FPA pixels due to the reduced width-to-height aspect relations. 47 In addition, it is difficult to fabricate random reflectors for shorter wavelength detectors relative to long wavelength detectors due to the fact that feature sizes of random reflectors are linearly proportional to the peak wavelength of the detectors. The quantum efficiency becomes a more difficult issue for OWIP multicolour FPA than for single colour. The typical operating temperature for QWIP detectors is in the region of 4 0 80K. The bias across each QWIP can be adjusted separately, although it is desirable to apply the same bias to both colours. As shown in Figure 2.30, the responsivity of both OWIPs is around 3()()-35() mA/W. It appears that the complex two-colour processing has not compromised the electrical and optical quality of either colour in the two-colour device since the peak quantum efficiency for each of the 2()-period OWIPs was estimated to be ~1()% in comparison with a normal single-colour OWIP with twice the number of periods which has a quantum efficiency of around 2 ( ) % . A pixel operability for each colour is > 9 7% in comparison with the value of > 99.9% routinely achieved for single-colour OWIPs. The NEDT value was 24 mK for the blue OWIP and 35 mK for the red OWIP. The difference was assigned to the poor transmission properties of the optics in the 11.2 Bm band. An accurate methodology is needed to design

Figure 2.29 Structure o.f two-colour stacked (_)iVIP ( ~(ft('r r(:l: 6 ~ ).

Comparison of photon and thermal detectors performance

57

the detector structure properly to meet different requirements. In the production process, the fabrication of gratings is still quite involved, and its efficiency is rather uncertain in small pixels and in pixels with thick material layers. To cover the MWIR range a strained layer InGaAs/A1GaAs material system is used. InGaAs in the MWIR stack produces a high in-plane compressive strain which enhances the responsivity. 6~.6~ The MWIR/LWIR FPAs fabricated by Sanders consist of an 8.6 ~tm GaAs/A1GaAs OWIP on top of the 4.7 l.tm strained InGaAs/GaAs/A1GaAs heterostructure. Recently, Gunapala et al. 69"7() have demonstrated the first 8 - 9 and 1 4 - 1 5 l.tm two-colour imaging camera based on a 6 4 0 x 486 dual-band OWIP FPA, which can be processed with dual or triple contacts to access the CMOS readout multiplexer. A single indium bump per pixel is usable only in the case of an interlace readout scheme (i.e., odd rows for one colour and even rows for the other colour) which uses an existing single colour CMOS readout multiplexer. However, the disadvantage is that it does not provide a full fill factor for both wavelength bands. The 6 4 0 • GaAs/A1GaAs gave excellent images with 99.7% of the LWIR pixels and 98% of VLWIR pixels working, demonstrating the high yield of GaAs technology. The estimated NEDT of LWIR and VLWIR detectors at 40 K are 36 and 44 mK, respectively. Due to BLIP, the estimated and experimentally obtained NEDT values of the LWIR detectors do not change significantly at temperatures below 65 K. The experimentally measured values of LWIR NEDT equal to 29 mK are lower than the estimated ones. This improvement is attributed to the 2D periodic grating light coupling efficiency. However, the experimental VLWIR NEDT value is higher t h a n the estimated value. It is probably a result of the inefficient light coupling in the 1 4 - 1 5 Bm region, readout multiplexer noise, and

Figure 2.30 Typical responsivity spectra at 40 K and a comnlon bias of 1.5 V. recorded simultaneously for two QWIPs in the same pixel (after ref. 6:5).

58 Handbookof Infrared Detection Technologies the noise of the proximity electronics. At 40 K the performance of both bands' detector pixels are limited by photocurrent noise and readout noise.

2.6 FPAs of thermal detectors IR semiconductor imagers use cryogenic or thermoelectric coolers, complex IR optics, and expensive sensor materials. Typical costs of cryogenically cooled imagers of around $ 5 0 0 0 0 restrict their installation to critical military applications which allow conduction of operations in complete darkness. Very encouraging results have been obtained with micromachined silicon bolometer arrays s'6'72'73 and pyroelectric detector arrays. 6'74-76 Several countries have demonstrated imagers with NEDT below 1 ()() mK, and the cost of simple systems is sometimes below $10 000. It is expected that high-performance imager system costs will be reduced to less than $1()()(), 7~ and above IR cameras will become widely available in the near future. Although developed for military applications, low-cost IR imagers are used in nonmilitary applications such as: drivers aid, aircraft aid, industrial process monitoring, community services, firefighting, portable mine detection, night vision, border surveillance, law enforcement, search and rescue, etc. 2.6.7 Micromachined silicon bolometers

The most popular thermistor material used in fabrication of the micromachined silicon bolometers is vanadium dioxide, V02. From the point of view of IR imaging application, probably the most important property of V02 is its high negative temperature coefficient of resistance (TCR) at ambient temperature, which exceeds 4% per degree for a single element bolometer and about 2% for FPA. There are two reasons for not using VOx (.x"> 2) with substantially higher temperature coefficient: first, reproducibility of properties suffers in the higher x-value films, second, heating becomes a problem with high resisitivity films. The final microbolometer pixel structure is shown in Figure 2.31. The microbolometer consists of a 0.5 l~m thick bridge of Si 3N4 suspended about 2 l~m above the underlying silicon substrate. The use of a vacuum gap of approximately 2.5 ~tm, together with a quarter wave resonant cavity between the bolometer and the underlying substrate, can produce a reflector for wavelengths near 10 ~tm. The bridge is supported by two narrow legs of Si 3N4. The Si3N4 legs provide the thermal isolation between the microbolometer and the heat-sink readout substrate and support conductive films for electrical connection. A bipolar input amplifier is normally required, and this can be obtained with biCMOS processing technology. Encapsulated in the centre of the Si3N4 bridge is a thin layer (500 A) of polycrystalline VOx. Honeywell has licensed this technology to several companies for the development and production of uncooled FPAs for commercial and military systems. At present, the compact 3 2 0 • microbolometer cameras are produced by Raytheon, Boeing, and Lockheed-Martin in the United States. The

Comparison of photon and thermal detectors performance 59

U.S. government allowed these manufactures to sell their devices to foreign countries, but not to divulge manufacturing technologies. In recent years, several countries, including the United Kingdom, Japan. Korea, and France have picked up the ball, determined to develop their own uncooled imaging systems. As a result, although the U.S. has a significant lead, some of the most exciting and promising developments for low-cost uncooled IR systems may come from non-U.S, companies, e.g., microbolometer FPAs with series p-n junction elaborated by Mitsubishi Electric. 77 This approach is unique, based on an allsilicon version of the microbolometer. The 2 4 0 • arrays of 501urn microbolometers are fabricated on industrystandard wafer (4 inch diameter)complete with monolithic readout circuits integrated into underlying silicon. Radford et al. 7~ have reported a 2 4 ( ) x 3 2 0 pixel array with 50 lam square vanadium oxide pixels and thermal time constant of about 40 ms, for which the average NETD (f/1 optics) was 8.6 mK. However, there is a strong system need to reduce the pixel size to achieve several potential benefits. The detection range of m a n y uncooled IR imaging systems is limited by pixel resolution, rather than sensitivity. Therefore, there is a strong system need to reduce the size of the pixels. Because the cost of the optics made of Ge, the standard material, depends approximately upon the square of the diameter, so reducing the pixel size reduces the cost of the optics. These reductions in optics size would have a major benefit in reducing the overall size, weight and cost of manportable IR systems. In addition the reduction in pixel size allows a significantly larger number of FPAs to be fabricated on each wafer. However, the NEDT is inversely proportional to the pixel area, thus, if the pixel size is reduced from 50 • 50 ~tm to 2 5 • 2 5 btm, and everything else remained the same, the NEDT would increase by a factor of four. Improvements in the readout electronics are needed to compensate for this. For future arrays, the f/1 NEDT performance of 25 l.tm pitch microbolometer FPAs is projected to be below 20 mK (see Figure 2.32).79'~() The development of highly sensitive 2 5 ~m microbolometer pixels, however, presents significant challenges in both fabrication process improvements and in pixel design. Microbolometer pixels

Figure 2.31 Bridge structure of Hone!twell microbolometer (after ref. 72).

60 Handbook of Infrared Detection Technologies fabricated with conventional single-level micromachining processes suffer severe performance degradation as the unit cell is reduced below 40 ~m. This problem can be mitigated to some degree if the microbolometer process capability (design rules) is improved dramatically. Table 2.7 summarizes the design and performance parameters for Raytheon's VOx microbolometers. A similar performance has been described by Altman and colleagues at BAE Systems (Lockheed Martin); 82.83 they reported a 6 4 ( ) x 4 8 0 FPA with 2 8 x 2 8 l,tm 2 pixels with NETD (f/1 optics) of below 55 mK. At present, several research programmes are focused towards enhancement of performance level in excess of 109 cmHz 1/2 W -1. It is anticipated that new materials will form the basis of the next generation of semiconductor film bolometers. The most promising material appears to be amorphous silicon (a-Si). The temperature coefficient of resistance values for a-Si, range from - 0 . 0 2 5~ -~ for doped, low resistivity films at room temperature to - 0 . 0 8 ~ - 1 for high resistivity materials. 6 So, although high TC1R values are attainable, they are accompanied by a high level of 1/f noise, s4 Properties of the films depend upon the method of preparation and the type of dopant. Amorphous hydrogenated silicon (a-Si:H) has a metastable state caused by defects arising from prolonged illumination (Staebler and Wronski effect). This is an undesirable feature that requires a specific annealing cycle during preparation (the methodology for reliability enhancement is described in ref. 8 5). If not removed, it adversely affects longterm reliability. Nevertheless, progress in the development of a-Si:H uncooled FPAs has been reported. 320 x 240 arrays with 4 5 lam pitch have been developed with an average NETD of 70 mK (for f/1 aperture and 50Hz imagery

Figure 2.,32 VOw.FPA development (after ref. 79).

Comparison of photon and thermal detectors performance 61 Table 2.7 Performance characteristics of Raytheon's VO~ microbolometers (after refs. 80 and 81)

Performance parameter

Capability (f/1 and 300 K scene)

Array configuration Pixel size (~m 2) Spectral response (~m) Signal responsivity (V/W)

320 x 240 5{}x50 8-14 > 2.5 x 1()' V/W or 50 mK/K......... < 20 < 150 p-p 1.() rms > 40

320 x 240 25x25 8--14 > 2.5xlO'V/W or 20 mK/K...... < 50 < 15() p-p 1.() rms > 1{}{}

or 25 mK/K......

>98 2{}{} 25

>98 15{} 25

>98 390 25

NEDT ~, f/1 {mK) Offset nonuniformity (mV) Output noise (mV) Intrascene dynamic range (~ f/1 (K) Pixel operability (%) Power dissipation (mW) Nominal operating temperature (~

640 x 480 25x25 8--14 > 2.SxlOTV/W

< 15() p-p 0.6 rms > 1{}{}

f r e q u e n c y ) . 85 Using a n i m p r o v e d t e c h n o l o g i c a l stack, fully c o m p a t i b l e w i t h i n d u s t r i a l process, good q u a l i t y 3 2 0 • FPA w i t h 35 ~tm pitch a n d f/1 optics h a s b e e n p r e s e n t e d w i t h NEDT close to 3 5 inK. 86 Because the typical resisitivity of a m o r p h o u s silicon films is several orders of m a g n i t u d e h i g h e r t h a n t h a t of VO• a-Si finds a p p l i c a t i o n in u n c o o l e d a r r a y s in w h i c h bias is c o n t i n u o u s r a t h e r t h a n pulsed w i t h o u t excessive J o u l e a n h e a t i n g of the FPA. ~'

2.6.2 Pyroelectric arrays The i m a g i n g s y s t e m s based on p y r o e l e c t r i c a r r a y s , u s u a l l y n e e d to be o p e r a t e d w i t h optical m o d u l a t o r s , w h i c h c h o p or defocus the i n c o m i n g r a d i a t i o n . This m a y be a n i m p o r t a n t l i m i t a t i o n for m a n y a p p l i c a t i o n s in w h i c h c h o p p e r l e s s o p e r a t i o n is h i g h l y desirable (e.g., guided m u n i t i o n s ) . Most of t h e p y r o e l e c t r i c m a t e r i a l s c o n s i d e r e d for t h e r m a l d e t e c t o r a r r a y s are the l e a d - b a s e d ' p e r o v s k i t e ' oxides s u c h as lead t i t a n a t e [PbTiO3-PT]. These m a t e r i a l s h a v e s t r u c t u r a l similarities w i t h the m i n e r a l p e r o v s k i t e (CaTiO3 }. The basic f o r m u l a is ABO3: w h e r e A is lead, O is o x y g e n a n d B m a y be one, or a m i x t u r e , of c a t i o n s e.g. lead z i r c o n a t e t i t a n a t e [Pb(ZrTi)O~-PZT], b a r i u m s t r o n t i u m t i t a n a t e [BaSrTiO3-BST], lead s c a n d i u m t a n t a l a t e [Pb{Sccj.~Ta{}.~)O3PST] a n d lead m a g n e s i u m n i o b a t e [Pb(Mgl/~Nb2/~)O~-PMN]. Often d o p a n t s are a d d e d to t h e s e basic f o r m u l a t i o n s to e n h a n c e or t u n e the m a t e r i a l properties. A b o v e Curie t e m p e r a t u r e , To, t h e s e m a t e r i a l s form a s y m m e t r i c n o n - p o l a r , cubic s t r u c t u r e . On cooling t h e y u n d e r g o a s t r u c t u r a l p h a s e t r a n s i t i o n to form a polar, ferroelectric phase. The a b o v e m a t e r i a l s c a n be f u r t h e r sub-divided into t w o g r o u p s . The ' c o n v e n t i o n a l ' p y r o e l e c t r i c m a t e r i a l s , s u c h as PT a n d PZT, o p e r a t e at r o o m t e m p e r a t u r e well below their Curie t e m p e r a t u r e w i t h o u t the need for an applied field. R e q u i r e m e n t s for d e t e c t o r t e m p e r a t u r e stabilization is m i n i m a l or c a n be

62 Handbookof Infrared Detection Technologies eliminated since there is little variation in detector performance over quite a large temperature range. It is, however, possible to operate ferroelectrics at or above To, with an applied bias field, in the mode of a 'dielectric bolometer'. This second group of materials (including BST, PST and PMN) have Tc slightly below the detector operating temperature, resulting in minimal pyroelectricity. In this case a constant dc field is applied during operation to induce a polarization and hence regain pyroelectric properties. Barium strontium titanate (BST) ceramic is a relatively well behaved material with a very high permittivity. Texas Instruments (Raython) has improved the performance of pyroelectric FPAs using a bias voltage applied to maintain and optimize the pyroelectric effect near the phase transition. ~7 Figure 2.33 shows details of the completed pyroelectric detector device structure. The 3 2 ( ) x 2 4 0 hybrid arrays with pixels 48.5 IJm are characterized by NEDT as low as 40 mK (array average) and the production average is between 70 mK and 8() mK. A demonstrated sustained production rate in excess of 5()() units per month is a small fraction of factory capacity. For the United Kingdom array programme lead scandium tantalate (PST) material has been chosen, s~ A hybrid pyroelectric/ferroelectric bolometer detector was the first to enter production, and is the most widely used type of thermal detector (in the U.S., the Cadillac Division of General Motors has pioneered this application, selling thermal imagers to the customer for just under $2()()()). 6 Although many applications for this hybrid array technology have been identified, and imagers employing these arrays are in mass production, no hybrid technological advances are foreseen. The reason is that the thermal conductance of the bump bonds is so high that the array NETD (f/1 optics) is limited to about 50 mK. The best NEDT achieved with a hybrid array is about 38 mK, which is consistent with thermal conductance of approximately 4 IJW/K. Pyroelectric array technology, therefore, is moving toward monolithic silicon microstructure technology. The monolithic process should have fewer steps and a shorter cycle time. Most ferroelectrics tend to lose their interesting properties as the thickness is reduced. However, some ferroelectric materials seem to maintain their properties better than others. This seems particularly true for PT and related materials, whereas for BST, the material does not hold its properties well in thin-film form. Thin-film ferroelectric (TFFE) detectors have the performance potential of microbolometers with minimum NEDT of about 1 mK. ~ Figure 2.34 shows NEDT calculated as a function of pixel pitch assuming a constant 30 Hz signal and a thermal time constant of about 15 ms. Reducing the pixel pitch from 50 IJm to 2 5 IJm requires a concomitant increase in thermal isolation by about a factor of four. This seriously impacts the possibility of high performance arrays with small pixels. The TFFE device approach appears remarkably similar to the VOx microbolometer structure developed by Honeywell. However, there are several key futures that distinguish it from that technology. 'ss-~ Since the device is a capacitor rather than a resistor as in a bolometer, the electrodes are located above and below the face of the pixel, are transparent and do not obscure the

Comparison of photon and thermal detectors performance

63

Figure 2. ~ ~ BST dieh'ctric l~olometer pixel ( ~(fter ref. 8 7).

active optical area. Usually, the electrical resistance of the leads can be quite large without degrading the signal-to-noise ratio, since the detector capacitance is approximately 3 pF. This enables the use of thin, poorly conducting electrode materials to minimize thermal conductance. A key feature of the design is that the ferroelectric film is self-supporting" there is no underlying m e m b r a n e necessary to provide mechanical support. In such a way, with the use of transparent oxide electrodes, the ferroelectric material can dominate thermal conductance. It is well known that absorption of IR radiation is accomplished by means of a resonant optical cavity. In a monolithic bridge structure, the cavity is located within the ferroelectric itself or in the space between ferroelectric and the ROIC. This can be realized in two ways. s9 9

9

The bottom electrode must be highly reflective, the top electrode must be semi-transparent, and the ferrolectric must be approximately 1 btm thick for optimal tuning of the cavity for 1 ()-12 btm radiation. Both electrodes must be s e m i - t r a n s p a r e n t , a reflective mirror must be present on the ROIC under each pixel, and the pixel must be located approximately 2 l.tm above the ROIC.

A key factor to performance of the ceramic thin films is the high temperature processing required in achieving the correct ferroelectric crystal phase. The TFFEs of interest are refractory, and require annealing at elevated temperatures to crystallize and develop good pyroelectric properties. Thermal treatments at temperatures that exceed about 45()~ may lead to an adverse interaction between the silicon and a l u m i n i u m interconnections. Various techniques for the deposition of thin ferroelectric films have been investigated including spin-on metal-organic decomposition, radio frequency m a g n e t r o n sputtering, dual ion

64

Handbook of Infrared Detection Technologies

102 ,. ,m9 v/'ek 101 v

E

Is UJ Z

10~

,

I I

101 100

10 ~

102 103 Thermal conductance (nW/K)

10 ~

Figure 2.34 The relationship between thermal isolation and performance of uncooh'd IR detectors in the thermal fluctuation noise limit (~ffter r~:L ,~9/.

beam sputtering, sol-gel processing, and laser ablation. Also a number of surface rapid thermal annealing techniques have been investigated to obtain optimum material response whilst leaving the underlying silicon substrate undamaged. ~{I Monolithic arrays produced to date have demonstrated poor sensitivity. Raytheon's group has demonstrated 32()x 24() array with an NEDT of about 400 mK with operability in excess of 9 5 %. Research group from DERA (UK) has developed 'integrated' and 'composite' detector technology. ~'9~1 In the first technology, the detector material was deposited as a thin film onto free standing micro-bridge structure defined on the surface of the silicon ROIC. The 'composite' technology combines elements of hybrid and integrated technologies (see Figure 2.35). Microbridge pixels are fabricated in a similar fashion to the integrated technology and are next formed onto a high density interconnected silicon wafer. The interconnect wafer uses materials that can withstand the intermediate high temperature processing stage during fabrication of thin ferroelectric films and contains a narrow conducting channel via for every pixel, permitting electrical connection to the underside. Finally, the detector wafer is solder-bump bonded to the ROIC as per the established hybrid array process. It is predicted that, using a PST film, an NEDT of 25 mK (50 Hz image rate and f/1 optics)can be achieved. 2.6.3 Thermoelectric arrays

Thermopile detectors, 91 while only of limited use for imaging applications, have a combination of characteristics that make them well suited for some low power applications. They are highly linear, require no optical chopper, and have D* values comparable to resistive bolometers and pyroelectric detectors. They

Comparison of photon and thermal detectors performance 65

operate over a broad temperature range with little or no temperature stabilization. They have no electrical bias, leading to negligible 1/f noise and no voltage pedestal in their output signal. However, much less effort has been made in their development. The reason is that their responsivity and noise are orders of magnitude less and thus their applications in thermal imaging systems require very low-noise electronics to realize their potential performance. Thermoelectric detectors found almost no use as matrix arrays in TV flame rate imagers. Instead, they are employed as linear arrays that are mechanically scanned to form an image of stationary or nearly stationary objects. The wide operating-temperature range, lack of temperature stabilization, and radiometric accuracy make thermopiles well situated for same space-based scientific imaging applications. Two-dimensional thermopile arrays have been reported only by two groups. 92'93 In both cases, a desire for low cost and manufacturability led to the use of polysilicon thermoelectric materials, which have relatively low thermoelectric figures of merit. Fote and co-workers ~4'9s have improved the performance of thermopile linear arrays by combining Bi-Te and Bi-Sb-Te thermoelectric materials. Compared with most other thermoelelectic arrays, their D* values are highest, as is shown in Figure 2.3 6. This technology has been developed to improve the performance of 2D arrays using a three-level structure with two sacrificial layers. In such way it is possible to improve fill factor and incorporate a large n u m b e r of thermocouples per pixel. Figure 2.37 shows the thermopile detector structure. The structure allows almost 100% fill factor and the model suggests that optimized detectors will have D* values over 109 c m H z l / 2 / W . Further efforts are continued to fabricate high performance 128 • 128 FPAs. 96 2.6.4 Status a n d trends of u n c o o l e d arrays

The information gathered here follows Kruse. ~ Table 2.8 illustrates the status and trends of uncooled arrays for military and commercial applications. As was

Figzlre 2.35 Schematic cross-section of the "composite" detector arra!l design (after reJ. 90).

66

Handbook of Infrared Detection Technologies

previously mentioned, microbolometer arrays containing 64()• pixels are under development. The pixel size is being reduced to 2"3 • 5 ~tm (in order to reduce the optics cost): 35 • 3 "3pm is an intermediate objective. The NEDT goal is 10 mK for high performance applications, instead low cost performance applications (e.g. security sensors) use 16()• 120 pixels, 5()• 50 pm, and NEDT of 100 mK. The commercial systems (microbolometer imagers and radiometers and ferroelectric imagers) derive from military systems that are too costly for widespread use. Imaging radiometers employ linear thermoelectric arrays operating in the snapshot mode: they are less costly than the TV-rate imaging radiometers employing microbolometer arrays. ~7~'~ In the large volume production, the cost of commercial systems will inevitably decrease (see Table 2.8). It seems likely that microbolometer FPAs will become dominant in uncooled detector technology in the near future. They do not require a radiation chopper and have very low thermal conductance which results in a high responsivity and a low NEDT. Moreover, changes in the shape of the lags of the monolithic construction allow trade-off between speed of response and NEDT. It is predicted that the performance of monolithic ferroelectric bolometers can be considerably improved, unlike hybrid pyroelectric detectors where only little development effort is under way. The hybrid pyroelelctric FPAs are not commercially available.

10 ~~

......."

9 Bi-Te/Bi-Sb-Te (this work) 9 Constantan/chromel 9 Silicon

~ .109

-

. .--"" 9 s

0 . 0 "

-

N

"I-

v

E o

V

"O 108

I

7 1 0

0

.

.

.

.

.

.

.

.

.

.

10 ~

.

.

.

.

.

I

10 ~

i

L

l ,

.

1

0 3

L L:

R e s p o n s e time (ms) Figure 2.36 Representative data from literature showing reported I)* values as a fiinction of response timeJor thin film thermopile linear arrays. The dashed line represents the Fote and ]ones results. Its slope indicates D* proportional to the square-root of response time. which is typical for thermopiles or bolometers with different geometries and the same material system (after ref. 9 ~ ).

Comparison of photon and thermal detectors performance

67

Figure 2.37 Schematic diagram of thermopile detector structure. The top diagram shows two pixels viewed from the top, with part of the left pixel cut awa!l to show the underl!tinq structure. The lower diagram shows a cross-section side view of two pixels (after ref. 9 6 ).

2.7 Conclusions The intention of this paper has been to provide investigations of the performance of both photon and thermal detectors, with emphasis on the material properties, device structure, and their impact on FPA performance, especially in the LWIR and VLWIR spectral regions. At present, HgCdTe is a widely used variable gap semiconductor and has a privileged position both in LWIR as well as VLWIR spectral ranges. Figure 2.38 shows a plot of the thermal detectivity (3()()K, ()~ FOV) versus operating temperature for the most prominent detector technologies. The thermal detectivity is used here to compare the various technologies for equivalent NETD irrespective of wavelength. The thermal D* figure of merit for photon detectors was obtained by equating the NETD of an ideal thermal detector for a given D* to the NETD of an ideal photon detector with a given D~ . The various regions P show the appropriate applications including 'low cost' uncooled therma 1 detectors, 'high performance uncooled' for night vision e n h a n c e m e n t and earth reconnaissance, 'tactical' for most imaging uses, and 'strategic' for various military-type instruments. Strategic sensors generally detect point targets, so the D* must be as high as possible within the constraint that the cooler must not cause overriding size, weight, reliability and cost issues. High performance near infrared has similar performance requirements, but can only provide a m i n i m u m of cooling because cost and weight minimization is critical. The extrinsic silicon

68

Handbookof Infrared Detection Technologies

Table 2.8

S t a t u s a n d t r e n d s o f c o m m e r c i a l u n c o o l e d arrays (after ref. 6)

Status Feature

Cost ($)

Commercial marketing of military thermal imagers Overspecified for commercial applications 320 x 2 4 0 pixel. 5 0 x 50 ~m bolometer arrays for thermal imagers 3 2 0 x 2 4 0 pixel, 50 x 50 ~tm bolometer arrays for imaging radiometers 120 x 1 pixel. 50 x 5013m thermoelectric arrays for imaging radiometers 3 2 0 x 2 4 0 pixel, 50 x 50 ~m hybrid ferroelectric bolometer a r r a y imagers for drivers' vision e n h a n c e m e n t

1 5 ()()()-2() ()()() 2() ()()()-5() ()()() 1 5 ()()() 2()()()-4()()()

Trends Development of new. low-cost thermal imagers and imaging radiometers Designed for specific commercial applications and meeting commercial requirements 160 • 120 pixel. 50 x 50 lam bolometer arrays for thermal imagers 160 • 120 pixel. 50 • 50 Bm bolometer arrays for imaging radiometers 160 • 120 pixel. 50 • 50 ~m bolometer arrays for driver's vision e n h a n c e m e n t systems 160 x 120 pixel. 50 x 50 Bm bolometer arrays for driver's vision e n h a n c e m e n t systems in extremely large volumes

< 1()()() < 5()()() < 2()()()

1 ()()()

detectors offer very high sensitivity, but at the very low operating temperature which is prohibitive in the most applications. The cryogenically cooled InSb and HgCdTe arrays have comparable array size and pixel yield at MWIR spectral band. However, wavelength tunability and high quantum efficiency have made HgCdTe the preferred material. Thus, the associated cooling and system power requirements can be optimally distributed. The monolithic PtSi Schottky barrier FPAs lead all other technologies with respect to array size (more than 1 ()~' pixels). However, the thermal mismatch barrier in hybrid FPAs has been overcome by developers fabricating InSb and HgCdTe arrays. Detector maturity is a function of accumulated experience and development effort, the complexity of the device required, and the inherent difficulty presented by the material technology. At present, HgCdTe photodiodes and BIB extrinsic silicon detectors are not fully mature. PtSi technology is mature and has reached a plateau. Other two detector technologies such as InSb and silicon bolometers are still evolving significantly as applications for larger array configurations and smaller pixel sizes continue to push the boundaries of technology. Despite serious competition from alternative technologies and slower progress than expected, HgCdTe is unlikely to be seriously challenged for highperformance applications and applications requiring multispectral capability and fast response. The recent successes of competing cryogenically cooled detectors are due to technological, not fundamental issues. The steady progress in epitaxial technology would make HgCdTe devices much more affordable in the near future. The much higher operation temperature of HgCdTe, compared to

Comparison ~f photon and thermal detectors performance

69

Figure 2. $8 Thermal D* versus operating temperature for d(~erent FI'A technologies (~(fter r~:f. 4 ~ ).

Schottky barrier devices and low-dimensional solid devices, may become a decisive argument in this case. The fundamental performance limits of HgCdTe photodiodes have not been reached yet. Continued development of the in situ vapour phase epitaxy methods (MBE and MOCVD) will allow bandgap engineering heterojunction devices of increasing quality and complexity. Also, continued development of epitaxial growth on alternative substrates such as silicon will reduce the cost of 2D arrays. Development of dual-band arrays will continue and three-band detectors will soon be demonstrated. To provide high resolution spectroscopic imaging larger HgCdTe FPAs will be used in Fourier-transform (FT) interferometers. Photodiodes will replace photoresistors for detection out to 15 btm since they are characterized by a more linear response. State of the art QWlP and HgCdTe FPAs provide similar performance figure of merit, since they are predominantly limited by the readout circuits. A low photoconductive gain actually causes a better NEDT of OWIP FPA than that of HgCdTe FPA with a similar storage capacity. The performance is, however, achieved with very different integration times. The integration time of OWIP devices is 1 0 - 1 0 0 times longer than that of HgCdTe arrays, and is typically 5-20 ms. Powerful possibilities of QWIP technology are connected with large LWIR and VLWIR FPA applications and with multicolour detection. Three-band and four-band FPAs will soon be demonstrated in near the future.

70

Handbook of Infrared Detection Technologies

Thermal detector arrays will increase in size and improve in thermal sensitivity to a level satisfying high performance applications at ambient temperature. It is supposed that the silicon microbolometer arrays and the monolithic ferroelectric arrays will capture the low-cost markets. Current uncooled bolometer FPAs have achieved NEDT of less than 1()mK (with 5()~m pixels and f/1 optics), opening the door to the use of less expensive slower optical systems. It is supposed that sales of IR thermal imaging equipment to the automobile market will begin rapidly to change the relative ratio between military/ government and commercial IR markets. Today only about 1 ()% of the market is commercial. After a decade the commercial market can grow to over 7()% in volume and 40% in value, largely connected with volume production of uncooled imagers for automobile driving.-'-' For large volume production for automobile drivers, the cost of uncooled imaging systems will decrease to below $1000. Of course, these systems will cover other segments of the transportation industry" trucks, trains, ships, barges, buses, and airplanes. For the same applications requiring uncooled detectors, the slow response speed is unacceptable. Recently, a number of concepts (e.g., a non-equilibrium device, 99 multi-junction HgCdTe photodiodes, lc1(I optical immersion) and new materials (InAsSb, InAs/GaSb-based type II superlattices) 1 have been proposed to improve the performance of photon detectors operating at near room temperature. The measurements show the possibility to achieve detectivity of ~ l x l ( ) 9 c m H z l / 2 / W at the 8 - 9 ~m range and potentially, the devices can be assembled in large FPAs. A new IR detector concept is micro electromechanical structures (MEMSs). This technology is a marriage of photolithography and mechanics. FPAs based on MEMS technology and a visible optical readout system may offer lower-cost LWIR imaging systems. ~(~1 Finally, considerable development of signal processing function into FPAs can be anticipated.

Appendix A1 HgCdTe photodiodes The Auger m e c h a n i s m is more likely to impose fundamental limitations to the LWIR HgCdTe detector performance. ~(J2 Assuming a non-degenerate statistic, the Auger generation rate is equal to

GA

.

. . . . 2r~1 + 2rl~7

2ri.tl

n+

(A1)

where n and p are the electron and hole concentrations, r ,41 i and r~t are the intrinsic Auger 1 and Auger 7 recombination times and y = r ,-17 i /r{t 1 is the ratio of Auger 7 and Auger I intrinsic recombination times, r!~1 is given by

Comparison olphoton and thermal detectors performance

.rAli

=

G (1 + bt)

(1 + 2/,)exp

(m,, , /,,,)IF1F212(kT/E,,) ~/2

1 + IJ el kT]

in seconds

71

(A2)

where # - m e ~rot, is the ratio of the conduction to the heavy-hole valence-band effective mass, ~, is the relative static dielectric constant, and F, and F2 are the overlap integrals of the periodic part of the electron wave functions. The value of IF1F21 is taken as a constant equal to ().2. The y term is of high uncertainty. According to Casselman and Petersen ~()2 for Hg~_xCdxTe over the range 0 . 1 6 ~ x ~ ( ) . 4 0 a n d 5()K~
2 r/ 217-2hctl/2

n

+ P/YJ

(A3)

The resulting Auger generation achieves its m i n i m u m for p - yl/211i. Since y > 1, the highest detectivity of Auger limited photodetectors can be achieved with a light p-type doping. The required y l / 2 n i p-type doping is difficult to achieve in practice for lowtemperature photodetectors (the control of hole concentration below the 5 x l O l s c m -3 level is difficult) and the p-type material suffers from some n o n f u n d a m e n t a l limitations such as contacts, surface, and Shockley-Read processes. These are the reasons why the low-temperature detectors are typically produced from lightly doped n-type materials. In contrast, p-type doping is clearly advantageous for LWIR and near-room-temperature detectors.

A2 InSb p h o t o d i o d e s

The standard m a n u f a c t u r i n g technique of InSb photodiodes begins with bulk ntype single crystal wafers with donor concentration about N,t-1 ()is cm-3. The Shockley-Read recombination centres are an intrinsic characteristic of n-type InSb. The electron and hole lifetimes are equal because in n-type material there is a single set of recombination centres. The electron and hole carrier lifetimes of high quality material at temperatures below 1 3()K obey the expression 1() r-

4.4 x 10 8

N,t

in seconds

(A4)

The thermal generation can be described by

Gt,, = n]

N,T

(A5)

72

Handbook of Infrared Detection Technologies

A3 GaAs/AIGaAs QWlPs The carrier generation rate from the QWs due to optical p h o n o n s may be obtained as 104 Gpl, o. -

Nowns-Z2

(A 6)

T

where Now is the n u m b e r of OWs, n+2 is the two-dimensional (2D) density of carriers originating from the excited sub-band state, and r is the lifetime of electrons in the upper state. It can be shown, that

ns2 - Lp2 (27rm * kT)

exp -

v,,-kT ~+,)

r -- 4hgpAEcoLp

(A7) (A8)

q2EphonI1

In the equations, Lp is the mean well spacing, m* is the electron effective mass, k is the Boltzman constant, Epl,o,, is the p h o n o n energy. I~ is a dimensionless integral close to 2 in the 8 - 1 2 ~m range for a quasibound upper state. Moreover, AEco is the cut-off energy defined as AEa, = hc/;.c = Vt, - El, (Vl, is the barrier energy of the OW, and ;.c is the cut-off wavelength ). 1/Ep = 1 / ~ - 1/~o ( ~ : and Eo is the dielectric permittivity at infinite and zero frequency, respectively), and

Ef. - E1 = n~lh2

(A9)

47rm*

is the Fermi energy of the well. and n~ is the 2D carrier density in the ground state. It should be noted that Gpl, does not depend on Lp. and is therefore a more f u n d a m e n t a l quantity of a OWIP. in contrast to n~e and r. The following parameters have been assumed: tn*=().()67mo. Ej,=63~o, Evl,o,,=36 meV, n~=2• 11 cm-2, andN0~v = 50.

A4 Extrinsic photoconductors At the low t e m p e r a t u r e of operation of impurity p h o t o c o n d u c t o r s (when kT << Ea and n << Nd, N~), the thermal equilibrium free-charge carrier in an n-type extrinsic semiconductor with a partially compensated singly ionized level is equal to 3 s.1 (~s Pith -- n,,,aj -- --~

Na

exp

-

Here N,. is the density of states in the conduction band. Na is the donor concentration, N,, is the compensating acceptor concentration, and E,~ is the bonding energy of the donor relative to the conduction band.

Conlparison of photon and thernlal detectors perfornlance

73

The majority carrier lifetime is determined by the density of empty (ionized) donor levels, and for low temperatures, such that n,,,,o < Na < Nd, is given by r9 -

(oTvthN,~)

(A11)

-~,

where ere is the capture cross-section for electrons into the donor level, and Vth = (8kT/zrm*)l/2 is the carrier thermal velocity. The shallow-level impurities (B and As) show typically crc~lO -11 cm 2, while the deep-level impurities (In, Au, Zn) show er,.=lO -13 cm 2. By comparison, the or,. of intrinsic photoconductors is about 10-17 c m 2" Practical values of ot for optimized extrinsic silicon photoconductors are in the range 1 0 - 5 0 c m -~. To maximize q u a n t u m efficiency, the thickness of the detector crystal should be not less than about 0.1cm. Typical q u a n t u m efficiencies are in range of 10 to 50% at the response peak. A5 Noise equivalent difference temperature Photon detectors Detectivity provides only a limited insight into the suitability of the IR detectors for specific IR systems. The objective criterion of the performance of imaging devices is the noise equivalent difference temperature (NEDT), i.e. the difference in temperature of the object required to produce an electrical signal equal to the rms noise voltage. The sensitivity of an IR system can be expressed in a variety of ways. 1~ In this paper, we follow the paper recently published by Kinch. 18 The limiting factor of detector sensitivity is the fluctuation in the relevant carrier concentration since this determines the minimum observable signal. This fluctuation is envisioned by assuming that the detector current is integrated on a capacitive node, giving a variance N1/2 __ g

.(Id -+- IB)Tint

,

(A12)

q

where g is the gain, Ia is the detector dark current, IB is the background flux current, and "tint is the integration time. The noise equivalent flux (also named as the minimum observable signal) is given by g A Q r l z i n t A -~ N 1/2

(A13)

where ~/represents the overall q u a n t u m efficiency of the detector, including the internal q u a n t u m efficiency. Using last two equations, the noise equivalent flux can be described as AO -- g(1 + Id/I~,)OB -~

N1/2

(A14)

74 Handbookof Infrared Detection Technologies Because AO=

-"

AT dQt3/dT

(A15)

so substituting above, we have for the noise equivalent difference temperature

~(1 + Ia/Ii3)Qu

NEDT -

(A16)

(dQI3/dT) N 1/2

where (dQR/dT)/QR=C, is the scene contrast. For unlimited available capacity N 1/2 varies as g, and AQ and NEDT are independent of the gain, g. However, for a limited m a x i m u m well capacity N, both bandwidth (Aft and NEDT will vary as g. These parameters depend on specific system and detector quantities such as area and bandwidth, where bandwidth is defined as Af= 1/2 r.

Thermal detectors In the case of thermal detector, the limiting temperature fluctuation is given by12.15 AT,,

\ 1 + w---2r~7J

(A1 7)

At low frequencies AT,,=(4kT2RthAf) 1/2. It appears that correlated double sampling of this limitation with an integrated time Tint, gives 18 2kT2[1 - exp(--rint/RthCth)]) 1/2 Cth

A T,, -

(AlS)

where it was assumed Af=l/RthCth. Comparing equation (A7) with a signal fluctuation change given by

ATs--

(dP)

--~ AT 1 - e x p (

(A 19 )

qn_t '~ R,,,

RthCthJ

we can obtain

2kT2 NEDT -

Cth[1 -- exp(--rint/RthCth)]

) 1/2 (dP/dT)Rth

(A20)

where (dP/dT) is the differential change in radiated power per scene temperature change in the spectral region of interest. The integration time should fulfil the condition ri,,t ,~ 2RthCth t o avoid image smearing. The m a x i m u m value of Rth is given by radiative coupling with the upper limit equal to 4 x 1 ()~ K/W for a 1 mil pixel size. For micromachined resonant cavity bolometers with f/1 optics, there is

Comparisonof photon and tlwrmaldetectorsperformance 75 a finite limit of detector thickness, typically t=3000 A. For 1 mil pixel size a volumetric specific heat is about 2 J/cm ~K.

References 1. M. Razeghi, Current status and future trends of infrared detectors, OptoElectron. Rev. 6, 1 5 5 - 1 9 4 (1998 ). 2. A. Rogalski, Infrared Detectors, Gordon and Breach Science Publishers, Amsterdam (2000). 3. A. Rogalski, K. Adamiec, and J. Rutkowski, Narrow-Gap Semiconductor Photodiodes, SPIE Press, Bellingham (2()0()). 4. A. Rogalski, Infrared detectors at the beginning of the next millennium, Sensors and Materials 12, 2 3 3 - 2 8 8 (2 ()00). 5. R.A. Wood and N.A. Foss, Micromachined bolometer arrays achieve lowcost imaging, Laser Focus World, 101-1 ()6 (June, 1993). 6. P.W. Kruse, Uncooled Thermal Imaging. Arrays. S!tstems, and Applications, SPIE Press, Bellingham (2001). 7. A. Rose, Concepts in Photoconductivit!l and Allied Problems, Interscience, New York (1963). 8. ]. Piotrowski and W. Gawron, Ultimate performance of infrared photodetectors and figure of merit of detector material, Infrared Phys. Technol. 38, 6 3 - 6 8 (1997). 9. B.F. Levine, Ouantum-well infrared photodetectors, ]. Appl. Phys. 74, 1111t81(1993). 10. ]. Piotrowski and A. Rogalski, New generation of infrared photodetectors, Sensors and Actuators A 6 7 , 1 4 6 - 1 5 2 (1998). 11. R.G. Humpreys, Radiative lifetime in semiconductors for infrared detection, InfraredPhys. 2 6 , 3 3 7 - 3 4 2 (1986). 12. P.W. Kruse, L.D. McGlauchlin and R.B. McOuistan, Elements of Infrared Technology, Wiley, New York ( 1962 ). 13. P.W. Kruse, The photon detection process, in Optical and Infrared Detectors, pp. 5-69, edited by R. J. Keyes, Springer-Verlag, Berlin (1977). 14. A. Rogalski, Comparison of the performance of quantum well and conventional bulk infrared photodetectors, Infrared Phys. Technol. 3 8, 2 9 5 - 3 1 0 (1997). 15. A. Smith, F.E. ]ones and R.P. Chasmar, The Detection and Measurement of Infrared Radiation, Clarendon, Oxford (1968). 16. R. Ciupa and A. Rogalski, Performance limitations of photon and thermal infrared detectors, Opto-Electron. Rev. 5, 2 5 7 - 2 6 6 ( 1997). 17. L. ]. Kozlowski and W. F. Kosonocky, Infrared detector arrays, in Handbook of Optics, Chap. 23, edited by M. Bass, E. W. Van Stryland, D. 1t. Williams and W. L. Wolfe, McGraw-Hill, Inc. New York ( 1995 ). 18. M.A. Kinch, Fundamental physics of infrared detector materials, ]. Electron. Mater. 2 9 , 8 0 9 - 8 1 7 (2000).

76 Handbookof Infrared Detection Technologies

19. T. Ashley and C. T. Elliott, Non-equilibrium devices for infrared detection, Electron. Lett. 2 1 , 4 5 1 - 4 5 2 (1985). 20. F. Fuchs, L. B~rkle, R. Hamid, N. Herres, W. Pletschen, R.E. Sah, R. Kiefer, and J. Schmitz, Optoelectronic properties of photodiodes for mid- and far-infrared based on the InAs/GaSb/A1Sb materials family, Proc. SPIE 4 2 8 8 , 1 7 1 - 1 8 2 (2001). 21. J. Piotrowski and A. Rogalski, Comment on 'Temperature limits on infrared detectivities ofInAs/InxGa~ _xSb superlattices and bulk HgxCdl_xTe', [J. Appl. Phys. 74, 4774 ( 1993)] ]. Appl. Phys. gO. 2 5 4 2 - 2 544 (1996). 22. P. R. Norton, Status of infrared detectors, Proc. SPIE 3 3 7 9 , 1 0 2 - 1 1 4 (1998). 23. W.F.M. Micklethwaite and A.J. Johnson, InSb: Materials and devices, in Infrared Detectors and Emitters: Materials and Devices, pp. 1 7 7 - 2 0 4 , edited by P. Capper and C.T. Elliott, Kluwer Academic Publishers, Boston (2000). 24. J. T. Wimmers and D. S. Smith, Characteristics of InSb photovoltaic detectors at 77 K and below, Proc. SPIE 3 64, 12 31- 31 ( 1983). 25. J. T. Wimmers, R. M. Davis, C. A. Niblack and D. S. Smith, Indium antimonide detector technology at Cincinnati Electronics Corporation, Proc. SPIE 9 3 0 , 1 2 5 - 1 3 8 (1988). 26. A. M. Fowler, I. Gatley, P. McIntyre, F. J. Vrba and A. Hoffman, ALADDIN, the 1 0 2 4 x 1024 InSb array: design, description, and results, Proc. SPIE 2 8 1 6 , ]50-160(1996). 2 7. J.B. Varesi, R.E. Bornfreund, A.C. Childs, W.A. Radford, K.D. Maranowski, J.M. Peterson, S.M. Johnson, L.M. Giegerich, T.J. de Lyon, and J.E. Jensen, Fabrication of high-performance large-format MWIR focal plane arrays from MBE-grown HgCdTe on 4" silicon substrates, ]. Electron. Mater. 30, 5 6 6 - 5 7 3

(2001). 28. M.B. Reine, Photovoltaic detectors in MCT, in Infrared Detectors and Emitters: Materials and Devices, pp. 2 79-312, edited by P. Capper and C.T. Elliott, Kluwer Academic Publishers, Boston (2000). 29. Y. Nemirovsky and G. Bahir, Passivation of mercury cadmium telluride surfaces, J. Vac. Sci. Technol. A 7 , 4 5 0 - 4 5 9 (1989). 30. W. E. Tennant, C. A. Cockrum, ]. B. Gilpin, M. A. Kinch, M. B. Reine and R. P. Ruth, Key issues in HgCdTe-based focal plane arrays: an industry perspective, ]. Vac. Sci. Technol. B l O , 1 3 5 9 - 3 6 9 ( 1 9 9 2 ) . 31. A. Rogalski and R. Ciupa, Long wavelength HgCdTe photodiodes: n+-on-p versus p-on-n structures, ]. App1. Phys. 7 7 . 3 5 0 5 - 3 5 1 2 (1995). 32. T.J. de Lyon, R.D. Rajavel, J.A. Vigil, J.E. Jensen, O.K. Wu, C.A. Cockrum, S.M. Johnson, G.M. Venzor, S.L. Bailey, I. Kasai, W.L. Ahlgren, and M.S. Smith, Molecular-beam epitaxial growth of HgCdTe infrared focal-plane arrays on silicon substrates for midwave infrared applications, ]. Electron. Mater. 27, 5 5 0 555(1998). 33. M. Kimata and N. Tsubouchi, Schottky barrier photoemissive detectors, in Infrared Photon Detectors, pp. 299-349, edited by A. Roga]ski, SPIE Optical Engineering Press, Bellingham (1995).

Comparisonof photon and thermaldetectorsperformance 77

34. M. Kimata, Metal silicide Schottky infrared detector arrays, in Infrared Detectors and Emitters: Materials and Devices, pp. 77-98, edited by P. Capper and C.T. Elliott, Kluwer Academic Publishers, Boston (2000). 35. N. Sclar, Properties of doped silicon and germanium infrared detectors, Prog. Ouant. Electr. 9 , 1 4 9 - 2 5 7 (1984). 36. P. R. Norton, Infrared image sensors, Opt. Eng. 30, 1 6 4 9 - 1 6 6 3 (1991). 3 7. S. Solomon, A. Tribble, N. Lum, J. Venzon, G. Domingo, A. Hofman, and M. Smith, High-background, long-wave Si:As IBC 32()x240 IR focal plane array, Proc. SPIE 2 8 1 6 , 1 6 1 - 1 6 8 ( 1 9 9 6 ) . 38. A. Rogalski, Assessment of HgCdTe photodiodes and quantum well infrared photoconductors for long wavelength focal plane arrays, Infrared Phys. rechnol. 4 0 , 2 7 9 - 2 9 4 (1999). 39. J. Bajaj, J.M. Arias, M. Zandian, J.G. Pasko, L.J. Kozlowski, R.E. De Wames, and W.E Tennant, Molecular beam epitaxial HgCdTe material characteristics and device performance: Reproducibility status, J. Electron. Mater. 24, 1 0 6 7 1076(1995). 40. S. D. Gunapala, J. K. Liu, J. S. Park, M. Sundaram, C. A. Shott, T. Hoelter, T. L. Lin, S. T. Massie, P. D. Maker, R. E. Muller, and G. Sarusi, 9 l-tm cutoff 256 x 256 GaAs/AlxGal_xAs quantum well infrared photodetector hand-held camera, IEEE Trans. Electron Devices 44, 5 1 - 57 ( 1997 ). 41. M.Z. Tidrow, J.C. Chiang, S.S Li, and K. Bacher, A high strain two-stack two-color quantum well infrared photodetector, Appl. Phys. Lett. 70, 8 5 9 - 8 6 1 (1997). 42. M.Z. Tidrow, W.A. Beck, W.W. Clark, H.K. Pollehn, J.W. Little, N.K. Dhar, P.R. Leavitt, S.W. Kennerly, D.W. Beekman, A.C. Goldberg, and W.R. Dyer, Device physics and focal plane applications of QWIP and MCT, Opto-Electron. Rev. 7 , 2 8 3 - 2 9 6 (1999). 43. L.J. Kozlowski, K. Vural, J.M. Arias, W.E. Tennant, and R.E. DeWames, Performance of HgCdTe, InGaAs and quantum well GaAs/A1GaAs staring infrared focal plane arrays, Proc. SPIE 3 1 8 2 , 2-13 ( 199 7). 44. B. F. Levine, K. K. Choi, C. G. Bethea, J. Walker, and R. J. Malik, New 10 gm infrared detector using intersubband absorption in resonant tunneling GaA1As superlattices, Appl. Phys. Lett. 50, 1 0 9 2 - 1 0 9 4 (1987). 45. H. Schneider, M. Walther, J. Fleissner, R. Rehm, E. Diwo, K. Schwarz, P. Koidl, G. Weimann, J. Ziegler, R. Breiter, and W. Cabanski, Low-noise OWIPs for FPA sensors with high thermal resolution, Proc. SPIE 4 1 3 0 , 3 5 3 - 3 6 2 (2()()()). 46. S.D. Gunapala, S.V. Bandara, J.K. Liu, E.M. Luong, N. Stetson, C.A. Shott, J.J. Bock, S.B. Rafol, J.M. Mumolo, and M.J. McKelvey, Long-wavelength 256 x 2 5 6 GaAs/A1GaAs quantum well infrared photodetector (QWIP) palm-size camera, IEEE Trans. ElectronDevices 4 7 , 3 2 6 - 3 3 2 (2()()0). 47. S.D. Gunapala, S.V. Bandara, J.K. Liu, E.M. Luong, S.B. Rafol, J.M. Mumolo, D.Z. Ting, J.J. Bock, M.E. Ressler, M.W. Werner, P.D. LeVan, R. Chehayeb, C.A. Kukkonen, M. Ley. P. LeVan, and M.A. Fauci, Recent developments and applications of quantum well infrared photodetector focal plane arrays, Opto-Electron. Rev. 8, 1 5 0 - 1 6 3 (2()() 1 ).

78 Handbookof Infrared Detection Technologies 48. S.D. Gunapala, S.V. Bandara, J.K. Liu, W. Hong, M. Sundaram, P.D. Maker, R.E. Muller, C.A. Shott, and R. Carralejo, Long-wavelength 6 4 0 • GaAs/A1GaAs quantum well infrared photodetector snap-shot camera, IEEE Trans. Electron Devices 45, 1 8 9 0 1 8 9 5 (1998). 49. S.D. Gunapala, J.S. Park, G. Sarusi, T.L. Lin, J.K. Liu, P.D. Maker, R.E. Muller, C.A. Shott, and T. Hoelter, 15-Bm 128 x 128 GaAs/AlxGa 1- xAs quantum well infrared photodetector focal plane array camera, IEEE Trans. Electron Devices 44, 4 5 - 5 0 (1997). 50. W. A. Beck and T. S. Faska. Current status of quantum well focal plane arrays, Proc. SPIE 2 7 4 4 , 1 9 3 - 2 0 6 (1996). 51. A. Manissadjian, P. Costa, P. Tribolet, G. Destefanis, HgCdTe performance for high operating temperatures, Proc. SPIE 3 4 3 6 , 150-161 (1998). 52. P. Norton, J. Campbell, S. Horn, and D. Reago, Third-generation infrared imagers, Proc. SPIE 4 1 1 0 , 2 2 6 - 2 3 6 (2()()()). 53. J.P. Chatard, LW MCT IRFPA cost optimization, Proc. SPIE 3 6 9 8 , 4 0 7 419(1999). 54. F. Bertrand, J.T. Tissot, and G. Destefanis, Second generation cooled infrared detectors. State of the art and prospects, in Physics of Semiconductor Devices, Vol. II, pp. 7 1 3 - 7 2 0 , edited by V. Kumar and S.K. Agarwal, Narosa Publishing House, New Delhi (1998). 55. T.J. DeLyon, J.E. Jensen, M.D. Gorwitz. C.A. Cockrum, S.M. Johnson, and G.M. Venzor, MBE growth of HgCdTe on silicon substrates for large-area infrared focal plane arrays: A review of recent progress. ]. Electron. Mater. 28, 7 0 5 - 7 1 1 (1999). 56. S.D. Gunapala and S.V. Bandara, Quantum well infrared photodetectors, in Handbook of Thin Devices, edited by M.H. Francombe, Vol. 2, pp. 63-99, Academic Press, San Diego (2()()()). 57. R.D. Rajavel, D.M. Jamba, J.E. Jensen, O.K. Wu, P.D. Brewer, J.A. Wilson, J.L. Johnson, E.A. Patten, K. Kasai, J.T. Caulfield, and P.M. Goetz, Molecular beam epitaxial growth and performance of integrated multispectral HgCdTe photodiodes for the detection of mid-wave infrared radiation, J. Crystal Growth 184, 1 2 7 2 - 1 2 7 8 ( 1 9 9 8 ) . 58. R.D. Rajavel, D.M. Jamba. J.E. Jensen. O.K. Wu, J.A. Wilson, J.L. Johnson, E.A. Patten, K. Kasai, P.M. Goetz and S.M. Johnson, Molecular beam epitaxial growth and performance of HgCdTe-based simultaneous-mode two-color detectors, ]. Electron. Mater. 2 7, 74 7 - 751 ( 1 9 9 8 ). 59. M. B. Reine, A. Hairston, P. O'Dette, S. P. Tobin, F. T. J. Smith, B. L. Musicant, P. Mitra, F. C. Case, Simultaneous MW/LW dual-band MOCVD HgCdTe 64 x 64 FPAs, Proc. SPIE 3 3 79.2()()-212 (1998). 60. M.A. Kinch, HDVIP | FPA technology at DRS, Proc. SPIE 4 3 6 9 , :366-578 (2001). 61. W. Cabanski, R. Breiter, R. Koch, K.H. Mauk, W. Rode, J. Ziegler, H. Schneider, M. Walther, and R. Oelmaier, 3rd gen. focal plane arrays IR detectors modules at AIM, Proc. SPIE 4 5 6 9 , 5 4 7 - 5 5 8 (2()() l ). 62. W.E. Tennant, M. Thomas, L.J. Kozlowski, W.V. McLevige, D.D. Edwall, M. Zandian, K. Spariosu, G. Hildebrandt, V. Gil, P. Ely, M. Muzilla, A. Stoltz, and J.H.

Comparison of photon and thermal detectorsperformance 79

Dinan, A novel simultaneous unipolar multispectral integrated technology approach for HgCdTe IR detectors and focal plane arrays, ]. Electron. Mater. 30, 590-594(200]). 63. J.P. Zanatta, P. Ferret, R. Loyer. G. Petroz, S. Cremer, J.P. Chamonal, P. Bouchut, A. Million, and G. Destefanis, Single and two colour infrared focal plane arrays made by MBE in HgCdTe, Proc. SPIE 4 1 3 0 , 4 4 1 - 4 5 1 (2000). 64. H.K. Pollehn and J. Ahearn, Multi-domain smart sensors, Proc. SPIE 3 6 9 8 , 4 2 0 - 4 2 6 (1999). 65. T. Whitaker, Sanders' OWIPs detect two colors at once, Compound Semiconductors 5(7), 4 8 - 5 1 (1999 ). 66. A.C. Goldberg, S.W. Kennerly, J.W. Little, H.K. Pollehn, T.A. Shafer, C.L. Mears, H.F. Schaake, M. Winn, M. Taylor, and P.N. Uppal, Comparison of HgCdTe and OWIP dual-band focal plane arrays, Proc. SPIE 4 3 6 9 , 5 3 2 - 5 4 6 (2OOl). 67. M.B. Reine, HgCdTe photodiodes for I1R detection: A review, Proc. SPIE 4288, 266-277 (2001). 68. Ph. Bois, E. Costard, J.Y. Duboz, and J. Nagle, Technology of multiquantum well infrared detectors, Proc. SPIE 3061,764-771 ( ] 997). 69. S.D. Gunapala, S.V. Bandara, A. Sigh, J.K. Liu, S.B. Rafol, E.M. Luong, J.M. Mumolo, N.Q. Tran, J.D. Vincent, C.A. Shott, J. Long, P.D. LeVan, 8-9 and 14-] 5 l~m two-color 640x486 quantum well infrared photodetector (QWIP) focal plane array camera, Proc. SPIE 3698, 687-697 (1999). 70. S.D. Gunapala, S.V. Bandara, A. Singh, J.K. Liu, B. Rafol, E.M. Luong, J.M. Mumolo, N.O. Tran, D.Z. Ting, J.D. Vincent, C.A. Shott, J. Long, and P.D. LeVan, 640x486 long-wavelength two-color GaAs/A1GaAs quantum well infrared photodetector (OWIP) focal plane array camera, IEEE Trans. Electron Devices 47, 963-971 (2000). 71. P.W. Kruse, Principles of uncooled infrared focal plane arrays, in Semiconductors and Semimetals, Vol. 4 7, pp. 17-42, edited by P. W. Kruse and D. D. Skatrud, Academic Press, San Diego ( 199 7 ). 72.1R. A. Wood, Monolithic silicon microbolometer arrays, in Semiconductors and Semimetals, Vol. 47, pp. 4 5 - 1 2 1 , edited by P. W. Kruse and D. D. Skatrud, Academic Press, San Diego ( 199 7). 73. R.A. Wood, Uncooled microbolometer infrared sensor arrays, in Infrared Detectors and Emitters: Materials atld Devices, pp. 1 4 9 - 1 7 4 , edited by P. Capper and C.T. Elliott, Kluwer Academic Publishers, Boston (2000). 74. C. M. Hanson, Hybrid pyroelectric-Ferroelectric bolometer arrays, in Semiconductors and Semimetals, Vol. 4 7, pp. 12 3-174, edited by P. W. Kruse and D. D. Skatrud, Academic Press, San Diego (1997). 75. D.L. Polla and J.R. Choi, Monolithic pyroelectric bolometer arrays, in Semiconductors and Semilnetals, Vol. 4 7, pp. 1 75-2() 1, edited by P. W. Kruse and D. D. Skatrud, Academic Press, San Diego(1997). 76. R.W. Whatmore and R. Watton, Pyroelectric materials and devices, in Infrared Detectors and Emitters: Materials alut Devices, pp. 9 9 - 1 4 7 , edited by P. Capper and C.T. Elliott, Kluwer Academic Publishers, Boston (2000).

80 Handbookof hlfrared Detection Technologies 77. T. Ishikawa, M. Ueno, K. Endo, Y. Nakaki, H. Hata0 T. Sone, and M. Kimata, Low-cost 3 2 0 • uncooled IRFPA using conventional silicon IC process, Opto-Electron. Rev. 7 , 2 9 7 - 3 ( ) 3 (1999). 78. W. Radford, D. Murphy, A. Finch, K. Hay. A. Kennedy, M. Ray, A. Sayed. ]. Wyles, R. Wyles, ]. Varesi, E. Moody, and F. Cheung. Sensitivity improvements in uncooled microbolometer FPAs. Proc. SPIE 3 6 9 8 . 1 1 9 - 1 3 ( ) (1999). 79. J. Anderson, D. Bradley, D.C. Chen. R. Chin, K. ]urgelewicz, W. Radford, A. Kennedy, D. Murphy, M. Ray, R. Wyles, ]. Brown, and G. Newsome, Low cost microsensors program, Proc. SPIE 4 3 6 9 , 5 5 9 - 5 6 5 {2()() 1 ). 80. D. Murphy, M. Ray, R. Wyles, J. Asbrock, N. Lum, A. Kennedy, J. Wyles, C. Hewitt, G. Graham, W. Radford, ]. Anderson, D. Bradley, R. Chin, and T. Kostrzewa, High sensitivity {25 Bm pitch) microbolometer FPAs and application development, Proc. SPIE 4 3 6 9 , 2 2 2 - 2 3 4 (2()() 1 }. 81. W. Radford, D. Murphy, A. Finch, K. Hay. A. Kennedy, M. Ray, A. Sayed, ]. Wyles, R. Wyles, ]. Varesi, E. Moody, and F. Cheung, Sensitivity improvements in uncooled microbolometer FPAs, Proc. SPIE 3 6 9 8 , 1 1 9 - 1 3 0 (1999). 82. M. Altman, B. Backer, M. Kohin. R. Blackwell, N. Butler, and ]. Cullen, Lockheed Martin 6 4 0 • 48() uncooled microbolometer camera, Proc. SPIE 3 6 9 8 , 137-143(1999).

83. M.N. Gurnee, M. Kohin, R. Blackwell. N. Butler, ]. Whitwan, B. Backer, A. Leary, and T. Nielson, Developments in uncooled IR technology at BAE Systems, Proc. SPIE 4 3 6 9 , 2 8 7 - 2 9 6 (2()()1). 84. B.I. Craig, R.J. Watson, and M.H. Unewisse. Anisotropic excess noise within a-Si:H, Solid-State Electrollics. 39, 8()7-812 {1996). 85. ].L. Tissot, ].L. Martin, E. Mottin, M. Vilain, ]. ]. Yon, and ].P. Chatard, 320• microbolometer uncooled IRFPA development, Proc. SPIE 4 1 3 0 , 4 7 3 - 4 7 9 (2000). 86. E. Mottin, ].L. Martin, ].L. Ouvrier-Buffet. M. Vilain, A. Bain, ].]. Yon, ].L. Tissot, and ].P. Chatard, Enhanced amorphous silicon technology for 32()• microbolometer arrays with a pitch of 35 btm. Proc. SPIE 4 3 6 9 , 25()-256 (20 01 ). 87. C. M. Hanson, Uncooled thermal imaging at Texas Instruments, Proc. SPIE 2 0 2 0 , 3 3 0 - 3 3 9 (1993). 88. R.K. McEwen and P.A. Manning, European uncooled thermal imaging sensors, Proc, SPIE 3 6 9 8 , 3 2 2 - 3 3 7 {1999). 89. ].F. Belcher, C.M. Hanson, H.R. Beratan, K.R. Udayakumar, and K.L. Soch, Uncooled monolithic ferroelectric IRFPA technology, Proc. SPIE 3 4 3 6 , 6 1 1 622(1998). 90. M.A. Todd, P.A. Manning, O.D. Donohue, A.G. Brown, and R. Watton, Thin film ferroelectric materials for microbolometer arrays, Proc. SPIE 4 1 3 0 , 1 2 8 - 1 3 9 (2000). 91. N. Teranishi, Thermoelelctric uncooled infrared focal plane arrays, in Semiconductors and Semimetals, Vol. 47, pp. 2() 3-218, edited by P. W. Kruse and D. D. Skatrud, Academic Press. San Diego i 1997 ). 92. T. Kanno, M. Saga, S. Matsumoto, M. Uchida. N. Tsukamoto, A. Tanaka, S. Itoh, A. Nakazato, T. Endoh, S. Tohyama, Y. Yamamoto, S. Murashima, N.

Comparison of photon and thermaldetectorsperformance 81

Fujimoto, and N. Teranishi, Uncooled infrared focal plane array having 128 x 128 thermopile detector elements, Proc. SPIE 2 2 6 9 . 4 5 0 - 2 5 6 (1994). 93. A.D. Oliver, W.G. Baer, and K.D. Wise. A bulk-micromachined 1024element uncooled infrared imager, Proc. 8th hit. Conf. Solid-State Sensors and Actuators (Transducers '95), and Eurosensors IX, pp. 6 3 6 - 6 3 9 1995. 94. M.C. Fote, E.W. Jones, and T. Caillat. Uncooled thermopile infrared detector linear arrays with detectivity greater than 1 ()~)cmHz 1/2/W, IEEE Trans. Electron Devices 4~, 1 8 9 6 - 1 9 0 2 (1998). 95. M.C. Fote and E.W. Jones, High performance micromachined thermopile linear arrays, Proc. SPIE 3 3 7 9 , 1 9 2 - 1 9 7 (1998). 96. M.C. Fote and S. Gaalema, Progress towards high-performance thermopile imaging arrays, Proc. SPIE 4 3 6 9 , 35()-354 (2()() 1 ). 97. P.W. Kruse, Application of uncooled monolithic thermoelectric linear arrays to imaging radiometers, in Setniconductors and Semimetals, Vol. 47, pp. 2 9 7 - 3 1 8 , edited by P. W. Kruse and D. D. Skatrud, Academic Press, San Diego (1997). 98. T. McManus and S. Mickelson, Imaging radiometers employing linear thermolelectric arrays, Proc. SPIE 3 6 9 8 , 352-36() (1999). 99. C.T. Elliott, Photoconductive and non-equilibrium devices in HgCdTe and related alloys, in Infrared Detectors alld Etnitters: Materials and Devices, pp. 2 7 9 312, edited by P. Capper and C.T. Elliott. Kluwer Academic Publishers, Boston

(2000).

l O0. J. Piotrowski, M. Grudziefl. Z. Nowak. Z. Orman, ]. Pawluczyk, M. Romanis, and W. Gawron, Uncooled photovoltaic Hg~ _• LWIR detectors, Proc. SPIE. 4 1 1 0 , 1 7 5 - 1 8 4 (2()()()). 101. T. Perazzo, M. Mao, O. Kwon, A. Majumdar, J.B. Varesi, and P. Norton, Infrared vision using uncooled micro-optomechanical camera, Appl. Ph!ls. Lett. 74, 3 5 6 7 - 3 5 6 9 (2001). 102. T. N. Casselman and P. E. Petersen, A comparison of the dominant Auger transitions in p-type (Hg,Cd)Te, Solid State Comnlutl. 33, 61:3-619 (198()). 103. R. Schoolar and E. Tenescu, Analysis ofInSb photodiode low temperature characteristics, Proc. SPIE 686, 2-11 (1986). 104. J.Y. Andersson, Dark current mechanisms and conditions of background radiation limitation of n-doped A1GaAs/GaAs quantum-well infrared detectors, J. Appl. Phys. 72, 6 2 9 8 - 6 3 0 4 ( 1 9 9 5 ) . 105. P. R. Bratt, Impurity germanium and silicon infrared detectors, in Semiconductors and Semimetals, Vol. 12. pp. 39-142. edited by R. K. Willardson and A. C. Beer, Academic Press, New York ( 19 77). 106. J.M. Lloyd, Thermal Imagillg Systems. Plenum Press. New York ( 1975 ).

This Page Intentionally Left Blank

Chapter 3

GaAs/AIGaAs based quantum well infrared photodetector focal plane arrays S. D. Gunapala and S. V. Bandara

3.1 Introduction The idea of using a q u a n t u m well to detect light can be understood by the basic principles of q u a n t u m mechanics. 1 An electron in a square q u a n t u m well is the classic particle-in-a-box of basic q u a n t u m mechanics. The electron cannot have just any energy inside the well: rather, it is constrained to reside in certain discrete energy levels, i.e., its energy is quantized. These allowed energy levels, which depend on the electron mass and the size and shape of the q u a n t u m well, can be calculated by solving the time-independent SchrSdinger wave equation. For a square q u a n t u m well, the energy levels depend, in a straightforward way, on the well dimensions (width and potential depth). At low temperatures, an electron will reside in the lowest energy level (the ground state) of the well. W h e n a photon strikes the well, it can impart its energy to the ground state electron and excite it to the next allowed energy level (the first excited state), a process called intersubband absorption. In order for this to happen, the photon energy must equal the energy separation between the ground state and the first excited state. A photon with a different energy is not absorbed because there is no allowed energy state available for the photon to excite the ground state electron. The simplest q u a n t u m well infrared photodetector (QWIP) design uses a simple square q u a n t u m well designed to hold just two states: a ground state deep inside the well, and the first excited state near the well top. A voltage bias can now be applied to sweep out the photoelectron: under constant illumination a steadystate photocurrent (a measure of the incident photon flux) thus flows t h r o u g h the detector. The spectral response of the QWIP is therefore quite narrow, its sharpness determined by the sharpness of the two energy states involved. Making a QWIP to detect light of a different wavelength is then simply

84 Handbook of Infrared Detection Technologies accomplished by changing the width and potential depth of the well in such a way that the two energy states are separated by the corresponding photon energy. Such a square quantum well can be created in the lattice-matched GaAs/ Al• material system by sandwiching a layer of GaAs between two layers of AlxGal_xAs. The bandgap of Al•215 being larger than that of GaAs, results in a square quantum well for electrons in the conduction band' the GaAs layer is the well layer; the AlxGal_xAs layers on either side are the potential barriers. The depth of the potential well can be precisely controlled by controlling the A1 mole fraction x in the AlxGal_• barrier layers. State-of-theart crystal growth techniques like molecular beam epitaxy (MBE) permit the epitaxial growth of such layers, on a typically six-inch diameter GaAs substrate. with ultra-high purity and with control of layer thickness down to a fraction of a molecular layer. This allows the width of the GaAs well layer to be precisely controlled, and the interfaces between the well and barrier layers to be made truly sharp to produce a textbook square quantum well. A controlled number of ground state electrons are provided by doping the GaAs well with Si (an n-type dopant) during the MBE growth. It is worth noting that the above design also creates a square quantum well (with the same width but a smaller potential depth) for holes in the valence band. Ground state holes can be provided by doping the GaAs wells with Be (a p-type dopant) instead of Si, resulting in a QWIP in which holes, rather than electrons, are the photocarriers. The much heavier hole has poor transport once it has been ejected out of the well by a photon, resulting in p-OWIPs being inferior in performance to n-QWIPs. We shall henceforth confine the discussion to electrons. It is important to note that the quantum well exists only in one spatial dimension: the growth direction normal to the layers. While the electron energies are quantized in this direction, the electrons are essentially free to move about in the planes of the GaAs well layers. This has important consequences for the coupling of incident light, as will be explained later. The OWIP is thus a vertical device in that there is vertical transport of the electrons, perpendicular to the layers. The quantum wells in this material system can be designed to detect light at wavelengths longer than 3 ~m (the wells cannot be made deep enough to detect shorter wavelength light). Several quantum wells (sandwiched between barriers) are usually grown stacked on top of each other to increase photon absorption. The upper limit on this number for a typical detector structure in this material system is around 50, about the number of wells and barriers that a photoelectron can traverse in an electric field without being captured by a well, downstream from the well from which the electron originated. Fewer quantum wells implies less photon absorption' more implies fewer photoelectrons collected. Such a stack of quantum wells is called a multi-quantum well (MQW) structure. The entire MOW stack is sandwiched between heavily-doped top (called the emitter) and bottom (called the collector) GaAs layers which provide electrical contacts to the device. The location of the first excited state relative to the top of the well can have a dramatic effect on device performance. By reducing the quantum well width it is

GaAs/A1GaAs based quantum ~vellinfrared photodetector focal plane arra!ts 8 5

possible to p u s h this state up above the well top and into the continuum resulting in a strong b o u n d - t o - c o n t i n u u m intersubband absorption. The major advantage of the bound-to-continuum OWIP is that the photoelectron can escape from the q u a n t u m well to the continuum transport states without being required to tunnel through the barrier. As a result, the voltage bias required to efficiently collect the photoelectrons can be reduced dramatically, thereby lowering the dark current (see below). Furthermore, since the photoelectrons do not have to tunnel through them, the AlxGa~_xAs barriers can be made thicker without reducing the photoelectron collection efficiency. Increasing the barrier thickness up to 500 A reduces the ground state sequential tunneling component of the dark current by an order of magnitude. Using all these improvements, Levine et al. 2 at AT&T Bell Laboratories successfully demonstrated the first bound-tocontinuum OWIP operating at l()13m with a dramatic improvement in performance. In addition to the photocurrent, all detectors including OWIPs produce a dark current which must be minimized to achieve high performance. The dark current is the current flowing through the detector when it is in the dark (i.e. with no photons impinging on it) and is ideally zero. In most applications the total current flowing through the detector is measured and there is no way to distinguish the dark current from the photocurrent. Though this dark current can be approximately subtracted in the image-processing electronics, a high dark current implies that the detector blinds itself even when it sees no photons: when it does see photons the image-processing electronics are swamped by the dark electrons with very little capacity left to process the photoelectrons. In QWIPs, the dark current originates from three different mechanisms. The dark current arising from the first process is due to q u a n t u m mechanical tunneling of ground state electrons from well to well through the AlxGa~_xAs barriers (sequential tunneling). This process is independent of temperature. Sequential tunneling dominates the dark current at very low temperatures ( < 30 K). The second mechanism is thermally-assisted tunneling which involves thermal excitation of a ground state electron followed by its tunneling through the tip of the barrier into the continuum energy levels. This process governs the dark current at medium temperatures. The third mechanism is classical thermionic emission (the emission of electrons over a finite potential barrier due to their finite temperature) and dominates the dark current at higher temperatures (> 45 K). Reducing the dark current due to this mechanism is critical to the commercial success of the OWIP since it enables the, highly desirable, higher temperature camera operation. The thermal generation rate of electrons is determined partly by their lifetime, which in turn is determined by the thickness of the AlxGal_xAs barriers. The dark current can be reduced by increasing the barrier thickness to the point beyond which any further increase would reduce the photocurrent (by a reduction in the efficiency of collection of photoelectrons). The thermal generation rate also depends on the number of electrons (i.e., doping density) in the well. Increasing this number increases the photocurrent" unfortunately, it also increases the dark current. The authors of this chapter found that the dark

86

Handbook of Infrared Detection Technologies

current could be reduced by m a n y orders of magnitude by lowering the well doping density, without significantly reducing the performance of the detectors. 3 For a OWIP designed to detect light at a desired wavelength, there is thus an optimum barrier thickness and well doping density which will minimize the dark current without sacrificing too much in terms of lost photocurrent. Another trick to reduce the dark current due to thermionic emission, and optimize the performance of QWIP is to use b o u l l d - t o - q u a s i b o u n d intersubband absorption (occurring when the first excited state is in resonance with the top of the well) as shown in Figure 3.1.4 This transition maximizes the intersubband absorption while maintaining excellent electron transport. The major advantage of this design lies in the fact that it increases the energy barrier to thermionic emission compared to the case of the bound-to-continuum OWIP. The energy barrier to thermionic emission is the potential height from the ground state to the well top. In the case of the b o u n d - t o - c o n t i n u u m OWIP, the energy barrier to thermionic emission is several meV less than the barrier to photon absorption (since the first excited state is several meV above the well top). In the case of the bound-to-quasibound OWIP, the two barriers are equal. If the two QWIPs are designed to detect the same photon energy, the bound-to-quasibound design thus has a several meV higher energy barrier to thermionic emission than the b o u n d - t o - c o n t i n u u m OWIP, an increase which will drop its dark current significantly (i.e., Ia :x e -:~I:/kT ~ e-1 for T= 7() K). One could push the first excited state deeper into the well to increase the energy barrier to therrnionic emission

Figure 3.1 Schematic diagram of the conduction band in a bol~nd-to-quasibound (.)V~'IP in an ex'ternall!t applied electric field. Absorption of lR photons can photoe.rcite electrons.l?om the ground state of the quantum well into the continuum, causing a photocurrent. Three dark current mechanisms are also shown: ground state tunneling ( 1 ): thermall!t assisted tunneling (2): and thermionic emission ( ~ ). The inset shows a cross-section transmission electron mierograph of a QWIP samph'.

GaAs/A1GaAs based quantum well infi'ared photodetector focal plane arra!ts

87

even further, but this would drop the photocurrent to unacceptably low levels. The bound-to-quasibound QWIP preserves the photocurrent while reducing the dark current. In order to be absorbed by the electrons in the q u a n t u m wells, the incoming light should have an electric field component in the q u a n t u m well direction, i.e., in the growth direction, normal to the layers. Only in this situation is the electric field of the light coupled to the quantized electron m o m e n t u m , enabling a photon to excite an electron and get absorbed in the process. Light being a transverse wave (whose electric field is perpendicular to the direction of travel), this selection rule means that light striking the layers normally (the most direct way to illuminate an imaging array of detectors) is not absorbed. If the light is sent through the thin (,-~500 l.tm) edge of the detector it can be absorbed since it now has a component of its electric field in the correct direction. The edge is sometimes given a 45 ~ wedge and the incident light focused normally on the polished edge facet, s Both schemes result in incident light being piped laterally t h r o u g h the detector. This clearly limits the configuration of detectors to linear arrays and single elements. For imaging, it is necessary to be able to couple light uniformly to two-dimensional arrays of detectors. An elegant way to couple normally incident light to an imaging array of OWIPs is to use a grating to bend some of the light inside the detectors. This is accomplished by putting a special reflector on the detector top and illuminating the detector from the back. If the reflector is a smooth mirror, it is useless: the normally incident light passes through the detector, strikes the mirror, and is reflected straight back out of the back side with none of the light being absorbed by the confined electrons. To be useful, the mirror has to be rough (on the scale of the wavelength of the light in the detector's GaAs material). This roughness may be either periodic r or r a n d o m 7 (Figure 3.2). A rough mirror scatters or sprays the incident light back in a cone (i.e., the roughness ensures that the angle of reflection no longer equals the angle of incidence). The details of the cone depend on the details of the roughness. This cone now strikes the bottom side. Those rays that are within a critical angle of the normal (17 ~ for the GaAs-air interface) refract or escape back into the air. The rest suffer total internal reflection with the back surface acting as a smooth mirror. The internally reflected rays are once again reflected off the top rough mirror. W h a t

Figure 3.2 (a) Two-dimensional light coupling grating on ()WIP pixel. (b) OWIP pixel with a random reflector.

88 Handbookof Infrared Detection Technologies happens next depends on whether the roughness of the top mirror is periodic or random. If it is periodic, the top mirror will scatter or bend these rays so that they are all normal to the q u a n t u m well layers again. These rays pass t h r o u g h the detector and out of the back side. A randomly roughened mirror, on the other hand, will randomly reflect or scatter all the rays internally reflected on to it from the bottom side each time, thereby allowing the incident light to bounce back and forth between the detector top and back surfaces several times. Only light within a 17 ~ (from normal) cone escapes out of the back side. Clever design can reduce the amount of light in the escape cone but cannot eliminate it altogether. For instance, if the random reflector is designed with two levels of rough surfaces having the same areas but located a quarter wavelength (;.c;~A~/4) apart, the normally reflected light intensities from the top and bottom surfaces of the reflector are equal and 180 ~ out of phase. This maximizes the destructive interference at normal reflection and lowers light leakage through the escape cone. On each pass through the detector, following the first reflection off the top rough mirror, some of the light is absorbed by the confined electrons since most of the light rays now have an electric field component in the correct direction (normal to the q u a n t u m well layers). A periodically roughened mirror thus yields two useful passes; a randomly roughened mirror offers several. One measure of the resulting efficiency of the detector in absorbing photons is its responsivity (the photocurrent for a given incident photon flux). Under identical conditions, a top reflector with one-dimensional periodic roughness will give a QWIP about the same responsivity that it would have if illuminated through a 45 ~ edge facet; both the reflector with two-dimensional periodic roughness as well as the random reflector, will approximately double this responsivity. Squeezing the m a x i m u m light trapping ability from the reflector requires increasing the detector aspect ratio, i.e.. the ratio of detector diameter to height, a design feature accomplished by thinning the ,-~5()() ~m thick GaAs substrate on top of which the detector material is grown, to sub-micron level. The resulting optical cavity is calculated and measured to improve the responsivity of a OWIP with a random reflector significantly over the 4 5 ~ case. s Thinning enhances the responsivities of the OWIPs with 1D and 2D periodically rough reflectors to be respectively about 2 and 4 times better than the 4"5~ case. The random reflector is fabricated on the OWIPs using standard photolithography and selective dry etching. The advantage of the photolithographic process over a completely random process is the ability to accurately control feature size and preserve the pixel-to-pixel uniformity necessary for very sensitive imaging focal plane arrays. Due to this high performance and the excellent uniformity of GaAs-based OWIPs, several groups have demonstrated two-dimensional long wavelength IR (LWIR) imaging arrays ~-2s operating at temperatures up to 77K. Infrared imaging systems that work in the 3 - 1 6 ~ m band have many applications. including night vision, surveillance, navigation, flight control, early warning systems, 26 Earth and planetary remote sensing, 2" industrial product quality control, etc. Furthermore, this technology has introduced novel thermographic techniques into medicine, which is clearly evident by the recent approval of a

GaAs/A1GaAs based quantum well infrared photodetector focal plane arrays

89

OWIP-based breast cancer detection instrument by the US Food and Drug Administration, for clinical use.

3.2 Detectivity D* comparison The blackbody detectivity D~ is basically the signal-to-noise ratio of a radiation detector normalized to unit area and operating bandwidth of the detector is given by, Ax/-A-~-f

(1)

i. with RB = ~I21R(2)W(2)d2

(2)

~/'IW (2) d2 where the responsivity R, can be written in terms of q u a n t u m efficiency and photoconductive gain g as (3)

a = (e/hv)o,,g

The photoconductive gain of QWIps can be written as (4)

g=L/l

where L is the hot electron mean free path and 1 is the thickness of the MOW region, and the temporal noise current in of a single element radiation detector is given by,

(5) where e is the charge of an electron, fl=2 for a photovoltaic detector (generation only) and fl=4 for a photoconductor (generation and recombination), the phoconductive gain #=1 for a photovoltaic detector and g(QWIP) is typically 0 . 2 - 0 . 5 (depends on the device structure). Ix~is the detector dark current and Ip, the detector photocurrent, is given by Ip = e~7,~g*A

(6)

where ~ is the photon flux. (See references 18 and 2 5 for details) and (dPR/dT) is the change in the incident integrated blackbody power in the spectral range of detector with temperature. The integrated blackbody power PR, in the spectral range from k I to k2, can be written as PB - Asin 2

cos4>

'J).l

W(2)d2

(7)

90

Handbook of Infrared Detection Technolo~lies

where 0, 4~, and W (2) are the optical field of view, angle of incidence, and blackbody spectral density respectively, and are defined by equations ( 8 ) and (9)

P(2) - W(2)sin2 (O/2 )AFcos(/)

(8)

where A is the detector area, 4~ is the angle of incidence, 0 is the optical field of view angle (i.e., sin 2 (0/2)=(4f2+1) --1 where f is the f number of the optical system; in this case 0 is defined by the radius p of the blackbody opening at a distance D from the detector, so that tan(0/2)=p/D), F represents all coupling factors and F = TI(1 - r)C where Tt is the transmission of filters and windows, r--28% is the reflectivity of the GaAs detector surface, C is the optical beam chopper factor (C=0.5 in an ideal optical beam chopper), and W(2) is the blackbody spectral density given by the following equation (i.e., the power radiated per unit wavelength interval at wavelength ~ by a unit area of a blackbody at temperature TR).

W().) - (27rc2h/)~ 5) (e hc/2kT~ - 1) -1

(9)

Let us consider a background limited condition. At this condition

ID < Ie

( 1 O)

By combining equations (1). (3), (4). (5), (6). and (9) the detectivity D* can be written as,

D* - G 1 v/rla ~,

(11)

Thus, D*IDEAL

/

D*OWlp

/ 2 r/a,IDEAL V/ r/~;a-~IP

(12)

The lowest absorption q u a n t u m efficiency (1/,,) of QWIP is typically 15% (including 30% reflection loss). The ~/,, of an ideal detector is 70% (assume 30% reflection loss). Thus equation (13) reduces to ~

OIDEAL D*o_wlP

=2.67

(13)

Thus, this analysis clearly shows the photoconductive gain is irrelevant at background limited performance (BLIP) condition, and therefore, the detectivities scales solely as a function of absorption q u a n t u m efficiencies of the detectors.

GaAs/AIGaAs based quantum well infrared photodetector focal plane arra!ls 91

3.3 Effect of nonuniformity The general figure of merit that describes the performance of a large imaging array is the noise equivalent temperature difference (NEAT). NEAT is the m i n i m u m temperature difference across the target that would produce a signalto-noise ratio of unity and it is given by 1r 17 NEAT -

Ax/-A-Af

(14)

D*R(dPn/dT)

Before discussing the array results, it is also important to understand the limitations on the FPA imaging performance due to pixel nonuniformities. ~s The total noise I,, of a focal plane array is given by,

i,e, - z,'; +

+ I,,) 2

(15)

Where u is the nonuniformity of the focal plane array, is given by, (7

(9"

#

Ip + ID

(16)

Where # is the mean total signal and a is the standard deviation of the histogram of total signal versus n u m b e r of pixels. Now the focal plane array detectivity or NEAT can be obtained by following equations,

DI~I,A =

(17)

In

NEATFpA =

v/AB D;,pa(dPR/dr)

(18)

Where I,, is the total noise of the focal plane array and it is given by equation (14). The figures of merit such as D*, NEAT, NEP, NEI, etc. are different representations of the basic signal-to-noise ratio of radiation detectors normalized in different ways. The signal-to-noise ratio of a focal plane array can be written as,

SNR -

IZ -

I,

Ie

V/i + ,,a (I,, + I1,)2

(19)

Under background limited condition (use equation (12)) this reduces to, SNR ~-

1 U

(20)

92 Handbookof Infrared Detection Technoloqies

This analysis clearly shows the importance of the array uniformity in the focal plane array total signal-to-noise ratio. This point has been discussed in detail by Shepherd 19 for the case of PtSi infrared FPAs 2(~ which have low response, but very high uniformity. The general figure of merit to describe the performance of a large imaging array is the noise equivalent temperature difference NEAT, including the spatial noise which has been derived by Shepherd, 2~Jand given by NEAT -

N,, dNb/dTb)

(21)

where Tb is the background temperature, and N,, is the total number of noise electrons per pixel, given by -

N

+

+

N, f

(22)

The photoresponse independent temporal noise electrons is Nt, the shot noise electrons from the background radiation is N~,, and residual nonuniformity after correction by the electronics is u. The temperature derivative of the background flux can be written to a good approximation as dNb = hcNb drb k)or 2

(2 3)

b

where )~=(),1+22)/2 is the average wavelength of the spectral band between 2a and ,<2. When temporal noise dominates. NEAT reduces to equation (14). In the case where residual nonuniformity dominates, equations (21) and (23) reduce to NEAT -

u~T 2 b 1.44

(24)

The units of the constant is cm K, 7. is in cm and Tt, is in K. Thus, in this spatial noise-limited operation NEATvcu and higher uniformity means higher imaging performance. Levine is has shown as an example, taking Tt,=300 K, 2=10 pm, and u-0.1% leads to NEAT=63 mK, while an order of magnitude uniformity improvement (i.e., u=O.Ol%) gives NEAT-6.3 mK. By using the full expression equation (11) Levine is has calculated NEAT as a function of D* as shown in Figure 3.3. It is important to note that when D*>~IO 1(~ cmv/Hz/W, the performance is uniformity limited and thus essentially independent of the detectivity, i.e., D* is not the relevant figure of merit. 2

3.4 640x512 pixel long-wavelength portable QWIP camera In order to detect LWIR radiation, we have designed the following MQW structure. Each period of the multi-quantum well (MQW) structure consists of a

GaAs/A1GaAs based quantum well infrared photodetector focal plane arrays

,oolL \

::X) ~" \ I t IL

,

;k

= lOpm

93

,

U-- 10 .3

u = 10 .4 ,,,

10 9

10~o

~

1011

1012

DETECTIVITY (cmd'--Hz/W)

Figure 3.3 Noise equivalent temperature difference NEAT as a function of detectivity D*. The effects of nonuniformity are included for u - 1 O- ~ and 10 4. Note that for D* > 101~ cm v/Hz/W detectivity is not the relevant figure of merit for FPAs. (Taken from ref. 18).

45 A well of GaAs (doped n - 5 • 1017 c m - ~) and a 500 A barrier of Alo. 3Gao.7As. Stacking m a n y identical q u a n t u m wells (typically 50) together, increases photon absorption. Ground state electrons are provided in the detector by doping the GaAs well layers with Si. This photosensitive MOW structure is sandwiched between 0 . 5 p m GaAs top and bottom contact layers doped n - 5 x l ( ) ]7 cm -3, grown on a semi-insulating GaAs substrate by molecular beam epitaxy (MBE). Then a 0.7 pm thick GaAs cap layer on top of a 300 A AI().3Ga().rAs stop-etch layer was grown in s i t u on top of the device structure to fabricate the light coupling optical cavity. The MBE grown material was tested for absorption efficiency using a Fourier Transform Infrared (FTIR) spectrometer. Figure 3.4 shows the measured absorption q u a n t u m efficiency of this material at room temperature. The epitaxially grow material was processed into 200 pm diameter mesa test structures ( a r e a = 3 . 1 4 • 1() -4 cm 2) using wet chemical etching, and Au/Ge ohmic contacts were evaporated onto the top and bottom contact layers. The detectors were back illuminated t h r o u g h a 45 ~ polished facet s and a responsivity spectrum is shown in Figure 3.5. The responsivity of the detector peaks at 8.5 pm and the peak responsivity (Rp) of the detector is 83 mA/W at bias VB -- --1.1 V. The spectral width and the cutoff wavelength are A 5 . / 5 . - 1()% and ,;.,.- 8 . 9 p m respectively. The measured absolute peak responsivity of the detector is small, up to about V R - - 0 . 5 V. Beyond that it increases nearly linearly with bias reaching Rp - 4 2 0 mA/W at V13 - - 5 V. This type of behavior of responsivity versus bias is typical for a bound-to-quasibound QWIP. The peak q u a n t u m efficiency was 1.4% at bias VR - - 1 . 1 V for a 45 ~ double pass. The lower q u a n t u m efficiency is due to the lower photoconductive gain at lower

94

Handbook of Infrared Detection Technologies 12 10 ILl O

8

e.

.2 L

6

O

"

4

6

7

8

9

10

11

W a v e l e n g t h (~Lm)

Figure ].4 Absorption quantmn e .~cienc!l o.lthe Qi~'II" material at room temperature.

0.1

B i a s = -1.1 V

A

<: ~.,

, m

(n 'o r u) (1)

0.08 0.06 0.04 0.02

-

0 6

7

8

Wavelength

9

10

(micron)

Figure 3.5 Responsivity spectrmn of a bound-to-quasibozind I.i~IR QiI'II ~ test structure at temperature T = 7 7 K. The spectral response peak is at Y,. ~ 1~m and the lon[l waveh, n#th t'lit-ofli.~ at 8.9/3 m.

operating bias, which is required to suppresses the dark current. Due to lower readout multiplexer well depth {i.e.. 1 1 x 1 ()~' electrons) a lower dark current is m a n d a t o r y to achieve a higher operating temperature and longer integration times. In BLIP condition the noise equivalent temperature difference NETD improves with the increasing integration time. In this case, the lowest operating temperature of 6 5 K was determined by the cooling capacity of the small Sterling cooler used in an Indigo Phoenix @ camera. The peak detectivity is defined as D~,- Rpx/-A-B/i,,, where Rp is the peak responsivity, A is the area of the detector and A = 3.14 x 1 ()-4 cm 2. The measured peak detectivity at bias V B = - I . 1 V and temperature T = 6 5 K is l x l ( ) ~ cmv/Hz/W. Figure 3.6 shows the bias dependence of peak detectivity as a function of temperature. These detectors show BLIP at bias V~ = - 1 . 1 V and temperature T = 72 K for 3()() K background with f~2 optics. Although r a n d o m reflectors have achieved relatively high q u a n t u m efficiencies with large test device structures, it is not possible to achieve the similar high q u a n t u m efficiencies with random reflectors on small focal plane array pixels due to the reduced width-to-height aspect ratios. In addition, it is

GaAs/A1GaAs based qttantum well infrared photodetector focal phme arra!ls

95

1.E+12 ~" 1.E+11 3 N ~:

1.E§

1 .E+09

-

60

300 K f/2

Background Optics Bias

-1.1 V

70

80

90

Temperature (K) Figure 3.6 Detectivit!t as a function of temperat ures at bias of 1.2 V.

difficult to fabricate r a n d o m reflectors for shorter wavelength detectors relative to very long-wavelength detectors (i.e., 1 5 [tm) due to the fact that feature sizes of r a n d o m reflectors are linearly proportional to the peak wavelength of the detectors. For example, the m i n i m u m feature size of the r a n d o m reflectors of 15 ~m cutoff and 9 btm cutoff FPAs were 1.2 5 and ().6 l~m respectively and it is difficult to fabricate sub-micron features by contact photolithography. As a result, the r a n d o m reflectors of the 9 ~m cutoff FPA were less sharp and had fewer scattering centers compared to the random reflectors of the 1 5 ~m cutoff QWIP FPA. Alternatively, r more IR light can be coupled to the OWlP detector structure by incorporating a two-dimensional periodic grating surface on top of the detectors which also removes the light coupling limitations and makes twodimensional OWIP imaging arrays feasible. This two-dimensional grating structure was fabricated on the detectors by using standard photolithography and CCI2F2 selective dry etching. After the 2D grating array was defined by the lithography and dry etching, the photoconductive OWIPs of the 64()x 512 FPAs were fabricated by wet chemical etching through the photosensitive GaAs/AlxGa~_xAs m u l t i - q u a n t u m well layers into the 0.5 ~tm thick doped GaAs bottom contact layer. The pitch of the FPA is 2 5 l.tm and the actual pixel size is 2 3 • 2 3 btme. The 2D gratings on top of the detectors were then covered with Au/Ge and Au for Ohmic contact and reflection. Figure 3.7 shows twelve processed OWIP FPAs on a three-inch GaAs wafer. Indium bumps were then evaporated on top of the detectors for Si readout circuit (ROC) hybridization. A single ()WIP FPA was chosen and hybridized (via indium bump-bonding process) to a 64()• 512 CMOS multiplexer (ISC 980 3) and biased at V/~ - - 1 . 1 V. At temperatures below 72 K, the signal-to-noise ratio of the system is limited by array non-uniformity, multiplexer readout noise, and photo current (photon flux) noise. At temperatures above 72 K, temporal noise due to the OWIP's higher dark current becomes the limitation. As mentioned earlier this higher dark current is due to thermionic emission and thus causes the charge storage capacitors of the readout circuitry to saturate. Since the QWIP is a high impedance device, it should yield a very high charge injection coupling

96 Handbook of Infrared Detection Technologies

Figure 3.7 Tweh,e 6 4 0 x ~ 12 Oi~'IP focal plane arra!!s on a three-inch GaAs wr

efficiency into the integration capacitor of the multiplexer. In fact, Bethea et al. ~ have demonstrated charge injection efficiencies approaching 9()%. Charge injection efficiency can be obtained from. '~

gmRDet r]inj -- I --1-gmRDet

1+

,1

j~C'DetRD,,t

where g,,, is the transconductance of the MOSFET and it is given by g,,, - elD,.t/kT. The differential resistance Rl2,.t of the pixels at - 1.1 V bias is 4.3 • 1 () 1(~ ohms at T - 65K and detector capacitance Ci~,.t is 3.()• l()-14 F. The detector dark current IDet = 1 . 5 pA under the same operating conditions. According to equation (25) the charge injection efficiency rli,,i - 9()% at a frame rate of 30 Hz. The FPA was back-illuminated through the fiat thinned substrate membrane (thickness ..~1300 A). This initial array gave excellent images with 99.92% of the pixels working (number of dead pixels ~2 5()), demonstrating the high yield of GaAs technology. The operability was defined as the percentage of pixels having noise equivalent differential temperature less than 1 ()()mK at 300 K background and in this case operability happens to be equal to the pixel yield.

GaAs/A1GaAs based quantum well infrared photodetector focal plane arraz3s 97

We have used the following equation to calculate the noise equivalent temperature difference NETD of the FPA. NETD -

v/AB D*B(dPB/dT)sin2(0/2)

(26)

where D~3 is the blackbody detectivity, dP~/dT is the derivative of the integrated blackbody power with respect to temperature, and 0 is the field of view angle (i.e., sin2(0/2)=(4f2+ 1)-1, where f is the f number of the optical system). Figure 3.8 shows the NETD of the FPA estimated (using equation (26)) from test structure data as a function of temperature for bias voltages Vl3 = - 1.1 V. The background temperature TI3 = 300 K, the area of the pixel A=(2313m) 2, the f number of the optical system is 2, and the flame rate is 30 Hz. Figure 3.9 shows the measured NETD of the FPA at an operating temperature of T=65 K, 16 ms integration time, bias V R = - I . 1 V for 300 K background with f/2 optics and the mean value is 24 mK. This agrees reasonably with our estimated value of 10 mK based on test structure data. The data taken from a test setup has shown mean NETD < 10 mK (the higher NETD of the portable imaging camera is due to the system noise, which includes the electronic noise and the Stirling cooler noise) at an operating temperature of T=65K and bias V ~ = - I . 1 V. for a 3OOK background. The uncorrected photocurrent non-uniformity (which includes a 1% non-uniformity of the ROC and a 1.4% non-uniformity due to the cold-stop in front of the FPA not yielding the same field of view to all the pixels) of the 327 680 pixels of the 6 4 0 x 5 1 2 FPA is about 14% {=sigma/mean). The non-uniformity after twopoint (17 ~ and 27 ~ Celsius) correction improves to an impressive 0.3%. It is worth noting that these data were taken from the first 6 4 0 x 512 OWIP FPA which we produced. Thus, we believe that there is plenty of room for further improvements of these FPAs. The net peak q u a n t u m efficiency of the FPA was 1.4% (lower focal plane array q u a n t u m efficiency is attributed to lower photoconductive gain at lower operating bias) and this corresponds to an 1 .E+03 Background Optics Bias

~" 1 . E + 0 2 E

300K

f/2

-1.1 V

~

/

~'~

I

'

tint

121

uJ Z 1.E+01 -

1 .E+00

I

I

60

70

1

80

Temperature (K)

90

Figure 3.8 Noise equivalent temperature difference NETI) estimated fi'om test structure data as a filnction of temperature for bias voltage V B = - 1.1 V. The background temperature TB= 300 K and the area of the pixel A=(23 i~m) 2.

98

Handbook of Infrared Detection Tecllnologies

60,000 50,000 40,000 E

30,000 20,000 10,000

0

I 10

1 20

1 30

k 40

I 50

I 60

I 70

I 80

I 90

NEDT (mK) Figure J. 9 NETD histogram of the 32 7 6 8 0 pixels of the 640 • ~ 12 arra!/ showing a high uni/brmit!l of the FPA. The uncorrected non-uniformit!l (-standard deviation ~mean) ~/the FPA is onl!l 14% including 1% nonuniformit!l of ROC and 1.4"/,, non-un(formit!l due to the cold-stop not being abh' to give the same.held o.f view to all the pixels in the FPA.

average of three passes of IR radiation (equivalent to a single 45: pass) through the photosensitive multi-quantum well region. It is worth noting that, under BLIP conditions, the performance of the detectors is independent of the photoconductive gain. and it depends only on the absorption q u a n t u m efficiency. A 640• OWIP FPA hybrid was mounted onto a 330 mW integral Sterling closed-cycle cooler assembly and installed into an Indigo Phoenix @ camerabody, to demonstrate a hand-held LWIR camera (shown in Figure 3.10). The Phoenix | infrared camera system has been developed by Indigo Systems Corporation to meet the needs of the research, industrial and ruggedized OEM communities. The system is comprized of a camera head and a selection of two video processing back ends. The camera head was made of Indigo's standard 6 4 0 • 512 format readout ISC 980 3. mated to long-wavelength OWIP detector materials. Two video processing units are the Real Time Imaging Electronics (RTIE) that provide conventional NTSC video as well as corrected parallel digital video out at video rates, and the Digital Acquisition System (DAS) that provides high-speed (40 MHz) raw digital data acquisition and output with limited real time video for system setup and focusing. The other element of the camera is a 100 mm focal length germanium lens, with a 5.5 degree field of view. It is designed to be transparent in the 7-1413m wavelength range, to be compatible with the QWIP's 8.:5 Bm operation. The digital acquisition resolution of the camera is 14-bits, which determines the instantaneous dynamic range of the camera (i.e., 16 384). However, the dynamic range of OWIP is 85 decibels. Its nominal power consumption is less than 4 "5watts.

GaAs/A1GaAs based quantmn well in l?ared photodetector focal phme arra!ls

99

Figure :~. 10 Picture of the 640 x 512 hand-held long waveh'ngth O I~'IP camera ( QWIP Phoenix~ ).

Video images were taken at a frame rate of 30 Hz at temperatures as high as T - 70 K, using a ROC capacitor having a charge capacity of 11 • 1()~ electrons (the m a x i m u m number of photoelectrons and dark electrons that can be counted in the time taken to read each detector pixel). Figure 3.11 shows one flame of a video image taken with a 9 ~m cutoff 64()x "512 pixel QWIP Phoenix | camera. The focal plane array performance data reported in this paper was taken with a laboratory measurement setup and OWIP Phoenix ~ camera. Estimates based on the single pixel data show that these FPAs should be able to provide 7 mK NETD with 30 ms integration time, which can be achieved at l. ] V bias. Focal planes were integrated with the close cycle stirling coolers, then these FPA/dewar assemblies were integrated with Phoenix ~ camera electronics. Thus we expect the OWIP Phoenix | camera to achieve a performance of NETD < 2 0 m K with shorter integration time ( < 16 ms with high bias), and a performance of NETD < l OmK with longer integration time at 30 Hz (30 ms integration with low bias).

3.5 640• 486 long-wavelength dual-band imaging camera The LWIR and very long-wavelength infrared (VLWIR) dualband QWIP device structure described in this section processed into interlace simultaneously readable dualband FPAs (i.e., odd rows for one color and the even rows for the other color). 26 The device structure consists of a 30 period stack, of VLWIR QWIP structure and a second 18 period stack of LWIR OWIP structure separated by a heavily doped O. 5 ~m thick intermediate GaAs contact layer. The first stack (VLWIR) consists of 30 periods of a 500 A AlxGal_xAs barrier and a 60 A GaAs well. Since the dark current of this device structure is dominated by the longer wavelength portion of the device structure, the VLWIR OWIP structure has been designed to have a bound-to-quasibound intersubband absorption 2s peak at 14.5 Bm. The second stack (LWIR) consists of 18 periods of a 500 A AlxGa~_xAs

100 Handbook of Infrared Detection Technologies

Figure 3.11 Oneframe of video image taken with the 9 I~m cutqfl640 x ~ 12 pixel O.WIP Phoenix ~ camera.

barrier and a n a r r o w 4 0 A GaAs well. This LWIR OWIP structure has been intersubband absorption peak at designed to have a b o u n d - t o - c o n t i n u u m 8.5 ~m, since photo current and dark current of the LWIR device structure is relatively small compared to the VLWIR portion of the device structure. This whole dualband OWIP structure is then sandwiched between (). 5 ~m GaAs top and bottom contact layers doped with tl= 5 • 1 ()1 - c m - 3, and has been grown on a semi-insulating GaAs substrate by MBE. Then a 3()()A AI~.3Ga~).7As stop-etch layer and a 1.0~tm thick GaAs cap layer were grown in situ on top of the device structure. GaAs wells of the LWIR and VLWIR stacks were doped with n - 6 • and 2 . 5 • cm -~ respectively. All contact layers were doped to n=5 • 1017 cm -3. The GaAs well doping density of the LWIR stack was intentionally increased by a factor of two to compensate for the reduced number of q u a n t u m wells in the LWIR stack. 2~ It is worth noting that the total current (dark c u r r e n t + p h o t o current) of each stack can be independently controlled by carefully designing the position of the upper state, well doping densities, and the number of periods in each MOW stack. All of these features were utilized to obtain approximately equal total currents from each MQW stack. Simultaneously measured responsivity spectrums of these vertically integrated dualband OWIPs are shown in Figure 3.12. Based on single element test detector data, the LWIR detectors show BLIP at bias VR=-2 V and temperature T - 72 K for a 300 K background w i t h f / 2 cold stop. The VLWIR detectors show BLIP under the same operating conditions at 45 K operating temperature. 28 Two different 2D periodic grating structures were designed to independently couple the 8 - 9 and 14-1 5 ~m radiation into detector pixels in even and odd rows of the FPAs. The FPA fabrication process is described elsewhere. 26 These dualband FPAs were tested at a background temperature of 300 K. with f / 2 cold stop, and at 30 Hz frame rate. As expected (due to BLIP), the estimated and experimentally obtained NEAT values of the LWIR d e t e c t o r s do not change

CaAs/AIGaAs based quantum well infrared photodetector focal plane arra!ts

101

0.6 VBIAS = 2V

0.4 v >I> cO Z

o

13_ 09

uJ cr

0.2

5

6

7

I

8

9

10

11

12

13

14

15

16

17

WAVELENGTH (#m)

Figure 3.12 Simultaneousl!l measured responsivit!! sl~ectrmn of verticall!! integrated LWIR and VLWIR dualband OWIP detector (Taken from reference 26 ).

significantly at temperatures below 65 K. The estimated NEAT of LWIR and VLWIR detectors at 40K are 36 and 4 4 m K respectively (See Figures 3.13 and 3.14). These estimated NEAT values based on the test detector data agree reasonably well with the experimentally obtained values. The experimental LWIR NEAT value is lower than the estimated NEAT value of 36mK. This improvement is attributed to the 2D periodic grating light coupling efficiency. On the other hand the experimental VLWIR NEAT value is higher than the estimated NEAT value of 44 mK. The authors believe this degradation is due to the inefficient light coupling at the 14-15 l~m region, readout multiplexer noise,

Figure 3.13 The uncorrected noise equivalent temperature d(ffl'rence (NEAT) histogram of 8-9 l~m detector pixels of the 640 • 486 dualband FPA. The mean NEAT is 29 mK.

102

Handbook of Infrared Detection Technologies

Figure 3.14 The uncorrected noise equivalent temperature tti!ti'rem'e (NEAT) histogram of l 4 - 1 5 ltm detector pixels of the 640 • 4 8 6 dtmlband FPA. The mean .\'EAT is 74 inK.

and noise of the proximity electronics. At 4() K the performance of both LWIR and VLWIR detector pixels of this dualband FPA are limited by photo current noise and readout noise. Video images were taken at a frame rate of 3() Hz. at temperatures as high as T= 74 K, using a ROC capacitor having a charge capacity of 9 • 1 ()~ electrons (the m a x i m u m n u m b e r of photoelectrons and dark electrons that can be counted in the time taken to read each detector pixell. Figure 3.1 5 shows simultaneously acquired 8 - 9 and 14-1 5 micron images using this two-color camera. 2s

3.6 640•

pixel broad-band QWIP imaging camera

A broadband QWIP device structure is designed by repeating a unit of several q u a n t u m wells with slightly different parameters such as well width and barrier height. 27 The positions of ground and excited states of the q u a n t u m well are determined by the q u a n t u m well width (L,,,) and the barrier height, i.e. the AI mole fraction (x) of the barrier 2- Since each single set of parameters for a boundto-quasibound q u a n t u m well 4 corresponds to a spectral band pass of about 1.5 ~m, three different sets of values are sufficient to cover a 1 ()-16 ~m spectral region. As shown in Figure 3.16. the MOW structure consists of m a n y periods of these three-quantum-well units separated by thick barriers. The device structure reported here involved 3 3 repeated layers of GaAs threequantum-well units separated by L t ~ 5 7 5 A thick AlxGa~ xAs barriers (See Figure 3.9). The well thickness of the q u a n t u m wells (see Figure 3.16) of threequantum-well units are designed to respond at peak wavelengths around 1 3, 14, thick and 151um respectively. These wells are separated by Lz, ~ 7 5 A AlxGal_xAs barriers. The AI mole fraction (xl of barriers t h r o u g h o u t the

GaAs/A1GaAs based quantum xvell itll)ared photodetector Jbcal plane arra!ls

1() 3

Figure :~. 15 Both pictures showy (flame-simHltaneousl!l acquired) t~vo-color images ~,i~11the 640 • 486 t~vocolor OWIP camera, hnage on the h'ft is from 14-1 ~ 12m infrared and the image on the right isJ?om 8-9 lxm infrared. Pix'el pitch of the FPA is 2 ~ 12m. The 14-1 ~ micron image is less sharp due to the difJ?action limited spot size being larger than the pix'el pitch of the FPA ( Taken l?om r~:ference 28).

Figure 3.16 Broad-band MOW structz~re is designed b!t repeating a zlnit of several quantum ~vells ~vilh slightl!l different parameters slwh as ~rell ~vidth and barrier height. The excited state energ!l levels broadened due to overlap of the ~vavefunctions associated ~vith excited states ofqlumtum ~vells separated b!l thin barriers.

structure was chosen such that the /.p=l 3 l~m quantum well operates under bound-to-quasibound conditions. ~J The excited state energy level broadening has been further enhanced due to the overlap of the wavefunctions associated with excited states of quantum wells separated by thin barriers. Energy band calculations based on a two-band model show excited state energy levels spreading about 2 8 meV.

104

Handbook of Infrared Detection Technologies

The sample was grown on a semi-insulating three-inch GaAs substrate by molecular beam epitaxy. It consists of the device structure described above sandwiched between top and bottom contact layers. Transport carriers (electrons) were provided by doping all GaAs wells and contact layers with Si. In order to measure the dark c u r r e n t - t e m p e r a t u r e curve, spectral responsivity (see Figures 3 . 1 7 - 3 . 1 9 ) and noise, 200 ~tm diameter mesas were fabricated using wet chemical etching and Au/Ge ohmic contacts were evaporated onto the top and bottom contact layers. The responsivity spectra of these detectors were measured using a 1()00 K black-body source and a grating monochromator. The detectors were back illuminated through a 45 ~ polished facet to obtain normalized responsivity spectra at different bias voltages. Then the absolute spectral responsivities were obtained by measuring total photocurrent from a calibrated black-body source. In Figure 3.18, responsivity curve at V j ~ = - 2 . 5 V bias voltage shows broadening of the spectral response up to A;. ~ 5.5 Bm, i.e. the full width at half

1 .E-06 "~ 1.E-07

E r (9 L

,-

1.E-08

1.E.09 ~-

O

~" 1.E-10~ r~

1 .E-11 1 .E-12

30

35

40

45

50

55

60

65

70

T e m p e r a t u r e (K)

Figure 3 . 1 7 Dark current verses temperature curve of 1 O- 1 5.4 lira broadband O_WIP at bias \:l~= - 2 . 5 V. Data were taken with a 2 0 0 ~tm diameter test device and normalized to 2 S • 2 S u m 2 pix'el.

0.5 0.4

~o. E

0.3 0.2 0.1

8

10

12

14

16

Wavelength (micron)

Figure 3.18 Responsivity spectrum of a broadband OIVIP test device at temperature T= 5 5 K. The spectral response peak is at 13.5 l~m and the long wavelength cutolJis at 1 ~.4 /2m.

GaAs/AIGaAs based quantum well infrared photodetector focal plane arrays

105

0.6

A

0.4

n

Iz

0.2

0~.

-v

0

1

2

3

4

Bias Voltage (V) Figure 3.19 Peak responsivit!t as a function of bias voltage at temperature T-- :~ ~ K.

m a x i m u m from 1 0 . 5 - 1 6 gm. This b r o a d e n i n g A2/2p ,-~42% is about a 4 0 0 % increase compared to a typical b o u n d - t o - q u a s i b o u n d QWIP. 4 Unlike n a r r o w - b a n d OWIPs, these detectors show spectral peak shifts from ,~ - 11.5 jJm to ,~. - 15.1 gm as negative bias voltage increased from VR - - 1 V to VB - - 5 V and similar behavior (~. - 11.5 IJm to ,~. - 14.7 IJm for VR - + 1 V to VB - + 5 V) was observed u n d e r positive bias voltages as well. This suggests that there is no substantial carrier depletion due to the applied electric field within the t h r e e - q u a n t u m - u n i t s because the direction of peak shift r e m a i n s the same u n d e r both positive and negative biases. The responsivity of the detector peaks at 13.5 l~m and the peak responsivity (Rp)of the detector is 2 5 0 m A / W at bias VR - 2 . 5 V. The bias dependent peak responsivity of the detector is s h o w n in Figure 3.19. The m e a s u r e d absolute peak responsivity of the detector is small, up to about VB--0.5 V. Beyond t h a t it increases nearly linearly with bias r e a c h i n g Re - 580 m A / W at V B = - 3 . 5 V. This type of behavior ofresponsivity versus bias is typical for a b o u n d - t o - q u a s i b o u n d OWIP. The peak q u a n t u m efficiency was 11% at bias V~--2.5 V for a 45 ~ double pass. The lower q u a n t u m efficiency is due to the lower well doping density (2 x 10 ~7 c m - 3) as it is necessary to suppress the dark c u r r e n t at the highest possible operating t e m p e r a t u r e . A peak q u a n t u m efficiency as high as 25% has already been achieved with regular well doping density (i.e., 1• cm-3). Due to lower readout multiplexer well depth (i.e., 11 • electrons) a lower dark c u r r e n t is m a n d a t o r y to achieve a higher operating t e m p e r a t u r e . In this case, the highest operating t e m p e r a t u r e of 45 K was determined by the well depth of the readout multiplexer. The dark c u r r e n t noise i,, of the device was m e a s u r e d using a s p e c t r u m analyzer at f - 55 K as a function of bias voltage. The noise gain g,, can n o w be obtained using the g - r noise calculated based on s t a n d a r d noise expression" i,, = x/4eI~g, AB w h e r e In is the dark c u r r e n t and AB is the b a n d w i d t h . Using experimental m e a s u r e m e n t s of noise and responsivity, one can n o w calculate specific detectivity D* form D * - R~/-A~/i,,, w h e r e A is area of the detector. Calculated D* value for the present device (,~. - 15.4 l~m) at, f - 55 K, and VR = 2.5 V is 3 x 10 l~ c m v ~ z / W . Even with broader response, this D* is c o m p a r a b l e to

106 Handbook of Infrared Detection Technoloqies previously reported D* of OWIPs with narrow spectral response. Figures 3.20 and 3.21 show the detectivity D* and the noise equivalent temperature difference (NEAT) as a function of the operating temperature of the device. It is well known that QWIPs do not absorb radiation incident normal to the surface unless the IR radiation has an electric field component normal to the layers of superlattice (growth direction) s. As we have discussed before s m a n y more passes of IR light inside the detector structure can be obtained by incorporating a randomly roughened reflecting surface on top of the detectors which also removes the light coupling limitations and makes two-dimensional OWIP imaging arrays feasible. This random structure was fabricated on the detectors by using standard photolithography and CCI,F, selective dry etching. After the r a n d o m reflector array was defined by the lithography and dry etching, the photoconductive OWIPs of the 64()• 512 FPAs were fabricated by dry etching through the photosensitive GaAs/AlxGa~_xAs m u l t i - q u a n t u m well layers into the 0.5 l,tm thick doped GaAs bottom contact layer. The pitch of the FPA is 2 5 l.tm and the actual pixel size is 2 3 x 2 3 lain2. The random reflectors on

1.E+12

A

1.E+11 N

-IE

~ 1.E+10

1 .E+09 30

40

50

60

70

T e m p e r a t u r e (K)

Figure 3.20 Detectivity as a fimction o.f temperatures at bias volta#e V~=-2. :~V.

1.E+04 1 .E+03 E a

uJ z

1.E+02 1.E+01 1.E+00 I 30

, 40

, 50

, 60

70

Temperature (K)

Figure 3.21 Noise equivalent temperature difference as a fi~nction o[temperatures at bias volta#e VB=-2.

~ V.

CaAs/A1GaAs based quantum well infrared photodetector focal plane arralls

1()7

top of the detectors were then covered with Au/Ge and Au for Ohmic contact and reflection. Figure 3.7 shows twelve processed OWIP FPAs on a three-inch GaAs wafer. Indium bumps were then evaporated on top of the detectors for Si readout circuit (ROC) hybridization. A single OWIP FPA was chosen and hybridized (via indium bump-bonding process) to a 64()x512 CMOS multiplexer (Indigo Systems 9803) and biased at VR = - 2 . 5 V (see Figure 3.22). At temperatures below 48 K, the signal-to-noise ratio of the system is limited by array nonuniformity, multiplexer readout noise, and photo current (photon flux) noise (see Figure 3.23). At temperatures above 48 K, temporal noise due to the OWIP's higher dark current becomes the limitation. As mentioned earlier this higher dark current is due to thermionic emission and thus causes the charge storage capacitors of the readout circuitry to saturate. Since the OWIP is a high impedance device, it should yield a very high charge injection coupling efficiency into the integration capacitor of the multiplexer. In fact Bethea et al. ~ have demonstrated charge injection efficiencies approaching 90%.

Figure 3.22 .4 size comparison of the 640 • ~ 12 long-wavelength QWIP FPA to a quarter.

30K

1.0E-07 -

35K

Q. E

"'Jr- 40K

,=I: 1.0E-09 -

Y.

c

o 'o L_

] OE-]] -

45K 50K

.

- e - - 55K

1.0E-]g 1.5

2

2.5

3

3.5

4

photo

Bias Voltage (V) Figure 3.23 Dark-current and the photo-current of 10-15.412m broadband QWIP as a fi~nction of bias voltage.

lO8 Handbookof InfraredDetectionTechnoloqies Charge injection efficiency can be obtained from equation (2 5 ), where g,,, is the transconductance of the MOSFET and is given by g,,, - eli)et/kT. The differential resistance RDet of the pixels at - 2 V bias is 5 . 4 • l~ ohms at T - 4 5 K and detector capacitance CDet is 1.4 • 1 O-14 F. The detector dark current IDet - 8 pA under the same operating conditions. According to equation (1) the charge injection efficiency rli,,i-99.5% at a frame rate of 30 Hz. The FPA was backilluminated t h r o u g h the flat thinned substrate membrane (thickness ~ 1 3 0 0 A). This thinned GaAs FPA membrane has completely eliminated the thermal mismatch between the silicon CMOS readout multiplexer and the GaAs based QWIP FPA. Basically, the thinned GaAs based QWIP FPA membrane adapts to the thermal expansion and contraction coefficients of the silicon readout multiplexer. Thus, thinning has played an extremely important role in the fabrication of large area FPA hybrids. In addition, this thinning has completely eliminated the pixel-to-pixel optical cross-talk of the FPA. This initial array gave very good images with 99.9% of the pixels working, demonstrating the high yield of GaAs technology. The operability was defined as the percentage of pixels having noise equivalent differential temperature less than lOOmK at 3OOK background with f/2 optics and in this case operability happens to be equal to the pixel yield. We have used equation (26) to calculate the NEAT of the FPA. where Dt*3 is the blackbody detectivity, dPR/dT is the derivative of the integrated blackbody power with respect to temperature, and 0 is the field of view angle (i.e., sine(0/ 2)-(4f2+1) -1, where f is the f number of the optical system). The background temperature T B - 30OK, the area of the pixel A=(231.tm) 2, the f number of the optical system is 2, and the frame rate is 30 Hz. Figure 3.24 shows the experimentally measured NEAT histogram of the FPA at an operating temperature of T -- 35 K, bias VR - - 2 . 5 V at 300 K background with f/2 optics and the mean value is 55 mK. This agrees reasonably well with our estimated value of 25 mK based on test structure data. The read noise of the multiplexer is 500 electrons. The factor of two short-fall of NEAT is mostly attributed to decrease in bias voltage across the detectors during charge accumulation (common in m a n y direct injection type readout multiplexers) and read noise of the readout multiplexer. The experimentally measured peak q u a n t u m efficiency of the FPA was 9.5% which agrees well with the 11% q u a n t u m efficiency estimated from the single element detector data. A 6 4 0 • 512 OWIP FPA hybrid was mounted onto an 84-pin lead-less chip carrier and installed into a laboratory dewar which is cooled by liquid neon, to demonstrate a LWlR imaging camera (FPA was cooled to 35 K). The other element of the camera is a 100 mm focal length AR coated germanium lens, which gives a 9.2~ 6.9 ~ field of view. It is designed to be transparent in the 8 12 ~tm wavelength range (which is not fully compatible with the 1 0 - 1 5 ~tm broadband OWIP array). An image processing station was used to obtain clock signals for the readout multiplexer and to perform digital data acquisition and non-uniformity corrections. The digital data acquisition resolution of the camera is 14-bits, which determines the instantaneous dynamic range of the camera (i.e., 16 384), however, the dynamic range of QWIP is 85 decibels.

GaAs/A1GaAs based quantum well infrared photodetector focal plane arrays

109

14000 12000 C

o

10000

O

8000

=

6000

r

4000

O

2000

0 -0.2

-0.1

0

0.1

0.2

0.3

NEAT (K) Figure 3.24 Noise equivalent temperature difference (NEAT) histogram of the 327 680 pix'els of the 6 4 0 x 5 1 2 array showing a high un(formit!t of the t:PA. The uncorrected non-uniformit!l (=standard deviation/mean) of this unoptimized FPA is onl!l 6.3% includin~l 1% non-uniformity of ROC and 1.4% nonuniformit# due to the cold-stop not being abh' to ~3ivethe sameJield of vicar to all the pixels in the FPA.

The measured mean NEAT of the OWIP camera is :35 mK at an operating temperature of f - 35 K and bias Vt~- - 2 . 5 V at 3 0 0 K background with f/2 optics (see Figure 3.24). This is in good agreement with expected focal plane array sensitivity due to the practical limitations on charge handling capacity of the multiplexer, read noise, bias voltage and operating temperature. The uncorrected NEAT non-uniformity (which includes a 1% non-uniformity of the ROC and a 1.4% non-uniformity due to the cold-stop in front of the FPA not yielding the same field of view to all the pixels) of the 327 680 pixels of the 6 4 0 x 5 1 2 FPA is about 6.3% ( - s i g m a / m e a n ) . The non-uniformity after twopoint (17 ~ and 27 ~ Celsius) correction improves to an impressive ().1%. As mentioned earlier, this high yield is due to the excellent GaAs growth uniformity and the mature GaAs processing technology. Video images were taken at a frame rate of 15 Hz at temperatures as high as T - 35 K using a ROC capacitor having a charge capacity of 11 x 1 ()~ electrons (the m a x i m u m number of photoelectrons and dark electrons that can be counted in the integration time of each detector pixel). Figure 3.2 5 shows a flame of video image taken with this 1 0 - 1 5 . 4 ~m 64()• 512 broadband OWIP imager. These high resolution images comparable to standard TV, demonstrate the high operability (i.e., 99.9%) and the stability (i.e., lower residual uniformity and l f f noise) of the 6 4 0 • long-wavelength OWIP staring array camera. It should be noted that these initial unoptimized FPA results are far from optimum. The light coupling gratings were not optimized (as described earlier) for maximum

1 ]0

Handbook of Infrared Detection Technolo~lies

Figure 3.25 This picture sho~vs one J?ame of video clip taken ~vith the 640• 512 pixel I 0 - I 5.41~m broadband QWIP focal plane arra!l. This ima~le sho~vs the liquid h, vel of a soda can and some.fingerprints on the C~II.

light coupling efficiency, no anti-reflection coatings were used on the backside of the FPA,

3.7 640• array

pixel spatially separated four-band QWIP focal plane

One unique feature of this spatially separated four-band focal plane array is that the four infrared bands are independently readable on a single imaging array. This feature leads to a reduction in instrument size, weight, mechanical complexity, optical complexity and power requirements, since no moving parts are needed. Furthermore, a single optical train can be employed, and the focal plane can operate at a single temperature. This four-band device structure was achieved by the growth of multi-stack QWIP structures separated by heavily doped contact layers, on a GaAs substrate. Device parameters of each QWIP stack were designed to respond in different wavelength bands. Figure 3.26 shows the schematic device structure of a fourcolor OWIP imager. A typical OWIP stack consists of a MOW structure of GaAs quantum wells separated by AlxGal_• barriers. The actual device structure consists of a 15 period stack of 3-5 pm QWIP structure, a 2 5 period stack of 8.5]Opm QWIP structure, a 25 period stack of ]()-12 ~tm OWIP structure and a 30 period stack of 13-15.:3 jam OWIP structure. Each photosensitive MOW stack was separated by a heavily doped (thickness ().2-().813m) intermediate GaAs contact layer (see Figure 3.26). Since the dark current of this device structure is dominated by the longest wavelength portion of the device structure, the VLWIR QWIP structure has been designed to have a bound-to-quasibound intersubband

GaAs/A1GaAs based quantmn well infrared photodetector focal plane arrags

111

Figure .3.26 Lager structure diagram ql four-band ()~'II ~ device and the 2I) periodic grating structure. Each pixel represents a 640 x 128 pixel area of the four-balld focal plane arra!t.

absorption peak at 14.0 ~m. Other OWIP device structures have been designed to have a bound-to-continuum intersubband absorption process, because the photo current and dark current of these devices are relatively small compared to the VLWIR device. This whole four-band OWIP device structure was then sandwiched between 0.513m GaAs top and bottom contact layers doped with n - 5 • 1017 cm- 3 and was grown on a semi-insulating GaAs substrate by MBE. The individual pixels were defined by photolithographic processing techniques (masking, etching, chemical vapor deposition, metal deposition, etc.). Four separate detector bands were defined by a deep trench etch process and the unwanted spectral bands were eliminated by a detector short-circuiting process. The unwanted top detectors were electrically shorted by gold coated reflective 2D etch gratings as shown in Figure 3.26. In addition to shorting, these gratings serve as light couplers for active QWIP stack in each detector pixel. The design and optimization of these 2D gratings to maximize QWIP light coupling were extensively discussed in ref. 5. The unwanted bottom detectors were electrically shorted at the end of each detector pixel row. Typically, in single-band QWIP FPAs. quarter wavelength deep (h -- 2p/4nGaAs) grating grooves are used for efficient light coupling. However, in the four-band FPA, the thickness of the quarter wavelength deep grating grooves are not deep enough to short-circuit the top three MQW QWIP stacks (e.g., three top OWIP stacks on 14-15.5 gm OWIP in Figure 3.26). Thus, three-quarter wavelength groove depth 2D gratings (h - 3;.p/4n~;,~.,) were used to short the top unwanted detectors over the 1 ()-12 and 14-15.5 microns bands. This technique optimized the light coupling to each OWIP stack at corresponding bands while keeping the pixel (or mesa) height at the same level, which is essential for the indium bump-bonding process used for detector array and readout multiplexer hybridization. Figure 3.27 shows the normalized spectral responsivities of all four spectral bands of this four-band focal plane array.

112

Handbook of Infrared Detection Technologies

Figure 3.2 7 Normalized spectral respo~lse of the lbltr-hand OWIt~.lbcal plane arrajt.

Few OWIP FPAs were chosen and hybridized to a 64()x 512 CMOS multiplexer (ISC 9803) and biased at V~ - - 1 . 1 V. At temperatures below 83 K, the signalto-noise ratio of the 4 - 5 l.tm spectral band is limited by array non-uniformity, multiplexer readout noise, and photo current (photon flux) noise. At temperatures above 45 K, temporal noise due to the 14-15.5/.tm QWIP's higher dark current becomes the limitation. The 8-1() and 1 0 - 1 2 ~m spectral bands have shown BLIP performance at temperatures between 45 and 83 K. The FPAs were back-illuminated through the fiat thinned substrate m e m b r a n e (thickness ~ 1 3 0 0 A). This initial array gave excellent images with 99.9% of the pixels working (number of dead pixels ~ 2 5()), demonstrating the high yield of GaAs technology. The operability was defined as the percentage of pixels having noise equivalent differential temperature less than 1 ()()mK at 30()K background and in this case operability happens to be equal to the pixel yield. A 640• pixel four-band OWIP FPA hybrid was mounted onto an 84-pin leadless chip carrier and installed into a laboratory dewar which is cooled by liquid helium to demonstrate a four-band simultaneous imaging camera. The FPA was cooled to 45 K and the temperature was stabilized by regulating the pressure of gaseous helium. The other element of the camera is a 1()() m m focal length AR coated g e r m a n i u m lens, which gives a 9 . 2 ~ ~ field of view. It is designed to be transparent in the 8 - 1 2 l.tm wavelength range. The SEIR | image processing station was used to obtain clock signals for readout multiplexer and to perform digital data acquisition and non-uniformity corrections. The digital data acquisition resolution of the camera is 14-bits, which determines the instantaneous dynamic range of the camera (i.e., 16 384), however, the dynamic range of OWIP is 85 decibels. Video images were taken at a frame rate of 30 Hz at temperatures as high as T - 45 K, using a ROC capacitor having a charge capacity of 11 • 1 ()r electrons {the m a x i m u m n u m b e r of photoelectrons

GaAs/A1GaAs based quantum well itffrared photodetector focal plane arra!ls

113

and dark electrons that can be counted in the time taken to read each detector pixel). Figure 3.28 shows one flame of a video image taken with a four-band 640 x 512 pixel OWIP camera.

3.8 QWIPs for low background and low temperature operation Although QWIP performs exceptionally well under high background levels, the detector exhibits anomalous behavior when it operates under stringent low irradiance and low temperature conditions {i.e. at extremely low photocurrent). Singh et al. have modeled and explained some of the irregular behaviors by treating OWIP as a resistance in series with a capacitance. 32'33 Arrington et al. 34 have shown experimentally the non-fiat frequency response curves of OWIP at low background and low operating temperature conditions. Under such conditions, responsivity decreases as the detector operating frequency increases. The roll-off frequency is completely determined by the device design and operational conditions. ~4 This behavior is empirically similar to dielectric relaxation effects observed in bulk extrinsic silicon and germanium photoconductors under similar operational conditions. ~s. ~, In principle, QWIP operates very similar to extrinsic bulk photoconductors. Electrons in the subbands of the isolated quantum wells can be visualized as electrons attached to impurity states in bulk photoconductors. As photogenerated electrons depart the active doped quantum well regions, they leave behind a space-charge buildup. Thus, in order to operate QWIP steadily, requires a sufficient dark or background photo current to replenish the depleted quantum wells. This is not an issue for QWIP detectors operating at high background conditions or high temperatures, because high photocurrent or high dark current can easily provide carriers to re-fill the space-charge buildup. However,

Figure 3.28 One frame of video image taken ~vith the 4-1 ~. ~ 12m cutoff four-band 640• ~ 12 pixel O.WIP camera. The image is barely visible in the 14-1 ~. ~ 12m spectral band due to the poor optical transmission of the anti-reflection layer coated germanium lens.

114

Handbookof Infrared Detection Technoloqies

at low operating temperatures for low-background irradiance levels, high resistivity of thick barriers in the active region impedes electrons from entering the detector from the opposite electrode. This could lead to a delay in refilling the space-charge buildup, and result in a lower responsitivity at high optical modulation frequencies. In order to overcome this problem we have designed a new detector structure, the Blocked Intersubband Detector (BID) with separate active quantum well region and blocking barrier. 37 As shown the Figure 3.29, the active region of BID consists of quantum wells separated by thin barriers, which creates sub-minibands due to large overlap of sub-level wave functions. Thus, space-charge buildup will get quickly refilled by electrons via sequential resonant tunneling along the ground state miniband from the emitter contact layer. A thick, impurity free blocking barrier is placed between the active region and collector to suppress the dark current of the device. We have fabricated a BID device consisting of 50 period MOW photosensitive regions and a 1 l~m thick A1.2Ga l_.2As blocking barrier layer. Each quantum well period contains a 60 A GaAs well and 1 ()()A A1.2Gal_.2As barrier. The whole device structure is sandwiched between two GaAs contact layers (Figure 3.29). The sample was grown on a semi-insulating GaAs substrate by molecular beam epitaxy. Transport carriers (electrons)were provided by doping all GaAs wells and contact layers with Si to a density of n=4 • 1017 cm-3. Mesas 20()13m in diameter were fabricated using wet chemical etching in order to measure dark current-voltage curves, spectral responsivity and noise. The responsivity spectra of these detectors were measured using a 10()0 K black-body source and a grating monochromator. To obtain normalized responsivity spectra at different bias voltages, detectors were back illuminated through a 45 ~ polished facet. Then the absolute spectral responsivities were calculated by measuring total photocurrent due to a calibrated black-body source. Figure 3.30 shows responsivity curves at different bias voltages from VI~ = -(). :3 to l. :3 V. Unlike in typical OWIPs, the spectral response of BID shows extra broadening because the photoexcitation occurs between sub-minibands instead of localized sub-levels. Figure 3.30 also illustrates the absorption quantum efficiency spectrum of the BID, measured using a 45 ~ wave guide geometry. Although the BID shows comparable absorption quantum efficiency, it shows much smaller responsivity than thick barrier OWIP devices. Therefore, the photoconductive gain of the BID (g = 0.06 at V = - 1 V bias voltage) is a few

Figure 3.29 Energy band diagram of GaAs/Al(;aAs Blocked Intersubband Detector (BID).

GaAs/A1GaAs based quantum well infrared photodetector focal plane arra~ts 115

Figure 3.30 Responsivity and absorption quantum efficienc!l spectra of the BID measured using a 4 ~~ wave guide geometry. Shown responsivity curves are for different bias voltages from VB=--O.5 to 1.5 V with 0.25 Vsteps.

times smaller (g ~ 0.2 - 0.5 at V = - 3 V bias voltage). This can be attributed to the lower electric field across the active region of the BID. Due to higher resistivity of the blocking layer, compared to the active superlattice, much of the applied bias voltage drops across the blocking layer. Thus, the majority of the photoexcited electrons relax within the superlattice before reaching the blocking layer. In contrast, the typical QWIP has uniform resistivity along the active layer, which allows a substantial uniform electric field across the device. In order to improve the carrier collection efficiency, we have designed a BID with a thinner blocking barrier. Although, this device operates at a lower bias voltage, the electric field distribution across the active layers is much higher than in the previous devices. Figure 3.31 shows the comparison of optical gains for two different devices obtained by measuring absorption quantum efficiency and responsivity. Figure 3.32 shows dark currents measured at V = - 1 V bias voltage and at different operating temperatures. The estimated detectivity D* at different operating temperatures, for a low background, with photon flux 4)= 4 • 108 photons/sec/pixel is also illustrated in Figure 3.32. This flux simulates 120 K space background withf/2-optics aperture.

3.9 Summary In summary, we have discussed the importance of FPA uniformity in NEAT, the general figure of merit that describes the performance of large imaging arrays. It is important to note that when D*>~IO l(~ cmv/Hz/W, the performance is uniformity limited and thus essentially independent of the detectivity, i.e., D* is not the relevant figure of merit. 2 Furthermore, we have demonstrated the 6 4 0 • pixel LWIR portable OWIP camera based on bound-to-quasibound device structure, the first 640• 512 pixel four-band FPA, 640• 486 pixel dual-band imaging camera, and

116

Handbook of Infrared Detection Technologies

0.5

.c_ 0 ,,,==

Blocking Layer Thickness 9600 A

0,4-

0.3

~. 0.2

O

0.1

0

i

I

i

!

4

8

12

16

20

Effective Electric Field (kVIpm) Figure 3.31 Optical gain obtained for two d(fferent B/D devices b!l measuring absorption quantum eSficienc!t and responsivity, hnprovement in optical gain is due to efficient carrier collection in the newest design.

E e-

~

L L

1 .E-07

.E+13

1.E-09-

.E+12 -

1.E-11

.E+11

1.E-13

.E+10 -

-

o .a

1 1.E-15 30

I

I

I

40

50

60

Temperature

(K)

.E+09 70

30

,

,

,

40

50

60

Temperature

70

(K)

Figure 3.32 BID dark current and peak detectivit!! D* against operating temperature. The dark current was measured with 200 I~m diameter mesa and normalized to a 30 • 30 i~m 2 pix'el. D* is calculated based on responsivit# and dark currents for a background with photon flux 4)=4 • 10 '~photons/s/pixel which simulates 120 K space background and f/2-optics.

640• pixel broad-band imaging camera. Furthermore, we have demonstrated that OWIP is able to operate at low temperatures, with a higher sensitivity under high background irradiance levels. The anomalous behavior of QWIP w h e n operating under stringent low irradiance levels at low temperatures was addressed. A new blocked intersubband detector (BID) structure, with active superlattice and a thick blocking barrier, has been proposed to overcome this problem. Experimental m e a s u r e m e n t s show similar absorption q u a n t u m efficiency, with a lower photoconductive gain. This can be attributed to a lower voltage drop across the active superlattice of the BID. The real advantage of the BID detector is that its infrared sensing photoemitter is a GaAs/A1GaAs based

GaAs/A1GaAsbasedquantum well infraredphotodetectorfocal plane arrays l 17 superlattice, thus its cutoff wavelength can be easily tuned by band gap engineering (i.e., tailorable cutoff wavelengths). Thus, if one seeks shorter wavelength operation such as 1() or 15 microns cutoff, the MOW based BID detectors will operate at much higher cryogenic temperatures such as 30K which can be achieved passively at space low background conditions.

Acknowledgements The research described in this paper was performed by the Jet Propulsion Laboratory, California Institute of Technology, and was jointly sponsored by the Breakthrough Sensors and Instrument Component Technology Thrust of NASA Cross Enterprise Technology Development Program, Air Force Research Laboratory, and the Ballistic Missile Defense Organization.

References 1. C. Weisbuch, Semicon. Semimet. 24, 1 - 1 3 3 (1987). 2. B. F. Levine, J. Appl. Phys. 74, R1 (1993 ). 3. S. D. Gunapala, B. F. Levine, L. Pfeiffer, and K. West, Dependence of the Performance of GaAs/A1GaAs Q u a n t u m Well Infrared Photodetectors on Doping and Bias, J. Appl. Phys. 69, 6517 ( 1991 ). 4. S. D. Gunapala, J. K. Liu, J. S. Park, T. L. Lin, and M. Sundaram, Infrared Radiation Detecting Device, US Patent No. 6,211.529. 5. S. D. Gunapala and S. V. Bandara, Q u a n t u m Well Infrared Photodetector (OWIP) Focal Plane Arrays, Setniconductors and Semimetals 62, 1 9 7 - 2 8 2 , Academic Press (1999). 6. J. Y. Andersson, L. Lundqvist, and Z. F. Paska, J. Appl. Phys. 71, 3600 (1991). 7. G. Sarusi, B. F. Levine, S. J. Pearton, S. V. Bandara. and R. E. Leibenguth, ]. Appl. Phys. 76, 4 9 8 9 (1994). 8. C. G. Bethea, B. F. Levine, V. O. Shen, R. R. Abbott, and S. J. Hseih, IEEE Trans. Electron Devices 3 8, 1118 ( 1991 ). 9. C. G. Bethea and B. F. Levine, B. F. In Proceedings of SPIE International Symposium on Optical Applied Science and Engineering, San Diego, CA ( 1992 ). 10. C. G. Bethea, B. F. Levine, M. T. Asom, R. E. Leibenguth, J. W. Stayt, K. G. Glogovsky, R. A. Morgan, J. Blackwell, and W. Parish, IEEE Trans. Electron Devices 40, 1957 ( 1993 ). 11. W. A. Beck, T. S. Faska, J. W. Little, J. Albritton, and M. Sensiper, Proceedings of the Second International Symposium on 2-20 l~m Wavelength Infrared Detectors and Arrays: Physics and Applications, Miami Beach, Florida (1994). 12. S. D. Gunapala, J. S. Park, G. Sarusi, True-Lon Lin, J. K. Liu, P. D. Maker, R. Muller, C. A. Shott, and T. Hoelter, 15 Bm 128 x 128 GaAs/AIGaAs Q u a n t u m Well Infrared Photodetector Focal Plane Array Camera, IEEE Trans. Electron Devices 44, 4 5 - 5 0 (1997).

118 Handbookof Infrared Detection Technologies 13. S. D. Gunapala, J. K. Liu, J. S. Park. M. Sundaram, C. A. Shott, T. Hoelter, True-Lon Lin, S. T. Massie, P. D. Maker, R. E. Muller, and G. Sarusi, 9 ~m Cutoff 2 5 6 x 2 5 6 GaAs/AlxGal_xAs Q u a n t u m Well Infrared Photodetector Hand-Held Camera, IEEE Trans. Electron Devices 44, 5 1 - 5 7 (1997). 14. S. D. Gunapala, S. V. Bandara, J. K. Liu, W. Hong, M. Sundaram, P. D. Maker, R. E. Muller, C. A. Shott, and R. Carralejo, Long-Wavelength 6 4 0 x 486 GaAs/AlxGal_xAs Q u a n t u m Well Infrared Photodetector Snap-shot Camera, IEEE Trans. Electron Devices 45, 1890 ( 1998 ). 15. J. Y. Andersson, J. Alverbro, J. Borglind, P. Helander, H. Martijn, and M. Ostlund, 3 2 0 x 2 4 0 Pixels Q u a n t u m Well Infrared Photodetector (QWIP) Array for Thermal Imaging: Fabrication and Evaluation. SPIT 3 0 6 1 , 7 4 0 - 7 4 8 (199 7). 16 K. K. Choi, A. C. Goldberg, N. C. Das, M. D. Jhabvala, R. B. Bailey, and K. Vural, S P I E 3 2 8 7 , 1 1 8 - 1 2 7 (1998). 17. R. Breiter, W. Cabanski, 1R. Koch. W. Rode. and J. Ziegler, SPIT 33 79, 423 (1998). 18. R. H. Kingston, Detection of Optical and Infrared Radiation Springer, Berlin (1978). 19. A. Zussman, B. F. Levine, J. M. Kuo, and J. de Jong, 1. Appl. Phys. 70, 5101

(1991). 20. F. D. Shepherd, In Infrared Detector and Arrays, SPIT 9 3 0 (SPIT Orlando, FL), p. 2 (1988). 21. J. M. Mooney, F. D. Shepherd. W. S. Twins. J. E. Murgia, and J. Silverman, Opt. Eng. ZS, 1151 (1989). 22. Grave and A. Yariv, Intersubband Transitions in Ouantum Wells, edited by E. Rosencher, B. Vinter, and B. Levine, Cargese. France (Plenum, New York), p. 15 (1992). 23. B. F. Levine, C. G. Bethea, K. G. Glogovsky, J. W. Stayt, and R. E. Leibenguth, Semicond. Sci. Technol. 6, C114 ( 1991 ). 24. M. T. Asom, C. G. Bethea, M. W. Focht, T. R. Fullowan, W. A. Gault, K. G. Glogovsky, G. Guth, R. E. Leibenguth, B. F. Levine, G. Lievscu, L. C. Luther, J. W. Stayt, Jr., V. Swaminathan, Y. M. Wong, and A. Zussman. Proceedings of the IRIS Specialty Group on Infrared Detectors I. 13 ( 1991 ). 25. V. Swaminathan, J. W. Stayt, Jr., J. L. Zilko, K. D. C. Trapp, L. E. Smith, S. Nakahara, L. C. Luther, G. Livescu, B. F. Levine, R. E. Leibenguth, K. G. Glogovsky, W. A. Gault, M. W. Focht, C. Buiocchi, and M. T. Asom, Proceedings of the IRIS Specialty Group on Infrared Detectors, Moffet Field, CA ( 1992). 26. D. Duston, BMDO's IS&T faces new hi-tech priorities, BMD Monitor, 1 8 0 183, May 19 (1995). 2 7. M. T. Chahine, Sensor requirements for Earth and Planetary Observations, Proceedings of Innovative Long Wavelength Infrared Detector Workshop, Pasadena, California, pp. 3-31, April 24-26 (1990). 28. S. D. Gunapala, S. V. Bandara, A. Singh, J. K. Liu, S. B. Rafol, E. M. Luong, J. M. Mumolo, N. 0. Tran, J. D. Vincent, C. A. Shott, J. Long, and P. D. LeVan, 6 4 0 x 4 8 6 Long-wavelength Two-color GaAs/A1GaAs Quantum Well Infrared Photodetector (OWIP) Focal Plane Array Camera IEEE Trans. Electron Devices

4 7 , 9 6 3 - 9 7 1 (2ooo).

GaAs/A1GaAs basedquantum well infraredphotodetectorfocal plane arrays 119 29. S. D. Gunapala, S. V. Bandara, ]. K. Liu, S. B. Rafol, C. A. Shott, R. ]ones, S. Laband, ]. Woolaway II, ]. M. Fastenau, and A. K. Liu, 9 IJm Cutoff 6 4 0 x 512 Pixel GaAs/Al• Quantum Well Infrared Photodetector Hand-held Camera, to be published in SPIE Proceedings 4 7 2 1 (2002). 30. S. V. Bandara, S. D. Gunapala, ]. K. Liu, E. M. Luong, ]. M. Mumolo, W. Hong, D. K. Sengupta, and M. ]. McKelvey, 10-16 IJm Broadband Quantum Well Infrared Photodetector, Appl. Phys. Lett. 72, 2427 (1998). 31. S. V. Bandara, S. D. Gunapala, ]. K. Liu, S. B. Rafol, ]. M. Mumolo, and D. Z. Ting, Array of OWIPs With Spatial Separation of Multiple Colors, NASA Tech Briefs 26 No. 5, 8a (2002). 32. A. Singh and D. A. Cardimona, SPIE, 2 9 9 9 , 46 (1997). 33. A. Singh and D. A. Cardimona, Opt. Eng. 38, 1424 (1999). 34. D. C. Arrington, ]. E. Hubbs, M. E. Gramer, Gary A. Dole, SPIE 4 0 2 8 , 288 (2000). 35. M. Ershov, S. Satou, and Y. Ikebe, 1. Appl. Phzds. 86, 6442 (1999) 36. N. M. Haegel, C. R. Brennan, and A. M. White, 1. Appl. Phys. 80, 1510 (1996). 37. S. Gunapala, S. Bandara, ]. Bock, M. Ressler, ]. Liu, ]. Mumolo, Sir Rafol, D. Ting and M. Werner, Large Format Long-wavelength GaAs/A1GaAs Multiquantum Well Infrared Detector Arrays for Astronomy SPIE 4 2 8 8 , 2 7 8 - 2 8 5 (2001).

This Page Intentionally Left Blank

Chapter 4

GalnAs(P) based QWIPs on GaAs, lnP, and Si substrates for focal plane arrays J. Jiang and M. Razeghi

4.1 Introduction 4.7.1 Overview of infrared detector

Every object emits infrared light. The hotter an object is, the more radiation it will emit. Detection and processing of infrared radiation can provide a wealth of information about an object that is not available in other regions of the spectrum. Infrared detectors are very important in both civilian and military applications and have been extensively studied over the past century. In general, infrared detectors can be categorized as either thermal detectors or photon detectors (as shown in Figure 4.1). Photon absorption in thermal detectors leads to an increase in temperature, resulting in a measurable change in certain material properties. The operation of the photon detectors, on the other hand, is based on the measurement of an electrical photocurrent generated by photon absorption in a semiconductor. A comparison between thermal and photon detectors is given in Table 4.1. The atmosphere has a few important transmission windows in the infrared region (as shown in Figure 4.2). These atmospheric windows are important for almost all infrared detector applications, allowing detection of objects at long distances. The short wavelength infrared (SWIR) window extends from the visible region up to 2.5 pm, the middle wavelength infrared (MWIR) window lies between 3 and 5 l~m, and the long wavelength infrared (LWIR) window ranges from 8 to 14 pm. The location of the LWIR window is fortunate since room temperature objects have a peak wavelength of almost 101~m. Detection of wavelengths larger than 141um, so called very long wavelength infrared (VLWIR), is mainly used for outer space application.

122 Handbook of Infrared Detection Technologies

I FerroelectrJc-PyroelectricI Resistive Thermal (Bo]ometers) ]

Infrared]

IV-V! ~PbSe,PbSnTe) ]

D

I

t ''-v'

I

t ![[-\: (In~. InAsSb,InTlSb.Ia,SbBi) ]

-~QuantumWells

Type II(InSl~'InA~Sb,IaAs;C_ralnSbl[

Figure 4.1 Infrared detector categorization based on their method of detection.

Near infrared Middle infrared " 44 3, i

i

! ......

-i

Far infrared ....

i

.

g~ m

i" I m

Q,

t ,,

IL t,tt t,__,,,,. ,.,tt',. .,,tt~

Wavelength (~m)

~

~

,,t .~

~,

Absorption Molecule

Figure 4.2 The transmission spectrum of the atmosphere over a horizontal 6000ft. path length at sea level. The regions of high transmission, called atmospheric windows, are evident. Quantum well infrared photodetector.

Almost all the real-world applications of the infrared detector are based on thermal imaging using the infrared focal plane array (FPA) camera. An FPA is an optical sensor placed at the focal plane of an optical system such as a camera, spectrometer or telescope. The infrared FPA is composed of an infrared detector array, which can be designed and manufactured to be sensitive to wavelength range from SWIR to VLWIR based on both thermal and photon detectors. Infrared FPAs are combined with a read out integrated circuit (ROIC) or multiplexer which allows the electronic access of every pixel in the array.

GalnAs(P) based OWIPs on GaAs. InP. and Si substrates for focal plane arrays Table 4.1

123

Pros and c o n s o f thermal and p h o t o n infrared detectors

Detectors

Advantages

Disadvantages

Thermal

9 Relatively low cost imaging possible 9 Sensitive over a wide infrared r a n g e 9 Light, rugged, reliable, and convenient to use 9 A d e q u a t e response time for some imaging applications 9 Easy to tailor the b a n d g a p of the alloys to cover the entire infrared region 9 High detectivity 9 Well developed theoretical and experimental results 9 Good material properties 9 Fast response

9 Slow response {ms o r d e r ) d u e to reliance on h e a t i n g of atomic s t r u c t u r e 9 Relatively low detectivity (D*~ 1 ()s cmHz 1,2/W )

Photon

9 Difficulty in device processing 9 High cost in g r o w t h and device processing

4.1.2 Quantum well infrared photodetector What is a Q WIP?

The q u a n t u m well infrared photodetector (OWIP) is a semiconductor infrared photon detector relying on intersubband absorption within either the conduction band (n-type) or the valence band (p-type). The idea of utilizing a q u a n t u m well for infrared detection was first presented by Esaki and Sakaka in 19 7 7 and can be explained by using the basic principles of quantum mechanics. 1 The quantum well is equivalent to the well-known particle in a box problem in quantum mechanics, which can be solved by the time independent SchrSdinger equation. The solutions to this problem are the eigenvalues that describe energy levels inside the q u a n t u m well in which the particle is allowed to exist. OWIPs are built from quantum wells of wide-bandgap materials that in bulk do not absorb in the MWIR or LWIR. However, electron (hole) excitation may occur between the ground state and excited states in a conduction (valence) band quantum well (see Figure 4.3), making intersubband absorption possible in the MWIR or LWIR regions. The quantum well structure is designed so that these photoexcited carriers can escape from the quantum well and can be collected as photocurrent. These detectors afford greater flexibility because the peak and cutoff wavelength can be continuously tailored within a range by varying layer thickness (quantum well width) and barrier composition (barrier height). With appropriate choice of the well and barrier material, the detection wavelength of a QWIP can be tailored to any wavelength from MWIR to VLWIR. Why QWIP?

So far, the 'king' of infrared detectors has been the HgCdTe intrinsic photodetector (MCT). HgCdTe is the most studied semiconductor material for IR detectors and it is the standard against which all of other IR photon detectors are matched against. 2'3'4 It has very high quantum efficiency and detectivity: at

124

Handbook of Infrared Detection Technologies

Figure 4.3 Schematic of intersubband absorption, which takes place entirely within the valence band (for ptype doping, from H~ to H2) or conduction band (for n-type doping, from E1 to E2) of a quantum well. respectively. 7 7 K, a reported q u a n t u m efficiency exceeding 70%, and a detectivity exceeding 1012 cmHzl/2W-1. Focal plane arrays (FPAs) as large as 6 4 0 x 4 8 0 and dualcolor 1 2 8 x 1 2 8 FPAs have been demonstrated, s'6 However, the future of this material is uncertain because of the difficulties associated with its growth, processing, and device stability (see Table 4.2). Yields as low as 1% for 128 • 128 arrays produce costs per array as high as $10 0 0 0 - - 6 0 000. 7 On the other hand, QWIPs use the III-V material system, which is very mature due to the previous development of GaAs- and InP-based lasers, LEDs, and microwave circuitry. The growth and fabrication of QWIPs can be borrowed directly from these technologies. The advantages of QWIPs are also shown in Table 4.2, including their use in high frequency applications, s Although most QWIPs must be operated at cryogenic temperature, advances in closed cycle cryo-coolers have reduced both the size and cost of such systems. QWIP FPAs show better or comparable array performance to HgCdTe at a much lower cost. 9 The first n-type OWIP device based on intersubband absorption was demonstrated by Levine et al. on GaAs/AIGaAs material system in 1987.1~ Its peak spectral response occurs at l O.8~tm. The first p-type QWIP was demonstrated by Levine et al. in 1991.11 Also in 1991, only four years after the first demonstration of the QWIP, Bethea et al. obtained the first infrared image using a ten-element linear scanning OWIP array. 12 W h y Al-ffee?

GaAs/A1GaAs historically has been the most studied III-V material system because of its simplicity. Today, most of the reported FPAs were based on GaAs/ A1GaAs OWIP structures. Al-free III-V materials, on the other hand, have some inherent advantages, like reduced surface recombination velocity, less stringent

GalnAs(P) based QWIPs on GaAs. InP. and Si substrates for focal plane arrays Table 4.2

125

C o m p a r i s o n b e t w e e n i n f r a r e d HgCdTe a n d QWIP d e t e c t o r .

Detector type

Advantages

Disadvantages

HgCdTe

9Excellent quantum efficiency 9 Very high detectivity

9 Poor array operability and uniformity 9 Material defects result in low RoA for many VLWIRpixels. Ro non-uniform 9 Unstable alloy and difficult material growth control 9 Radiation-hard arrays are difficult due to narrow bandgap and defects in material 9 Low yield and high costs for large-area arrays 9 Reproducibility is poor due to large sensitivity of bandgap to composition

9 Bandgap can be adjusted to vary detection wavelength

QwIP

Mature III-V growth technology 9Lower quantum efficiency than intrinsic HgCdTe 9 Wide-bandgap material is better for radiation-hard applications. 9Excellent array uniformity 9 High RoA allows long integration time 9 Narrow photoresponse spectrum 9Multi-color arrays demonstrated 9Very fast response for ultra-high frequency applications

9 Requires lower sensor temperature than intrinsic detectors for ). < 12 Hm 9 Require light coupling for n-type QWIPs

processing r e q u i r e m e n t s , a n d a b s e n c e of DX centers. To take a d v a n t a g e of these properties, people started to g r o w OWIPs w i t h o t h e r m a t e r i a l systems, like I n G a P / G a A s , G a l n A s P / I n P , G a I n A s P / G a A s , et a/. 13'14'1s Table 4.3 gives a c o m p a r i s o n b e t w e e n Al-free a n d Al-based s e m i c o n d u c t o r materials. W i t h the a d v a n t a g e s of Al-free m a t e r i a l s given in Table 4.3, high-reliability a n d low-cost infrared OWIP FPAs based on these m a t e r i a l s y s t e m are h i g h l y expected. In addition, the Al-free OWIP g r o w n on silicon s u b s t r a t e also s h o w e d better d e t e c t o r p e r f o r m a n c e t h a n t h a t of Al-based OWIP on silicon (see Section 4.6), w h i c h t r a n s l a t e s to better m o n o l i t h i c i n t e g r a t i o n infrared FPA based on Al-free QWIPs. As a result, the Al-free GalnAs(P) m a t e r i a l system, as a n i m p o r t a n t a l t e r n a t i v e to A1GaAs m a t e r i a l s y s t e m for OWIP s t r u c t u r e , has been receiving m o r e a n d m o r e a t t e n t i o n in r e c e n t years. In the rest of this c h a p t e r , we will focus on the GaInAs(P) OWIP s t r u c t u r e g r o w n on v a r i o u s substrates, s u c h as GaAs, InP, a n d silicon.

4.1.3 S t a t e - o f - t h e - a r t

The detectivity (D*) is the p r i m a r y figure of merit used in e v a l u a t i n g the single e l e m e n t p h o t o d e t e c t o r p e r f o r m a n c e . Today. the D* for 4 pm a n d 9 pm

126

Handbookof Infrared Detection Technologies

Table 4.3

C o m p a r i s o n b e t w e e n Ai-free and Al-based s e m i c o n d u c t o r m a t e r i a l s 16't7

Al-free material

Al-based materials

9 No oxidation, which results in higher device reliability 9 Simple device processing

9 Passivation needed for device fabrication

9 Low material growth temperature 9 Reduced surface recombination 9 DX-defect centers associated with AI are avoided 9 Potential for growth of monolithic integration with Si based device.

9 Higher material growth temperature which results in interdiffusion problem 9 Poor device reliability

OWIPs measure about 1.5 • 1012 cm-Hzl/2/W and 2x 1()~ cm-Hzl/2/W at 77 K, respectively. 1s.19 In addition, thanks to the narrow absorption peak exhibited by QWlPs, multicolor operation is possible by simply stacking different types of q u a n t u m wells. Tidrow et al. reported a four-color OWIP detector based on stacked InGaAs/ A1GaAs and GaAs/A1GaAs m u l t i - q u a n t u m wells (MQWs). 2~ OWIP FPAs using a different material system have been demonstrated by m a n y groups. 9'21'22 These FPAs have shown excellent imagery in the LWlR atmospheric window. Large format LWlR OWIP FPAs up to 6 4 0 x 4 8 0 pixels have been commercialized by the Jet Propulsion Laboratory (JPL). 23.24 JPL also demonstrated a 15 pm VLWlR OWIP FPA. 2s For the portable market, a handheld LWIR infrared camera based on a 2 5 6 x 2 5 6 GaAs/AIGaAs FPA was realized. 26 For multi-color FPAs, two-color LWIR/VLWlRFPAs were demonstrated. This FPA gave excellent images with 99.7% of the LWIR pixels and 98% ofVLWIR pixels working. 27 As an example, Figure 4.4 shows a thermal image taken by a 2 5 6 • infrared OWlP FPA camera. The handprint on the book can be seen very clearly.

4.2 Fundamentals of QWIP 4.2.1 Intersubband absorption Traditional interband optical absorption involves photoexciting carriers across the band gap E~, i.e., promoting an electron from the valence-band ground state to the conduction-band excited state (see Figure 4.5). An intrinsic semiconductor photon detector utilizing the interband absorption is only sensitive to the photons whose energy is larger than the semiconductor's energy gap. For the detection of MWIR and LWIR regions of spectrum, a narrow E~ material must be used. It is well known that these low band gap materials are more difficult to grow and process than large band gap semiconductors such as

GalnAs(P) based QWIPs on GaAs. InP. and Si substrates for focal plane arrays

127

Figure 4.4 Thermal image taken by a 256 • ~60WIPfocal plane array camera. The handprint on the book can be seen clearly. (Courtesy of Dr. Murzy ]habvala ).

conduction band absorbed photon

~

-

valence band

Figure 4.5 intrinsic semiconductor detectors, an excited electron must cross the entire bandgap to reach the conduction band.

GaAs. To take advantage of the superior growth technology that exists with the wide bandgap semiconductor material, intersubband absorption inside the quantum wells is employed for detection of MWIP, or LWIR. For intersubband absorption, photon absorption takes place from ground state to excited states in quantum wells, as shown in Figure 4.3. The energy separation between the two bound states, E1 and E2 or H1 and H2, is much smaller than the bandgap of the well or barrier material,which translate to a much longer intersubband absorption wavelength. Thus, by controlling the location of the confined energy levels, the absorbed infrared wavelength can be

128 Handbookof Infrared Detection Technologies controlled. For a one-dimensional periodic q u a n t u m well structure, the position of confined energy levels can be computed by the Kronig-Penney model. 2s'29' 3o The relationship between the energy states and the well and barrier characteristics is given by:

1(

cosxL - coskwcoshabb + -~ rl - 1 )

sinkwsinha,,b

(1)

where X is a solution parameter, L is the period of superlattice thickness (well plus barrier width), k = v / 2m,,,E/~-. m;,,is the effective mass of the well material, E is 9 , the energy, /i is the reduced Planck's constant, w is the well thickness, ab = v/2m*b(Vo- E)ffi2, m~ is the effective mass of the barrier material, V(, is the conduction band offset, b is the barrier thickness, and 71 - m,*,,ab/m~,k. The values of allowed energy levels E can be obtained by numerical solution of equation ( 1 ). From (1), we can see that the absorbed infrared wavelength can be tailored by the bandgap and thickness of well and barrier material. For n-type QWIPs, which utilize an intersubband transition between F-valley derived states, the oscillator strength of a m u l t i - q u a n t u m well structure is given by:

f12 ~ I(FllT"-p[F2}[ 2

2

where F1 and F2 are the electron envelope functions, -K is the polarization vector for the incident infrared light, and - f the m o m e n t u m operator. If we assume the q u a n t u m well growth direction is along the z-axis, then the envelope functions should depend on z only, with F1-Fl(Z) and F2-F2(z). It is seen readily that for normal incidence of infrared radiation, for which 7=(e,.. e.,/, 0), results in zero absorption. One remarkable note to this q u a n t u m selection rule is that it only applies to n-type doped q u a n t u m wells. For p-type q u a n t u m wells, there is strong mixing among the heavy holes and lights holes at non-zero wave vector in the valence band which makes absorption of normal incidence infrared light possible. 31.32 One requirement for detection using q u a n t u m wells is that the well material must be doped. Because the energy of an absorbed photon is less than the bandgap, it cannot produce an excited photocarrier by itself. With the addition of carriers in the well, excitation only needs to happen from the first bound state to the next bound state. For n-type doping, excitation occurs entirely within the conduction band" for p-type doping, excitation occurs entirely within the valence band. Photocurrent is produced w h e n an excited photocarrier is able to escape the well whereupon an externally applied bias can sweep it out to be collected by one of the contacts. For n-type QWIPs, depending on the position of the first excited state in the q u a n t u m well 33 and barrier layer structure (see Figure 4.6), intersubband transitions in OWlPs can be classified to four types" bound-to-bound state (BB), 1~ bound-to-quasibound state (BQB), 34 b o u n d - t o - c o n t i n u u m state (B-C), 3~ and bound-to-miniband (B-M). 36 These structures all exhibit different degrees of

GalnAs(P) based QWIPs on GaAs. InP. and Si substrates for focal plane arrays

129

Figure 4.6 Schematic of energ!t band diagram for B-B. B-C. BOB. and B-M OWIPs.

electron confinement within the q u a n t u m well. In general, less confined upper states exhibit a larger photoresponse bandwidth and weaker peak detectivity. The B-B OWIPs have intersubband absorption occurring between two bound states contained within the q u a n t u m well. However, the carriers excited to the higher bound state still need a way to escape from the well and get swept out. The two possible ways are tunneling through the barrier and thermionic emission. This sort of OWIP has a narrow absorption spectra and large dark current. The large dark current is related to the thin barriers used to facilitate the tunneling of photocarriers out of the q u a n t u m well. In a B-C QWIP, the intersubband absorption is based on the bound-toc o n t i n u u m states transition. The peak absorption is relatively small because the wave functions of the upper excited states spread over the barrier region. The absorption spectrum is expected to be wider due to the broad final states band above the barrier. In a BQB QWIP, the wave function of the upper excited states is aligned with the top of barrier. Since the barrier height also limits the cut-off wavelength, there is a sharp cut-off for this type of detector. In a B-M QWIP, the transition is from the localized bound ground state in enlarged wells to a resonant-coupled miniband of superlattice barriers. Since the well width in a B-M QWIP is m u c h wider than that of B-C and BOB QWIPs, there is a large overlap of wave functions between the initial and final states, and the interaction is strong. A blue shift of the peak detection wavelength is obtained by

130

Handbook of Infrared Detection Technologies

aligning the excited state to the top of the miniband, while a maximum bandwidth is achieved when the excited state is lined up in the middle of the miniband. Thus, B-M QWIPs offer more flexibility and fine tuning of the detection wavelength and spectral bandwidth than other types of OWIPs. For p-type QWlPs, the absorption and responsivity are much lower than those of n-type OWIP due to the large hole effective mass and low hole mobility. Reduced absorption stems from the absorption coefficient, which is inversely proportional to the effective mass of holes. On the other hand, a key benefit, as mentioned before, is the ability to absorb normal incidence radiation. To improve the intersubband absorption of the p-type QWlP, tensile strained-layer (TSL) or compressive strained-layer (CSL) quantum wells have been utilized. 37-4~J The strong mixing between light and heavy hole states substantially increases the normal incidence absorption in the strained-layer p-type OWIPs. In TSL p-type QWIPs, the intersubband absorption occurs between the ground light-hole (LH1) state to heavy-hole continuum states. Since the ground states are occupied by the light holes with smaller effective mass, the linear optical absorption is greatly enhanced over the unstrained case. In CSL p-type QWIPs, the intersubband absorption occurs between the ground (HH1) heavy-hole state to excited heavy-hole states (e.g. HH 3). Apart from the reduction of hole effective mass by compressive strain in the quantum well, the conducting holes behave like light holes with higher hole mobility due to the resonance between heavyhole and light-hole states. The energy band diagram for TSL and CSL p-type OWIP is shown in Figure 4.7. 4.2.2 Q W i P p a r a m e t e r s

Single OWIP detectors are usually rated by four figures of merit" absorption, dark current, responsivity, and detectivity. The photocurrent I t, generated in a QWIP can be written as Ip - npqv

(3)

Figure 4.7 Schematic energy band diagram for p-t!tpe tensile strained-la!ler (TSL) and compressively strained-layer ( CSL ) QWIP.

GalnAs(P) based O_WIPs on (;aAs. InP. and Si substrates for focal plane arra!ts

1] 1

where np is the volume density of photogenerated carriers, q the electron charge, and v is the transport velocity through the OWIP. For example in n-type OWIPs, np is given by Hp

z

Pcos0c~ Pe l'I, hid

where P is the incident optical power on the detector, p,, is the probability of escape from the q u a n t u m well, 0 is the angle between the incident light and the direction perpendicular to the q u a n t u m wells, and rL is the recapture lifetime. From this information we can calculate the detector's responsivity, which is defined as the ratio of the electrical current generated by the detector to the total optical power absorbed. Expressed mathematically, the peak responsivity Rp m the m a x i m u m value of responsivity as a function of photon energy - - is Rp =

Ip

Pcos0

=

q O.p,,g hv

(4)

where 77,, is double-pass absorption q u a n t u m efficiency and g is the photoconductive gain. Given the length of the m u l t i - q u a n t u m well region as 1. r/,, and g are defined as -

1 - exp(-2cd) 2

(s)

and VrL = _ra = _L

g-

1

(6)

1

rT

where r r = l / v is the transit time and L is the hot electron mean free path. The overall q u a n t u m efficiency 17is calculated as 77 - rl,,pe -

Rp h 1) qg

(7)

This gives an indication of how many carriers are being generated, which may contribute to the photocurrent, versus how many photons are absorbed. Another very important type of current in a OWIP. the dark current, can be calculated from I,(v)

- ,*(V)qv(V)A

(8)

where n*(V) is the effective number of electrons excited from the well into the c o n t i n u u m state, v ( V ) is the average transport velocity, and A is the detector area. n * ( V ) i s expressed as: 41

132

Handbook of Infrared Detection Technologies

~ f(E)T(E, V)dE n* ( V) - ~ 2m*Lp J E~

(9)

where Lp is the superlattice period. T(E. V) is the tunneling current transmission factor for a single barrier, andf(E)is the Fermi factor, given by

f(E) --

1 + [ e x p ( E - E1 - E,:)/kT]

where E is relative to the bottom of the q u a n t u m well, Et: is the Fermi level, k is Boltzmann's constant, and T is the temperature. The average transport velocity v(V) is given as

p.F -

V/1 + (#F/v,) 2 where # is the mobility, F is the average electric field, and vs is the saturated drift velocity. Related to the dark current is the noise current, which ultimately becomes the limiting factor for the sensitivity of a detector: the signal (responsivity) of a detector can be externally amplified, but the noise current would be amplified as well, leaving the signal-to-noise ratio unchanged. Several kinds of noise have been identified in photodetectors, though there are just two that can dominate the noise current in a QWIP. 42 The first is called generation-recombination noise, or GR noise. It is composed of r a n d o m thermal excitation and decay of carriers in the q u a n t u m wells, leading to fluctuations in the n u m b e r of carriers. The second is Johnson noise, which is a strictly thermal noise, and causes fluctuations in the velocity of carriers. Since noise is a r a n d o m process, the addition of noises obeys the root-mean-square (rms) system, that is, the square of the total noise equals the sum of the squares of the component noises. The total noise current i,, in a OWIP, then, is

In 29

_

i2R 27 llohnson .2

The GR noise current and the Johnson noise current are given by

icR -- v/4qlagAf and

~

( 1 O)

T

where Af is the m e a s u r e m e n t bandwidth and R is the series resistance of the QWIP.

GalnAs(P) based QWIPs on GaAs. InP. and Si substrates for focal plane arra!ts 13 3

Lastly, the specific detectivity gives the signal-to-noise ratio of the detector. It is an indication of how small a signal a detector can pick up. The specific detectivity D* is given by D* = Rp x/-AAf 9 In

(12)

It is a standardized figure of merit for both photovoltaic and photoconductive detectors and is independent of detector area or peak absorption wavelength. The general figure of merit of a 0WIP FPA is the noise equivalent temperature difference (NEAT). NEAT is the m i n i m u m temperature difference across the camera target that would create a signal-to-noise ratio of unity. It's given by 4 ~ NEAT =

v~Af

(13)

D*(dPB/dT)

where D* is the detector detectivity and P I~ is the background photon power. When D* is larger than 101() cm-Hzl/2/W, the NEAT is limited by the uniformity of FPA. 44 4.2.3 Comparison of n-type and p-type Q WIPs

Today, almost all the OWIP FPAs are made of n-type doped OWIP detectors (see Section 4.1.3). The main advantages for n-type 0WIPs are higher responsivity and detectivity due to the higher electron mobility, higher optical absorption and photoconductive gain related with the smaller electron effective mass compared with p-type OWIPs. Today, the values of detectivity for n-type OWIPs are generally higher than that of p-type QWIPs at the same peak wavelength. The obvious drawback in n-type OWIPs is the zero normal incidence absorption due to the q u a n t u m mechanical selection rules. Different light coupling schemes (such as gratings) are thus needed for the n-type QWIPs to obtain better detector performance. This usually complicates the fabrication process of 0WIP FPAs and also limits the pixel size. The normal incidence absorption feature of p-type QWIPs allows the fabrication of grating-less FPAs. Another advantage of p-type 0WIPs is the Fermi Table 4.4

n-type

p-type

Comparison b e t w e e n n-type and p-type QWIP detector

Advantages

Disadvantages

9Higher responsivity and detectivitv due to the higher electron mobility and low electron effective mass 9Detectionrange from SWIRto VLWIR 9Normal incidence absorption 9Grating-less FPA 9Lowerdark current

9Light-couplingrequired 9ComplicatedFPA processing and limited pixel size 9Lowerresponsivity and detectivity 9Difficultyfor VLWIRdetection

134 Handbook of Infrared Detection Technologies level pinning effect at higher doping density, in which the Fermi level above the ground bound state in the OW is nearly independent of the doping density in the OW. Thus, it is possible to dope the OW at a higher doping density to increase the linear absorption coefficient and the responsivity without increasing the dark current in the detector. 4.2.4 Growth, fabrication and device characterization of a single QWIP device

The most widely used growth technologies for OWIP devices are molecular beam epitaxy (MBE) and metalorganic chemical vapor deposition (MOCVD). MBE is an ultra-high vacuum thin film technology developed in the early 1970s. 4s Both gas source (GSMBE) and solid source (SSMBE) can be used for material growth. The advantages of this growth technology are an ill sitll controllable growth process, extremely precise control over layer thickness and doping profile, and high uniformity ( > 3 inch). Its main disadvantage is the high expense. MOCVD has established itself as a unique and important epitaxial crystal growth technique which yields high quality low dimensional structures for fundamental semiconductor physics research and devices. 4~' This growth technique is attractive in its ability to grow uniform layers, sharp interfaces, and for its commercial applications. The details of MBE and MOCVD technology have been described in many publications. 4 s-4 r In general, a OWIP structure is grown on a semi-insulating substrate (e.g. InP, GaAs) and consists of 2 0 - 5 0 periods of multiple q u a n t u m well (MOW) sandwiched between a top and bottom contact layer. The q u a n t u m well is doped n or p-type, as desired. Once a OWIP structure has been grown either by MBE or MOCVD, it must undergo a fabrication process to create a working detector. The basic fabrication procedures are: 1. Cleaning of sample in the order of trichloroethylene (TCE), acetone, methanol, and DI water. 2. Positive photolithography to define the detector pixel and light coupling structure (if needed). 3. Etching of the exposed semiconductor, either by dry etching {e.g. electron cyclotron resonance-enhanced reactive ion etch (ECR-RIE)) or wet chemical etching. 4. Negative photolithography for the metal contact definition. 5. Metal contact evaporation. 6. Lift-off process and alloying. Z Device packaging (die-bonding and wire-bonding). A finished GalnAs/InP OWIP mesa detector is shown in Figure 4.8. As discussed in Section 4.2.1 for n-type QWIPs, since there is no absorption for normal incidence, a light coupling scheme needs to be applied on the OWIP device for detection of the desirable normal incidence illumination. For a single n-type OWIP detector, as one step of device fabrication, a 45 ~ facet is usually polished on the substrate for light coupling. Different coupling schemes,

GalnAs(P) based QWIPs on (;aAs. InP. and Si substrates for focal plane arra!ls

13 5

Figure 4.8 A schematic diagram of the GalnAs/InP QWIP on InP substrate. This is how the device appears after fabrication, with the mesas etched and the metal contacts deposited.

for mull-element detectors, like diffraction gratings, 4~ random scattering reflectors, 49 microlenses, ~(~ and corrugated structures ~1 have also been developed. Several device characteristics, like the absorption spectra, relative spectral response, blackbody responsivity, noise current, and dark current, are usually measured to evaluate the performance of OWlP devices. The packaged OWIP device is usually tested inside a liquid nitrogen cryostat. The absorption coefficient given by an absorption spectra m e a s u r e m e n t can be used to calculate the room temperature (T=300K) and low temperature ( T = 7 7 K ) absorption q u a n t u m efficiency using r/~.~2 The relative spectral response gives information about a detector's peak absorption wavelength, cut-off wavelength, and spectral linewidth. Both the absorption spectra and relative spectral response can be performed with a Fourier Transform Infrared (FTIR) spectrometer. The absolute magnitude of the responsivity can be accurately determined by measuring the photocurrent, Ip, with a calibrated blackbody source that is at a specific temperature. Noise current m e a s u r e m e n t can be performed on a fast fourier transform (FFT) spectrum analyzer. The dark current of the OWlP can be measured with a semiconductor parameter analyzer and a dark shield placed around the OWIP inside the cryostat. Once the noise current and dark current are known, the overall q u a n t u m efficiency, photoconductive gain, and specific detectivity of a detector can be calculated by the formulas given in Section 4.2.2.

4.3 Fabrication of infrared FPA 4.3.1 Infrared FPA fabrication steps

As we mentioned in Section 4.1.1, most of the infrared detector applications are based on FPA cameras. So, how to make an operational infrared FPA also becomes an important issue for OWIP researchers. In this section, we will give a

136

Handbook of Infrared Detection Technologies

Figzire 4.9 Schematic qf ir(lrared FPA fabrication process.

GalnAs(P) based O WIPs on (;aAs. InP. and Si substrates for focal plane arrays

137

general description of the infrared FPA fabrication process. A schematic of infrared FPA fabrication is shown in Figure 4.9. Many technologies and much equipment are involved in this complicated fabrication process. The details of this process are described below step by step. 1. Growth of an infrared detector structure on the appropriate substrate. The growth is performed by MOCVD or MBE technology. This step basically determines the final FPA~sperformance. Different characterization will be performed to ensure that the detector's properties meet the requirements. 2. Fabrication of detector pixels. This includes the UV-photolithography for the sample patterning and the etching (like ECR-RIE dry etching or wet chemical etching) for the pattern transfer. For large format FPAs, the pitch size (array spacing) is usually between 30 and 50 l~m. The detector pixel size is even smaller. 3. Detector pixel passivation. Si02 or Si3N4 are common choices for the passivation material. The deposition of Si02 and Si 3N4 can be performed by plasma enhanced chemical vapor deposition (PECVD). The passivation layer will neutralize the surface states and insulate the covered areas. 4. Patterning and etching of the passivation layer. The purpose of this step is to make openings for the metal connection to each detector pixel. The opening size depends on the pixel size. 5. Fabrication of indium solder bumps. This is the most complicated step during the FPA fabrication and will be described in more detail in the next section. Indium bumps will be used as the interconnection between detector and ROIC pixels. 6. Flip-chip bonding of the detector substrate with a Si-based ROIC. The schematic of the alignment procedure is shown in Figure 4.10. An optical head will be brought between the upper and lower chips. The surface image of both upper chip and lower chip will be displayed on a monitor with the help of two video

Figure 4.10 Schematic offlip-chip bonding procedure.

138

Handbook of Infrared Detection Technoloqies

Figure 4.11 An indimn joint ~(fter reflow process.

cameras. The alignment can be done either manually or automatically. After fine alignment and parallelism, the optical head will be removed and the upper and lower chip will be pressed together with preset pressure and temperature. The indium joint needs to be refiowed at the preset temperature within an inert atmosphere. A refiowed indium joint is shown in Figure 4.11. 7. Underfill dispensing between FPA and ROIC. Underfill is usually based on an epoxy system. 53 The epoxy underfill provides the necessary mechanical strength to the detector array and readout hybrid, prior to the thinning process. It also reduces the effect of global thermal expansion mismatch between the detector array and readout chip and protects the FPA hybrid, from moisture, ionic contaminants, and hostile operating environments such as shock and vibration. 8. Detector substrate thinning process. This is usually performed by abrasive polishing or wet chemical etching. In some cases, the substrate is completely removed. The purpose of this step is to further reduce or completely eliminate (for substrate removal) the thermal expansion mismatch between detector array and readout chip. This basically to allow the OWlP FPAs to go through an unlimited number of temperature cycles without any indium bump breakage or delamination. Furthermore, the substrate thinning process also eliminates the pixel-to-pixel optical cross-talk and significantly enhances the optical coupling of infrared radiation into the OWIP pixels. 54 9. Packaging of the FPA hybrid. The hybrid will be loaded onto a lead-flee ceramic chip carrier (LCCC) by the die-bonding process and input/output metal pads on the readout chip will be connected to the LCCC pins by a wire-bonding process. 10. The final step of OWIP FPA fabrication is the test and evaluation of the FPA hybrid's performance. The test system is usually composed of a liquid-nitrogen Dewar, a cold clamp assembly, a camera head, digital electronics, and computerbased testing software.

GalnAs(P) based QWIPs on GaAs. InP. and Si substrates for focal plane arra!ls

139

4.3.2 Indium solder bump fabrication steps

Indium has become the most important mating material for infrared FPAs due to the fact that it stays ductile at liquid helium temperature, is easy to work with, and forms a good bond at low temperature. Highly-uniform and tall indium bumps are required for the infrared FPA fabrication. It has been shown that taller indium bumps can increase the FPA hybrid's reliability and thus its operating life. ~~ An indium solder bump is chiefly fabricated by two technologies: evaporation and electroplating. The comparison between the two technologies is shown in Table 4.5. In this section, the low-cost electroplated indium bump technology will be described. Figure 4.12 shows the schematic representation of the fabrication process for an electroplated indium solder bump. ~6-~s The details of an indium solder bump fabrication process are given below step by step. 1. Passivation of detector pixels. This is same as step three in the FPA fabrication described above. 2. Deposition of a Ti shortcut layer. This layer serves as a current path for the later electroplating process. 3. Thick photoresist ( > 10 ~m) lithography. The opening on the thick photoresist provides the mold for the indium electroplating process. Very strict control of photoresist spinning and UV-exposure is required to assure a good mold profile, s9 4. Electroplating of Under-Bump-Metallurgy (UBM) layers. A high reliability solder bump interconnect is comprised of two features: the UBM, also known as the Ball Limiting Metallurgy (BLM), and the solder ball itself. UBM is usually composed of several different metal layers and it provides the following features: good adhesion to the detector pixel metal pad, an effective solder diffusion barrier, and a solder wettable metal of appropriate thickness. 6~ For indium solder bumps, Ti/Au can be used as UBM layers. 5. Electroplating of indium solder bumps. This is performed in an appropriate indium plating bath (e.g. indium sulfamate) under a controlled current density. The plated indium bumps usually have a mushroom-shaped profile before reflow (see Figure 4.13 (a)). Table 4.5

Comparison between evaporated and electroplated solder bumping technology Advantages

Disadvantages

Evaporated solder bumping technology

9 Highly uniform bump 9 Excellent metallurgical control of bump

Electroplated solder bumping technology

9 Lower equipment costs 9 Simple fabrication process

9 High cost due to expensive evaporation equipment 9 Limited to material with high vapor pressure 9 Complicated fabrication process 9 More pollution to environment. Why? 9 Low t h r o u g h p u t 9 Non-uniform bump height 9 Induced subsurface wafer stress 9 Poor alloy control

140 Handbookof Infrared Detection Technologies

Figure 4.12 Schematic of the indium solder bump fabrication process.

6. Removal of the thick photoresist. This can be done with Acetone or photoresist stripper. 7. Reflow of the indium solder bumps. The purposes of reflow are to increase bump height by reshaping the indium into a sphere (see Figure 4.13(b)) and to facilitate the flip-chip bonding alignment. Reflowed indium bumps have also been shown to have higher reliability. 61 Reflow is usually performed in a rapid thermal process oven (RTP) under N2 atmosphere. Some novel reflow methods utilizing ECR-RIE and low-pressureMOCVD were also demonstrated. 62'123 These methods can effectively remove the indium oxide on the surface of indium bumps to ensure the high quality reflow. 8. Removal of Ti shortcut layer. This can be done by wet chemical etching with dilute HF (2.5 %). 4.3.3 ROIC for infrared FPA

As we described in Section 4.3.1. infrared FPAs are mated via indium bump technology to silicon based readout integrated circuits (ROIC) which allow electronic access to every single detector pixel in the array. The ROIC serves as an electrical interface between the infrared FPA and off-focal plane signal

GalnAs( P) based OWIPs on GaAs. InP. and Si substrates for focal plane arrays

141

Figure 4.13 (a) Mushroom-shaped indium sohh'r bump before reJlo~v. (b) Indium solder ball after reflow process.

processing electronics. A ROIC is usually a hybrid between a multiplexer and an array of low power circuits which, at a minimum, provide a precise bias voltage to the detectors and extract the photocurrent from them with the least possible noise. During the FPA operation, the integration capacitor in the ROIC unit cell stores the charges provided by an individual detector photocurrent within a fixed integration time. The resulting signal is multiplexed to a serial stream of sampleand-hold signals within the fixed readout time. Additional focal plane circuitry is usually required to improve the uniformity, signal-to-noise, and versatility of the FPA. In most ROICs, the signal is pre-amplified at the front-end stage to avoid further corruption by other noise sources. A number of readout preamplifiers, such as self-integrator (SI), source-follower detector (SFD), direct injection (DI), source-follower direct injection (SFDI), and capacitor feedback trans-impedance amplifier (CTIA), have been utilized for infrared FPA applications. (~3 Figure 4.14 shows a ROIC unit cell with the most popular DI type preamplifier. 64 The detector bias voltage is controlled by the Vbias_adj gate and the signal is multiplexed out from the sample and hold capacitor during the readout period. The anti-bloom gate keeps the input circuit from saturating. Today, most ROICs are built by advanced sub-micron cmOS technology which has the advantages of higher operation frequency and capability for random access. 65 Many types of large format ROICs are currently commercially available. 66'67'126 The general requirements of ROICs for infrared FPAs are: stable detector bias; small input resistance: low nonlinearity: low power consumption; and large dynamic range. (~ For a specific type of infrared detector, the ROIC has to be tailored to the detector characteristics.

4.4 p-type QWIPs Hole intersubband absorption is attractive because the strong mixing of heavy and light holes in the valence band (at k~:()) allows the desirable normal incidence illumination geometry to be used. To fully utilize the advantage of

142 Handbook of Infrared Detection Technologies

Figure 4.14 Simplified schenmtic of a ROICmlit cell based on a direct injection input circuit.

p-type intersubband transition, p-type OWIPs based on different material systems have been studied by m a n y groups. In this section, we will give some examples of p-type OWlP for detection of different wavelength range. All of the OWIP structures reported here are based on Al-free GalnAs(P) material systems. 4.4.1 p-type MWIR QWIPs p - t y p e GaAs/GalnP

J. Hoff et al. reported MWIR p-type OWIP based on a lattice-matched (no strain) GaAs/GaInP MQW structure. 69 Three p-type OWIP structures were grown on semi-insulating GaAs substrate by a low-pressure MOCVD reactor. Each structure contained 50 GaAs q u a n t u m wells separated by nominally 280 A wide Gao.51 Ino.49 P barriers. These superlattices were sandwiched between thick GaAs layers used for top (0.5 l~m) and bottom (1.() l~m) contacts. All GaAs layers were doped with Zn to a net acceptor concentration of 1 x 1 ()lg cm-3. Different GaAs well widths (22A, 32A, and 55A) were used for the three different structures. 4 0 0 x 4 0 0 l~m mesa detectors were fabricated using photolithography and wet chemical etching, and 1 5 0 x l S ( ) l ~ m square Au/AuZn electrodes were evaporated and alloyed. The devices were front-illuminated under normal incidence. The spectral response shifts towards longer wavelength as the well width is reduced, which is expected as the hole ground state is pushed up towards the top of the barrier. Dark current was measured as a function of bias for the 22 A well at a different temperature (see Figure 2 in reference 69). At low bias, tunneling is negligible and the dark current originates from thermionic emission above the top of the barrier. The activation energy of holes in the q u a n t u m well can be calculated directly from the dark current data. Since dark current is dominated by thermionic emission from the q u a n t u m well at temperatures greater than 4 5 K, there is an Arrhenius dependence of dark current with temperature.

GalnAs( P ) based OWIPs on GaAs, InP, and Si substrates for focal plane arrays

l n ( I D / T ) oc - E A / k T

14 3

(14)

Where EA is the activation energy out of the quantum well. The EA revealed from this method for the three samples are 2 3 5 , 3 0 0 . and 330 meV for the 22, 32, and 55 A wide well QWIP respectively. These activation energies correspond to wavelength of 5.28, 4.13, and 3.75 l~m, respectively. Those values closely match to observed cut-off wavelength in photoresponse curves (see Figure 1 in reference 69). The detectivity of 8 x l O ~ cmHzl/2W-] at 2.5 l~m and T=77K was obtained for the 22 A well QWIP. This low detectivity is related to its small gain and absorption quantum efficiency (,-, 1.4 x 1 ()- 3) at 2.5 Bm. Each of these devices was found to be background limited at liquid nitrogen temperatures ( 77 K) or higher. Typical p-type GaAs/A1GaAs OWIPs peak within 3 0 - 5 0 meV above the cut-off wavelength. 7~ The magnitude of the photoresponse decreases with decreasing wavelength thereafter. However. the peak response in these GaAs/ GaInP QWIPs is typically 2 0 0 m e V or more above the cut-off wavelength. Moreover, the 22 A wide well sample appears to have two peaks to its spectral response. These unique spectral shapes were proven, through the device modeling by a Kane Hamiltonian, to be caused by the influence of the spin splitoff band. Therefore, since there is significant coupling between the heavy hole and the split-off and between the heavy hole and the light hole, the observed spectral shapes result from the dual influence of split-off extended state absorption with light-hole extended state absorption. The lower energy peak seen in the 22 A wide sample corresponds to the appearance of the second lighthole state in the continuum of the standard q u a n t u m well. The 32 A and 55 A wide well samples are wide enough for the second light-hole state to be buried in the standard quantum well so it cannot contribute to photoresponse. p-type GaAs/GalnAsP QWIP

To increase the cut-off wavelength, a lattice-matched GaAs/ Gao.71Ino.29Aso.39Po.61 OWIP was grown by LP-MOCVD with 50 periods of 30 A wide GaAs quantum wells separated by 280 A wide GaInAsP (Eg=l.8 eV at T= 300 K) barriers. 72 The effect of the quaternary alloy was to reduce the valence band offset relative to lattice matched GaAs and thus increase the cut-off wavelength. A photovoltaic effect is observed from the photoresponse curve (see Figure 1 in reference 72). This arises from an asymmetric quantum well potential profile. Such asymmetry could be related to either a structural difference in the two q u a n t u m well interfaces or to dopant migration during the material growth. The photoresponse has a peak around 4 lum with broad maxima. Compared with the GaAs/GaInP OWIP reported in the last section, the peak wavelength increased slightly. This OWIP remains background limited up to a detector temperature of 100 K for biases between - 7 . 5 V and +2.5 V. The activation energies derived from Arrhenius plots of clark current, 170, 204. and 226 meV for bias voltages +9, - 9 , and 0.1 V respectively, agree with the cut-off wavelengths observed in the photoresponse curve. 72

144 Handbookof Infrared Detection Technologies p-type GalnAs/InP QWIP Sengupta et al. reported a lattice-matched GaInAs/InP p-type QWIP. 7~ The

QWIP structure was grown by GSMBE on InP substrate and consisted of 30 periods 10 A uniformly Be-doped (p= 3 x 1 () 1~ cm- 3) Ga(~.47In(}.s ~As wells and 500 A thick, uniformly Be-doped (1 x 1 ()l 7 cm- ~) InP barriers. 200 ~m diameter mesa was fabricated by wet chemical etching. The 4 5 ~ incidence responsivity at 80 K was measured as a function of wavelength at different bias voltage (see Figure 2 in reference 73). The peak photoresponse shifts slightly with bias and an increase in magnitude of response was observed with positive bias voltage. They attributed the increase in response with magnitude of the bias voltage to the increased photoexcited carrier escape and the increased carrier drift velocity. 4.4.2 p-type LWIR Q WIPs p-type GalnAsP/GalnAsP QWlP

To achieve LWIR detection, J. Hoff et al. proposed a fully quaternary ptype O W I P . 74'7~ The result, according to theory, should be a QWIP with a cut-off wavelength of approximately 1() ~m. p-type lattice-matched Gao.87Ino.13Aso.74Po.26/Ga(~.~,2In(~.3sAs(j.22P(j.7~ QWIP with 30 A wells and 2 8 0 A barriers was grown by LP-MOCVD on GaAs substrate. One remarkable feature of this q u a t e r n a r y - q u a t e r n a r y QWIP is that the photoresponse is extremely broad, from 2.5 lam to 1 () btm. 74 This unique spectral shape occurs because of the dual influence of the light-hole extended state and the split-off extended state. It also indicates that high quality quaternaryquaternary superlattices should be inherently capable of multi-color photoresponse.

4.5 n-type QWIPs Compared to p-type QWIPs, it is easier to realize longer infrared wavelength detection with n-type OWIPs due to the smaller conduction band discontinuity. Also, due to the smaller electron effective mass and higher mobility, n-type OWIPs have shown excellent detector performance. In this section, we give some n-type QWIP examples that cover the wavelength range from MWIR to VLWIR. In addition, a multi-color QWIP example is also given at the end. 4.5.1 n-type t WIR Q WIPs

LWIR is the most important atmospheric window since room temperature objects have a 10 l.tm peak wavelength. LWIR QWlPs based on m a n y different material systems have been demonstrated. In this section, detector performance of three n-type GaInAs/InP OWlP samples will be described.

GalnAs(P) based OWIPs on GaAs, InP. and Si substrates for focal plane arra!ts

145

n - t y p e GalnAs/InP Q W I P

GalnAs/InP OWIP was firstly fabricated by Gunapala et al. in 1991. 76 Compared to the most popular GaAs/AlxGal_xAs OWIPs, the Al-free GaInAs/InP QWIPs have the following advantages" 1. more straightforward device fabrication process since no passivation is required; 2. higher electron mobility due to the lower effective electron mass: 3. binary InP barrier layers have an inherently lower defect density which results in lower dark current. Samples (A, B, and C) were grown on semi-insulating InP subsrate by a LPMOCVD. 77 Each QWIP structure consists of 20 periods of 60 A GaInAs wells, separated by 500 A InP barriers. Only the middle 50 A of each well was doped with Si and doping densities are 1.7• 5• and 1.7x1018 cm -3 for sample A, B, and C, respectively. 4 0 0 x 4 0 0 Bm OWIP mesa detectors were fabricated by photolithography and ECR-RIE dry-etching. Since each mesa has 4 5 ~ sidewall after dry-etching (with BCI ~/C12/Ar etch chemistry), the OWIPs were measured with normal incidence with front-illumination geometry. The dark current for the three OWIPs was measured as a function of bias voltage (see Figure 3 in reference 77).The large increase of dark current with the doping density in the well is due to the decrease in activation energy. The peak detection wavelength was measured at 9.013m for all three samples (see Figure ] in reference 77). One noteworthy aspect to point out is the exponential dependence of the responsivity with applied bias (see Figure 2 in reference 77). '

I

'

I

100 ~~ >

j~

10

oo

/JJ

jl

ill /

Z

~mm /

. m

oo c

'

1

m~ m m~

/

0.1 0.01

.f

!

nh

1 E-3

, 0

I

2

,

Bias (V)

I

4

,

-

6

Figure 4.15 Peak responsivit!l of a corrugated (;alnAs/InP C)iVII~with corrugation period 10 I~m. T= 77 K.

146 Handbookof Infrared Detection Technologies This behavior is very different from what is usually observed in GaAs/AIGaAs QWIPs, whose responsivities increase linearly with bias then saturate at higher bias. A likely cause is avalanche gain, which is due to impact ionization of electrons excited out of the quantum wells. 7s Sample B yields a maximum detectivity of 3 . 5 x l O l ~ at an operating bias 0.75V. Further, by changing the doping density by a factor of 10, the detectivity changes by over a factor of 20 (see Figure 6 in reference 77). With a more effective light coupling scheme, the detector performance of GaInAs/InP OWIPs can be improved dramatically. Figure 4.15 shows a peak responsivity curve for a GaInAs/InP OWIP which is covered by corrugation on its detector mesa. 79 This QWlP has the identical structure parameters as sample B, and the corrugation period is 1()~m. The corrugation is created with dryetching by a ECR-RIE and has approximately 60 ~ sidewall. Compared with a peak responsivity of 33.2 A/W at a bias voltage of + 5 V for sample B, 77 a record high responsivity of 218 A/W was obtained at bias of + 5 V (mesa top positive) for this corrugated GaInAs/InP QWIP. 4.5.2 n-type VLWIR Q WIPs

n-type InGaAsP/InP Due to the smaller conduction band offset between GaInAsP and InP, peak photoresponse wavelengths ranging from 8-2()IJm and beyond are possible by changing the composition of the quaternary, g() Two lattice-matched GaInAsP/ InP QWIPs (A and B: E,=O.82eV) and one GaInAs/InP (C, the same structure parameter as sample A from Section 4.4.1) was grown by LP-MOCVD. gl The GaInAsP/InP OWIP structure consists of 20 periods of 65 A GaInAsP wells, separated by 500 A InP barriers. The bandgap of well material for sample A and B are 0.95 eV (2=1.3 ~tm) and 0.80 eV (,;.=1.55 lam). The n doping level in the well is 1.7 x 10 ~7 c m - 3. The relative response curves were measured by a FTIR at 80 K (see Figure 4 in reference 81 ). For a bias of - 1 V (mesa top negative), the 50% cut-off wavelengths for the three samples are 9.3, 10.7, and 14.2 ~m for A, B, and C, respectively. The FWHM (as Ak/)~) of the three samples are 36%, 10%, and 5.5% for A, B, and C, respectively. The switch from a bound-to-bound to a bound-to-continuum transition between sample B and A is evident. The measured peak detectivity for the three samples were 1.1 x l 0 s cmHzl/2W -1 (T=3OK), 2x l 0 s cmHzl/2W-~ (T=80 K), and 8x 1 0 9 cmHzl/2W -1 (T=80 K) for sample A, B, and C, respectively. n-type InGaAs/GaAs QWIP The longest detection wavelength for OWIP was reported by Perera et al. on a InGaAs/GaAs QWIP with cut-off wavelength at 35 ~tm. 82 The OWIP structure consisted of 20 periods of 9 8 A In().()sTGa().gj3As wells (Si doped to n=4xlOl~ -2) and undoped 4 1 8 A GaAs barriers. This QWIP structure showed a bound-to-quasibound intersubband transition. 240 • 240 ~m detector mesas were fabricated by wet chemical etching. The peak photoresponse was

GalnAs(P) based QWIPs on GaAs. InP, and Si substrates for focal plane arrags

14 7

measured about 35~tm at T=4.2K and bias O.15V. The measured peak at responsivity and detectivity were 0.3 A / W and l.O• T=4.2 K. 4.5.3 Multi-color QWIPs

Compared to QWIPs based on GaAs substrates, a major advantage of InP-based OWIPs is that both 3-5 ~m and 8-20/3m spectral bands can be realized with lattice-matched n-type devices (see Figure 9 in reference 83 ), which is not true for GaAs-based QWIPs. A two-color n-type QWIP was grown on semi-insulating InP by GSMBE.s4 The MWIR MOW consisted of 20 periods of 3 5 0 A Ino.s2Alo.48As barriers and 35A Int).s3Gat~.47As (n-doped to 5 x 1 0 1 7 c m -3) quantum wells. The LWIR MOW consisted of 20 period 500 A InP barriers and 55A Ino.s3Gao.47As wells (n-doped to 5xl()lTcm-~). 4 ( ) 0 x 4 0 0 p m detector mesas were fabricated by ECR-RIE dry etching. Since no middle metal contact was made, this QWIP operated as a voltage tunable two-color detector. At T= 7 7 K, for bias less than 7 V, only the 8.5 Jam peak was observed and for bias larger than 7 V, the 4 lum peak was observed.

4.6 Low cost QWIP FPA integrated with Si substrate 4.6.7 Overview of QWIPs on Si

Currently, most QWIPs are grown on GaAs (GaAs/A1GaAs heterostructure) and InP (GaInAs/InP h e t e r o s t r u c t u r e ) s u b s t r a t e s . Unfortunately, the readout integrated circuits (ROICs) for FPAs are silicon-based. As seen in Section 4.3, complicated techniques, like flip-chip bonding, are necessary to hybridize the FPAs with the silicon ROIC. GaAs/A1GaAs OWIPs directly grown on Si have been reported, ss-87 OWIPs directly grown on Si can make use of large-area substrates, with higher thermal conductivity and mechanical strength. Most importantly, this arrangement makes monolithic integration with the ROIC possible. The major obstacle to this goal is the large lattice mismatch between silicon and III-V materials (4.1% between GaAs and Si and 8.1% between InP and Si), which creates very high densities of threading dislocation in the epitaxial layer. These dislocations dramatically deteriorate the III-V material's optical and electrical qualities. To reduce the threading dislocation densities, many growth techniques, such as ex situ or in situ thermal annealing, low temperature nucleation, lateral epitaxial overgrowth, and insertion of a very thick, graded, or superlattice buffer, have been developed to improve the quality of GaAs and InP grown on Si. ss-91 Thermal annealing effectively reduces threading dislocation densities both in InP and GaAs epitaxial layers grown on Si. Annihilation and coalesence of dislocations are caused by dislocation movement under the thermal stress induced by annealing. 9~ Due to the larger immobile In atom in InP, stress induced

148 Handbook of Infrared Detection Technologies by thermal annealing can reduce threading dislocation density in InP-on-Si with less interdiffusion problems than in GaAs. 4.6.2 Growth of GalnAsllnP QWlP-on-Si

In this section, growth process of a OWIP-on-Si will be described in detail. Two 20 period GaInAs/InP OWIP structures were grown on Si substrate and on InP substrate by LP-MOCVD, respectively. 1~ The Si substrates used are GaAs-on-Si manufactured by Kopin Inc. with a GaAs coating thickness of 1 ~m. The {1 ()0) Si substrate has a 4 ~ miscut toward the [()11] direction. This misorientation is necessary to prevent the formation of anti-phase domains in the growth of III-V materials o n Si. 46 The growth of this QWIP-on-Si starts with substrate deoxydation and deposition of two different thin buffer layers. The substrate is heated up to 550~ under AsH3 flow, followed by growth of a thin GaAs buffer layer at 510C. After that, the substrate is cooled to 4()()~ and a 5()0 A thick InP nucleation layer is grown at growth rate of 3() A/min. Next, a thick buffer is grown, followed by a thermal annealing cycle. The temperature is raised to 500~ and a 6 ~m InP thick buffer layer is grown at growth rate of 180 A/rain. After every 1 ~m of InP. the growth is interrupted and an in situ thermal cyclic annealing (TCA) is performed. From room temperature, every annealing cycle consists of a temperature increase to 55()~ for 3 rain, followed by a decrease to room temperature. This procedure is repeated 3 times for every TCA. After TCA, the temperature is raised to 500~ and the next 1 ~m InP layer is grown. PHi/He is flown during the whole TCA process. Threading dislocation densities and rms surface roughness, for a 2 0 x 2 0 l.tm field estimated by atomic force microscopy (AFM) image lc~4 for thick InP buffer layer on Si, are 2 • 10 5 c m - 2 and 42 A respectively. After the growth of the entire 6 ~m InP thick buffer layer, the temperature is decreased to 480~ and a GaInAs/InP is grown. The details of the GaInAs/InP OWIP structure was described elsewhere. 77 X-ray diffraction spectra for QWIPon-Si and OWIP-on-InP are shown in Figure 4. ] 6. Satellite peaks generated by the periodicity of the MOW structure were observed. Satellite peaks were observed to the ] 7 th and 8 ~h order for ()WIP-on-InP and QWIP-on-Si samples respectively. More satellite peaks correspond to better interface quality and layer thickness uniformity. 4.6.3 Detector performance of GalnAsllnP QWIP-on-Si

Figure 4.17 shows the dark current Id measured as a function of bias voltage for a 20 period GaInAs/InP QWIP-on-Si sample. 4()()• detector mesas are fabricated with the process described in Section 4.2.4. For comparison, the dark current of a GaInAs/InP OWIP-on-InP with an identical structure is drawn on the same figure. At a bias voltage of - 1 .()V, the Ia of the QWIP-on-Si sample is 2.039 l.tA at T = 7 7 K . With the temperature change from 77K to 120K, the I,lOf the OWIP-on-Si increases only one order of magnitude. This behavior is very

GalnAs(P) based OWIPs on (;aAs. InP. and Si substrates for focal plane arrays

149

Figure 4.16 The measured (002) X-ray d(ffraction spectra.for ()WIP-on-Si and OWIP-on-InP samples.

different compared to the QWlP-on-InP. which changes by three orders of magnitude from 7 7 K to 120 K. This temperature-insensitivity can be attributed to the threading dislocations present in the QWlP-on-Si. The threading dislocations contribute a large temperature-insensitive leakage current under bias. Figure 4.18 shows a comparison of I,l at a bias of - 1 V for both the QWlPon-Si and the QWIP-on-InP as a function of temperature. For the purpose of comparison, a leakage current of 2.0 ~A was subtracted from the Ia for the QWIP-on-Si sample. Figure 4.18 shows almost the same I,t versus T relationship within the whole temperature range for the two samples. ~ s The GaInAs/InP QWIP-on-Si detector has a peak response at 7.80 ~tm, while the peak response of the GaInAs/InP OWIP-on-InP is at 8.15 ~tm. The small blueshift (0.35 ~tm) is due to the presence of some residual strain in the epitaxial layers which leads to a small change of q u a n t u m well parameters. This blueshift was also observed in a GaAs/A1GaAs OWIP on Si. ~~'~ 7 The 77 K peak responsivity for the QWIP-on-Si and QWIP-on-InP as a function of bias is shown in Figure 4.19. The two detectors have identical peak responsivities up to a bias voltage of (). 75 V. Beyond that. the responsivity of the QWIP-on-Si starts to trail off. At a bias voltage of 4.0 V, the peak reponsivity of 29.96 A/W for QWIP-on-InP is about 18 times higher than the 1.64 A/W responsivity for the QWIP-on-Si. The large increase of responsivity at higher bias voltage is attributed to an avalanche mechanism in the QWIP detectors. 78 Due to the presence of threading dislocations, the efficiency of the impact ionization

150

Handbook of Infrared Detection Technologies

~20~-~a mple C QWIP-On-Si ,

103

.

,

.

,

9

,

~

.

i1

,

......... l]

77___~

103

~- 10-~

101 120 77

10_9 1011

05

=

,

-4

-2

0 Bias (V)

2

-In 4

10 -1

lo10-7 10 -9 1011

Figure 4.17 Dark current measured as afimction of bias voltage for QWIP-on-Si and QWIP-on-InP sample.

Figure 4.181 d at bias - 1 V versus Tfor QWIP-on-InP sample (solid squares) and QWIP-on-Si sample (open circles, subtract 2.0 i~A for Id).

process in QWIP-on-Si seems lower than that of the OWIP-on-InP. Similar tendencies are also observed for M-based GaAs/AIGaAs OWIPs-on-Si, although the InP-based OWIPs-on-Si have three times higher responsivitiy. Lastly, the GaInAs/InP QWIP-on-Si detector showed a high detectivity of 2.3 • 109 cmHzl/2/W at a field of 0.89 kV/cm at T=77 K. This detectivity is one

GalnAs(P) based QWIPs on GaAs. InP. and Si substrates for focal plane arrays i

151

9

sample QWIP-on-lnP_

> (/) E 8. (/)

/~

n~ -

sample QWIP-on-Si e ~ e j

~ q l ~

OjO /

0.1

0.01

1E_3

u

0

,

,

2

,

I

4

Bias (V) Figure 4.19 Tilepeak responsivity comparison between OWIP-on-InP (squares) and OWIP-on-Si( circles).

order of magnitude lower than that of the QWIP-on-InP, which is 3 . 3 • 10 l~ cmHzl/2/W. 4.6.4 How to fabricate a monolithic integrated FPA with Si substrate

With the successful growth of a GaInAs/InP OWIP on Si substrate, a monolithically integrated infrared focal plane array becomes feasible. In Figure 4.20, we show a proposed fabrication process based on n-type GaInAs/InP OWIPs. 1. Fabrication of ROIC on Si wafer. 2. Passivation of ROIC surface. 3. Thinning and polishing the backside of Si ROIC wafer. 4. Growth of n-type GaInAs/InP OWIP structure on the backside of Si substrate. 5. Fabrication of detector array by standard lithography and etching process. 6. Passivation of surface of detector array. 7. Lithography and etch through the Si substrate. 8. Deposition of connection metal between every detector pixel and its corresponding ROIC.

4.7 New approaches of QWIP N e w characterization techniques of QWIPs

Despite the rapid development of the quantum well (OW) infrared technology, the intrinsic properties of the QWIPs have not been directly measured under the operating conditions of the detector. However, these details would be extremely useful for detector optimization. New techniques need to be explored for better QWIP characterization. K.K. Choi et al. utilized the surface corrugation to probe

152

Handbook of Infrared Detection Technologies

Figure 4.20 Schematic of the proposed monolithic FPA fi~brication process.

the absorption coefficient and the photoconductive gain of a OWIP under different operating conditions. 9s.~, With a series of varied corrugation period QWIPs, by measuring the detector responsivity as function of period of corrugation, absorption coefficient ~ and gain g can be deduced by curve fitting accurately. Novel QWlP devices

For most applications of n-type QWIPs, effective light-coupling schemes are very important. Based on a corrugated structure. N.C. Das et al. incorporated dielectric or metal coverage on the top to improve performance of the corrugated quantum well infrared photodetectors (C-QWlP) in two wavelength regimes. 97 With non-absorbing Si3N4 coverage, the responsivity of the C-OWIP was improved 3.3 times in the 8 Ftm range and 1.8 times in the 14 ~m range. K.K. Choi et al. improved the C-OWIP detector performance by adding a center vertical trench in each of the corrugations. 9s Ting et al. also proposed a submonolayer QWIP structure for improved n-type normal absorption. ~8 With the insertion of a submonolayer of large lattice mismatched material in the quantum well, the wave function within the quantum well will be modified and therefore make the normal incidence possible. Development of two-color or multi-color OWIP detectors is highly desirable for future high performance IR systems. Without the stacking of different MOW, a postgrowth bandgap engineering technique was recently utilized to tune the wavelength of OWIPs and to achieve multi-color detection. L. Fu et al. used single

GalnAs(P) based QWlPs on GaAs. InP. and Si substrates for focal plane arrays

15 3

high-energy proton implantation and rapid thermal annealing to tune the spectral response ofOWIPs. A redshift as large as 1.8 l~m was observed. 99 With integration of OWIPs with other semiconductor devices, m a n y novel applications are made possible. Liu et al. reported a OWIP-LED for pixelless largearea imaging applications. 1~176OWIP (MWIR or LWIR) and LED (near-IR emission) structures are grown on top of each other. Under forward bias, the decrease of serial resistance of OWIP with the detection of MWIR or LWIR radiation increases the bias voltage on the LED, thus increasing its near-IR (NIP,) emission. This device is actually a MWIR and FWIR to NIR converter. The resulting NIR emission from theLED can be imaged easily using the well developed Si-CCD. Down to nanoscale

It is well know that dipole selection rules forbid absorption of photons polarized in the plane of the wells. Different light coupling schemes have been developed. To solve this problem fundamentally, people start to create the lateral electron confinement with different nanoscale structures, like q u a n t u m dot, q u a n t u m wire, and q u a n t u m grid, in addition to the usual vertical confinement by the material layers. S. Kim et al. demonstrated an InGaAs/InGaP q u a n t u m dot detector for MWIR detection. 1~176 The InGaAs q u a n t u m dots were formed by intrinsic strain due to a lattice mismatch. Normal incidence photoconductivity was observed at a peak wavelength of 5.5 ~m. X.O. Liu et al. demonstrated a 9 l~m A1GaAs/GaAs q u a n t u m wire infrared detector which was based on a V-grooved substrate. 1~ L. P. Rokhinson et al. demonstrated a q u a n t u m grid infrared detector, l~ The grid pattern, with a very narrow linewidth in the OWIP active region, was fabricated by electron-beam lithography and reactive ion beam etching. The coupling efficiency increase at very narrow linewidth, gave evidence for lateral q u a n t u m confinement.

4.8 Conclusions In this chapter, we have given a brief review of current Al-free GalnAs(P) QWIPs grown on InP, GaAs, and Si substrates. Device physics, fabrication, and m e a s u r e m e n t have been described in detail. Fabrication processes for infrared FPAs and indium solder bumps are also described in detail. After more t h a n 10 years of research and development, infrared OWIP detectors have moved from the laboratory to the thermal imaging market. Today, OWIP research is still very active and requires more development to reach the QWIP's full potential. Higher operating temperature and multi-spectral OWlP cameras are the most straightforward goals. With new processing technology, a m u c h smaller and larger format infrared FPA based on Al-free OWIP structures is likely to be available on the market soon. Even more promising are the same FPAs monolithically integrated on Si, which, as discussed (or should have been), can lead to a simpler process and a significant reduction in camera costs.

154 Handbookof InfraredDetectionTechnologies

References 1. L. Esaki and H. Sakaki, IBM Tech. Disc. Bull. 20, 2 4 5 6 ( 1977). 2.W. Mason and ]. R. Waterman, Appl.Phys.Lett. 74, 1633 (1999). 3. K. Adamiec, A. Rogalski and ]. Rutkowski, J.Tech.Phys. 38, 431 (1997). 4. A. Rogalski, Infra.Pbys.Technol. 3 8 , 2 9 5 (1997). 5. T. Tung, L. V. DeArmond, R. F. Herald. P. E. Herning, M. H. Kalisher, D. A. Olson, R. F. Risser, A. P. Stevens, and S. ]. Tighe, Proc. SPIE 173 5, 109 ( 1992). 6. T. ]. Lyon, ]. A. Vigil, ]. E. ]ensen, O. K. Wu, ]. L. Johnson, E. A. Fatten, K. Kosai, G. Venzer, V. Lee, and S. M. Johnson, J. Vac. Sci. Tecbnol. B 16, 1321 (1998). 7. T. Naegele, Electronics 7, 76 ( 1989). 8 H. C. Liu, C. Y. Song, A. Shen, R. Dudek, M. Gao, E. Dupont, Z. R. Wasilewski, M. Buchanan, P .H. Wilson, B. ]. Robinson, D. A. Thompson, G. ]. Brown, F. Szmulowicz, and J. Ehret, Proc. SPIE 4 2 8 8 , 130 (2001 ). 9. S. D. Gunapala, S. V. Bandara, J. K. Liu. E. M. Luong, S. B. Rafol, J. M. Mumolo, D. Z. Ting, J. J. Bock, M. E. Ressler, M. W. Werner, P. D. LeVan, R. Chehayeb, C. A. Kukkonen, M. Levy, P. LeVan, and M. A. Fauci, Infra. Phys. and Technol. 4 2 , 2 6 7 (2001). 10. B. F. Levine, K. K. Choi, C. G. Bethea, J. Walker, and R. J. Malik, Appl. Phys. Lett. 50, 1092 (1987). 11. B. F. Levine, S. D. Gunapala, J. M. Kuo, S. S. Pei, and S. Hui, Appl. Phys, Lett. 59,1864(1991). 12. C. G. Bethea, B. F. Levine, V. O. Shen, R. R. Abbott, S. J. Hseih, IEEE Trans. Electron. Dev. 38, 1118 (1991). 13. J. Hoff, C. Jelen, S. Slivken, E. Mihel, O. Duchemin, E. Bigan, M. Razeghi, G. Brown, and S. M. Hegde, Appl. Phys. Lett 6~, 1130 (1994). 14. J. Hoff, S. Kim, M. Erdtmann, R. Williams, J. Piotrowski, E. Bigan, and M. Razeghi, Appl.Phys.Lett. 67, 22 (1995 ). 15. S. D. Gunapala, B. F. Levine, D. Ritter, R. Harem, and M. Panish, Appl. Phys. Lett. 60, 637 ( 1992). 16. M. Razeghi, H. Yi, J. Diaz, S. Kim, and M. Erdtmann, Proc. SPIE 3 0 0 1 , 243 (1997). 17. C. Jelen, S. Slivken, ]. Hoff, M. Razeghi, and G. J. Brown, Appl. Phys. Lett. 70,360(1997). ] 8. G. Hasnain, B. F. Levine, D. L. Sivco, and A. Y. Cho, Appl. Phys. Lett. 56, 770(1990). 19. S. D. Gunapala, S. V. Bandara, J. K. Liu, E. M. Luong, N. Stetson, C. A. Shott, J. J. Bock, S. B. Rafol, J. M. Mumolo, and M. J. McKelvey, IEEE Tran. Electron. Dev. 47, 326 (2000 ). 20. M. Z. Tidrow, X. Jiang, S. S. Li, and K. Bacher, Appl. Phys. Lett. 74, 1335 (1999). 21. S. D. Gunapala, S. V. Bandara, J. K. Liu, W. Hong, M. Sundaram, P. D. Maker, R. E. Muller, C. A. Shott, and R. Carralejo, IEEE Trans. Electron. Dev. 45, 1890(1998).

GalnAs(P) basedQWIPs on (;aAs. InP. and Si substrates forfocal plane arra!ls 15 5 22. H. Schneider, M. Walther, C. Sch6nbein. R. Rehm, J. Fleissner, W. Pleteschen, J. Braunstein, P. Koidl, G. Weimann, J. Ziegler, and W. Cabanski, Physica E 7, 101 (2000). 23. C. A. Kukkonen, M. N. Sirangelo, R. Chehayeb, M. Kaufmann, J. K. Liu, S. B. Rafol, and S. D. Gunapala, Infra. Ph!ls. Technol. 42, 39 7 (2001 ). 24. C. A. Kukkonen, S. D. Gunapala, S. V. Bandara, J. K. Liu, and J. Lorens, Proc. SPIE 3 6 9 8 , 706 (1999). 25. S. D. Gunapala, J. S. Park, G. Sarusi, T. L. Lin, J. K. Liu, P. D. Maker, R. E. Muller, C. A. Shott, and T. Hoelter, IEEE Trans. Electron. Dev. 44, 45 (1997). 26. S. D. Gunapala, J. K. Liu, J. S. Park, M. Sundaram, C. A. Shott, T. Hoelter, T. L. Lin, S. T. Massie, P. D. Maker, R. E. Muller. and G. Sarusi, IEEE TrAns. Electron. Dev. 44, 51 (1997). 2 7. S. D. Gunapala, S. V. Bandara, A. Singh, J. K. Liu, S. B. Rafol, E. M. Luong, J. M. Mumolo, N. O. Tran, J. D. Vincent, C. A. Shott, J. Long, and P. D. LeVan, Proc. SPIE 3 6 9 8 , 687 (1999). 28. P. Yuh and K. L. Wang, Phys. Rev. B 3 8, 38 ( 1988 i. 29. G. Bastard, Phys. Rev. B 24, 5693 (1981 ). 30. G. Bastard and J. A. Brum, IEEEJ. Ouantum Electron. OE-22, 1625 (1986). 31. J. R. Hoff, M. Razeghi, and G. J. Brown, Phys. Rev. B 54, 10773 (1996 ). 32. Y. C. Chang and R. B. James, Phys. Rev. B 3 9 , 1 2 6 72 (1989). 33. S. D. Gunapala and S. V. Bandara. Microelectronics J. 30, 1057 (1999). 34. S. Gunapala and K. M. S. V. Bandara, Physics of Thin Fihns 21, Aademic Press. New York, 113 ( 1995). 35. B. F. Levine, C. G. Bethea, G. Hasnain, V. O. Shen, E. Pelve, R. R. Abbott, and S. J. Hsieh, Appl. Phys. Lett. 56, 851 (1990). 36. L.S. Y u a n d S . S. Li, Appl. Phys. Lett. 59, 1332 (1991). 37 J. Chu and S. S. Li, IEEE ]. Ouant. Elect. 33(7), 1104 ( 199 7). 38. Y. H. Wang, S. S. Li, J. andChu, P. Ho, Appl. Phys. Lett. 64, 727 (1994). 39. J. Chu, S. S. Li, and P. Ho, Appl. Phzds. Lett. 69, 1258 (1996). 40. Y. H. Wang, J. Chu, S. S. Li, and P. Ho, J. Appl. Phys. 76, 6009 (1994). 41. B. F. Levine, C. G. Bethea, and G. Hasnain, Appl. Phys. Lett. 56, 851 (1990). 42. W. A. Beck, Appl. Phys. Lett. 63, 3589 ( 1993). 43 A. Zussman, B. F. Levine, J. M. Kuo, and J. de Jong, J. Appl. Phys. 70, 5101

(199]).

44. B. F. Levine, J. Appl. Phys. 74, R 1( 1993 ). 45. A. Y. Cho, The Technology and Physics of Molecular Beam Epitaxy. New York: Plenum Press ( 1985). 46. M. Razeghi, The MOCVD Challenge Vol 1. Bristol: Adam Hilger (1989). 4 7. M. Razeghi, The MOCVD Challenge Vol 2. Bristol: Adam Hilger (1995). 48. G. Hasnain, B. F. Levine, C. G. Bethea, R. A. Logan, J. Walker, and R. J. Malik, Appl. Phys. Lett. 54, 2515 (1989). 49. G. Sarusi, B. F. Levine, S. ]. Pearton, K. M. S. Bandara, and R. E. Leibenguth, Appl. Phys. Lett. 6 4 , 9 6 0 ( 1994 ). 50. N. T. Gordon, SemicondSci. Technol. 6, C106 (1991 ).

15 6 Handbookof InfraredDetection Technologies 51. C. J. Chen, K. K. Choi, M. Z. Tidrow, and D. C. Tsui, Appl. Phys. Lett. 68, 1446(1996). 52. G. Hasnain, B. F. Levine, C. G. Bethea, R. R. Abbott, and S. J. Hsieh, I. Appl. Phys. 67, 4361 (1990). 53. J. H. Lau, Low Cost Flip-Chip Technologies for DCA, WLCSP, and PBGA Assemblies, New York: McGraw-Hill (2000). 54 .S.D. Gunapala, S. V. Bandara, A. Singh, J. K. Liu, S. B. Rafol, E. M. Luong, J. M. Mumolo, N. O. Tran, J. D. Vincent, C. A. Shott, J. Long, and P. D. LeVan, IEEE Transactions on Electron Devices, 47, 963 (2000). 55. M. ]. Varnau, S. Yeh, Flip Chip/BGA Workshop, Binghamton, New York (October 1996 ). 56. T. Flynn, C. W. Argento, and ]. O'Brien, Proc. SPIE 3 9 0 6 , 8 (1999). 57. A. B. Frazier and M. G. Allen, ]. Microelectromech. Syst. :2, 87 ( 1993). 58. E. K. Yung and I. Turlik, IEEE Trans. Components, Hybrids, and Manufacturing Technology 14, 549 ( 1991 ). 59. W. W. Flack, S. White, and B. Todd, Proc. SPIE 3 6 7 8 , 49 (1999). 60. J. H. Lau (Ed.) Flip Chip Technologies. New York: McGraw-Hill (1996). 61. Y.-H. Kim, J.-H. Choi, K.-S. Choi, H. C. Lee, and C.-K. Kim, Proc. SPIE ] 0 6 1 , 60 (1997). 62. J. Jiang and M. Razeghi (unpublished). 63. H. H. Martijn, U. Halldin, P. Helander, ]. Alverbro, and ]. Y. Anderson, Proc. SPIE 3698, 789 (1999). 64. N. Y. Aziz, R. F. Cannata, G. T. Kincaid, R. J. Hansen, ]. L. Heath, W. J. Parrish, S. M. Petronio, and ]. T. Woo]away, Proc. SPIE 3698, 766 (1999). 65. F. Bertrand, ]. L. Tissot, and G. Destefanis, Proc. SPIE 3316, 713 (1998). 66. R. F. Cannata, R. ]. Hansen, A. N. Costello, and W. ]. Parrish, Proc. SPIE 3698,756(1999). 67. N. Y. Aziz, G. T. Kincaid, W. J. Parrish, J. T. Woolaway, J. L. Heath, Proc. SPIE 3 3 6 0 , 80 (1998). 68. H. Kulah and T. Akin, Proc. SPIE 3 6 9 8 , 778 (1999). 69. J. Hoff, X. He, M. Erdtmann, E. Bigan, and M. Razeghi, ]. Appl. Phys. 78, 2126(1995). 70. F. Szmulowiz and G. Brown, Appl. Phys. Lett. 66, 1659 (1995). 71.F. SzmulowizandG. Brown, Phys. Rev. B 52, 13203 (1995). 72. ]. Hoff, S. Kim, M. Erdtmann, R. Williams, I. Piotrowski, E. Bigan, and M. Razeghi, Appl. Phys. Lett. 67, 22 ( 1995). 73. D. K. Sengupta, S. L. Jackson, D. Ahmari, H. C. Kuo, ]. I. Malin, S. Thomas, M. Feng, G. E. Stillman, Y. C. Chang, L. Li, and H. C. Liu, Appl. Phys. Lett. 69, 3209(1996). 74. J. Hoff, C. Jelen, S. Slivken, G. ]. Brown, and M. Razeghi, Proc. SPIE 2 6 8 5 , 62(1996). 75. I. R. Hoff, M. Razeghi, and G. Brown, Phys. Rev. B 54, 10773 (1996). 76. S .D. Gunapala, B. F. Levine, D. Ritter, R. Harem, and M. B. Panish, Appl. Phys. Lett. 58, 2 0 2 4 (1991 ). 77. M. Erdtmann, A. Matlis, C. Jelen, M. Razeghi, and G. Brown, Proc. SPIE 3948,220(2000).

GalnAs(P) basedOWIPs on GaAs. InP. and Si substratesfor focal plane arrays 15 7 78. B. F. Levine, K. K. Choi, C. G. Bethea, J. Walker, and R. J. Malik, App1. Phys. Lett. 5 1 , 9 3 4 (1987). 79. J. Jiang and M. Razeghi, (unpublished). 80. S. D. Gunapala, B. F. Levine, D. Ritter, R. A. Hamm, and M. B. Panish, App1. Phys. Lett. 6 0 636 (1992). 81. M. Erdtmann, J. Jiang, A. Matlis, A. Tahraoui, C. Jelen, M. Razeghi, and G. Brown, Proc. SPIE 3 9 4 8 , 227 (2000). 82. A. G. U. Perera, S. G. Matsik, H. C. Liu, M. Gao, M. Buchanan, W. J. Schaff, and W. Yeo, Infra. Phys. and Technol. 42, 157 (2001 ). 83. M. Razeghi, M. Erdtmann, C. Jelen, F. Guastavinos, G. J. Brown, and Y. S. Park, Infra. Phys. and Technol. V42, 135 (2001 ). 84. C. Jelen, Gaxln~_.,.AsyP~_,j-Based n-type Long Wavelength Quantum Well Infrared Photodetectors: Growth, Characterization, and Fabrication. NU Doctoral Thesis, Evanston (1998). 85. D. K. Sengupta, W. Fang, J. I. Malin, J. Li, T. Horton, A. P. Curtis, K. C. Hsieh, S. L. Chuang, H. Chen, M. Feng, G. E. Stillman, L. Li, H. C. Liu, K. M. S. V. Bandara, S. D. Gunapala, and W. I. Wang, AppI. Phys. Lett. 71, 78 (1997). 86. E. R. Brown, F. W. Smith, G. W. Turner. K. A. Mclntosh, and M. J. Manfra, Proc. SPIE 173 5 , 228 ( 1992 ). 87. D. K. Sengupta, M. B. Weisman, M. Feng, S. L. Chuang, Y. C. Chang, L. Cooper, I. Adesida, I. Bloom, K. C. Hsieh, W. Fang, J. I. Malin, A. P. Curtis, T. Horton, G. E. Stillman, S. D. Gunapala, S. V. Bandara, F. Pool, J. K. Liu, M. Mckelvey, E. Luong, W. Hong, J. Mumolo, H. C. Liu, and W. I. Wang, J. Electron. Mater. 2 7 , 8 5 8 (1998). 88. T. E. Crumbaker, H. Y. Lee, M. J. Hafich, and G. Y. Robinson, ]. Vac. Sic. Technol. BS, 261 ( 1990). 89. Y. Ababou, P. Desjardins, A. Chennouf, R. Leonelli, D. Hetherington, A. Yelon, G. L'EspOrance, and R. A. Masut, J. App1. Phys. 8 0 . 4 9 9 7 (1996). 90. M. Yamaguchi, A. Yamamoto, M. Tachikawa, Y. Itoh, and M. Sugo, App1. Phys. Lett. 53, 2293 (1988). 91. S. Naritsuka, T. Nishinaga, M. Tachikawa, and H. Mori, Jpn. J. Appl. Phys. 34,1432(1995). 92. M. Razeghi, J. Jiang, C. Jelen, G. J. Brown, submitted for Proceeding of SPIE Photonic West 2 0 0 2 Symposium for Photodetectors Materials and Devices VII Conference, San Jose, CA (Jan 2002). 93. M. Erdtmann, GalnAs/InP quantum well infrared photodetectors on Si substratefor low-costfocalplane arrays. NU Doctoral Thesis, Evanston (2001). 94. J. Jiang, C. Jelen, M. Razeghi, and G. J. Brown, IEEE Photonic Tech. Lett. 14, 372(2002). 95. K. K. Choi, C. J. Chen, and D. C. Tsui, ]. Appl. Phys. 88, 1612 (2000). 96. C. J. Chen, K. K. Choi, L. Rokhinson, W. H. Chang, and D. C. Tsui, Appl. Phys. Lett. 75, 3210 ( 1999). 9 7. N. C. Das and K. K. Choi, IEEE Trans. Electron. Dev. 47, 653 (2000). 98. D. Z.-Y. Ting, S. V. Bandara, S. D. Gunapala, J. K. Liu, S. B. Rafol, and J. M. Mumolo, InfraPhys. and Technol. V 4 2 , 2 0 5 (2001 ).

158 Handbookof Infrared Detection Technologies 99. L. Fu, H. H. Tan, C. Jagadish, Na. Li, N. Li, X. Liu, W. Lu, and S. C. Chen, Appl. Phys. Lett. 78, 10 (200]). 100. H. C. Liu, Proc. SPIE 3 9 7 5 , 35 (20()()) 101. S. Kim, H. Mohseni, M. Erdtmann, E. Michel, C. Jelen, and M. Razeghi, Appl. Phys. Lett. 7 3 , 9 6 3 (1998) 102. S. Kim, M. Erdtmann, and M. Razeghi, J. Korean Phys. Soc. 35, 303 (1999). 103. S. Kim, M. Erdtmann, and M. Razeghi, Proc. SPIE 3 6 2 9 , 371 (1999) 104. X. O. Liu, N. Li, Z. F. Li, W. Lu, S. C. Shen, Y. Fu, M. Willander, H. H. Tan, C. Jagadish, and J. Zou, lpn. I. Appl. Ph!ls. 39, "3124 (20()()). 105. L. K. Rokhinson, C. J. Chen, D. C. Tsui, G. A. Vawter, K. K. Choi, Appl. Phys. Lett. 7 4 , 7 5 9 ( 1 9 9 9 ) . 106. M. J. Varnau and S. Yeh, Flip Chip/BGA Workshop, Binghamton, New York (October 1996). 107. T. Flynn, C. W. Argento, and J. O'Brien, Proc. SPIE 3 9 0 6 , 8 (1999). 108. A. B. Frazier and M. G. Allen, J. Microelectromech. Syst. 2, 87 ( 1993 ). 109. E. K. Yung and I. Turlik, IEEE Trans. Components. Hybrids, and Manufacturing Technology 14, 549 (1991 ). 110. W. W. Flack, S. White, B. Todd, Proc. SPIE 3 6 7 8 . 4 9 (1999). 111. J. H. Lau (Ed.), Flip Chip Technologies. New York: McGraw-Hill (1996). 112. Y.-H. Kim, J.-H. Choi, K.-S. Choi, H. C. Lee, and C.-K. Kim, Proc. SPIE 3061,60(1997). 113. J. Jiang and M. Razeghi (unpublished). 114. H. H. Martijn, U. Halldin, P. Helander, J. Alverbro, J. Y. Anderson, Proc. SPIE 3 6 9 8 , 789 (1999). 115. N. Y. Aziz, R. F. Cannata, G. T. Kincaid, R. J. Hansen, J. L. Heath, W. J. Parrish, S. M. Petronio, and J. T. Woolaway, Proc. SPIE 3 6 9 8 , 766 (1999). 116. F. Bertrand, J. L. Tissot, and G. Destefanis. Proc. SPIE 3 3 1 6 , 7 1 3 (1998). 117. R. F. Cannata, R. J. Hansen, A. N. Costello, and W. J. Parrish, Proc. SPIE 3698,756(1999). 118. N. Y. Aziz, G. T. Kincaid. W. J. Parrish, J. T. Woolaway, and J. L. Heath, Proc. SPIE 3 3 6 0 , 8 0 ( 1 9 9 8 ) . 119. H. Kulah and T. Akin, Proc. SPIE 3 6 9 8 , 778 ( 1999 ).

Chapter 5

InAs/(Gain)Sb superlattices: a

promising material system for infrared detection L. BOrkle and F. Fuchs

5.1 Introduction InAs and GaSb constitute a nearly lattice matched material system offering great flexibility in the design of optoelectronic devices. Because of the negative band overlap and similarities to the HgTe/CdTe material system, InAs/GaSb superlattices have been considered by Sakaki and Esaki as early as 1978.1 In 198 7 the material system was proposed for infrared detection devices by Smith and Mailhiot. 2 Intense work focussing on mid infrared laser devices started in the early 90s, 3 followed by first reports on detection devices with promising electrooptical properties comparable to the established Mercury-Cadmium-Telluride (MCT) material system. 4's However, for the practical use of InAs/(GaIn)Sb superlattices in infrared imaging devices, stable and reproducible growth conditions and device processing technology have to be established. Most demanding is the development of a stable and robust passivation. Regarding the latter, up to now only one report has been available."

5.2 Materials properties 5.2.1 Bandstructure of InAs/(Galn)Sb

superlattices

Because of their small difference in lattice constant, InAs and (GaIn)Sb form an ideal material system for the growth of semiconductor heterostructures. E.g., Gal_xInxSb with an indium concentration of ] 5% grows compressively strained on a GaSb substrate with a lattice mismatch of A a / a - 0.94%, while InAs is

160 Handbook of Infrared Detection Technologies under tensile strain with a lattice mismatch of A a / a = - 0 . 6 2 % . In an InAs/ (GaIn)Sb superlattice the compressively strained Gal_xInxSb layers can compensate for the tensile strain in the InAs layers. The average lattice mismatch of an InAs/(GaIn)Sb superlattice is given by ( Aa~

\aj

_ 2_

~2dsL

1

(1)

nMLaO

where ao is the lattice constant of the GaSb substrate, dsL is the superlattice period, and nML designates the number of atomic monolayers in a superlattice period. For a certain layer thickness ratio and indium mole fraction x, it is possible to achieve strain compensation on a GaSb substrate. InAs is a low band gap semiconductor with a direct energy gap of (). 35 eV, while GaSb has a direct gap of ().74eV at room temperature. 7 The relative alignment of conduction and valence bands is also of utmost importance. InAs and (GaIn)Sb establish a broken gap type II system, where the conduction band edge of InAs is lower in energy than the (GaIn)Sb valence band edge. The relative alignment of the band edges is shown in Figure 5. l(a). The band gap energy of the ternary Gal_xInxSb alloy depends on the In molar fraction x, and is well described by 8

E~(x) -

813.3 - 991.3x + 413x 2

(2)

[meV]

while the valence band offset of Gal_xlnxSb relative to the binary GaSb compound is given by

800 813.3

f

+67.4 - - - -

7

712.1

673.9

400 ...................

r(9 UJ

o

t

231.4

-T

J

T

GaSb InAs

40.3

NN~ LH

-28.8

349.5

410 -400

+29.2

Gao 8~lno1rSb

(a) unstrained

InSb

LH +31 7 ~.~-.-.-.-..---, -17.7 HH

InAs

Gao 851no~Sb

(b) strained

Figure 5.1 Band alignment between the constituents of an InAs/(Galn)Sb superlattice in: (a) unstrained condition: and (b) coherently strained to the GaSb substrate. The conduction band edge of unstrained InAs was chosen as zero energy.

InAs/(Galn)Sb superlattices: a promising material s!lstemfor infrared detection 161

AEv(x) = 234.7x-

78.7x 2

[meV]

(3)

In the case of the binary compound GaSb the band overlap with InAs is about 140 meV. 9 Because of the type II band alignment, the effective band gap of InAs/ (GaIn)Sb superlattices is smaller than the band gaps of its constituents. The band gap of an InAs/(GaIn)Sb superlattice can be continuously adjusted in a range between 0 and about O. 3 eV. -~ In particular, in the case of thick individual layers the superlattice can become semi-metallic with a vanishing band gap energy. Since electrons are mainly located in the InAs layers of the superlattice, whereas holes are confined to the (GaIn)Sb layers, optical transitions occur spatially indirectly and, thus, the optical matrix element for such transitions is relatively small. In addition, the wave function overlap strongly decreases with increasing superlattice period, which is inconvenient for detection applications in the far infrared. In order to compensate for the spatial separation of electrons and holes, the superlattices are commonly grown with ternary (GaIn)Sb layers instead of binary GaSb. By introducing indium into the binary GaSb material, these layers become compressively strained. The important influence of strain on the band alignment in the superlattice is shown in Figure 5.1 (b). Biaxial strain can be decomposed into a hydrostatic and a uniaxial component. The compressive hydrostatic strain component in the (Gain)Sb layers raises the band edge energies, while the uniaxial component leads to a splitting of heavy and light hole states, such that the heavy hole state becomes the valence band ground state. On the other hand, the tensile strain in the InAs layers leads to a similar effect for InAs but with opposite signs, thus leading to an overall reduction of the superlattice band gap. The band structure of an InAs/(GaIn)Sb superlattice is shown schematically in Figure 5.2. The band gap of the superlattice is determined by the energy difference between the electron miniband E1 and the first heavy hole state HH 1 at the Brillouin zone centre and can be varied continuously in a range between 0 and about 0.3 eV by adjusting the individual layer thicknesses and the indium molar fraction in the ternary (GaIn)Sb layers. 2 The dispersion of the heavy hole states exhibits a strong anisotropy, while the other subbands are nearly isotropic. The electronic structure of InAs/(GaInSb) superlattices can be calculated with satisfactory accuracy by using a three-band envelope-function approximation (EFA) formalism. 1~ Table 5.1 shows the results of such a calculation for a superlattice comprising 13 monolayers InAs and 8 monolayers Ga~.~sln~.~sSb. The superlattice is coherently strained to the lattice constant of the GaSb substrate and shows a residual net strain of Aa/a = - 0 . 8 9 • 1()-~. The zero of the energy scale is set to the value of the unstrained InAs conduction band minimum. Values in parentheses show the deviation to the results of a full EFA 8 • 8k.p EFA calculation. ~ For the electron miniband E~ and the first heavyhole band HH1, which determine the superlattice band gap, the values of the three-band EFA calculation are only some millielectronvolts larger than those of the 8 • 8 band calculation. Hence, the deviation for the fundamental band gap is

162

Handbook of Infrared Detection Technolo#ies

Figure 5.2 Band structure of a superlattice consisting of 1 ~ monola!lers InAs and 8 monolayers Gao.~ slno. 1sSb.

Table 5.1 R e s u l t s o f a t h r e e - b a n d EFA c a l c u l a t i o n for a s u p e r l a t t i c e c o m p r i s i n g 15 m o n o layers I n A s ($.995 n m layer t h i c k n e s s ) a n d 8 m o n o l a y e r s Gao.sslno.lsSb layers ( 2 . 4 6 2 n m layer t h i c k n e s s ) . The s u p e r l a t t i c e is c o h e r e n t l y s t r a i n e d to a GaSb s u b s t r a t e a n d s h o w s a r e s i d u a l n e t s t r a i n o f A a / a = - O . 8 9 x l O --3. T h e zero o f Energy is c h o s e n at t h e c o n d u c t i o n b a n d m i n i m u m o f u n s t r a i n e d InAs. Values in p a r e n t h e s e s s h o w t h e d e v i a t i o n to t h e r e s u l t s o f a full 8 xSk.p EFA c a l c u l a t i o n ll Band

EL HH1 LH1 HH2

Zone center q = () [meV] 242.3 9"3.6 -67.2 -205.3

{+3t (+5) (+42) (+18)

Zone edge q: = 7l'/ds-l. [meV] 386.1 9-5.6 -174.6 -203.7

1+7t {+51 I+36) i+17)

Miniband width AE [meV] 143.8 ().(14 1()7.4 1.6

1+41 1-().()1 ) (-16) t-l)

Effective mass m-, 0.026 -96.5 -().()32 2.4

(-().(1(12 t (+16.71 [ +().()()2) (+().61)

only 2 meV, which corresponds to a relative error of about 1%. However, for higher subbands the deviation from the results of a full 8 • 8 band calculation drastically increases. For superlattices with a band gap energy in the mid and far infrared, the individual layer thicknesses are typically in a range between :3 and 10 monolayers, leading to a strong coupling between electron wave functions of adjacent potential wells. An example for the distribution of the electron and hole wave functions is shown in Figure "3.3 for a superlattice consisting of 13 monolayers InAs and 8 monolayers Ga~.ssIn~.l sSb. The probability density of the electrons is concentrated in the InAs layers. However, due to the strong coupling in the electronic system there is a significant delocalization of the electron wave function over the superlattice. Typical values of the electron miniband width easily exceed 1 ()()meV. On the other hand, the heavy holes are strongly localized in the (GaIn)Sb layers. Thus. as a consequence of the type II

InAs/(Galn)Sb superlattices: a promising material S!lstem for infrared detection 9

I

'

9

InAs

0.6

I

9

(Galn)Sb

I

9

I

InAs

9

!

16 }

" i

(Galn)S

r ,%.,,,,. ..___.,

0.4

o,I

c-

O

13 r ::3

/

0.2

\

0.0

i/ i

9

| |

!

2

f~-~"-

L

6

I

8

,

I

10

112

Position z [nm] Figure 5. ] Probability densit!t distribution o.f the first electron miniband El and the first heavy hole state HH1 in a 13 ML InAs/8 ML Gao.~cslno. 1~Sb superlattice.

band alignment of the material system, electrons and holes are spatially separated. This is particularly disadvantageous for optical absorption, where a significant overlap of electron and hole wave functions is needed. The delocalization of electrons and, consequently, the optical absorption in an InAs/ (GaIn)Sb superlattice is promoted by a reduction in the electronic confinement. This can be achieved either by growing thinner (Galn)Sb barriers or by introducing more indium into the (GaIn)Sb layers, causing a reduction in the energy of the barrier material. ~(~ On the other hand, the effective masses in an InAs/(GaIn)Sb superlattice are not directly dependent on the band gap energy, as it is the case in a bulk semiconductor, j2 Compared to the electron effective mass of HgCdTe, the electron effective mass (m*,~O.()2 in an InAs/(GaIn)Sb superlattice is larger, leading to a reduction in tunnelling contributions in the leakage currents of an IR photodiode. Also related to the larger effective mass in the superlattice is a higher combined density of states, which in turn compensates for the smaller optical matrix element of the superlattice. Thus, despite the type II character of the optical transitions in an InAs/(Galn)Sb superlattice, the optical absorption is comparable to that of HgCdTe. In addition, larger minority carrier lifetimes are expected because of smaller Auger recombination rates compared to MCT. 13

5.2.2 X-ray characterization High-Resolution X-ray diffraction is an important tool for the structural characterization of the MBE grown superlattices. In Figure 5.4 the X-ray diffraction pattern of an InAs/(GaIn )Sb superlattice, close to the (004) diffraction order of the GaSb substrate, is shown. The superlattice peaks can be observed up to the seventh order, indicating an excellent materials quality. From the angular

1 6 4

Handbook of Infrared Detection Technologies

t-

'

,

'

w

,

,

,

lOSl

lo o

10 [_

.~

1(~

,

\

2

lO4

8

,

,106

SL-O GaSh

9

~o~[

)1

SL-%AGasbI1 w

9

10 2

i

I I

lo'

I

j/ I

5

30.3 30.4

+5 .

I . 24

.

/

26

.

.I 28

,

10 3

10~ 10~ . 22

;

.

I

30

Incidence Angle

,

I

32

34

,

,

36

j

+6 -.., .~ 38

o~ [~

Figure 5.4 High-resolution X-raft diffraction pattern of an InAs/((;ahl)Sb superlattice close to the (004) reflection of the GaSh substrate. The superlattice has a period of 4.56 nm and a residual lattice mismatch of + 0 . 8 8 x 1 0 -~

spacing Am of adjacent peaks the superlattice period dsI. can directly be calculated using 2

dsL = 2 Ao)cos|

(4)

where 2 is the wavelength of the x-ray beam and O designates the average diffraction angle between the peaks considered for evaluation. The net strain in the superlattice (-~) •

-- sina)6aSb sine.OSL

(5)

can be determined from the relative position of the zero-order diffraction peak of the superlattice a)sL with respect to the position of the (004) peak a)G~Sb of the GaSb substrate. E.g., in Figure 5.4, the zero-order diffraction peak of the superlattice is found at a smaller angle than the peak of the GaSb substrate (coSL < a)caSb). Thus, the average lattice constant of the superlattice in growth direction is larger than the lattice constant of the GaSb substrate and the net strain of the superlattice is compressive with a lattice mismatch of +0.88 • 1()- 3. For a residual mismatch below 3 • 1 ()- 3, strain relaxation is not observed in the growth of a 1 l.tm thick superlattice stack. In the inset of Figure 5.4 the diffraction pattern close to the substrate reflection peak is shown at an enlarged scale. The oscillations that can be observed between the diffraction peaks are the so called 'pendellasung-fringes', which can be considered as the X-ray analogue to optical

InAs/( GaIn)Sb superlattices: a promising material s!!stemfor infrared detection 16 5

interference fringes. Such X-ray fringes can only be observed in very homogenous samples. Strain or thickness inhomogeneities in the order of A a / a > 10 -5 and A t / t > 3 x 10-~ lead to a significant reduction in the visibility of the interference fringes. 14 The oscillation period of the fringe pattern can be used to determine the overall thickness of the superlattice stack. For a full structural characterization of the superlattice the individual thicknesses of the InAs and Gal_xlnxSb layers as well as the In molar fraction x have to be determined. Information about the individual layer thicknesses in a superlattice can usually be gained by evaluating the envelope of the X-ray diffraction pattern. However, because of the very short periodicity of our IRsuperlattices, the superlattice diffraction pattern is spread out over a very wide angular range, and therefore not enough diffraction peaks can be observed to provide significant information on the individual layer thicknesses. The error of such an evaluation in the case of short period InAs/(GaIn)Sb superlattices remains too large. However, the missing structural parameters can be obtained by making use of photoluminescence measurements, providing the fundamental band gap of the superlattice system. 1s The fundamental band gap is very sensitive to both the InAs layer thickness and the In molar fraction in the ternary (GaIn)Sb layers. The EFA band calculation described in the previous section can be used to extract the individual layer thicknesses as well as the In molar fraction with satisfactory accuracy from the superlattice band gap determined from IR photoluminescence measurements and the superlattice period d~L and net strain zXa/a obtained from X-ray characterization. 5.2.3 Interfaces

InAs/(GaIn)Sb superlattices can be considered as a pseudo-quaternary system. At the transition from a Gal_xlnxSb to an InAs layer both the group II! and the group V elements change. Thus, depending on the shutter sequence during the MBE growth two interface types with either Ga-As or In-Sb interface bonds can be established (Figure 5.5), which differ strongly in interface bond length and strain introduced into the superlattice stack. The type and quality of interface bonds in a binary InAs/GaSb superlattice can be examined using R a m a n spectroscopy. In a R a m a n spectrum both InSb- and GaAs-bonds show characteristic modes by which the different interface types can be identified. 17 Because of their larger lattice constant with respect to GaSb, In-Sb interface bonds introduce compressive strain into the superlattice, whereas Ga-As interfaces induce tensile strain. In order to clarify the influence of different interface types on superlattice period and net strain, Figure 5.6 compares the Xray diffraction patterns of two InAs/GaSb superlattices, both having identical layer thicknesses. The diffraction pattern shown in the upper half of the figure, stems from a sample grown with InSb-like interfaces only. The superlattice has a period of 5.8 nm and is compressively strained with a strong lattice mismatch of A a / a = +4.1 • 10-3. Thus, in this sample the tensile strain of the InAs layers is overcompensated by the compressive strain of the InSb interfaces. By replacing

166

Handbook of Infrared Detection Technologies

Figure 5.5 Illustration of the possible interface t!lpes at the transition between a GaSh and an InAs la!ler.

105 ~ 100Periods: i E_

oO

0 ]GaSb

-1

10 3 r -~

.~a/a= +4.1x10 n

II

2 x InSb-like IF

I1

I

]1

II

I1 II

/

-~

,..-,.n=

+1 .

!

+2

t-

8 >,, tO

_c

105 ~- 50 Periods: 3 x InSb-like IF r 1 x GaAs-like IF

29

-1 . I

30

GaSb L0 ! m 11

3a/a = -0.4x10 .3 dS L = 11 1 nm

+1

,

"

31

i :-

-

32

Incidence Angle o [o] Figure 5.6 X-ray diffraction pattern of two InAs/(;aSb superlattices. B!! replacing ever!t one out of four InSblike interfaces with a GaAs-like interface, the lattice mismatch is reduced from +4.1 x 1 O- ~ to - 0 . 4 x 1 O-

every one out of four InSb-like interfaces with a GaAs-like interface, the lattice mismatch of the superlattice is reduced to - ( ) . 4 • 1()-~ (Figure 5.6 bottom). In addition, by exchanging only every fourth interface bond type, the periodicity of the superlattice stack is doubled and. as a consequence, the angular spacing between adjacent superlattice peaks in the diffraction pattern is reduced by a factor of 2.

InAs/( Galn)Sb superlattices: a promising material s!tstem for infrared detection

16 7

5.2.4 Sample homogeneity An important issue for the realization of large area focal-plane arrays (FPA), like, e.g., a 2 5 6 x 2 5 6 FPA with an active area of around 1 cm 2, is the homogeneity of the epitaxial layer across the wafer. Figure 5.7 shows the band gap variation of an InAs/(GaIn)Sb superlattice across a two-inch wafer as determined from photoluminescence spectroscopy measurements. The structural parameters of the superlattice exhibit an excellent homogeneity with a decrease in superlattice period by only 3% and a change in the lattice mismatch from - 1 . O to - 1.3 • 10 -3 when going from the center to the edge of the wafer. The change in the structural parameters is accompanied by an increase in the superlattice band gap, which can be attributed to an increase in confinement in the individual superlattice layers. Within the diagonal size of a 2 5 6 x 2 5 6 FPA, which is roughly 15 mm, the superlattice band gap increases by about 2 meV. This corresponds to a relative change in the band gap energy of less than 2%.

5.2.5 Residual doping For the design optimization of IR photodiodes, the residual doping of InAs/ (GaIn)Sb superlattices is a crucial parameter. The influence of the growth temperature on the residual doping of InAs/(GaIn)Sb superlattices was investigated, with a series of samples grown at substrate temperatures ranging from 360 to 440 ~C. 16 The dependence of the residual doping on growth temperature, as determined from magnetotransport measurements, is shown in Figure 5.8. Superlattices grown at substrate temperatures of 360 and 380~ exhibit residual n-type doping with donor concentrations of 2 x 1 ()~ 6 and 6 x 101 s c m - 3, respectively. At a growth temperature of 400~ the superlattices change to residual p-type 136 Diagonal size of a 256 x 256 FPA 134

E E J

132 O0

130

n

9

9

128

mL_ .~!~-

0

_ .0

AE < 2 % g

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

,

I

5

,

1'0

,

I

15

,

20

Distance from Center [mm] Figure 5.7 Band gap variation of an InAs/ ( GaIn )Sb superlattice across a two-inch wafer.

168

Handbook of Infrared Detection Technologies I

4X1016 all-"

'---'

9

I

'

I

'

i

'

- " - n-type - o - p-type

I

0

2x10 ~6

c-'-

-.~

lx1016

"o

8x10 ~s

~,

6xlff s

~ O

4x10 is I

360

,

I

380

,

I

400

,

I

,

420

Growth Temperature

I

440

[~

Figure 5.8 Dependence of the residual doping of lnAs/( (;aln )Sb superlattices on growth temperature.

doping with an acceptor concentration of 6 • 1 ()1 5 cm- 3, and growth temperatures exceeding 400~ result in a further increase in the acceptor concentration. In addition, a reduction of the in-plane electron mobility from 15 000 down to about 4 0 0 0 c m 2 / V s can be observed when the growth temperature for the superlattices is increased from 360 to 44()~ It is well known that the in-plane electron mobility in InAs/(GaIn)Sb superlattice is limited by interface roughness scattering and, thus, the electron mobility/~,, ~x A -2 is conversely proportional to the square of the height fluctuations, A, at the superlattice interfaces. 1s The reduction in the electron mobility by a factor of 5 with the increase in growth temperature can consequently be attributed to roughly a doubling of the height fluctuations A caused by increased interface roughness. 16 The conversion of the InAs/(GaIn)Sb superlattices from residual n- to p-type doping can be traced by photoluminescence (PL) measurements. For superlattices grown at temperatures below 4()0~ a strong increase in PL intensity is observed when the growth temperature is increased, whereas at temperatures above 400~ the PL intensity remains almost constant. As shown in Figure 5.9, for n-type samples, the PL intensity is a direct measure for the background donor concentration. An increase in the n-type background leads to a significant reduction in the PL intensity of the InAs/(GaIn)Sb superlattice. In contrast, for samples with p-type background the PL intensity is almost independent of acceptor concentration, as long as the latter does not exceed ~1 x 1017 c m - 3 . 1 6 Thus, the optimum growth temperature for low gap photodiodes can be found by using the PL intensity as a monitor. The optimum growth temperature for InAs/(GaIn)Sb superlattices is found at the lowest temperature which leads to a

InAs/(Galn)Sb superlattices: a promising material sflstem for infrared detection

"-----Co

t-

169

C)'

1000

O* c-

--J Q.

100 o

9 n-type samples p-type samples ,.I

1016

i

|

i

|

i

. l.l

1017

-..,,. 10~8

Doping Concentration [cm-3] Figure 5.9 Photoluminescence intensity of lnAs/ ( GaIn )sb superlattices as a flmction of doping concentration. The samples with doping concentrations below 1 • l()~; cm -~ are nominall!l undoped, all others are intentionally doped.

PL intensity close to the maximum. Using this optimization procedure, a residual background in the mid 1015cm -3 range can be obtained within a growth window of + 10~

5.3 Superlattice photodiodes 5.3.1 Diode structure

Figure 5.10 shows the cross-section of a typical InAs/(GaIn)Sb superlattice diode mesa. The epitaxial layers were grown by molecular beam epitaxy (MBE) at substrate temperatures around 400~ on undoped (001) oriented two-inch GaSb substrates. The group V to group III beam-equivalent-pressure (BEP) ratio was set between 2 and 4.5 for the growth of (GaIn)Sb and close to 6 for InAs. 14 Overall strain compensation of the superlattice stack was achieved by appropriate interface engineering. For growth on GaSb substrates, the smaller lattice constant of InAs is compensated by the larger lattice parameter of Gal_xInxSb (0.1 < x < 0.3), with the detailed strain balance depending on the InAs to (GaIn)Sb layer-thickness ratio. However, the interfaces across which both the group-III and group-V atoms change, have to be dealt with separately. For InAs/(GaIn)Sb superlattices, two types of interfaces with either Ga-As or In-Sb interface bonds can be realized, which differ strongly in interface bond length and strain introduced into the superlattice stack. Strain compensation was achieved by growing the layers with

170 Handbook of Infrared Detection Technologies

Figure 5.10 Cross-section of an InAs/( Galn )Sb superlattice diode mesa.

alternating GaAs-like and InSb-like interfaces, terminating each individual layer with its group-V element and starting the following layer with a monolayer of the respective group-III element. ~r. ~9.2~ For a band gap energy in the 8 - 1 2 l.tm atmospheric window the period of the superlattices is typically in a range between 5 and 7 nm corresponding to 1 5 - 2 0 monolayers, and the indium molar fraction in the ternary (GaIn)Sb layers was set close to 20%. A 5 0 0 n m thick, 3 x 1() is cm -3, Be-doped, GaSb layer forms the p-contact of the diode. The following InAs/(GaIn)Sb superlattice comprizes 150 periods and is approximately 7 5 0 n m thick. The first 60 periods of the superlattice are intentionally p-doped with an acceptor concentration of NA "-, 5.7 x 1016 cm -3. The p-n junction of the diode is located between the 30 low p-doped (Na ~-, 2.3 x 101~ cm -3) and the following 40 nominally undoped, but residually n-type (ND "-~ 1 x 1016 cm -3) periods. The final 20 periods of the superlattice are highly n-doped with a donor concentration of ND '~' 1.8 x 1017 cm -3. The superlattice is capped with a 20 nm thick, Si-doped (N/) ~-, 2 x 1018 cm -3) InAs layer which forms the n-contact of the device. Be and Si are used as p- and n-type dopants for the superlattice. In the n-doped superlattice regions the InAs-layers are doped with Si, while the (GaIn)Sb-layers remain undoped. The p-doping of the superlattice is achieved by doping the (GaIn)Sb superlattice layers with Be and leaving the InAs layers undoped. Both contact metallizations of the diode are 500 nm thick. In the case of the pcontact it consists of a series of Ti/Pt/Au-layers, whereas for the n-contact, a Ti/ A u / P t / A u - c o n t a c t metallization is used. In order to protect the device from ambient influences, the superlattice diode mesas are coated with a SiN-based passivation. Figure 5.1 l(a) shows the band alignment simulated at thermal equilibrium for the above device at 77 K. The Fermi level is indicated by the dashed line. In the simulation, the one-dimensional Poisson equation was numerically solved along

InAs/( Galn )Sb superlattices: a promising material s!tstem for infrared detection

.0,'

9

9

!

9

i

9

i

9

171

i

0 5f GaSh InAs/(Galn)Sbsuperlattice I InAs

ot-il,....i'

-0

9

1

!

9

i

p-region

9

i

9

i

9

n-region

!

m

ii o Iii

O[

=d

1

~

10TM

(b)

~. f

0

(c)

,

1

/ 0.0

8

0.2

~

0.4

0.6

i 1 0.8

Position z [IJm] Figure 5.11 Simulation of a InAs/(Galn)Sb superlattice diode with a band gap of 145 meV in thermal equilibrium at 77 K. The Figure shows: (a) the band alignment: (b) the electric field: and (c) the carrier concentration as afunction of position in the diode.

the growth direction, treating the InAs/(GaIn)Sb superlattice as a bulk semiconductor. The superlattice properties such as relative band alignment and effective masses were determined from a three-band EFA calculation. The strength of the electric field as a function of position in the diode is shown in Figure 5.1 ] (b). Apart from the built-in electric field at the p-n junction, which is indicated by the dashed line, electric fields also appear at the intersection between different doping levels. The m a x i m u m electric field at the p-n junction at thermal equilibrium is Fo= 13 kV/cm and the built-in potential is Vb~= 124 mV. Assuming a triangular-shaped field dependence, the depletion layer width of the diode is given by d = 2 V~i/Fo = 191 nm. Figure 5.1J(a) also clearly shows that both the n- and the p-contact layers can serve as diffusion barriers for the minority carriers in the diode. Because of the long diffusion length of electrons in the superlattice (which is on the order of several microns and, hence, is beyond the dimensions of the diode) minority electrons in the p-doped superlattice region are reflected at the large band discontinuity in the conduction band between the InAs/(GaIn)Sb superlattice and the p-GaSb contact layer, leading to an increase in the dynamic impedance RoA of the superlattice diode. 21

172

Handbook of Infrared Detection Technologies

5.3.2 Diode processing

The processing of the mesa photodiodes requires four lithographic steps. These steps are illustrated in Figure 5.12. In the first processing step, the top contact of the diodes is defined by a lift-off technique (Figure 5.12(a)), followed by wet chemical etching of mesa structures using citric acid in the second step (Figure 5.12(b)). Since citric acid etches the InAs/(GaIn)Sb superlattice with a high selectivity with respect to GaSb, the etching process stops when the p-GaSb contact layer is reached. Figure 5.13 shows an electron microscopy image of a wet chemically etched mesa sidewall. Following the definition of the bottom pcontact metallization (Figure 5.12(c)) a SiN based passivation layer is deposited, which protects the diode mesa from ambient influences (Figure 5.12(d)). In the final processing step the passivation layer is opened in order to enable contact between the diode n- and p-contacts (Figure 5.12(e)). Figure 5.14 shows a microscope image of a processed InAs/(GaIn)Sb superlattice diode. In the figure, the diode mesa and the two contact pads of the diode can be clearly distinguished. For front side illumination of the device, a square area is left open in the top contact metallization. 5.3.3 Photo response

The cut-off wavelengths of InAs/(GaIn)Sb superlattice photodiodes can be tuned to cover the whole mid-IR spectral range (3-2OBm). Figure 5.15 shows the (a) Top Contact

(d) Passivation

(b) Defenition of Mesa

(e) Etching of Contact Holes

(c) Bottom Contact

Figure 5.12 Schematic of the individzlal steps in the processing of lnAs/( GaIn )Sb sz~perlattice diode mesa.

InAs/( GaIn)Sb superlattices: a promising material system for infrared detection

17 3

Figure 5.13 Electron microscope image of a wet chemicall!t etched mesa sidewall.

Figure 5.14 Optical microscope image of a processed InAs/ ( GaIn )Sb superlattice photodiode.

spectral responsivity of different photodiodes with superlattice band gaps ranging from 0.1 to 0.2 7 eV and having an active layer thickness of 0.8 l~m. All photodiodes exhibit high responsivities at normal incidence with peak values typically exceeding 2 A/W. 5.3.4 I-V measurements Current-voltage characteristics

The current mechanisms in a p-n junction can be differentiated by their characteristic temperature dependence. While the diffusion current changes with temperature as n2, the generation-recombination current is proportional to ni. Therefore, at high temperatures, the dark current in a diode is limited by diffusion currents, while generation-recombination currents can dominate at lower temperatures.

174 Handbookof Infrared Detection Technologies

Figure 5.15 Spectrally resolved photo response of several InAs/( (;aln)Sb superlattice diodes with band gaps rangingfromO. 1 to0.27 eV.

Figure 5.16 shows the current-voltage characteristics of an InAs/(GaIn)Sb superlattice diode with a band gap of 143 meV in a temperature range between 20 and 200 K. At 7 7 K, the forward branch of the diode I--V curve exhibits an ideality of/~ = 1.34, while at voltages above ,~(). 13 V the diode current is limited by the diodes series resistance of Rs-= 2.26 x 10 -3 f2cm 2. Towards higher sample temperatures the diffusion current more and more prevails against the generation-recombination current and thus, the ideality of the diode continues to decrease with increasing temperature and reaches a value o f / ~ - - 1.18 at 150 K. On the other hand, at lower temperatures the g-r current dominates significantly, leading to an ideality of/~ = 2 at 20 K. The reverse bias branch of the diode I - V curves can be subdivided into two voltage regimes. For small reverse biases up to approximately - 0 . 4 V the diode leakage current shows a strong dependence upon temperature. When the temperature is increased from 20 to 200 K the diode current density increases by more than nine orders of magnitude. While at high temperatures, diffusion currents dominate the leakage of the superlattice diode, which is obvious from the almost flat current-voltage characteristic at small reverse biases. Below 1 0 0 K the generation-recombination currents increasingly prevail, and the characteristic flat current-voltage characteristic of diffusion currents with a saturation current density Jo is replaced by the I - V characteristics of g-r currents, which is slowly increasing with reverse bias. At temperatures below 50 K, defect assisted tunnelling currents gain an increasing influence on diode leakage. At voltages below - 0 . 4 V the current-voltage characteristics of the superlattice diode only exhibit a weak dependence upon sample temperature,

InAs/( GaIn )Sb szIperlattices: a promisiug material s!istem for infrared detection

175

Figure 5.16 Current-voltage characteristics of an hlAs/((;ah~)Sh sllperlattice diode with a band gap of 143 meV in a temperature range between 20 and 2OO K.

and a strong increase of the diode current with increasing reverse bias can be observed. This behaviour is caused by electrons tunnelling from the valence band in the p-region across the superlattice band gap into the conduction band in the n-region of the diode and is known as Zener tunnelling. 22

Dependence ot the dynamic impedance R()A on temperature and band gap Further insight in the leakage current mechanisms in an InAs/(Galn)Sb super]attice diode can be gained by investigating the temperature dependence of the dynamic impedance at zero bias

dl

R(,A - ~ 1 : - o

(6)

as shown in Figure 5.17 for a diode with a band gap of 14 3 meV. The diode shows diffusion limited performance according to R~jA :x exp(E~/kT) down to 120 K. However, below about 75 K the dynamic impedance exhibits a temperature dependence according to RoA :x exp(E,t/2kT ), suggesting that the electrical performance of the diode is limited by g-r currents. The deviation of the measured data from g-r limitation at temperatures below 5() K can likely be attributed to an increasing influence of defect-assisted tunnelling currents at lower temperatures. The effective suppression of such tunnelling currents can be attributed to both a low density of defect states in the superlattice band gap, as well as a large out-of-plane reduced effective mass of ().02 6 m~)in the superlattice as compared with bulk semiconductors with the same band gap. The solid curve in Figure 5. ] 7 shows the result of a fit of the total dynamic impedance

176

Handbook of Infrared Detection Technolo~lies

Figure 5.17 Temperature dependence of the d!!namic impedance at zero bias R()A Jor an InAs/((;aln)Sb superlattice diode having a band gap of 14 ~ me ~'.

(R(,A)

+(R(,A)-'(,R

(7)

to the experimental R()A values, where the generation-recombination lifetime rcR in the depletion region, as well as the minority carrier lifetimes r,, and rh were used as fitting parameters. The dashed lines in the plot represent the contributions of the generation-recombination currents 2

rcR vl,~ (RoA)c,R-- edni

(8)

and the diffusion currents at electrically reflecting contact layers-'

(R()A)dif-

kT ( w 1,

+

w,, ) - 1

(9)

to the total dynamic impedance of the diode, respectively. For simplicity, equal minority carrier lifetimes were assumed for electrons and holes in the fitting procedure. The superlattice band gap was set to Eq-- 143 meV as determined from a measurement of the spectral responsivity of the diode. From the leastsquares fit a minority carrier lifetime of r , , - - r n - - l ( ) n s and a generationrecombination lifetime of r ( ; ~ - 2 8 n s were determined. For comparison, it should be noted in this context that for an InAs/Ga().7sln().2sSb superlattice with a band gap of 1 4 0 m e V Grein et al. theoretically predicted an Auger limited minority electron lifetime of r,, - 700 ns. 2 ~ This discrepancy can be explained by the fact that g-r processes seem to dominate the minority carrier lifetime of the

InAs/(Galn)Sb superlattices: a promising material s!!stem for infrared detection

177

superlattice. In addition, the contact layers may not be perfectly electrically reflecting, leading to an underestimation of the determined carrier lifetimes in the fitting process. Figure 5.18 plots the dynamic impedance R~A of several InAs/(GaIn)Sb superlattice diodes at 77 K as a function of their band gap. For comparison, the gray shaded area indicates the dynamic impedance range achieved with state-ofthe-art HgCdTe photodiodes. 2s The band gap energies of the superlattice diodes cover the whole range of the atmospheric transmission window between 8 and 12 ~m. The solid line in the figure shows the dynamic impedance as calculated with (8) and (9) based on the lifetimes determined previously. The dashed lines in the figure give the contributions of the diffusion and g-r currents to the total dynamic impedance of the diodes, respectively. Both contributions depend exponentially on the superlattice band gap. The transition between diffusion and g-r limited behaviour takes place for superlattice band gaps between 75 and lOOmeV at 7 7 K . While the dynamic impedance of diodes with a superlattice band gap below 75 meV is diffusion limited at 77 K. the electrical performance of diodes with a band gap above l()()meV is limited by g-r currents. A superlattice diode with a band gap of 143 meV (corresponding to a cut-off wavelength of 8.7~tm) exhibits a dynamic impedance of R o A = 1.Sk~2cm 2, while a diode with a band gap of 1 ()() meV (corresponding to a cut-off wavelength of 12.4 jam) reaches an R ~ A of 1 ()f2 cm 2. Promising data at even longer cut-off wavelengths were reported by Mohseni et al. 24 Hence, the dynamic impedances of InAs/(Galn)Sb superlattice diodes are comparable to the ones of state-of-theart HgCdTe photodiodes. 2 s.2~

Figure 5.18 D!tnamic impedance R~A ~[several InAs/( (;aln)Nb superlattice diodes at 77 K as a fllnction ~[ their band gap. The band gap energies ~I the diodes cover the ~vhoh' range of the atmospheric transmission window between 8 and 12 IJm. The gra!j shaded area indicates the d!lnamic impedance range achieved ~vilh state-of-the-art HgCdTe photodiodes.2 ~.2~,

178

Handbook of Infrared Detection Technologies

Surface leakage currents A major aspect in the development of photovoltaic detectors for the mid and far infrared is the development of a suitable passivation process. The function of an ideal passivation is to protect the photodiode against ambient influences without degrading its electrical and optical properties. Since the contribution of surface leakage currents to the total diode current, increases with decreasing size of the devices, a passivation becomes especially important for the very small photodiodes in a focal plane array (FPA). Under the assumption that the surface leakage current is proportional to the diode perimeter p the total diode current is given by

1(v) - ?(v) + j (V)p

(lo)

where the proportionality constant jp denotes the surface leakage current density per unit length and I a is the bulk contribution to the diode current, which is proportional to the diode cross-sectional area A. For the dynamic impedance at zero bias this yields 6

Ro---A ~(RoA) =

+ ~"

(11)

where the first term defines the bulk and the second term the surface contribution to the diode dynamic impedance. The parameter -1

d-

\av

(12) V=O

is the dynamic impedance of the surface leakage current per unit length jP at zero bias and, therefore, describes surface leakage behaviour of the diode. The value of r{') depends, among other things, on the passivation of the diode, band gap and temperature. The larger r Pll the smaller is the contribution of surface leakage currents to the overall diode leakage current. Since the second term in (11) is reciprocally dependent on the size of a diode mesa, surface leakage currents particularly dominate in small diodes. In the course of process development, a multitude of different coatings were tested for their suitability as a passivation of the mesa sidewalls in InAs/(GaIn)Sb superlattice diodes. In order to assess the influence of surface leakage on the superlattice diodes, current-voltage measurements on series of diodes with different perimeter-to-area ratios were performed. The series included diodes of different sizes having perimeter-to-area ratios in a range between ().011 and 0.048 ~m -a. Measurements on diodes with an even larger p/A value are difficult, since devices with a mesa size smaller than 8 0 • pm 2 cannot be contacted with a ball bonder. Figure 5.19 shows a graph of the reciprocal dynamic impedance (RoA) -1 at 77 K as a function of the perimeter-to-area ratio E/A for a series of differently passivated diodes having a 143 meV band gap. The different passivations are

InAs/(GaIn)Sb sziperlattices: a promising material s!!stem for infrared detection

179

Figure 5.19 Reciprocal dynamic impedance (R~A)-l at 77 K plotted as a fllnction of the perimeter-to-area ratio p/A for a series of differentl!l passivated InAs/( GaIn )Sb szlperlattice diodes having a 14 3 meV band gap.

denoted with the letters A, B, and C. Using ( 1 1 ) the surface leakage currents can be separated from the bulk dynamic impedance of the diodes. The bulk contribution to the total dynamic impedance is obtained from the intercept of the linear fit with the y axis, while the slope determines the contribution of the surface leakage currents. For the bulk dynamic impedance a value of (RoA)a= 1 4 6 0 f a c m 2 is obtained in all three cases, indicating that the bulk dynamic impedance is actually dependent only on the bulk properties of the diodes. On the other hand, the slopes of the graphs, which determine the surface contribution, differ significantly for the different passivations. Diodes with passivation A ) exhibit the worst leakage current behaviour with an 4j value of 104 kf2 cm, whereas the least degradation of the electrical performance of the superlattice diodes by surface leakage (~') = 843 kf2 cm) is introduced by passivation C. The influence of a passivation on surface leakage in even smaller diodes, e.g. like the ones in a FPA, can be estimated by extrapolating the graphs of the linear fits in Figure 5.19 to higher p/A values. The size of the individual photodiodes in the 2 5 6 • FPA (Figure 5.20) developed at the IAF is 38• 2, which corresponds to a perimeter-to-area ratio of ().1051.tm -~. Thus, for the photodiodes in a FPA it can be expected that with passivation A the bulk dynamic impedance (RoA) A is reduced by more than an order of magnitude by surface leakage currents, whereas with passivation C the bulk dynamic impedance is only degraded by a factor of three. In order to experimentally investigate the surface leakage currents in FPA pixels, a 2 5 6 x 2 5 6 FPA was hybridized to a silicon fan-out structure by a flipchip technique. The electrical contacts between the photodiodes and the silicon fan-out chip were realized by indium bumps. Figure 5.21 shows the results of measurements on a fan-out sample having a superlattice band gap of 120 meV. At 77K, the bulk contribution to the total dynamic impedance of the

180 Handbook of Infrared Detection Technologies

Figure 5.20 InAs/(Galn)Sb photodiodes in a 256 x 2 5 6 FPA. The FPA pitch is 40 I~m and the size of the individual diodes is 38 x 38 l~hi2 co rresponding to a perimeter- to-area ratio of O. 105 t2 m - i

'

0.04 -

I

9

-E]-77K -o-67

I

9

I

'

I

'

I

'

(RoA)A = 5 7 ~ c m 2

K (RoA)A = 549 ~cm 2

30

0.03 40 "i" v

0.02 ~ ~ ~ ~ : ~ ~

256x 256 FPA 60

Eg= 120 meV

loo

0.01

I

0.00

,__N

80

0.02

0.04

0.106 '

p/A Figure 5.21 Results of measurements on a 120meV.

0/08

0.10

200 500 0.12

[pm-1]

256x256 fan-out samph' with a superlattice band gap of

InAs/(Galn)Sb superlattice diodes is (R()A)4- 57 ~ c m 2. By extrapolating the linear fit, one would expect a dynamic impedance of R()A -- 3 3 ~ cm 2 for a pixel in a 2 5 6 • FPA. The actually measured dynamic impedance of 27 ~ c m 2 lies slightly below that value. However, it is still only a factor of 2 below the bulk dynamic impedance ( R o A ) A. With respect to 77 K, the bulk dynamic impedance at 67 K has increased by one order of m a g n i t u d e and reaches a value of (RoA)4= 549 f2 cm 2. The dynamic impedance of a FPA pixel is R()A - 108 f2 cm 2 and, hence, is degraded by a factor

InAs/(Galn)Sb superlattices: a promising material system for infrared detection

181

of 5 with respect to the bulk value. Thus, by increasing the diode temperature from 6 7 K to 77 K the bulk leakage current increases by an order of magnitude, whereas the surface leakage currents only double. At 67 K the contribution of the surface leakage currents to the overall diode current is therefore larger than at 77K. In order to reduce free carrier absorption and thermal stress in the 2 56 x 2:36 FPA hybrid the GaSb substrate was thinned to 30 ~m. Figure :3.22 shows the measured spectral responsivity of a FPA pixel under backside illumination through the thinned substrate. The fringe pattern that is superimposed on top of the spectral response of the photodiode is due to interference of the light reflected by the front and backside surfaces of the thinned GaSb substrate. 5.3.5 C - V measurements

The capacitance of a reverse-biased p-n junction is determined by the capacitance of the space charge in the depletion layer. The space charge capacitance of a diode is given by 27

c.(v)

-

-d(V)

(1])

where ~ is the dielectric constant, and d(V) is the width of the depletion layer. Bias-dependent capacity measurements can therefore be used to determine the doping level in a diode. Such measurements were performed on InAs/(Galn)Sb 5

15 10 I rrT"

I

Wavelength [Iam]

'1

~ ~

r

. ...,,

=3

2

I

256 x 256 FPA-Hybride Thinnedto 30 Iam Anti-reflectioncoated

. .....,

>

o0 r

8.

T=77K O'.1

0.2 I

O'.3 0'.4 0'.5 Photon Energy [eV] ,

|

i

,

0.6 I

,,

Figure 5.22 Spectral responsivit!l of a 2 56 • 2 56 t:PA pi.rel under backside illumination. The (;aSb substrate is thinned to 30 I~tn and tire photodiode is illuminated through the thinned substrate in order to reduce free carrier absorption and thermal stress.

182

Handbook of Infrared Detection Technologies

superlattice diodes using an HP 4192A impedance analyser. Figure 5.2 3 shows the measured reciprocal quadratic capacity of an InAs/(GaIn)Sb superlattice diode with a band gap of 143 meV as a function of applied bias. Two voltage regimes can be distinguished in the plot. At reverse biases below - 0 . 2 5 V a linear fit of the data yields a reduced doping level of Nred- 1.33 • ] 016 cm- 3 and a builtin voltage of Vbi--O.041 V. The deviation of the measured data from the linear behaviour around zero bias can be attributed to an increasing influence of the diffusion capacitance on the overall measured diode capacitance in that voltage regime. At reverse biases above - 0 . 2 5 V the slope of the curve in Figure 5.23 is smaller, thus leading to a higher reduced doping of Nred= 1.92 x 1016 cm- 3. The occurrence of these two voltage regimes can be explained by the doping profile of the device. The diode n region consists of a weakly doped region near the p-n junction (20 superlattice periods) and a heavily doped region towards the n contact (20 superlattice periods), whereas the p region of the device is relatively wide (70 superlattice periods) and intrinsically p-doped. At small reverse biases, the depletion layer only extends into the weakly doped n region. Because of the low doping level, the depletion layer expands rapidly into the weakly doped n region with increasing reverse bias, which according to equation (l 3) corresponds to a large slope in Figure 5.23. At a voltage of approximately - 0 . 2 5 V the depletion layer reaches the heavily-doped n region, and from then on expands more slowly with increasing reverse bias. The slower expansion of the depletion region results in the smaller slope in Figure 5.23. Because of the high n-doping level ND>>NA, the acceptor concentration approximately equals the reduced doping level NA~Nred in this voltage regime. In order to determine the exact doping levels of the diode, one has to take into account that in the n-doped regions of the device the nominal n-doping ~T~omis partly compensated by the residual p-doping NA of the superlattice, and therefore N D - Np)~ N3. In addition, in the investigated diode the nominal n-doping '

I

4x106

% (3 ,<

.

~o0~r~

~r

.

.

I

.

'

~ "I)

I

"

I

d= 1..92• 10~6cm-3

3x106 2x108 lx106

f 200 kHz I

-0.6 Figure 5.23

,

,

Ned = I

1.33• 10~6cm3 " ~ o o , ~ . \ ,

I

-0.4 -0.2 Voltage V [V]

,

0.0

C-V measurement performed on an InAs / ( (;aln )Sb superlattice diode at 90

K.

InAs/(GaIn)Sb superlattices: a promising material s!istem for infrared detection

183

NTnom(2) = 10 NT"n~ 1) is a factor level in the heavily doped s e m i c o n d u c t o r region ,,D "'D of 10 higher t h a n in the lower doped region. Using this information, the doping (2) levels NA, N(D1), and N D in the diode can be calculated from the two reduced doping c o n c e n t r a t i o n s determined in the C - V m e a s u r e m e n t . For the residual ptype doping of the InAs/(Galn)Sb superlattice a value of N A - 1 . 9 9 x 1016 cm -3 -- 6.01 x was obtained and the n o m i n a l n-type doping levels are NTn~ .,~ 1016 cm -3 and N~ ~ 6.01 x 10 ~7 cm -~. Using these values, an effective (1) - 4 . ( ) 2 x 10 16 c m - 3 and NI)(2) - 5.81 x l 0 17 cm -3 donor c o n c e n t r a t i o n of N D can be calculated for the weakly and heavily n-doped s e m i c o n d u c t o r regions, respectively. ,

5.3.6 Noise measurements

Noise m e a s u r e m e n t s on the InAs/(Galn)Sb superlattice diodes were performed using a spectrum analyser w h i c h enables a direct m e a s u r e m e n t of the noise c u r r e n t as a function of frequency. During the m e a s u r e m e n t the sample was dispersed into liquid nitrogen. In order to reduce signal distortion, the diode c u r r e n t was pre-amplified directly at the sample holder and then fed into the s p e c t r u m analyser. Figure 5.24 shows the voltage-dependence of the noise c u r r e n t of an InAs/ (GaIn)Sb superlattice photodiode with a detection cut-off at 8.7 ~m at 7 7 K. The size of the investigated diode mesa was 1 5 0 • 300 ~tm 2. The noise c u r r e n t was m e a s u r e d in the frequency range between 1 ()()() Hz and 1195 Hz by averaging over 2 0 4 8 individual m e a s u r e m e n t s . The standard deviation of the m e a s u r e m e n t s was typically a r o u n d 15 %. The full curve in Figure 5.24 shows the calculated shot noise

i

9

i

I

9

i

9

i

i

'

!

o Measureddiode noise Shot noise

10 lo

t

1 .j

.~. 1011 T= 77K

1012

/

10-13 10-14

f= 1000...1195 Hz I

-0.5

,

I

-0.4

,

I

-0.3

,

I

-0.2

,

I

-0.1

,

I.

0 0

,

01

.1

Voltage V IV] Figure 5.24 Noise current as a fimction of applied voltage of an InAs/( GaIn )Sb superlattice photodiode with a detection cut-off at 8.7 iJm measured in the fi'equenc!l range between 1000 Hz and 1195 Hz at 77 K.

184 Handbook of Infrared Detection Technologies

12 -

(14)

2elAf

of the superlattice diode. By comparing the calculated values with the measured data points we find that, in the investigated frequency range, the diode noise close to zero bias is given by shot noise, whereas at reverse voltages exceeding - 0 . 1 3 V a significant deviation from the ideal dependence can be observed. The frequency characteristic of those contributions dominating at larger reverse biases clearly shows a 1/f dependence (Figure 5.2 5). Such additional 1/f noise contributions can either be associated with surface leaking currents, 28 or are attributed to tunnelling processes via defect states in the semiconductor band gap. At low frequencies the frequency dependence of the measured noise current is well described by a f-~.69 law, while at higher frequencies the slope in Figure 5.25 follows a f-o.34 relationship. The transition between these two noise regimes shifts to higher frequencies with increasing reverse biases. Such differences in the frequency dependence can be explained by tunnelling processes via different defect states. The thermally limited detectivity Dr*i, of the photodiode can readily be calculated from the measured diode noise current I,~ using

/aAS

(15)

where A is the area and R; denotes the responsivity of the device. The result of such a calculation for an InAs/(GaIn)Sb superlattice photodiode with a detection cut-off at 8.7~tm is shown in Figure :3.26. The calculation is based on a

Figure 5.25 Frequency dependence of the measured noise current of an InAs/(Galn)Sb superlattice photodiode with a detection cut-off at 8.7 I~m as a function of applied bias.

InAs/(Galn)Sb superlattices: a promising material system for infrared detection

185

responsivity of R;. - 2 A/W (compare Section :3.3.3). The maximum detectivity of 1.4 • 1012 cm x/Hz/W is reached in the photovoltaic operation mode at zero bias. In this case the detector noise can also be interpreted as Johnson noise

,

~/~A

(16)

Dth -- R2 V 4 k T

of the zero bias dynamic impedance R o A of the photodiode. 21 Close to zero bias, the detectivity of the photodiode is well above the fundamental background limit of D~Lm -- 1.2 • 1011 cm x/-Hz/W, whereas in the range where 1 ffnoise is the dominant contribution, the detectivity sharply drops below the background limit D~L m. Since the noise behaviour of InAs/(GaIn)Sb superlattice diodes in photovoltaic operation mode is limited by the Johnson-noise of the zero bias dynamic impedance RoA, the thermally limited detectivity of such devices can also be directly calculated using (16). Figure 5.2 7 shows the detectivity of a superlattice diode with a cut-off wavelength of 8.7 ~tm in photovoltaic operation as a function of temperature. Background limited performance is obtained at temperatures below 90 K. At a temperature of 7 7 K the diode exhibits a dynamic impedance of R o A - 1 . S k c m 2 corresponding to a Johnson-noise limited detectivity of Dth -- 1.2 • 1012 cm x / ~ / W . This value is already one order of magnitude above the background limit D~3LrP. In Figure 5.28 the detectivity of different InAs/ (GaIn)Sb superlattice photodiodes is plotted as a function of their respective band gap. A photodiode with a cut-off at 1 2 . 4 p m reaches a dynamic impedance of RoA - 10 f2 c m 2 at 7 7 K, corresponding to a detectivity of D~h - 9.7 • 101~ cmx/Hz/W which is still well above the background limit. Thus, InAs/(GaIn)Sb superlattice photodiodes with cut-off wavelengths in the 9

lo '2

i

9

i

9

i

9

T=77K RI= 2A/W

i

9

l

".

~

!

lO~ c~ "5

10~o

-0.5

,

-0.4 I

~

-0.3 I

,

-0.2 I

~

-0.1 I

,

0.0 i

Voltage V [V] Figure 5.26 Experimentally determined detectivity of an InAs/(Galn)Sb superlattice photodiode with a a detection cut-off at 8.7 lzm as afunction of applied bias.

186

Handbook of Infrared Detection Technologies

T e m p e r a t u r e T [K] 1O0 50

200 150 9 I

10TM

9

E .o.

9

I

9

I

E = 143 meV =2

1013

T-

I

10' 2

.Ic

>,

10~

-

"5 0 ~

io ~

I

,

5

I

,

10

ll5

,

I

Inverse Temperature

!

,

20

25

IO00/T [K-~]

Figure 5.27 Detectivity of an InAs/(Galn)Sb superlattice photodiode with a cut-off at 8.3 I~m as a function of temperature.

Wavelength 11 10

12 I

~"

1012

'

I

9

Xc [pm] 9

I

8

I

T= 77K

'

QO

oO

R = 2A/W

9

1(~

::3

3

E .o.

8 .~ >

o

10~

o'SS~

...... O

_ . - - - - - -

. . . . . b.-- --~ . . . . . . . . . . . . .

o

9

I

1oo

,

I

125 Band g a p E

,

I

150

[meV]

Figure 5.28 Detectivity of lnAs/( Galn )Sb superlattice photodiodes having different band gap energies.

atmospheric transmission window between 8 and 12 pm exhibit background limited performance in photovoltaic operation. The dynamic impedance of such devices is comparable to those of state-of-the-art HgCdTe diodes with the same band gap energies. It can be expected that superlattice diodes having comparable q u a n t u m efficiencies to HgCdTe photodiodes will be realized in the future by developing devices with thicker superlattice stacks.

InAs/( Galn)Sb superlattices:a promising materialSZlStemfor infrareddetection 18 7

5.4 Summary and outlook It has been shown that the performance of infrared detectors comprising InAs/ (GaIn)Sb superlattices in the active layer can be comparable to M(TT photodiodes. Large-sized IR photodiodes with cut-off wavelengths in the third atmospheric window (8-12 ~tm wavelength) with background-limited performance were fabricated. The development of processing technology is still in progress, showing first successful results on passivated focal plane arrays hybridized on silicon fan-out structures. The present technology is based on two-inch GaSb substrates. However, three-inch substrates are already commercially available. From a fundamental point of view, the antimonide-based materials system exhibits advantages regarding tunnelling currents because of a higher effective electron mass which undergoes only a weak dependence on the effective band gap. In addition, advantages are expected because of smaller Auger recombination rates compared to MCT. Especially for applications at very long wavelengths and under low background conditions, InAs/(GaIn)Sb superlattice photodiodes are expected to outperform M(TT based imagers.

References 1. H. Sakaki, L. L. Chang, G. A. Sai-Halasz, (7. A. Chang, and L. Esaki, Twodimensional electronic structure in InAs-GaSb superlattices, Solid State Commun. 26 (9), 589 (1978). 2. D. L. Smith and (7. Mailhiot, Proposal for strained type II superlattice infrared detectors, J. Appl. Phys. 62 (6), 2545 (1987). 3. D. H. Chow, R. H. Miles, T. (7. Hasenberg, A. R. Kost, Y. H. Zhang, H. L. Dunlap, and L. West, Mid-wave infrared diode lasers based on GaInSb/InAs and InAs/A1Sb superlattices, Appl. Phys. Lett. 67 (2 5), 3700 (199:3). 4. J. L. Johnson, L. A. Samoska, A. (7. Gossard, J. L. Merz, M. D. Jack, G. R. Chapman, B. A. Baumgratz, K. Kosai, and S. M. Johnson, Electrical and optical properties of infrared photodiodes using the InAs/Gal_xInxSb superlattice in heterojunctions with GaSb, J. Appl. Phys. 80 (2), 1116 ( 1996). 5. F. Fuchs, U. Weimar, W. Pletschen, J. Schmitz, E. Ahlswede, M.Walther, J.Wagner, and P. Koidl, High performance InAs/Gal_xInxSb superlattice infraredphotodiodes, Appl. Phys. Lett. 71 (22), 3251 (1997). 6. L. Biirkle, F. Fuchs, R. Kiefer, W. Pletschen, R. E. Sah, and J. Schmitz, Electrical characterization of InAs/(GaIn)Sb infrared superlattice photodiodes for the 8 to ]2 l.tm range, Mat. Res. Soc. Syrup. Proc. 607, 77 (2000). 7. O. Madelung (ed.), Intrinsic Properties of Group IV Elements and III-V, II-VI, and 1-VII Compounds, Vol. III/22a, Landolt-B~rnstein New Series, SpringerVerlag, Berlin ( 198 7). 8. (7. G. van de Walle, Band lineups and deformation potentials in the modelsolid theory, Phys. Rev. B 3 9 (3), 18 71 ( 1988 ). 9. I. Vurgaftman, J. R. Meyer, and L. R. Ram-Mohan, Band parameters for III-V compound semiconductors and their alloys, ]. Appl. Phys. 89 ( 11 ), 5815 (2001).

188 Handbookof InfraredDetectionTechnologies 10. F. Szmulowicz, E. Heller, K. Fisher, and F. Madarasz, Optimization of Absorption in InAs/InxGal_xSb superlattices for long-wavelength infrared detection, Superlattices and Microstructures 17 (4), 373 ( 1995). 11. F. Szmulowicz, Private communications. 12. G. Bastard, Wave Mechanics Applied To Semiconductor Heterostructures, Les Editions de Physique, Les Ulis C6dex ( 1986). 13. C. H. Grein, P. M. Young, M. E. Flatt6, and H. Ehrenreich, Long wavelength InAs/InGaSb infrared detectors: Optimization of carrier lifetimes, J. Appl. Phys. 78 (12), 7143 (1995). 14. F. Fuchs, L. B/irkle, W. Pletschen, ]. Schmitz, M. Walther, H. G611ich, N. Herres, and S. Mfiller, InAs/Gal_xInxSb infrared superlattice diodes: Correlation between surface morphology and electrical performance, Proc. SPIE 3794, 41 (1999). 15. F. Fuchs, L. B/irkle, R. Hamid, W. Pletschen, E. Sah, R. Kiefer, and J. Schmitz, Optoelectronic properties of photodiodes for the mid- and far-infrared based on the InAs/GaSb/A1Sb materials family, Proc. SPIE 4288, 171 (2001 ). 16. L. B/irkle, F. Fuchs, J. Schmitz, and W. Pletschen, Control of the residual doping of InAs/(Galn)Sb infrared superlattices, Appl. Phys. Lett. 77 (11), 1659 (2000). 17. N. Herres, F. Fuchs, J. Schmitz, K. M. Parlor, J. Wagner, J. D. Ralston, and P. Koidl, Effect of interracial bonding on the structural and vibrational properties oflnAs/GaSbsuper]attices, Phys. Rev. B 53 (23), 15688 (1996). 18. C. A. Hoffman, J. R. Meyer, E. R. Youngdale, F. J. Bartoli, and R. H. Miles, Interface roughness scattering in semiconducting and semimetallic InAsOa~_xlnxSb superlattices, Appl. Phys. Lett. 63 (16), 2210 ( 1993 ). 19. J. Wagner, J. Schmitz, F. Fuchs, U. Weimar, N. Herres, G. Trhnkle, and P. Koid], Structural characterization of InAs-(Galn)Sb superlattices for IR optoe]ectronics, Mat. Res. Soc. Syrup. Proc. 421, 39 (1996). 20. J. Wagner, J. Schmitz, N. Herres, F. Fuchs, and M. Walther, Spectroscopic e]lipsometry for characterization of InAs/Gal_xInxSb superlattices, I. Appl. Phys. 83 (10), 5452 (1998). 21. M. B. Reine, A. Sood, and T. J. Tredwell, Photovo]taic Infrared Detectors, in R. K. Willardson and A. C. Beer (ed.), Mercury Cadmium Telluride, Vol. 18 of Semiconductors and Semimetals, chapter 6, pp. 201, Academic Press, New York (1981). 22. U. Weimar, F. Fuchs, E. Ahlswede, J. Schmitz, W. P]etschen, N. Herres, and M. Wa]ther, Tunneling effects in InAs/GaInSb superlattice infrared photodiodes, Mat. Res. Soc. Symp. Proc. 4 8 4 , 1 2 3 (1998). 23. C. H. Grein, P. M. Young, H. Ehrenreich, and T. McGill, Auger Lifetimes in Ideal InGaSb/InAs Superlattices, J. Electron. Mat. 22 (8), 1093 (1993). 24. H. Mohseni, M. Razeghi, G. J. Brown, and Y. S. Park, High-performance InAs/GaSb superlattice photodiodes for the very long wavelength infrared range, Appl. Phys. Lett. 78 (15), 2107 (2001). 25. A. Rogalski, Heterostructure infrared photovoltaic detectors, Infrared Physics and Technology 41 (4), 213 (2000).

InAs/(Galn)Sb superlattices: a promising material system for infrared detection

189

26. J. L. Johnson, The InAs/GaInSb strained layer superlattice as an infrared detector material: An overview, Proc. SPIE 3 9 4 8 , 118 (2000). 27. S. M. Sze, Physics of Semiconductor Devices, 2nd edition, John Wiley & Sons, New York ( 1981 ). 28. S. P. Tobin, S. Iwasa, and T. J. Tredwell, ] / f noise in (Hg,Cd)Te photodiodes, IEEE Trans. Electron Devices ED-2 7 ( 1 ), 43 ( 1980).

This Page Intentionally Left Blank

Chapter 6

GaSb/InAs supperlattices for infrared FPAs M. Razeghi and H. Mohseni

6.1 Type-II heterostructures 6.1.1 Historical review

In 19 77, Sai-Halasz and Esaki suggested type-II band alignment and some of its interesting physical behavior. ~ Soon after that they reported the optical absorption of type-II superlattices, 2 and later the semimetal behavior of the superlattice. 3 The applications of such a superlattice were proposed only several years after. 4 The flexibility of the material to cover a huge infrared range (2 to > 50~tm) and the possibility of a reduced Auger recombination rate s captured the attention of m a n y groups. Type-II heterojunctions have been found in m a n y applications of electronic devices such as resonant tunneling diodes (RTD) and hot electron transistors. However. perhaps the most important applications have been in optoelectronics, and m a n y significant results have been achieved from type-II modulators, ~ detectors, 7"s9'~(~ and laser diodes. 11.12 6.1.2 Definition of type-II band alignment

The band alignment of any heterojunction can be categorized as type-I, type-II staggered or type-II misaligned. In type-I heterojunctions, one material has lower energy for electrons and the holes and therefore both carriers are confined in that layer. In type-II heterojunctions, however, the electrons are confined in one material and the holes in the other. In the extreme case, which is called typeII misaligned, the energy of the conduction band of one material is less than the valence band of the other one.

192

Handbook of Infrared Detection Technologies

Figure 6.1 Type-I, type-II staggered, and t!lpe-II misaligned heterojunction and some of the material with these band alignments.

6.1.3 Features of type-II band alignment and their applications The special band alignment of the type-II heterojunctions provides three important features that are shown in Figure 6.2. These features are used in m a n y devices to improve the overall performance of the device. The first feature is that a superlattice with the type-II band structure can have a lower effective bandgap than the bandgap of each layer. This is an important issue for the applications in the mid and long infrared wavelength range, since one can generate an artificial material (the superlattice) with a constant lattice parameter but different bandgap. Recently, very successful detectors 7 and lasers 12 have been implemented in the 2 - 1 5 B m wavelength range with InAs/GaInSb superlattices lattice-matched to GaSb substrates. The second feature is the spatial separation of the electrons and holes in a typeII heterojunction. This p h e n o m e n o n is a unique feature of this band alignment and is due to the separation of the electron and hole potential wells. As a result of such spatial separation, a huge internal electrical field exists in the junction without any doping or hydrostatic pressure. High performance optical modulators have been implemented based on this feature. 6 The third feature is the zener-type tunneling in a type-II misaligned heterojuntion. Electrons can easily tunnel from the conduction band of one layer to the valence band of the other layer, since the energy of the conduction band of the former layer is less than the energy of the valence band of the later layer. Unlike a zener tunneling junction which requires heavily doped layers, no doping is necessary for such a junction. Therefore, even a semimetal layer can be implemented with very high electron and hole mobilities since the impurity and ion scattering are very low. This feature of type-II heterojunctions has been successfully used for resonant tunneling diodes (RTDs) and recently for the implementation of type-II q u a n t u m cascade lasers. 1

(;a~b/InAs supperlattices for infrared FPAs

19 3

Figure 6.2 Unique features of t!lpe-II heterojunctions and superlattices.

6.2 Type-II infrared detectors 6.2.1 Principle of operation The active layers of photovoltaic and photoconductive type-II detectors are made from superlattices with a type-II band alignment. Similar to a type-I superlattice, the allowed energy states form the 'minibands', due to the coupling of electrons and holes in adjacent wells. However, unlike type-I superlattices, one can adjust the bandgap of type-II superlattices from a finite value to virtually zero. These superlattices resemble a direct gap semiconductor, since the minimum of the miniband in m o m e n t u m space is located at zero. Knowing the band structure and the optical absorption process in type-II superlattices, one can practically use the conventional photovoltaic and photoconductive structures to realize high performance type-II detectors. In the following sections, we shall explain the band structure of the type-II superlattices as well as their optical absorption process.

6.2.2 Band structure of type-II superlattices Let us study the minibands of a type-II s u p e r l a t t i c e in a more detail. Figure 6.3 shows an example of the miniband energy profile versus the superlattices period. Note that the effective mini bandgap of the superlattice, that is the energy between the minimum of E1 and maximum of HH1 decreases as the period of the superlattice increases. This feature provides a wide tunability range for the design of IR detectors. The mini gap energy shrinks to zero at a period of about 150 A which represents the transition between semiconductor and semimetal in the superlattice. The miniband width for electrons and holes decreases as the period of the superlattice increases, since the coupling between the adjacent well decreases. This means that for a photovoltaic device, where the charge transport is perpendicular to the superlattice interfaces, the mobility decreases for longer period superlattices. Another important fact is that a set of superlattices can be designed with different layer thickness and a constant energy gap by reducing

194

Handbook of Infrared Detection Technolo#ies ]

l'-

.........

~'

1'

i

r"

ga2

-0,2

H3

H2

-O.4

..

0

SO

100

d c ,l

150

200

_

.

250

Figure 6.3 The miniband energy profile of t!lpe-II superlattices with equal InAs and GaSb la!ters versus the period of the superlattice.

the electron well width and increasing the hole well width appropriately. Although the energy gaps of the members of the above set are identical, their band structures are totally different. We used this property to en#ineer the band structure of the type-II superlattices such that devices based on these superlattices can operate at higher temperatures.

6.2.3 Optical absorption in type-II superlattices Selection rules Assuming an in-plane effective mass to the electrons and holes and retaining the decoupling of mi=_+3/2 and m j = + l / 2 states, the dispersion relation of electron, light-holes, and heavy-holes can be written as: -)

EHH.n

(k•

=

-Eg

-

HHn

ELH,n

(k_,_) =

-Eg

-

LH.

)

)

h-k~_ Ec.n(kt_) = En 4 - ~ 2me

h2k~_ 2M,1 /i2k2 2m.

(1)

(;aSbllnAs suplmrlatticesJbr infrared FPAs

195

w h e r e HH~, LHn and E, are the n th e n e r g y level in the well and Mr,, m~, and mc are the effective mass of each particle. The e n e r g y reference is the b o t t o m of the c o n d u c t i o n band. Using this method, the polarization selection rule of the i n t e r b a n d t r a n s i t i o n in type-II and type-I q u a n t u m wells will be as the following table w h e r e i

I-I -

-

(S p.,.lX

>-<

S 1': Z)

S p,,, Y > - <

IT/()

(2)

This value is related to the Kane matrix element Ep by

(3)

Ep -- 2mo FI 2

The value of Ep for most of the s e m i c o n d u c t o r s is b e t w e e n 20 and 23 eV. The second selection rule involves the envelope function q u a n t u m n u m b e r . In a type-I q u a n t u m well the transition b e t w e e n level n and m is only valid i f n + m is even. However, the strongest transitions are at m = n w h i c h leads to h i g h e r overlap of the w a v e f u n c t i o n s . In a type-II q u a n t u m well, the transition will be s t r o n g e r for h i g h e r values of n and m. The reason is that the electrons and holes at h i g h e r e n e r g y states have h i g h e r t u n n e l i n g probabilities, and h e n c e the tails of the electron and hole w a v e f u n c t i o n s have a higher overlap.

interband absorption in type-II

The absorption coefficient for the t r a n s i t i o n from the first heavy-hole level to the first electron energy level, ~un ~-v: ~(m). is p r o p o r t i o n a l to the density of the states and the optical m a t r i x element between HH ~ and E ~: O/HHI__+EI(O) )

a

Ep

--

e

mcM1

2

A-7-1<xklx,) I , _ _ _, •

mc + M 1

Y ( / T w - Eg - El - H H 1 )

(4)

47r 2e 2

n

ncmomf2

Table 6.1 P o l a r i z a t i o n s e l e c t i o n r u l e s for h e a v y - h o l e s , l i g h t - h o l e s a n d s p i n - o r b i t s p l i t - o f f t o conduction band transition with different polarization and propagation directions Propagation direction

E .,. Polarizali....

-E, I>olarizali....

E. l',,larizali,m

Type of transition

z

n/v/2 rI/v/6 n/v/3 _ _

n/v/2 n/v/6 n/x/3 n/v/2 n/v/6

2n/v/6

HH,~Em LH,,~E,, SO,,--'Em HH.-~Em LH,,-~Em

-

Fl/v/3 -

Fl/v/3 -

SOn-~Em HH,,-+Em

2n/v'6

LH,1--~Em

FI/v/3

S(),,~Em

x y

n/v/2 n/v'6 FI/v/3

-

-

196

Handbookof Infrared Detection Technologies

where Ep, mc, M1, El, and HH1 are defined in equations {1) and (2), n is the refractive index and c is the speed of light in v a c u u m , and Y is the step function. Z1 e )~1h are the electron and hole w a v e f u n c t i o n s in the first energy state. Therefore, the absorption of light in a type-I q u a n t u m well d o m a i n is step-like in frequency as s h o w n in Figure 6.4. In a type-II q u a n t u m well one should calculate the overlap of the valence and conduction band energy states. However, the problem can be simplified if we assume t h a t the p e n e t r a t i o n of the heavy holes is negligible into the electron well (this is an acceptable assumption since the effective mass of heavy holes is considerably high.) The heavy hole w a v e f u n c t i o n s can be written as h

2

Xr (Z) -- - ~ sin[kv(z - L/2)]Y(z - L/2)

(5)

2 xh(z) -- ~ s i n [ k v ( z + L / 2 ) ] Y ( - z - L/2)

w h e r e Xh is the w a v e f u n c t i o n of the left and right states and is normalized to 1 over the length +)v/2, w h e r e )v is the length w h e r e w a v e f u n c t i o n s decrease close to almost zero and it is m u c h bigger t h a n L. the width of the well. k,, is the w a v e n u m b e r in the z direction. Now we can define the even and odd w a v e f u n c t i o n s as h

-- ~

h

1

Xeven(Z)

1

[xh(z) qL xh(z)]

(6)

Xodd(Z) -- - ~ [xh(g) -- xh(z)]

the w a v e f u n c t i o n of electrons ~ 1e in the E 1 energy level can be written as x] (z) x] (z) -

AcCOS(kcz) Bcexp[~Cc(Z-L/2)]

]zl~
(7)

Izl > L/2

where kc and ~Ccare the w a v e n u m b e r of electrons inside and outside the well and Ac and Bc are found from the normalization of the w a v e f u n c t i o n s to unity. The overlap of the odd part of the hole w a v e f u n c t i o n and electron w a v e f u n c t i o n vanishes to zero while the even part gives h

e

,f2

(XevenlX]) -- 2 V ~Bc ~

kv

(8)

(;aSb/InAs SUl~perlattices for infrared FPAs

197

Figure 6.4 The absorption coefficient of a t!lpe-I quantum well in the.frequenc!t domain shows a step-like increase of the absorption coefficient.

Now the absorption coefficient of the type-II well can be expressed as _ oeHnl -~ E1(co)

2e2EpB~ mcM1 [ - x } ncmomh2~c m7 + ~I1 1 + x 2 + arctan(x)

(9)

x 2 - 2m'--2(/ira - Eg + A,, - El)

/i2Kc2

where Av is the offset of the valence band at the junction. Figure 6.5 shows the absorption coefficient in the frequency domain. In comparison to a type-I q u a n t u m well, the absorption coefficient does not increase rapidly at the onset of the energy gap since the overlap of the electron and hole wavefunctions is increasing at higher energies.

6.2.4 Modeling and simulation of type-II superlattices

Energy band modeling The modeling of the superlattice energy bands is a necessary step in the realization of type-II infrared detectors since it provides crucial information about the bandgap of the superlattice and the electron and hole wavefunctions. Based on this information, one can design a superlattice structure for a specific cut-off wavelength with maximized absorption coefficient and carrier lifetime. The basis for such modeling is the envelopefllnction approx'imation in which the band structure of a periodic or non-periodic heterojunction (superlattices or

198

Handbook of Infrared Detection Technologies

Figure 6.5 The absorption coefficient of a t!lpe-II quantmn well it) th(' l?equenc!t domain shows a slow increase in the absorption coefficient at the transition en('rg!l.

q u a n t u m wells) can be modeled. In this approximation, we assume that the electron or hole wavefunctions consist of two parts: a periodic part due to the regular crystal periodicity, and an envelope part due to the heterojunction. The envelope modulates, and is assumed to be m u c h larger than, the periodic part. Assuming a heterojunction of material A and B, the electron wavefunction. q,(r), can be written as */r)

{lo)

-

where r is the position vector in real space, fl~'x'B} (r) is the envelope function in layer A and B, and ul (r) is the periodic part of the Bloch function, and I runs over as m a n y bands as are included in the analysis. The envelope functions can be decomposed into components that are in-plane and perpendicular to the A-B junction: f}A,m (r• z) - ~ 1

e(ik:r )x(1A,B ) (z)

(11)

where r• and z are the perpendicular and parallel to the growth direction vectors, k• ky) is the perpendicular wavevector, S is the area of the sample, and xIIA'BI(z) is the parallel envelope function for band 1. The main goal is the calculation of the X functions. The Hamiltonian for the heterojunction is

p2 H -- 2mo + VA(r)YA + VB(r)Yt~

(12)

( ; a S b / h l A s s u p p e r l a t t i c e s f o r infrared F P A s

199

where p is the m o m e n t u m , m{~ is electron mass, Va(r) and VB(r) are the atomic potential in layers A and B. and Y,\= 1 in layer A and zero otherwise, and YB_=I in layer B and zero otherwise. We have

(13)

ul.{)(r ) Hut.o(r) - /e(a)v kLl.o ~a + u(B)yB) '~I.{~

where Ul,o(r) is the periodic wavefunction and El.{)is the energy of band 1 at k = 0 . Now applying the Hamiltonian to q/(r) and simplifying it, we find that the Z functions should fulfill the following for energy E:

DX = EX

(14)

where X=(ZI, Z,,,.... ) is a I xN vector of Xl for N different bands and D is an N x N matrix with elements Dl.,,, as following: Dr.,,,-

/~2 32

~(B)v h2k2 EI.r )YA + ~l.o ~t~ 2t - - - * 2mr

2mt~ Oz 2

ilk: &,,, + ~ (lip__, Ira> (15)

-~(llp=l,,z>2m{j a: where ~a and ~32 are the first and second order derivative operators, ! and m are two different bands, and al.,,, is the dirac-delta function (zero for l~=m and one for l=m). Now it is clear that matrix D is just the k . p matrix of the bulk materials A and B, except that: k-_ is replaced with ~ a n d k-_2 with ~a-~ and EI.~ depends on w h e t h e r one is in layer A or B. Since the conduction, heavy-hole, light-hole and spin orbit bands have considerable interaction in n a r r o w - g a p type-II superlattices, the eight-band k . p matrix (N=8) was chosen. The 8 • 8 matrix at k• is decomposed into two 4 x 4 matrices ofD Tand Dl:

IoT OlEKTI IT1 0

D~

XI

- E X X!

(16)

Since the spin-up and down are identical at k• eigen-energies are twice degenerate. Matrix D is 1

Go(z) + - -2m,~ pzFp~. ()

5

II 1

(DT=DI=D and Xl=Xl=X),

- \. ]Flp~ ()

~

Emt(z) - 2m~j P.,.i'[1 - -Y2 )P,

\/ ~Flp~. (}

I~7)

D=

~.~

-~

Flpz

Flpz

() I)

El ~l(z)

2m~ p'(YI + 2y2 )Pz

v~

- -

PzY2P~

mc--~PzY2Pz t's~(z} - ~

I

PzYl P~

where FI, 71, 72, and F. are semiconductor parameters and can be found in semiconductor data books such as Landolt-Bornstein. Ec(z), Etttt(z), Eijt(z), and Eso(Z) are the energy of the conduction, heavy-hole, light-hole, and spin-orbit sp|it-offbands at k = 0 in the bulk semiconductors. These are functions of position

200 Handbookof Infrared Detection Technologies since at different z values, different layers of semiconductor, with different band energies, exist. Although the energy of different bands at k - 0 is readily k n o w n for most of the binary and t e r n a r y semiconductors, the band lineup of two semiconductors at their heterojunction needs to be calculated.

Band alignment modeling The theoretical calculation of the band lineups at semiconductor heterojunctions has been a difficult task, especially since experiments show a wide range of measured values. Although m a n y different models have been suggested 14 for such calculations, they are not convenient for our modeling, since they require huge computations. Model-solid theory is provides a simple yet accurate method for the calculation of the band lineups. It also does not require, a posteriori, the strain effect, since strain will directly appear in the deformation potentials. Assuming that as is the lattice constant of the substrate and ae is the lattice constant of the epi-layer, the strain field parallel to the junction is

Ell -

as ae

1

(18)

The strain in the perpendicular direction is c• -- - D e l l

(19)

and the value of D for different crystal orientation is DO01 _ 2c12 c11

Dl1()_ Cll -+- 3 c 1 2 - 2c44. Cll -or-c12 --1-2C44 D 111 = 2 C11 -t--2C 12 -- 2C44

(2o)

Cll --}-2C12 + 4C44 where c l a, c 12, and c44 are elastic constants of the epilayer. W h e n the strain is along [0() 1 ], the position of the heavy-hole, light-hole, and spin-orbit bands can be calculated from A() 1 6E()()l EHH -- Ev,av + av(2gLL+ c• + T A()

1

][ )

9

] 1/2

ELH -- Ev.av + a,.(2clt + c~) ---~-) +-~6E,,,,z +-~ A?) + A,,aE,,,,~ +~(SE,,,,l) 2

(21)

A() 1 1[ , 9 1 ~/2 Eso - Ev.~,,r+ a,.(2ell + c~_) --(-~-)+,46E,,,,1 - 5 A~ + ZS,,SE,,,,1 + ~(SE,),,l) 2

where Ev.~,, is the average valence band energy and A()is the spin-orbit to valence band gap, av is the hydrostatic deformation potential of the valence band

(;aSb/InAs supperlatticesfor infrared FPAs 201

and ~Eool-= 2 b ( 8 • 81l) where b is the shear deformation potential. The conduction band can be calculated from Ec -- Ev,av + (av + ar

A() + 8• + - - ~ - + Eg

where ac is the hydrostatic deformation potential of the conduction band and Eg is the bandgap of the semiconductor.

6.3 Experimental results from type-II photoconductors 6.3.1 Uncooled type-II photoconductors in the )~ = 8-12 ~tm range The need for uncooled photon IR detectors

Currently available photon detectors have low operating temperatures, and hence require cryogenic coolers. However, in most of the applications, these coolers are not desirable because of their short lifetime and the added power consumption, weight, volume and costs. Commercially available uncooled IR imaging sensors use ferroelectric or microbolometer detector arrays. These sensors are inherently slow and cannot detect rapid scene changes needed for many applications. Some of the applications which require a fast detector response time (z < 30 ms) are: flee-space communication, proximity fuzes, active infrared countermeasure systems, missile detection/situational awareness for highly maneuvering airborne platforms, LIDARs, gated-imaging, and night vision systems. Thus there is a need for the development of high-speed uncooled detectors in order to meet the requirements of present and future applications.

Problems of currently available photon detectors Although photon detectors have gigahertz bandwidths, their high temperature detectivity is severely degraded due to several physical limitations. The existing infrared photon detectors can be categorized as interband, which are mostly HgCdTe and InAsSb, or intersubband q u a n t u m well infrared detectors(OWIP). There are some fundamental limitations, namely a fast Auger recombination rate in the interband detectors and a high thermal generation rate in the intersubband detectors, which drastically decrease their performance and ability for high operating temperature. Moreover, the difficulty of the growth, nonuniformity due to high sensitivity to the composition, and large tunneling currents in HgCdTe as well as the required sophisticated processing for normal incidence light coupling in n-type OWIPs, are the other drawbacks of the currently available IR photon detectors. Advantages of type-II superlattices for uncooled IR detection In comparison to HgCdTe, the higher effective mass of electrons and holes and the slower Auger recombination rate 16.17 lead to lower dark current and higher

202 Handbookof Infrared Detection Technologies operating temperature in type-II superlattices. Another advantage of a type-II superlattice is the possibility of bandgap engineering. Unlike bulk material or type-I superlattices, one can modify the energy of the conduction and valence minibands of a type-II superlattice with a high degree of freedom. Recently, II-VI HgTe/CdTe and III-V InAs/GaxInl_xSb type-II superlattices have shown very promising results in the long wavelength ranges. 7"s' 1g. ~ Experimental results Superlattices were grown by molecular beam epitaxy (MBE) on semiinsulating GaAs substrates. The reactor is an Intevac Modular Gen II MBE machine with uncracked As and Sb, and elemental Ga, In, and A1 source material. A 41~m GaSb buffer layer was grown directly on three-inch GaAs substrates. The wafer was then broken into ,~ 1 cm 2 pieces and indium-mounted to molybdenum blocks. InAs is found to have a very n a r r o w window for planar growth, while high quality GaSb can be grown in a wider range of growth conditions when reflection high energy electron diffraction (RHEED) showed a l x 3 reconstruction pattern. The optimum growth conditions for InAs layers were found to be: T=400~ according to a pyrometer, a V to III incorporation rate ratio ~3, and a growth rate of (). 5 monolayer/s. In this condition, RHEED showed 2 x 4 reconstruction patterns. The pyrometer is calibrated with the temperature of the transition from a 1 x 5 to a 1 x 3 reconstruction pattern in the GaSb buffer layers. Based on the theoretical modeling and simulation we chose the optimum structure for a room temperature detector at )~= 11 pm. The structure consisted of a t - 2 l~m superlattice with 48 A InAs, 30 A GaSb and one monolayer of InSb at the interfaces, as it is shown to improve the optical and electrical quality of the superlattice. 2~ Finally, the superlattice was capped with a thin 200 A GaSb layer. The spectral photoresponse of the device was measured using a Galaxy 3000 FTIR spectrometer system. The samples were illuminated t h r o u g h the front side with normal incidence. The absolute response of the photodetectors was calculated using a blackbody test set, which is composed of a blackbody source (Mikron 305), preamplifier (EG&G PA-6), lock-in amplifier (EG&G 5209), and chopper system (Stanford Research System SR540). Figure 6.6 shows the responsivity of the device in the 2 - 1 7 pm wavelength range at 78 K and 300 K with an in-plane electrical field of 5 V/cm. To assess the temperature dependence, the current responsivity of the device was measured at )~=10.6 pm wavelength from 78 K to room temperature at a constant electrical field. Figure 6.7 shows the responsivity of the detector at ~=10.6 l~m versus the detector temperature. In order to see whether current responsivity follows a power function, we fit the data to an allometric function. An allometric fit has a general form of y - A x B where x is the variable and A and B are the fitting parameters. This fit shows that the responsivity of the detector is nearly proportional to T- 1.9 3. This is an u n u s u a l behavior, since responsivity of the n a r r o w gap material is usually an exponential function of temperature at higher temperatures where Auger recombination is the dominant recombination mechanism. Theoretically,

GaSb/InAs supperlattices for infrared FPAs

203

10 2

~

10

g, 10-' 2

4

6

8

10

12

14

16

Wavelength (lim) Figure 6.6 The responsivity spectra of the device at 78 K and ~00 K with an in-plane electrical field of 5 V/cm.

Zegrya et al. 21 showed that the Auger recombination rate is a power function of the temperature (proportional to T 2) in type-II heterostructures compared to the exponential function in the bulk semiconductors. Since the current responsivity is proportional to the carrier lifetime, which is dictated by the Auger recombination rate, this power dependency of the responsivity indicates a good agreement with theoretical predictions. The effective lifetime of the carriers was also extracted from the responsivity and Hall measurements on a t=0. Slim thick superlattice lg

(23)

ref - - E ( l i e -+- lip)

where 1=2 mm is the device length, g is the photoconductor gain, and E= 5 V/cm is the electrical field. The gain of the device can be calculated from Rihc

g-

(24)

~nq

where Ri=2 mA/W is the current responsivity, h is the Planck constant, c is the speed of light, )v=10.6 lim is the wavelength of the light, 1"1 is the q u a n t u m efficiency, and q is the electron charge. Assuming an internal quantum efficiency near unity and negligible reflection from the bottom of the superlattice and unpolished backside of the substrate, the q u a n t u m efficiency can be calculated from rl-(1-r)(1-e

-at)

(25)

204

Handbook of Infrared Detection Technologies I

10

8

"~

\ .

I

"

;L= 10.6 bu~

I

=

\

\\

\

"

I

Measured data

AIIometric Fit

,,, \ \

6

"

\

i,

\\\\

\

"""""-IL

0

.

50

i

...

100

,

I

,

I

,

I

150 200 250 Temperature (K)

~

i

300

Figure 6.7 The current responsivity of the device versus temperature at ).= 10.6 I~m at constant voltage bias. The squares are the measured points and the line is an Allometric fit (AT R) to the points which shows that responsivity is nearly proportional to T -2.

where r is the top surface reflection coefficient, : z = l . 8 x l 0 3 cm -1 is the absorption coefficient of the superlattice and t=0.5 l~m is the thickness of the superlattice. Assuming r0.3, the q u a n t u m efficiency, photoconductive gain, and carrier lifetime can be calculated from above formulas as: r1=6.02%, g=3.9• -3, and Zer=26.8 ns. The effective lifetime is about an order of magnitude longer than the carrier lifetime in HgCdTe photoconductors with similar bandgap and carrier concentration at room temperature, x2 Since Auger recombination is the dominant recombination mechanism at room temperature, we believe that the enhancement of carrier lifetime is due to the suppression of Auger recombination in the type-II superlattice. Noise m e a s u r e m e n t s

The noise was measured with a fast fourier transform (FFT) spectrum analyzer (Stanford Research System SR 760) and a low noise, wide band pre-amplifier with 54 dB voltage gain (EG&G PA-IO0). Figure 6.8 shows the input noise spectrum of the FFT analyzer, the output noise spectrum of the shorted preamplifier, and the output noise of the pre-amplifier, when it was connected to the detector. The detector was biased by the pre-amplifier at Vb=5 volts. The mean-square noise of the detector can be modeled as

--

+

+ vL, f

(2 6)

GaSb/InAs supperlattices for infrared FPAs

205 ,

1

0"4 1 ~.,,.

t

1 (I s ~"

.....

",'...

':;;,. tI

Output noise of the pre-amp

!

-....... Output noise of the pre-amp connected to the detector

.

! I

".'.',

10.6

10 -v 10-8

10 .9

0

2000

4000

6000 8000 10000 Frequency (Hz)

12000

Figure 6.8 The measured input noise spectrum of the fast Fourier transform ( FFT) spectrum analyzer, the output noise spectra of the pre-amplifier with a shorted input and the output noise spectra of the pre-amplifier when it is connected to the detector at ~00 K.

where Vn is the overall noise of the detector, Vj is the J o h n s o n - N y q u i s t noise, V 1/f is the 1/f noise and Vc;R is the g e n e r a t i o n - r e c o m b i n a t i o n noise. The value of the Johnson noise can be calculated as

V~ - 4kTRAf,

(2 7)

where k is the Boltzmann constant, T is the t e m p e r a t u r e and R = 7 6 ohms is the resistance of the device. The value of Johnson noise for the device at room t e m p e r a t u r e is 1.12 nV/Hz 1/2. The 1/fnoise can be approximated as

V2/r - V2R fl/f f'

(28)

where fa/fis a constant which depends on the sample, and fis the frequency. This shows that at high e n o u g h frequencies, 1/f noise can be negligible compared to the other two types of noise. Then the value of the g e n e r a t i o n - r e c o m b i n a t i o n noise can be extracted from the total noise of the device in this range and the value of the Johnson noise. The noise equivalent circuit model 23 was used to extract the noise of the photodetector as V,~=I.7nV/Hz ~/2 above l()kHz. From the above equations, the value of the g e n e r a t i o n - r e c o m b i n a t i o n noise can be calculated as Vc;R-1.28nV/Hz 1/2. The g e n e r a t i o n - r e c o m b i n a t i o n noise can be approximated as

206

Handbook of Infrared Detection Technologies

2Vb l+b (np 17Af ) 1/2 VGR = (lwt)]/2 bn + p n + p 1 + O2172

(29)

where w = 4 m m is the detector width, o is the a n g u l a r frequency and o17 << 1 in this experiment, and b is the ratio of p~, to pp. From the V~;R equation, the value of the carrier lifetime can be calculated as 17=1 7 ns which is close to the value of the lifetime extracted from the optical response m e a s u r e m e n t . It should be noted that the calculation of the carrier lifetime from the electrical noise of the device is not an accurate technique and may only provide a rough estimation for the carrier lifetime. Response time measurement As was explained earlier, one of the motivations for the realization of uncooled photon detectors is the need for uncooled detectors with response times 17< 30 ms. Therefore, it was very i m p o r t a n t to measure the response time of the type-II uncooled detectors. However, conventional methods such as using mechanical choppers could not achieve time accuracy below the millisecond range. We used a room t e m p e r a t u r e q u a n t u m cascade laser, developed at COD, 24 as a n a r r o w b a n d and high speed infrared source in our m e a s u r e m e n t . The schematic diagram of the setup is s h o w n in Figure 6.9. The pulse generator and laser driver were inside an Avtech AVR-4A-PW which is capable of generating high power electrical pulses with fall time of about 5 ns. The q u a n t u m cascade laser was uncooled and operated at k=8.5 lam with a negligible time delay. An EG&G PA-1 ()() low-noise pre-amp was used to amplify the detector signal. Unfortunately, the pre-amp is not very fast and has a fall time of more t h a n 40 ns. The output signal of the pre-amp was measured with a Tektronix TDS 520B digital oscilloscope. It shows a 9()%-10% fall time of about 68 ns, as s h o w n in Figure 6.10, for the whole setup, and hence the detector has a fall time below 40 n s . 2 6 Device performance and comparison with state of the art The performance of infrared detectors is usually compared, based on their detectivity which is an indication of their sensitivity. Knowing the responsivity

Figure 6.9 The schematic diagram of the time response measurement setup.

(;aSb/InAs supperlattices for infrared FPAs

207

Figure 6.10 The overall response olthe setup.

and the noise of the devices, we could calculate their detectivity. The uncooled devices show a detectivity of about 1.3x l()S cmHzl/2/W at •-1 1 ~m which is higher t h a n the detectivity of commercially available HgCdTe at similar wavelengths and temperature 2s (see Figure 6.] 1 ). Unlike HgCdTe, these type-II superlattices are grown on conventional GaAs substrates, and hence highly uniform material can be grown on three and five-inch wafers, readily. In comparison to the thermal detectors, such as microbolometers, type-II superlattices have similar detectivity, but are showing at least five orders of magnitude faster response. 2~'. Although thermal detectors with higher detectivity are possible, the price that one has to pay is the speed. The change of the temperature of a bolometer is riP~ AT - Gv/1 + m2~c2

(30)

where AT is the temperature change, Pc~ is the IR power that comes to the bolometer surface, r I is the percentage of the IR power that is absorbed, G is the thermal conductance of the bolometer, m is the frequency of the IR emission, and z is the thermal response time of the bolometer and is ~=C/G, where C is the thermal capacitance of the bolometer. It is easy to find that, at low frequencies, m~<
208 Handbookof Infrared Detection Technologies

Figure 6.11 The detectivity of type-II superlattices compared to the theoretical limit and experimental detectivity ofHgCdTe (MCT) detectors at 700 K at/,.~ 11 I~m.

effort to improve the responsivity of the thermal detectors is based on the reduction of the thermal conductance, G. However, this will increase the response time of the device if one cannot reduce its thermal capacitance, since ~=C/G. Unfortunately, the thermal capacitance of the microbolometers cannot be reduced further since it requires a thinner or smaller device that is not practical. In order to decrease the thermal conductance of the device, the length of the legs that are supporting the sensitive layer can be increased (see Figure 6.12). Figure 6.13 shows the relative frequency response of pixels with different leg lengths presented by Raytheon. Although increasing the leg length by a factor of four decreases the NEDT by nearly a factor of two, the 3 dB frequency knee also decreases by a factor of four. 27 6.3.2 Cooled type-II p h o t o c o n d u c t o r s for ~

> 20 Hm

Motivation

High performance infrared detectors with cutoff wavelengths above 16 pm are very much needed for space-based applications such as deep-space astronomy and pollution monitoring. Currently available detectors with high quantum efficiency in this wavelength range are Mercury Cadmium Telluride (MCT) and extrinsic silicon detectors. However, due to the high non-uniformity 28 of HgCdTe, detector arrays with acceptable uniformity can only be realized with extrinsic silicon detectors at long wavelength. Although high detectivity and

(;aSb/hlAs slipperlattices for infrared FPAs

209

Figure 6.12 Schematic diagram of microbolometers with: ( a ) orginal pixel legs: and ( b ) two times longer legs.

good uniformity have been achieved with this type of detector, they have to be cooled below 10 K. Consequently, a three-stage cryo-cooler is required which is heavy, bulky, and has a short lifetime. These drawbacks are especially important for space applications since they significantly increase the launch cost. Theoretical 29 calculations and our experimental results show that InAs/ Gal_xInxSb type-II superlattices have a similar absorption coefficient to MCT, and therefore detectors with very high q u a n t u m efficiencies are possible. However, unlike MCT the uniformity of the growth is not an issue in this material system due to the strong bonding of III-V compound semiconductors. Experimental results

The structures were grown on semi-insulating GaAs substrates. First, GaSb buffer layers with a thickness of about 21Jm were grown directly on the GaAs substrates using a similar set of growth conditions given in the previous section. The growth was optimized for low background doping levels (,~ 5 x 1 () 1~ cm- ~), and high surface smoothness ( ~ 4 A rms roughness) in this layer. The superlattices consist of InAs and GaSb layers with one monolayer of InSb at both interfaces and were grown directly on GaSb buffer layers. The growth was terminated by a 100 A GaSb cap layer. Spectral photoresponse was measured using a Galaxy 3000 FTIR spectrometer system. The samples were illuminated through the front side at normal incidence. Absolute response of the photodetectors was calculated using a blackbody setup, which is composed of a blackbody source (Mikron 305), preamplifier (EG&G PA-6), lock-in amplifier (EG&G 5209), and chopper system (Stanford Research System SR540). Figure 6.14 shows the responsivity of the device in the 3-2 5 l~m wavelength range at 80 K with an in-plane electrical field of 20 V/cm. As was mentioned before, a wide range of cutoff wavelength can be covered with this material system. Figure 6.1 5 shows the spectral responsivity of three devices based on three different superlattice structures covering a 12-2 5 IJm range.

210

Handbook of Infrared Detection Technologies 10.0

l x Leg Pixel

O..

3 d B Freq = 15 H z

..,2

2x Leg Pixel

O.. v

3 d B Freq = 7 Hz

1.0

C 0

rr cD

._> 3 d B Freq = 6 H z

0.1

.

l

.

.

.

l

i

9

9 l

9

9

9

9

l

i

i

l

100

10

1 C h o p p i n g F r e q u e n c y (Hz)

Figure 6.13 The relative frequenc!l response of different test pixels with standard Ra!ltheon Sb- 1 51 leg length and 2 times and 4 times longer legs.

15 111"I'1'I

'

I

'

I

'

9

I

Wavelength (lam) 5

10 I

T=80K 100

J

10

E~o%68.5mev

~

I

5O

I

100

,

I

150

,

I

200

9

I

.

250

Energy (ev) Figure 6.14 The spectral response of the device.

I

300

,

I

350

9

400

(;aSb/InAs supperlattices for infrared FPAs

10 J

9i ,

30

20

i

i

i".

211

Wavelength (pm) 10 i

T=gOK <

~ 10 2

/

.~.

r~

~

/~

10~

InAs/GaSb(54_A/5 TM - InAs/GaSb(63A/57A)

l 10~

.•

10 0

i

40

,

I

i

60

80

,

I

100

i

I

I

I

120

140

160

9

I

180

200

Energy (mev)

Figure 6.1 _5 The spectral responsivity of three superlattices with cutoff wavelengths from 12 to 2 5 Ixm at 8OK.

One of the most important issues for detectors with long cutoff wavelengths is the uniformity of the material that is translated to the uniformity of the energy gap. Such uniformity is a crucial requirement for high resolution focal plane arrays (FPA) where the device area can be several square centimeters. We studied the uniformity of the growth over a three-inch wafer diameter by growing three samples at the center, middle, and the edge of a three-inch block simultaneously. Figure 6.16 shows the relative spectral response of these samples. The 10% cutoff energy varied by only about 5 meV from the center to the edge, showing excellent uniformity over a very large area. 3o

6.4 Experimental results from type-II photodiodes 6.4.7 U n c o o l e d type-II P h o t o d i o d e s in the ~ = 8 - 7 2 Hm range Motivation

The need for uncooled photon d e t e c t o r s in the mid and long wavelength IR ranges, as well as the limitation of currently available photon detectors have been detailed in Section 6.3. Although we demonstrated a high performance uncooled photoconductor in the long wavelength range, based on type-II superlattices, this device is more suitable for single-element detectors and cannot be easily used for two-dimensional (2D) focal plane arrays (FPA). Unlike a photoconductor, the current of a photodiode flows perpendicular to the surface, and hence it is scaled by the area of the device. For example, the current of a

212

Handbook of Infrared Detection Technoloqies

20 16 i

,

!

Wavelength (~trn) 8

12 ,

!

--A

i

T=80K

9 C (R=l.5

1

g

(R=0)

...... B (R=I")

[/::

")

~C67.2mev

0.1 t'i ! ..... _E_c-69 _.4_mev .................................

0.010

1;0

.

. 1;0

.

260

.

2;0

300

Energy (meV) Figure 6.16 Relative spectral response of three samples from the center ( A ). one inch from the center (B). and 1.5 inches from the center ( C) of a three-inch diameter u,~(fer block.

50 lam x 50 }am p h o t o c o n d u c t o r at a given voltage bias is equal to the c u r r e n t of a I c m x I cm device, while the c u r r e n t of a 50 lamx 50 ~tm photodiode is 4() 0()0 times less t h a n the c u r r e n t of a 1 cm ><1 cm device. This fact becomes very crucial for currently used large area FPAs, with. for example, 5 1 2 x 5 1 2 = 2 6 2 144 elements. Moreover, photodiodes can operate even at zero bias w h i c h not only reduces the bias and heat dissipation r e q u i r e m e n t s significantly, but also eliminates the 1/f noise. This type of noise increases inversely with frequency, and can be the d o m i n a n t source of the total noise for low frequency applications such as IR imaging systems.

Experimental results The structures were grown using cracked As and Sb sources. The cracking zone temperature for both cells was 9()()~ The deposition rates of the material were calibrated with dynamic RHEED oscillation to within 1%. First, 1 lam of GaSb contact layer doped with Be ( N , \ = l x l ( ) l g c m -3) was grown on a GaSb-p substrate. Then a stack of five devices was grown, each of which consisted of 20 periods of p-type InAs/GaSb:Be (39 A/4()A), 20 periods of nominally undoped InAs/GaSb ( 3 9 A / 4 0 A ) , and 20 periods of n-type InAs:Si/GaSb ( 3 9 A / 4 0 A ) superlattices. GaSh layers in this superlattice had a graded doping from 10 is cm -3 to 2 x ] O ]7 cm -3. In the n-type superlattice. InAs layers were doped with Si (ND-2 x 101 s cm- ~). The shutter sequences were designed such that both interfaces were InSb type. Finally, the growth was capped with a ().01 lam InAs:Si layer (ND=I x 1() is cm-3). The growth temperature was about 52()~ for the GaSb and 395~ for the superlattices according to a pyrometer calibrated with the surface reconstruction transition temperatures of GaSb (,-~39()~ and InSb (,-.,3 80 ~ C).

CaSb/InAs supperlattices for infrared FPAs

213

Absolute spectral responsivity was calculated from the measured spectral response of the devices, using a Fourier transform infrared (FTIR) spectroscopy system, and the device's photoresponse to a calibrated blackbody setup. The peak responsivity was Ri=O.14 A/W at ;k-7 pm leading to a Johnson noise limited detectivity of D*-1.2 • l()S cm Hz~/2/W at room temperature. Figure 6.17 shows the detectivity of a device versus I1R wavelength and energy. Although the I1R path length in the atmosphere was only about 15 cm, C 0 2 and water vapor absorption features are visible in the spectrum. Although non-equilibrium HgCdTe and InAsSb detectors with high differential resistance under reverse bias have been demonstrated recently, ~ a high 1/f noise degrades the performance of these devices below several megahertz by two to three orders of magnitude. Consequently, these nonequilibrium devices cannot be used for low frequency applications such as IR imaging systems. B2 The operation of type-II photodiodes under zero bias ensures that the main noise component is the thermal (Johnson) noise and 1/f noise is eliminated. Our experimental m e a s u r e m e n t s indicate that even under a considerable reverse bias, type-II detectors do not have a high frequency 1/f noise. Figure 6.18 shows the frequency spectrum of the amplified output of a detector. The device was under a - 0 . 2 volt bias and illuminated by the chopped IR radiation of a blackbody. The chopper frequency was f~)-396 Hz, blackbody temperature and aperture diameter were TBB=8()()K and DBB=2.54cm, and the detector was located d= 15 cm away from the blackbody aperture. Although the m e a s u r e m e n t includes the noise of the pre-amplifier (Analog Device AD797) and the FFT spectrum analyzer (Stanford Research System SR 76()). the knee of the 1 I f noise is below ,-, 1 O0 Hz. Under the given parameters of the inset of the Figure 6.18, the signal to noise ratio (SNR) was more than 44dB with a bandwidth of Af= 1 ()() Hz around f()= 3 9 6 Hz.

300250

Energy (meV) 150

200

1.2x108~' . . . .

A ~

'

T=300K

ias-ovo,t

~"1.0xl 08

)~=8~rn

:~8.0x10'1- ~'~

T

1

N

.0xl0

4.0x O

\

I ~176

4

\ .

.

5

.

.

6

.

.

.

.

.

,

7 8 9 10 Wavelength (!urn)

,

1'1

12

Figure 6.17 The detectivity of a device versus tile wavelength and energ!t of the IR radiation. Features at 4._3 lzm and 6 - 7 lzm are due to the C02 and water vapor absorption during the measurement.

214

Handbook of Infrared Detection Technologies

20

l - -

I

I

;

TBB=800K; DBB=I inch; d=15cm; Aopt= 1.37xl 0-3cm 2

RoA=1.36x10-2 f~cm2 Bias =-0.2 Volts

-20

SNR=44. l d B @ 396Hz; Af= 100Hz

rn "{3 v

-4

O -40 . . . . . . . . . . O 13_ 9 N

-60

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

!

I

E

' 'I'li

I.,.

o -8o

Z

, ,

,

-100 ............................

-120

0

L 50

l oo

1So

200

2go

3oo

l

3so

I

400

4g0

5oo

Frequency(Hz) Figure 6.18 Frequency spectrum of the amplified output of a detector illuminated bfl a chopped IR radiation of a blackbod!t. The detector was biased at - 0 . 2 ~"and was located d= 1 S cm awa!lfrom the blackbod!l aperture.

We used a q~tantum cascade laser (QCL), operating at room temperature, 33 as a high-speed source of IR radiation at X-5 Bm to study the response time of the uncooled type-II devices. Figure 6.19 shows the current of the OCL as well as the output of the pre-amplifier versus time. The inset shows the schematic diagram of the measurement setup. The laser threshold current marks the current above which the laser starts emitting. Considering the fall time of the pre-amplifier ( ~ 5 0 ns) and the fall time of the output signal of the pre-amplifier ( ~ 1 1 0 ns), the detector response time was about (1102-5()2)1/2--,--~1()0 ns using the sum-of-squares approach. Comparison with state of the art

The performance of state-of-the-art uncooled HgCdTe and microbolometers was detailed previously. We demonstrated uncooled detectors based on type-II superlattices with a 50% cutoff wavelength of 8 ~tm. The detectivity of the devices with zero bias and without optical lenses or anti-reflection coatings was 1 . 2 x l O S c m H z l / 2 / W . The response time of the detectors, measured with a q u a n t u m cascade laser at room temperature, was about 1 O0 ns. The following table compares the performance of these devices with available HgCdTe and microbolometer detectors.

GaSb/InAs supperlattices for infrared FPAs

215

Figure 6.19 The current of the OCL laser (right axis ) and tile amplified output of the detector (left axis ) versus time. Inset shows the measurement setup. Note that the laser emits when current is above the threshold level.

6.4.2 Cooled type-II photodiodes in the ~ > 14 pm range

Motivation

The application of infrared detectors in the VLWIR range as well as the physical limitations of currently available devices were detailed previously. Obviously an interband detector is the best candidate for high q u a n t u m efficiency and higher operating temperature, if a good uniformity is available. The experimental results of the previous chapter suggest that type-II superlattices have superior uniformity compared with the available interband detectors in the VLWIR range. However, realizing that the material quality cannot be enhanced on a lattice mismatch substrate0 we started the work on GaSb substrates. Photovoltaic devices were the natural choice due to the high conduction of the undoped substrate as well as the benefits of the photovoltaic devices for imaging applications. E x p e r i m e n t a l results

The material is grown by an Intevac Modular Gen II molecular beam epitaxy (MBE) equipped with As and Sb valved cracker sources on p-type GaSb substrates. The photodiode structures were grown at 395~ according to a calibrated pyrometer. First, a 1 pm GaSb b u f f e r ~ c o n t a c t layer doped with Be ( p ~ l x l O 18cm-3) was deposited. Then a 0.5 pm thick InAs/GaSb:Be (p~1• is cm -3 to 3 • -~) superlattice was grown followed by a 2 pm thick nominally undoped superlattice. Finally. a 0.5 pm thick InAs:Si/GaSb (n~l• -3) superlattice was grown and capped with a IOOA thick InAs:Si ( n , ~ 2 • top contact layer. The growth rate was 0.5

216

Handbook of Infrared Detection Technologies

Table 6.2 The performance of uncooled type-ll superlattice photodiodes compared with the uncooled HgCdTe and microbolometer at 8 ~ m Parameters

Type-II SL

HgCdTe

Detectivity

~ 108 cmHz 1'2/W

~ 1 ()s cmHz I

Microbolometer

2/W

,~ 10 ~ cmHz l'

)~=8 ~tm; T = 3 0 0 K RoA kc=8 ~tm: T = 3 0 0 K

1 . 4 x 10 -2

~ 1 x 1 ( ) 4 fZcm 2

_

Uniformity

Very good

Poor

Very good

Zero bias o p e r a t i o n R e s p o n s e time

Possible ~1() - s s

Possible ~1 ()-s s

Not possible ~1()-~ s

f2cm 2

2/W

monolayer/s for InAs layers and 0.8 monolayer/s for GaSb layers. The V/III beam-equivalent pressure (BEP) ratio was about 4 for InAs layers and about 1.2 for GaSb layers. The cracker temperature for As and Sb cells was 800~ The selected thickness of InAs and GaSb layers were determined for specific cutoff wavelengths using a four-band superlattice k.p model. For devices with a cutof wavelength of nearly 16 ~m, the thickness of InAs layers was 54 A and the thickness of the GaSb layers was 40 A. Structural quality of the epitaxial layers was assessed using high resolution Xray diffraction. Figure 6.20 shows the typical X-ray diffraction pattern of the photodiode structures. The mismatch between the average lattice constant of the superlattice and the GaSb substrate was below 0.06%, while the full width at half maximum (FWHM) of the satellite peaks was below 70 arcsec for the grown devices. The surface morphology of the samples was also studied with an atomic force microscopy (AFM). Theoretical calculations show that surface and interface roughness lead to defect-like energy states inside the superlattice energy gap and broadening of the band edges. 34 Moreover, experimental results show a strong correlation between the surface roughness and electrical performance of InAs/ Gal_xInxSb superlattice photodiodes. Figure 6.21 shows the gray-scale surface morphology of a sample. Wide atomic steps are visible which is an indication of excellent surface smoothness. We could routinely grow samples with a root mean square (rms) surface roughness below 4 A over a 2 0 B m x 2 0 ~ m area which is among the best reported values for this material system. The transmission electron microscope (TEM) images of the material provided by Wright Patterson Air Force Base (WPAFB) also showed very good structural quality of the material. Figure 6.22(a) shows the bright field TEM image of the undoped area of a superlattice, while Figure 6.22(b) shows the dark field TEM image of the p-type layers of the same sample. Standard photolithography, wet etching, lift-off, and metallization were used to fabricate 400 l~m• lam mesas with 150 ~m• 150 ~m top contacts and 300 ~m • 150 l~m bottom contacts. Similar to the uncooled devices, very smooth side walls with a nearly 60 ~ angles could be etched routinely (see Figure 6.2 3). The processed samples were indium bonded to a copper heatsink and attached to the cold finger of a helium refrigerator equipped with KRS- 5 infrared windows. Since the refrigerator is a high-power device, it can easily induce a substantial

GaSb/InAs supperlattices for infrared FPAs

100000

10000

217

0' GaSb Substrat~ I i ~-

,

.

, . InAs Cap

-|1

,

I!

+ 1,

.

,

.

,

+2

"~

-5

10

-4

+3

'

+4 +5

28

29

30

31

Omega (Degrees)

32

33

Figure 6.20 High resolution X-ray diffraction of an InAs/GaSb ( ~4 A / 4 0 A) superlattice grown on a GaSb substrate. F W H M of the satellite peaks is below 60 arcsec, and the mismatch to the substrate is below O. 06 %.

a m o u n t of noise to the measurements. Proper grounding was used to reduce the noise level, however the average noise was considerably higher than a liquid nitrogen cryostat. Figure 6.24 shows the measured and modeled current densities versus the applied bias for devices with a cutoff wavelength of )~c= 16 ~m. The calculated current density, which consists of tunneling, generation-recombination, and diffusion current densities, shows good agreement to the measured values for forward and reverse biases. We assumed an effective mass of m,=O.O 3 mo for electrons and mh=0.4 mo for holes based on previous theoretical calculations 34 and experimental results. Based on the experimental measurements on similar devices, 3s we also assumed an electron mobility parallel to the growth direction Of~e-lO00 cm2/Vs. The mobility of the holes is not significant in the diffusion current, since the device has a n+-p junction. The fitting parameters for the model were carrier lifetime ~e=~h=220 ns, unintentional background doping level p=2.1 • 10 is cm -3, and generation-recombination lifetime in the depleted layer "~GR=0.6 ns. In contrast to HgCdTe, tunneling is not significant even at high values of reverse bias due to the higher effective mass of the electrons in type-II superlattices. However, generation-recombination current is the dominant source of dark current for these devices at T=8()K, and hence further improvement of the growth should increase RoA and detectivity. Also, we could observe the W a n n i e r - S t a r k oscillation in the Zener tunneling current of these devices. Figure 6.25 shows the differential resistance of the device versus reverse voltage bias. An oscillation with a peak-to-peak separation

218

Handbook of Infrared Detection Technologies

Figure 6.21 AFM image of the surface of the device. Atomic steps with several micron width are clearly visible.

Figure 6 . 2 2 (a) bright and (b) dark field TEM images of a superlattice photodiode.

(;aSb/InAs supperlattices for infrared FPAs

219

Figure 6.23 SEM image of a mesa cross-section produced by wet etching. The sidewalls are very smooth with an angle of about 60 ~

of about 29.9 mV is clearly visible. Using the same formalism as previously, we could extract the thickness of the depleted layer to be dp=120nm, and the background doping level to be p=2.1 • 1015 cm-3. This value shows an excellent agreement to the calculated background doping level based on the dark current modeling. This agreement and the fact that the background calculation based on the Wannier-Stark oscillation can be calculated without the involvement of unknown parameters, indicates that the above value must be very close to the real background level. Absolute spectral responsivity was calculated from the measured spectral response of the device, using a Fourier transform infrared (FTIR) spectroscopy system, and its photoresponse to a calibrated blackbody setup. Figure 6.26 shows the typical spectral responsivity of the detectors with kc =16 Hm. The absorption features of CO2 and H20 are due to the small difference in the optical path length of the background measurement and the detector measurement. The peak responsivity for the sample is about 3.5 A/W which leads to a quantum efficiency of ~35% at 12~tm. Adding anti-reflection coating can reduce the 30% surface reflection of the device and the q u a n t u m efficiency will reach to nearly 50%. The 90% to 10% cutoff energy of the responsivity is only about 13 meV at T=80 K which is close to the thermal limit of 2kT= 12 meV. As it will be detailed later, such a sharp cutoff is a unique characteristic of our superlattices compared to the published results of other groups working on type-II superlattice detectors.

220

Handbook of Infrared Detection Technologies

100

I

"E 1~ V

>,

'

I

'

I

'

I

'

I

'

!

'

I

Measured Points Calculated

]

T=80K

Tunneling+Diffusion+GR

1

= ~

co r

Diffusion

~0.1 C-

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

- ,,

L_

:30.014 0 "

G I ~ *"

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

.

Tunneling

!

~" ,~. /

"~.

1E-3- 0 4 0 - O'3 s .

.

\ .

.

I

-0:30 -0:2s - 0 2 0

'

O' -

,, '

-0

I

'

-o'os

'

0'00

'

Bias (Volts) Figure 6.24 The measured and modeled dark current density of a device with ,;.,.= 16 #m at T=80 K. The model consists of the tunneling, generation-recombination and diffusion components of the dark current and indicates that generation-recombination is the dominant source of the dark current around :ero bias.

Some of the samples with longer cutoff wavelengths were sent to the Wright Patterson Laboratory, since their FTIR system is purged with nitrogen and a very accurate spectral response can be measured. The low temperature m e a s u r e m e n t of these samples showed a clear set of features in the spectral response of these devices. In order to explain such behavior, we modeled the energy dispersion of the superlattices with a simple four-band k . p model. Figure 6.27 shows the results of such modeling for a sample with a 24 pm cutoff wavelength. The large superlattice period has led to strongly localized energy states evidenced by the small energy change of the bands over the whole m o m e n t u m space. Consequently, the density of states is approaching a two-dimensional configuration. In this situation, the optical absorption increases at the beginning of a transition between two bands and decreases at the edge of the Brillouin zone. The calculated energies of such transitions as well as the measured photoresponse of the detector at T= ] () K is shown in Figure 6.28. It is clear that the photoresponse increases at the onset of a transition and decreases at the edge of the Brillouin zone. Theoretical calculations predicts that unlike type-I q u a n t u m wells, the selection rule in type-II q u a n t u m wells do not require that the s u m m a t i o n of the q u a n t u m numbers between the relative states must be an even number. Figure 6.29 shows that there are strong transitions between the C1 and HH2 which is in agreement with the above.

GaXb/InAs supperlattices for infrared FPAs

1060

9

I

221

"

T=80K

1040 1020 1000 "~

980

940.

9

I

,

I

9

I

,

I

9

I

,

I

i

.

i

9

-0.68 -0.66 -0.64 -0.62 -0.60 -0.58 -0.56 -0.54 Bias (Volts) Figure 6.25 The differential resistance of a device with ).,= 16 tim at T=80 K versus reverse bias. shows a clear oscillation due to the Wannier-Stark oscillation in the Zener tunneling current.

Energy (meV) 4.0

400

300

,.,

.

200

,

100

.

3.5 '3.0 v

>, 2.5

.m

>

. B

~ c-

2 09

,-

1.0

&

t,....

O

0.5 0.0 -0.5

'

i

4

,

I

6

,

|

8

,

I

10

,

I

,

12

I

14

,

I

16

,

I

18

,

20

W a v e l e n g t h (l.r Figure 6.26 Absolute spectral responsivit!l of a device with 2,= 16 ~ m at T=80 K. The 90% to 10% cut-off energy width is about 12 meV.

222 Handbook of Infrared Detection Technolo~lies 9

I

"

I

"

I

"

I

C;1

200

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

100

o

-

. . . . . . . . .

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

-100 -200 9

LH~

-300[..

LH 3

-400

....

0.0

i ~ " : : ;

0.2

. . . .

'

0.4

'

'

0.6

'

|

0.8

I

1.0

k z (TT/d) Figure 6.27 The energy dispersion in tile k: direction of a superlattice with 21ML of lnAs and 1 ~MLof (;aSb calculated with the four band k.p model.

Comparison with state o[ the art

In this part, we shall compare the performance of the devices developed in this work for the VLWIR range operation with the other high performance type-II superlattices as well as HgCdTe and extrinsic silicon detectors. Although several groups are actively working on type-II photodiodes, we could not find any published data about the absolute optical responsivity and detectivity of their devices in the VLWlR range. However, the relative spectral response of the devices reported by other groups was compared with the spectral response of the devices developed in this work. Unlike most of the other groups, we used only binary materials in our superlattices. We believe that the use of binary layers in these superlattices has significantly enhanced the uniformity and reproducibility of the energy gap. The 90%-10% cutoff energy width of devices with a wide range of cutoff wavelengths is only about 2kT which is at least four times smaller compared to similar devices based on InAs/ Gal_xIn• ( 0 . 2 5 < x < 0 . 3 5 ) superlattices. 7'~'3~'~7"3~ Figure 6.29 shows the spectral response of devices with cutoff wavelengths from 8 to 20 l~m versus the energy of the photons. The slope of the cutoff edge is nearly independent of the cutoff wavelength and is nearly 14 meV for these detectors at T=80 K. This value is very close to the thermal broadening limit or 2 k T - 12 meV at 80 K, considering the Fermi distribution of the energy of the electrons in the conduction and valence minibands. These measurements were confirmed with similar measurements at the Wright Patterson Laboratory. For example, their results

(;aSb/'hlAs supperlattices for infrared FPAs

0.25

mHH=0.4m0" InAs=21ML GaSb=13ML; T=9.'

51

"

'

"

9~

0.20

.9, q l , , - , - , , , I P -

i'

\v

0"10 t

j

, /

il

"

'

"

y~-----~-

9

9

i/-~

~:

HH2-C1

HHI-C1

0.15

Eoffset =160meV

" ' I LH3-C " HH3-C I'

'

223

/

\

\j

if

o.o51_J

0.00t i

0

J! ,

I

1O0

,

I

200

,

I

300

,

I

,

400

i

500

.

i

600

.

700

Energy (meV) Figure 6.28 The features of the measured photoresponse of a device with a cutoff wavelength of 24 i~m at T= 9.5 K show good agreement with the calculated transition energies at critical momentum values.

show that one of our detectors with Xc=24 l~m has a 9()%-10% cutoff energy of about 15 meV at T=9.5 K. The study of the cutoff wavelength of the InAs/Gal_xInxSb superlattices shows that the higher the value of x, the more sensitive the cutoff wavelength is to both the value ofx and the thickness of the layers. Figure 6.30 shows the cutoff wavelength of InAs/Gal_xInxSb superlattices for a 7 ML (monolayer) thick Gal_xInxSb layer and 7-15 ML thick InAs layers versus different In mole fractions x calculated by our four-band k.p model. For example, this graph shows that for a superlattice with a 1()lam cutoff wavelength, the slope of the curves or dX/dx increases at higher values of x. Similarly, the cutoff wavelength is more sensitive to the InAs thickness at higher values of x. For example, at x=0.4, if the thickness of the InAs layer changes from 13 ML to 15 ML, the cutoff wavelength changes by ~ 6 ~m. While at x=0.3, similar change leads to a 2.5 l.tm change in the cutoff wavelength of the superlattice. Indeed, the study of the results of the other groups shows considerable correlation between the 90% and 1()% cutoff energy and the indium mole fraction independent of the superlattices ~cutoff wavelengths (see Figure 6.31 ). The available data on HgCdTe devices in the VLWIR range is limited since most of the high performance devices have military applications. Therefore, it is not possible to have a conclusive comparison between type-II devices and the best HgCdTe devices available. Although the proper design of type-II superlattices with Xc=161~m has improved the value of the RoA product significantly, it is still about one order of

224 Handbook of Infrared Detection Technologies

,10 -40. 30 "

20 '

Wavelength

I

'

10 !

T=gOK

9

,o-

l I

50

.

I

75

,

I

100 125 Energy (meV)

9

I

150

9

175

Figure 6.29 Spectral response of photodiodes with d(fferent lnAs/(;aSb superlattices in their active la!lers. The thickness of the GaSb layer is 40 A for all of the superlattices, while the thickness of the InAs la!ter is shown for each device.

magnitude less than the R~A product of HgCdTe with a similar cutoff wavelength. This is mainly due to the higher background carrier concentration in type-II superlattices. The improvement of the growth has decreased this value from ~ 5 x l O 1 6 c m -3 to the current value of ~ 2 x l ( ) l S c m -3 over the past five years. Therefore, it is foreseeable that the current value can be decreased further to the ~ 5 x 1014 cm -3 range which is close to the background level of current HgCdTe material. Detectivity is proportional to the square-root of R~A, and consequently type-II devices show about three times lower detectivity compared with HgCdTe devices at ~= 16 lam. Most of the available data indicate that the uniformity of HgCdTe over a large area degrades rapidly as the cutoff wavelength increases above 1413m. Some of the most recent data provided by Lockheed Martin show a variation of more than 16meV in the energy gap over a 40cm• area for HgCdTe with a cutoff wavelength of about 15 l.tm. 39 In comparison, we demonstrated only a 5 meV variation in the energy gap over a 70cm• area of type-II superlattices with a cutoff wavelength of about 19 l~m. 3~ Extrinsic silicon detectors are the most commonly used detectors in the VLWIR range. Some of the reasons for their popularity are: excellent uniformity over large areas; high q u a n t u m efficiency: and high detectivity. However, they have

GaSb/InAs supperlattices for infrared FPAs

225

Figure 6.30 Cutoff wavelength of lnAs/Gal _.,.In.,.Sb superlattices for 7 ML thick Gal _,.In,.Sb and 7-15 ML thick InAs layers versus different In mole fractions, x.

two major drawbacks that are especially significant for space base applications. These are very low operating temperature and very low speed at a low background photon flux. In comparison with extrinsic silicon, type-I! superlattices show a m u c h higher operating temperature with similar q u a n t u m efficiency. This is an inherent advantage of the intrinsic detectors to extrinsic detectors and is explained briefly here. Under a low excitation condition, the detectivity of an intrinsic detector can be simplified as 4~ D . * - r l ) v ( T i n ~ 1/2 2hc k,tinni/ m

- -

(31)

where ~:in is the intrinsic carrier lifetime, ti, is the intrinsic detector thickness, ni is the intrinsic carrier concentration, r I is the q u a n t u m efficiency, h is the Planck constant, c is the speed of light, and )v is the wavelength. Similarly, the detectivity of an extrinsic detector can be calculated as

Dex

2hc \texnth

(3 2)

where 17exis the extrinsic carrier lifetime, t~x is the extrinsic detector thickness, and nth is the concentration of thermally generated carriers. The ratio of the detectivities is then

226 Handbook of Infrared Detection Technologies

Figure 6.31 The 90% to 10% cutoff energll of devices with different indiunl nlole fraction, versus their cutoff wavelengths.

D i n _ ('qntexnth) 1/2 D~x k Textinni

(33)

To obtain high q u a n t u m efficiency, the thickness of the detectors must be on the order of their inverse optical absorption coefficient :x, therefore , ( ,~1/2 Din _ .Tin0~innt___h Dex \ "CexO~exni,/

(34)

where (~in is the optical absorption coefficient of intrinsic material and ~ex is the optical absorption coefficient of the extrinsic material. The optical absorption coefficient of extrinsic detectors is nearly two orders of magnitude lower than intrinsic detectors. For example, ~ex "-' 15 cm- 1 for Si:Ga versus ~in " " 1800 c m - 1 for type-II superlattices with k c - 1 6 l.tm. This fact combined with nth/ni~,lO0, results in nearly two orders of magnitude higher detectivity for intrinsic detectors at a given temperature. For example, the detectivity of Si:Ga at 80 K is about ,-~108 cmHzl/2/W (extrapolated data), 41 while it is 1.5 • 1() l~ cmHzl/e/W for type-II superlattices with a similar cutoff wavelength and temperature. Another disadvantage of extrinsic silicon detectors is their low speed due to a long dielectric relaxation time. Since extrinsic detectors need to be cooled to very low temperatures, the carrier concentration in the silicon can be as low as 1012 cm -3. This means that the material has such a high resistivity that the internal resistor-capacitor (RC) time constant, or the so-called dielectric

(;aSb/InAs supperlattices for infrared FPAs

227

relaxation time constant Zrel~,x~t~,,n,is very large. The optically generated carriers in an extrinsic detector remain excited nearly the same length of time as Trelaxation, during which the detector is practically saturated. The value of this time constant can be calculated as 41 Trelaxation --

gtex qrl~sTexla

(3 5)

where ~ is the permittivity of the material, t~x is the device thickness, ~ is the background photon flux, ~ex is the carrier lifetime, and ~t is the carrier mobility. The typical parameters for extrinsic silicon are r1=30%, t~x-().O5cm, andg=8• At a low radiation of ~ - 1 ( ) 12 photons/cm2s, which is the case for most space based applications, the value of ~r~l~x~tion is nearly 0 . 0 8 4 s leading to a cutoff frequency of about 12 Hz! Even with a 300 K background, the cutoff frequency of the extrinsic silicon detectors is a few kHz. In comparison, type-II and other intrinsic detectors have cutoff frequencies in the tens of megahertz range, independent of the background photon flux.

6.5 Future work The technology developed during this work has led to the demonstration of typeII detectors with some of the 'best' and 'first' reported results. However, these detectors are still far from their optimal performance. Although we have decreased the background carrier concentration by one order of magnitude to its current record level of 2 • 10 ~5 c m - 3, this value is still considerably higher than the best HgCdTe material. Therefore, an even better control of the interfaces, stability of the growth parameters during the growth, and optimization of the MBE shutter sequences is required. At the current value of the background doping level, the major source of the dark current is due to diffusion and not the surface leakage, hence the developed processing techniques seem to be adequate. However, once the material quality is improved, the device processing and passivation is an important technical issue that must be addressed. Although the wet and dry etching techniques that are developed provide very smooth surfaces, the chemical property of the finished surface has not yet been studied. This characteristic determines the surface leakage current, which could be a considerable source of the device dark current and noise. Passivation is a step that can affect the surface chemistry significantly, and hence can alleviate some of the above problems. Another method that can increase the detectivity and the impedance of the devices is the e n h a n c e m e n t of the optical-to-electrical area ratio using microlenses. Theoretically, this can increase the detectivity by about one order of magnitude without any negative side effects. Mass transfer and binary mask methods can be used to realize arrays of microlenses on the back of the substrates. The major difficulty is the alignment of the lenses and the detectors, since the substrates are opaque and conventional mask aligners cannot be used.

228

Handbook of Infrared Detection Technologies

Although the Auger recombination rate could be decreased by engineering the band structure of the type-II superlattices, the Auger process has not been completely eliminated yet. Figure 6.32 shows the energy dispersion of the minibands in the different m o m e n t u m spaces. It is clear that the Auger recombination is not possible at kx=() and kv-(). However, because of the inplane dispersion, the in-plane m o m e n t u m of the electrons is not necessarily zero and hence Auger recombination is still possible. By inducing q u a n t u m confinement in the x and y directions, the dispersion in these directions can be quantized. This can eliminate the Auger recombination and hence the carrier lifetime will be limited by other mechanisms such as radiative and non-radiative recombination, which are several orders of magnitude slower than the Auger recombination at room temperature. 1 ~total

=

1

t ~

1

+

1

"lTAuger 17radiative Tnon-radiative

(36)

Since the responsivity of the detector is proportional to the carrier lifetime, the responsivity will also increase dramatically which, in the absence of any noise e n h a n c e m e n t , will lead to a m u c h higher detectivity. Although infrared detectors based on self-assembled q u a n t u m dots have already been demonstrated, 42 the non-uniformity of the dot size leads to linewidth broadening and reduction of the signal. Here a method is proposed which can produce high quality and highly uniform q u a n t u m dot structures. The starting material is a superlattice. Stacks of high quality q u a n t u m dots can be developed by etching pillars and consequent dielectric coating. In order to achieve q u a n t u m size effect, one needs to confine the electrons within tens of nanometers. However. surface leakage current will be a severe problem for such a small device, since the ratio of the surface to the volume increases dramatically. In order to circumvent this problem, one can use a gated device shown in Figure 6.33. Although the diameter of the device is 5 0 - 2 0 0 nm, the effective confinement diameter can be adjusted by the gate voltage to a m u c h smaller

Figure 6.32 The energy dispersion of the minibands in the different momentum spaces.

(;aSb/InAs supperlattices for infrared FPAs

229

value. Moreover, the allowed energy states of the electrons can be changed by the gate voltage and hence the cutoff wavelength of the device. The sensitive area of such a q u a n t u m pillar is very small and not useful for m a n y applications. Figure 6.34 shows a realization scheme that can provide a large detector area containing a large number of closely packed pillars. Below, the feasibility of this approach is demonstrated. Low energy e-beam lithography was used to produce top metal contacts. High quality metal contacts with diameter in the range of 1()()()-1()()nm were successfully defined on the surface of the samples as shown in Figure 6.35. Reactive ion etching (RIE) was used to produce uniform anisotropic etching of the pillars through the GaSb test material. The dry etching parameters was

Figure 6.33 Schematic diagram of the gated pillars ~vith an adjustable lateral confinement.

Figure 6. J4 Realization of the pillars with advanced metalization and passivation techniques.

230 Handbook of Infrared Detection Technologies

Figure 6.3 5 Top metal contacts defined b!l e-beam lithograph!t on the samples: (a ) 500 nm diameter contacts: and (b) 1 O0 nm contacts.

Figure 6.36 (a) Reactive ion etching was usedjbr the formation of un(form two-dimensional arra!ls of pillars. (b) SEM image of the cleaved edge of a mesa covered b!l a z~nfform la!ler of Si ~N4 with a thickness of about 5Onto.

similar to V.G.3. Figure 6.36(a) shows the SEM image of an array of 5()()nm pillars produced by the dry etching. The vertical to horizontal etching ratio is in excess of five. We were able to produce two-dimensional arrays of pillars with excellent uniformity over thousands of square microns. Finally, passivation of the material was studied. Passivation is one of the most important steps for the realization of these q u a n t u m devices. The essential issues are the uniformity of the dielectric and the coverage of the device surface. Plasma enhanced chemical vapor deposition (PECVD) was used to form uniform layers of silicon nitride on these structures. Figure 6.36 shows SEM image of a cleaved edge of a mesa covered with a 50 nm Si 3N4 layer by this method.

(;aSb/InAs supperlattices for infrared FPAs

2 31

References 1. G. A. Sai-Halasz, R. Tsu, and L. Esaki APL 3 0 , 6 5 1 (1977). 2. G. A. Sai-Halasz, L. L. Chang, J. M. Welter and L. Esaki, Solid State Commun. 27,935(1978). 3. G. A. Sai-Halasz and L. Esaki, Phys. Rex,. B 18, 2812 ( 19 78). 4. D. L. Smith and C. Mailhiot, J. Appl. Ph!ls. 62, 2 545 (1987). 5. E. R. Youngdale, J. R. Meyer. C. A. Hoffman and F. 1. Bartoli, Appl. Phys. Lett. 64,3160(1994). 6. H. X i e a n d W . I. Wang, J. Appl. Ph!ls. 76, 92 (1994). 7. ]. L. Johnson, L. A. Samoska, A. C. Gossard, J. Merz, M. D. Jack, G. R. Chapman, B. A. Baumgratz, K. Kosai and S. M. Johnson, I. Appl. Phys. 80, 1116 (1996). 8. F. Fuchs, U. Weimer, W. Pletschen, J. Schmitz, E. Ahlswede, M. Walther, J. WagnerandP. Koidl, Appl. Phys. Lett. 71, 32-51 (1997). 9. H. Mohseni and M. Razeghi, Photon. Technol. Lett. 13, 517 (2001 ). 10. H. Mohseni, M. Razeghi, G. Brown and Y. S. Park, Appl. Phys. Lett. 78, 2107(2001). 11. B. H. Yang, D. Zhang, Rui Q. Yang, C.-H. Lin, S. J. Murry and S. S. Pei, Appl. Phys, Lett. 72, 2 2 2 0 (1998). 12. C. L. Felix, J. R. Meyer, I. Vurgaflman, C. H. Lin, S. J. Murry, D. Zhang and S.S. Pei, Phot. Tech. Lett. 9 , 7 3 4 ( 1 9 9 7 ) . 13. C. H. Lin, R. Q. Young, D. Zhang, S.J. Murry, S. S. Pei, A. A. Allerman and S. R. Kurtz, Elec. Lett. 3 3 , 5 9 8 (1997). 14. C. G. Van de Walle and R. M. Martin, Phys. Rev. B 3 5, 8154 ( 198 7). 15.C.G. VandeWalle, Phys. Rev. B39, 1871 (1989). 16. C. H. Grein, P. M. Young, H. Ehrenreich and T.C. McGill, ]. Electronic Material 22, 1093 (1993). 17.H. Mohseni, V.I. Litvinov and M. Razeghi, Ph!ls. Rex,. B 58, 378 (1998). 18. D. L. Smith, T. C. McGill and J. N. Schulman, Appl. Ph!!s. Lett. 43, 180 (1983). 19. C. Mailhiot and D. L. Smith, J. Vat'. Sci. Technol. A7(2 ), 44 -5 (1989). 20. N. Herres, F. Fuchs, J. Schmitz, K. Pavlov, J. Wagner, J. Ralston, P. Koidl, C. Gadaleta and G. Scamarcio, Phys. Rex,. B 5 3 . 1 - 5 6 8 8 ( 1 9 9 6 ) . 21. G. Zegrya and A. Andreev, Appl. Phys. Lett. 6 7 , 2 6 8 1 (199-5). 22. J. Piotrowski, W. Galus and M. Grudzien, Infrared Phys. 31, 1 (1990). 23. D. Wang, G. Bosman, Y. Wang and S. Li, J. Appl. Phys. 77, 1107 (199-5). 24. S. Slivken, A. Matlis, C. Jelen, A. Rybaltowski, J. Diaz and M. Razeghi, Appl. Phys. Lett. 74, 173 (1999). 25. H. Mohseni, J. Wojkowski and M. Razeghi, IEEE J. Of Quantum Elect. 3 5, 1041 (1999). 26. H. Mohseni andM. Razeghi, Proceedin#s 1999-ISDRS, -563 (1999). 2 7. W. Radford, D. Murphy, A. Finch. K. Hay, A. Kennedy, M. Ray, A. Sayed, J. Wyles, R. Wyles and J. Varesi, SPIE Proceeditl#s 3 6 9 8 , 119 (1999). 28. A. Rogalski, Infrared Ph!lsics and Technolog!140, 2 79 ( 1999 ).

232 Handbookof Infrared Detection Technoloqies 29. D. Chow, R. Miles, J. Schulman, D. Collins and T. McGill, Semicond. Sci. Technol. 6, C4 7 (1991 ). 30. H. Mohseni, A. Tahraoui, J. Wojkowski, M. Razeghi, G. J. Brown, W. C. Mitchel and Y. S. Park, App1. Phys. Lett. 77, 1:572 (2000). 31. C. Elliott, N. Gordon, D. Wilson, C. Jones, C. Maxey, N. Metclafe and A. Best, J. of Modern Opt. 45, 1601 (1998). 32. C. Elliott, N. Gordon, R. Hall, T. Phillips, C. Jones and A. Best, J. Electronic. Materials 26, 643 (1997). 33. S. Slivken, A. Matlis, C. Jelen, A. Rybaltowski, J. Diaz and M. Razeghi, Appl. Phys. Lett. 74, 173 (1999). 34. G. Bastard, Phys. Rev. B 25, 7584 (1982). 35. L. Burkle, F. Fuchs, R. Kiefer, W. Pletschen. R. E. Sah and J. Schmitz, Mat. Res. Soc. Syrup. Proc. 607, 77 (20()0). 36. M. Young, D. Chow, A, Hunter and R. Miles, App1. Surface Sci. 1 2 3 / 1 2 4 , 395(1998). 37. F. Fuchs, U. Weimar, E. Ahlswede, W. Pletschen, J. Schmitz and M. Walther, Proc. SPIE 3 2 8 7 , 14 (1998). 38. R. Miles andJ. Wilson, SPIE 2 8 1 6 , 76 (1996). 39. S. Tobin, M. Hutchins and P. Norton, J. Electronic. Materials 29, 781

(2000).

40. A. Rogalski, Infrared Photon Detectors, (SPIE Bellingham WA, 1995), p. 40. 41. D. Schroder, Charge-coupled Devices, (edited by D. Barbe, Springer-Verlag, Heidelberg, 1980), p. 57. 42. S. Kim, H. Mohseni, M. Erdtmann, E. Michel, C. Jelen and M. Razeghi, Appl. Phys. Lett. 7 3 , 9 6 3 (1998).

Chapter 7

MCT properties, growth methods and characterization Randolph E. Longshore

7.1 Preface Mercury Cadmium Telluride (MCT) is a very important semiconducting material developed over the past 44 years. MCT is used to fabricate infrared (IR) detectors for military and scientific applications. Military applications include the use of IR detector focal plane arrays (FPA's) in thermal imaging systems and in ground and space surveillance IR imaging systems. Scientific applications include IR detector arrays in telescope sensors for astronomy and IR detectors for spectrometers and radiometers. This chapter will describe MCT characteristics, MCT material growth techniques and MCT material characterization methods. The focus of this chapter will be on recent developments and emerging technologies for MCT. For example, Molecule Beam Epitaxy (MBE) thin layer MCT growth methods are being developed which will allow the design and fabrication of new devices. These devices include Avalanche Photodiodes (APDs), High Operating Temperature (HOT) IR detectors, very long wavelength IR (VLWIR) detectors using strain-layer super lattice (SLSL) designs, and Multi-spectral Focalplane Arrays (MFPAs).

7.2 Introduction MCT is very useful as an IR detector material, primarily because of two of its features compared to other IR detector materials: 1. MCT is a direct bandgap semiconductor and 2. the bandgap of MCT can be adjusted by varying the ratio of the amount of the HgTe compound to the amount of CdTe in the particular MCT composition.

234 Handbookof Infrared Detection Technologies Because of the direct bandgap, MCT has a high adsorption coefficient for the IR, resulting in high q u a n t u m efficiency in a relatively thin IR detector structure (approximately 10 ~tm thick). The direct bandgap allows operation of the IR detector at a higher temperature than other long wavelength IR detectors, such as extrinsic silicon photoconductors. The variable bandgap feature allows the tuning of a detector's cutoff wavelength in the wavelength region from 0.7 micrometers (lam) to slightly greater than 25 ~m by adjusting the MCT composition. Requirements on MCT materials are driven by the desire to produce IR detectors with optimum performance at a low cost. Optimum performance means high operability (i.e. few dead pixels), high quantum efficiency, uniform responsivity, low noise and detector stability. For the production of low cost, large area IR FPAs, large, uniform, defect flee, MCT wafers are needed. Three basic MCT material requirements are: 1. large wafer size, 2. small variations in composition and doping across a large wafer, and 3. very few defects across a large wafer. Other material requirements are determined from the specific IR device that is to be fabricated with the MCT material. For example, heterojunction IR arrays and multispectral IR detector arrays require multilayer material structures with uniformity in layer parameters, a sharp interface between each layer (or, depending on the device design, a controlled doping profile and/or controlled composition profile across layers) and stability of the layers. In addition, low cost production requires reproducibility of material parameters over many growth runs and rapid, non-destructive material evaluation methods. 7.2.1

Brief history

Research and development of MCT materials and devices have been documented in several books covering the periods from approximately 1970 to the present. 1-7 Developments of MCT before 1970 were not published in detail because of the potential military applications. Most of the early research was performed at military laboratories or commercial laboratories under government supervision. In the early 1980s, universities became more involved in the research and development of MCT. W. D. Lawson and co-workers at the Royal Radar Establishment in England did the initial studies on mixtures of HgTe and CdTe in 1958. s Others involved in the early development of MCT included French, United States, Polish and former Soviet Union, investigators. 1'2 A review of the state of the art for MCT for the mid 1980s is given by Herman and Pessa. 9 A history timeline for MCT material and IR detector development through the 1990s is given by Norton. 1~ Several journals and proceedings give information on the development and status of MCT technology. The Proceedings of the US Workshop on the Physics and Chemistry of Mercury Cadmium Telluride was published in the Journal of

MCT properties, growth methods and characterization

235

Vacuum Science and Technology during the period from 1981 to 1993. Currently, the workshop is published in the Journal of Electronic Materials. The workshop topics have been broadened to cover all II-VI materials. The Proceedings of the International Conference on II-VI Compounds is published in the Journal of Crystal Growth. Each year there are several proceedings published by SPIE that address MCT IR detector materials and MCT IR detectors and sensors.

7.3 MCT characteristics and material properties 7.3.1 Composition~crystal

structure

MCT is a ternary alloy consisting of a solid solution of the two compounds, HgTe and CdTe. A MCT composition is typically denoted as Hg 1_xCdxTe where the x (or x-value) is the mole fraction of CdTe and thus ( l - x ) is the mole fraction of HgTe. The Hgl_xCdxTe alloy system has a zincblende crystal structure which is in the form of two interpenetrating face-centered cubic lattices. In this crystal structure, each lattice site is surrounded by four nearest neighbors in which the Te-ion forms a sublattice with four nearest neighbors of the other sublattice. The other sublattice consists of randomly distributed Cd-ions and Hg-ions. (see Figure 7.1) There are five possible selections of the Cd and Hg as nearest neighbors to the Te-ion: 1. four Cd-ions, 2. four Hg-ions,

Figure 7.1 MCT zincblende crystal structure and ( 1 1 1 ) crystal surfaces.

236 Handbook of Infrared Detection Technologies 3. two Cd-ion and two Hg-ions, 4. three Cd-ions and one Hg-ion, or 5. three Hg-ions and one Cd-ion. These five structures occur randomly t h r o u g h o u t the MCT material in a perfect crystal. For the (111 / crystal plane, MCT has a B type surface (Te-rich) and an A type surface (Cd-rich) as shown in Figure 7.1. The lattice constant, ao, for MCT depends only slightly on the x-value, < O. 31% between x=O to x = l varying nonlinearly by approximately 2 0 x l O - l ~ m . 11 Hence, the lattice constant for MCT does not follow Vegard's law where the lattice constant varies linearly with composition. Lattice constant (in Angstroms) as a function of composition is given by the following equation: 12 ao -

6.4614 + (8.4x + 11.68x 2 - 5.7x3)1() -3

/1)

7.3.2 Bandgap CdTe is a direct bandgap semiconductor, whereas HgTe is a semimetal with zero bandgap. For x greater than approximately 0.15, the MCT alloy is a direct bandgap semiconductor. A simplified schematic of the MCT bandgap, (Eq), is shown in Figure 7.2. The band structure shown at the F point at the center of the Brillouin zone for a zincblende crystal structure consists of the conduction band, (F6), a light hole

l~g

F8 ]

Figure 7.2 MCT simplified band structure.

,\ICT properties, grou,th nlethods and characterization

2 37

band, (F8), a heavy hole band, (F8) and spin-orbit split-off band. (F7). Detailed band structure calculations for MCT can be found in 'The properties and applications of the Hgl_xCdxTe alloy system'. 13 The band structures for the conduction band, the heavy hole valence band and the light hole valance band are described by the following equations: 14

f2k 2 - -

..

_)

-3F-

~

2 D1o

]i2k 2

E,,,,(k) =

(3)

2mhh

E/,(k)---~-

1-

h2k 2

~=o(k) - - A -+ 2.,0

(2)

l+~3E~//

ti2k 2 4-~ 2 mo

(4)

k2p2

3(<, + A)

(5)

where E q is the bandgap energy. P is the Kane matrix element, E~. is the conduction band energy, E;,;, is the heavy hole band energy, E,, is the light hole energy, Eso is the spin orbit split-off band, k is the w a v e n u m b e r , nlo is the free electron mass and m;,;, is the heavy hole effective mass. Values for these p a r a m e t e r s are P = 8 . 4 • 10 -8 eV cm and A= 1 . 0 8 - ( ) . 12x eV. ~3 The bandgap of MCT varies with composition and temperature. Several empirical equations have been reported. 2'1 s.l~ E q is empirically related to t e m p e r a t u r e (T) and x for MCT by the following equations:

Eal(x,T)=-0.25+

1 . 5 9 x + 5.233 x 1()-4(1 - 2 . ( ) 8 x ) T + O . 3 2 7 x 3

Eg2(x, T) -- - 0 . 3 0 2 + 1 . 9 3 x - ().81x 2 + ().8 32x 3 + 5.3 5(1 - 2x) 10-4T

(6)

(7)

E~3(x, T) = - 0 . 3 0 2 + 1.93x - 0.81x 2 + 0.8 32x 3

+ 5 3511 - 2.,~)10 -~ ( - 1 8 2 2 + r ~ 9 ]~ ~75 -7 ~ - ) where Ea is in eV and T is in K. The most recent equation, E:~3, is plotted in Figure 7.3 for bandgap versus t e m p e r a t u r e and x-value.

238 Handbook of Infrared Detection Technologies

0.360"380"420.320.....340.260......... 280."220."....240"' ".............................................................................................................. 40.3 ".................................. 0.2

0.18

0.140.................... 12 100 0.1 0.08

0.16

i--

0.06

0.04

0.15 0.17 0.19 0.21 0.23 0.25 0.27 0.29 0.31 0.33 0.35 0.37 0.39 0.41

0.02

0

x-value

Figure 7.3 MCT bandgap versus x and temperature.

Since the cut-off wavelength, ;.,., for an IR detector can be approximately related to Ea by the equation" ;~,.- 1.24/E~

(9)

;~c values versus temperature and x can be determined. Figure 7.4 shows the variations in )~c with temperature and x for MCT. This is very useful in determining the MCT material parameter in designing an IR detector. Furthermore, the variation of ;., versus x can be determined. For example, for the MWIR (3-5 ~tm), there is little change in cutoff wavelength with change in xvalue. However, at the longer wavelengths, ( > 10 pm), the cutoff wavelength is very sensitive to changes in x-value. This defines the limit to which composition control must be held across a MCT wafer area for a uniform IR detector cut-off wavelength. In MCT crystal growth, the x-value for LWIR MCT must be controlled accurately to better than ().()()3 to maintain less than a micrometer variation in cutoff wavelength across the wafer. 7.3.3 Intrinsic carrier concentration

Carrier concentration variations with temperature and composition have been measured and empirical formulas developed. ]~'~7 Figure 7.5 shows a plot of the intrinsic carrier concentration (ni) versus temperature and composition using the most recent equation, hi2.

AICT properties, growth methods and characterization 239

24 23 21

~\\\

20 18 17

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

14

_~ 13 ~12

.,

9 8 7 6 5 4 3 2 1

300~(~..~ ~.... 9

q

0.19

0.17

0.21

0.23

0.25

0.27

0.29

0.31

0.33

0.35

0.37

0.39

0.41

x-value

Figure 7.4 MCT cutoff wavelength versus x and temperature.

nil - (5.585 - 3.82x + 1.753 x I()-~T - 1 . 3 6 4 x l()-3xT) • lO14E3/4r3/2exp(_Eg" ~ " \2kT] ni2 -

(A + B x + CT + D x T + Fx 2 + GT 2) x 1 ()14E,1T3/4 3/2exp \~_~j//"~-Eq

(1())

(11)

w h e r e A = 5 . 2 4 2 5 6 , B = - 3 . 5 7 2 9 , C = - 4 . 7 4 0 1 9 , D - 1 . 2 5 9 4 2 x 10 -2 , F = 5 . 7 7 0 4 6 and G = - 4 . 2 4 1 2 3 x 1() -r This figure shows the intrinsic carrier c o n c e n t r a t i o n at 300 K and for lower t e m p e r a t u r e s c o m m o n for the operation of IR detectors. Intrinsic carrier c o n c e n t r a t i o n values are useful in material analyses and in the design of Ill detector.

7.3.4 Doping and impurities Impurities h a v e been extensively studied for MCT. is Several elements from g r o u p IA and IB h a v e been d e t e r m i n e d to be acceptors on metal sites (Cd or Hg substitution). Several g r o u p IIIB elements h a v e been d e t e r m i n e d to be donors on metal sites, and some elements from g r o u p VIIB have been d e t e r m i n e d to be donors on Te-sites. Group IVB elements act as impurities or d o p a n t s for MCT. The activation of d o p a n t s depends on the MCT g r o w t h p a r a m e t e r s . Impurities s h o w n to be donors in MCT are B, A1, Ga, In, Si, O. CI, Br, and I. Impurities s h o w n to be acceptors in MCT are Li, Cu, Ag, Au, As, Sb. and P.

240 Handbook of Infrared Detection Technolo[lies 1.E+18 1.E+17

1.E+16 1 .E+15 1.E+14 ..... 1 .E+I 3 c

E o 1.E+12

'~

1.E+11

1 .E+IO

1 .E+09 1 .E+08 1 .E+07

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

1 .E+06 0.15

0.17

0.19

0.21

0.23

0.25

0.27

0.29

0.31

0.33

0.35

0.37

0.39

x value

0.41 ~-~

Figure 7. ~ MCT intrinsic carrier concentration versus .r and temperature.

Indium has been shown to be an effective donor in MCT. Results have varied as to its doping efficiency. Recently, Denisov has measured indium donors introduced into MCT with approximately a 1()()% doping efficiency. ~9 Others have had mixed result with indium. Capper is and Dingrong et al. 2~ concluded that indium was only partly active as a donor. Koppel et al. 21 have shown that the process by which indium is introduced into LPE MCT, influences it behavior as a donor in MCT. For the MOVCD growth method, indium cannot be used effectively since the indium alkyl reacts with the tellurium alkyl. Iodine has been found to be a donor dopant for MCT with the MOCVD growth method. Arsenic is used for p-type doping in both MOCVD and MBE grown material. Point defects (vacancies and interstitials) and lattice defects (dislocations and grain boundaries) can act as donors or acceptors in MCT or influence the electrical properties of dopants in MCT. For example, typical bulk grown and Terich LPE grown MCT are p-type because of the high concentration of Hg vacancies that behave as acceptors in the MCT crystal. Also, dopants can collect at precipitates or lattice defects and become neutralized. 7.3.5 Carrier mobility Mobility depends on scattering from ionized impurities, longitudinal optical phonons (LO) and defects. In MCT, electron mobility is two to three orders of magnitude greater than the heavy-hole mobility. Experimentally, electron mobility is 2 • for x~=().2 at a temperature of 77K. Heavy-hole

AICT properties, growth methods and characterization

241

mobility is measured in the range 1 ()()-3()() cm2/Vs. 1"2 Minority carrier mobility has also been measured using the Haynes-Shockley 22 approach. In this method, a light pulse generates excess carriers and the carriers are swept across the thin MCT layer by an electric field. The velocity of this carrier packet is measured and the mobility calculated from the ratio of the drift velocity to the electric field across the layer. Hole mobility in p-type MCT for x={).29 of 180 cm2/Vs at 18 5 K was measured and a hole mobility for x={).22 of 46{)cme/Vs at 9 0 K was measured. 23 Also, hole mobility in p-type MCT for x = 0 . 2 1 5 at 77K was measured to be 57()cmX/Vs. 24 The temperature dependence of the hole mobility was given by the equation, 24 /,,, = 1 0 S T - 2.78

(12)

where /x~ is the hole mobility in cm2/Vs and T is the temperature in K. Measurement of the ambipolar mobility of electrons versus temperature in p-type MCT for x = 0 . 2 9 and .28 5 was made using the travel time m e a s u r e m e n t method for the excess carrier packet. 2s Electron mobilities were measured in the 2 x 104 to 4 • range for temperatures below 2{){)K. An equation representing electron mobility at 3(){) K is given in cm2/Vs as 12 1

/*"

(8.754 x l{}-4x - 1.{}44 x 1{}-4)

(13)

Temperature dependent relationships for hole and electron mobilities have been expressed as 26 9 x l{}8b

/~,,(x, T) -

where

and

(14)

T2,~

a -

and b -

l*,,(x, T) - ,,,(x, T) 1 {}{}

-

(15)

(16)

Detailed calculations of mobility in n-type MCT have been made by Meyer and Bartoli. 27 Their model includes scattering from ionized impurities, neutral impurities, acoustic, piezoelectric, LO. and disorder scattering. Their model has been use to determine the compensation densities in p-type and n-type MCT.28-3{} 7.3.6 Carrier lifetime

There are four recombination processes that define the carrier lifetime in MCT. These are the two fundamental direct or band to band recombination

242

Handbookof Infrared Detection Technolo#ies

mechanisms, radiative recombination and Auger recombination, and two nonfundamental or indirect recombination mechanisms" Shockley-Read (SR) recombination and surface recombination. 1~ In the radiative recombination process, an electron in the conduction band recombines with a hole in the valence band giving up the excess energy as a photon. In the Auger recombination process, an electron in the conduction band recombines with a hole in the valence band giving up the excess energy to an electron in the conduction band. This second electron is excited to a higher energy level in the conduction band. For the Shockley-Read mechanism, recombination is t h r o u g h a level or recombination center in the energy gap between the conduction band and the valence band. This is a two-step process in which the center captures a carrier of one type and subsequently attracts a carrier of the opposite type, resulting in its annihilation. These levels in the energy gap are due to impurities or lattice defects. Surface recombination is due to surface defects or charge layers at which carriers recombine. A surface recombination velocity characterizes the surface recombination mechanism. Surface recombination is the limiting carrier lifetime for IR detectors that are not properly passivated at its surfaces. Minority carrier lifetimes have been measured in bulk and epitaxial materials. Carrier lifetimes in the 50 ns-l.512s range and surface recombination velocities in the range 3 0 0 - 2 0 , 0 0 0 cm/s have been measured in thin layer n-type (carrier : C - 0 . 2 . 31 Surface c o n c e n t r a t i o n - 5 x l O ~s cm -3) photoconductors with recombination velocity varied greatly with surface treatment. Excess carrier lifetimes have been measured in p-type MCT by photoconductive decay. 32 Measured carrier lifetime values varied from 2 3 to 100 ns for p - 1 0 1 6 cm3 material with x = 0 . 2 2 5 . Lifetime in photodiodes limited by SR centers have been measured by diode pulse recovery and EBIC techniques. 3~Auger limited lifetimes have been reported for high quality MCT. 34-3~ Freezeout and background photon flux effects on recombination mechanisms in HgCdTe have been analyzed by Schacham and Finkman. 37 They found that the temperature dependence of excess carrier lifetime for p-type MCT is very different from the excess carrier lifetime for n-type MCT because of freezeout. They also showed that the Auger 7 process dominates recombination in p-type MCT.

J

J LH

LH

Ra d iative

A ugerl

A uger7

Figure 7.6 Radiative, Au#er 1 and Au#er 7 recombination processes.

MCT properties, growth methods and characterization

243

The theory of the carrier recombination mechanisms has been described in detail for MCT. 33'3s'37-39 Radiative recombination is determined by the capture probability, B. At temperature T and for MCT with a dielectric constant e, the recombination rate, B, is given by the equation 3

B = 5.8

x

10-138~c

+ m~;

too.

1 + m* + --7 ml,

3

(17)

( 3 ~ ) ~ - ( E ~ + 3kTG + 3.W5k2T2)

in units of cm3/s, kT and E~ are in eV and where e:,,: is the static dielectric constant. The radiative recombination time, ra, is given by the equation 1

(18)

rR = B[no + Po + ac]

where no and Po are the electron and hole densities and ac is the number of excess carriers. Equations for the intrinsic Auger lifetimes, rai, are 4~

38x10_18 2 1 ( ( l + 2 r ) Eo) 9 e:~ (1 + r)2(1 + 2r)exp (1 + r) k-T rai = 3

(19)

IF1Fel 2 kT 2

where r - me~m*h and F~ F2[ is the wave function overlap integral. The extrinsic Auger lifetime (Auger 1 mechanism for n-type MCT), rA1, is 2 n 215Ai 72A1 ~- (tl o -Jr-po)(tlo -~- (~C) --t- ~(Po q- (~C)

(20)

where f l - v / 7 ( l ++2 r ) e x P ( 2 -(1(1-r)+ r)~Eg)

(21)

For the Auger 7 mechanism, for p-type MCT, a similar equation is used 2nayrA1 TA 7 -- ( tlo -}- Po -}- (~c) ( no Jr- Po )

since b<< 1 for MCT and where

(22)

244

Handbook of Infrared Detection Technologies

y~ 6

(23)

3k__l

~-g3

Recently, Auger lifetimes have been calculated using an extremely accurate MCT bandstructure 41 by Krishnamurthy and Casselman. 42 From these new results, the Auger 1 and Auger 7 calculated values agree much better than prior results. For LWIR, a value for 7 is calculated to be in the range from 3 to 6. And approximate equation for SR lifetimes, rs~, where the density of SR centers is small is given as 4 3.44

rsR =

"rpo(tlo -ff rll -ff Sc) -ff r,,o(Po -4--Pl -4- &')

(no + po + ac)

(24)

where

nl exp[ '

ands1 NexpE ]kT

and where Nc and N,, are the density of states for the conduction and the valence bands, Ec and E,, are the energy values at the conduction and valence band edge, and Et is the energy of the SR center. Also

rno = ~

1

O'nVt17n Nt

and

rpo =

1 O'pvthpNt

where N t is the density of SR centers and a,, and ap are the capture cross sections for electrons and holes respectively. The carrier thermal velocity, Vth, for electrons and holes is given by, vti,(8kT/mn) 1/2. For n-type MCT and p-type MCT, these equations can be approximated as:

rsHRn- rpo(1 + ,,,,"l'~] and = r,,o (1 +

rsH~p-

r,,o (1

+ P'~p,,]

exp( E':-Et,)kT )

(25)

To include surface recombination, the lifetime, r~,:r, is used: d r~f - 2S where d is the layer thickness and S is the surface recombination velocity. The net lifetime is given by the following expression,

(26)

MCT properties, growth methods and characterization

1 1 . . . . "!5

TR

t- ~

1

T A1

+ ~

1

TA 7

1 2S + -+ -"cS a d

245

(2 7)

7.3.7 Defects

Defects are a major problem for MCT material. Defects can reduce carrier lifetime, act as dopants, cause dead pixels, and provide current leakage paths that increase device dark current and 1/f-noise. Some common defects include: crystal lattice defects (i.e. small angle grain boundaries, dislocations); inclusions (i.e. Te inclusions); point defects and material damage (i.e. cracks, pits, scratches). There are m a n y causes for these defects such as: poor control of the growth parameters; poor substrate quality: poor substrate orientation and poor substrate lattice matching: contamination in the materials: contamination in the dopant sources; and contamination in the growth chamber. Kresel gives an account of the effects of defects on detectors. 45 Primary defects to be concerned about are: point defects such as vacancies and interstitials; dislocations and stacking faults; precipitates: and grain boundaries. These defects can reduce the carrier lifetime by acting as recombination centers and can cause increased dark current in photovoltaic devices if the defects are near or within the junction region. Recent experimental results indicate the reduction in photodiode performance due to the present of defects. 46 Photodiodes containing defects such as dislocation clusters, pinholes, striations, Te inclusions and heavy terracing, were found to have poor performance. Leakage currents in photodiodes began to increase for dislocation densities above 5 • s cm -2. At this dislocation level for the photodiodes tested, there were approximately 12 dislocations per photodiode. Hg interstitials were found to have a large impact on photodiode performance. Hg interstitials were found to induce a bias dependent dark current. Hg vacancies were found to reduce photodiode performance but to a lesser extent than the Hg interstitials. Hg vacancies appear to act as SR centers thus increasing the diffusion current in the photodiodes. There was also a bias current dependence found for the Hg vacancies but the m e c h a n i s m was not determined.

7.4 MCT crystal growth methods For high performance device fabrication, nearly perfect, large area crystalline MCT material is required. There have been four primary methods developed for the growth of high quality MCT crystals, Bulk Crystal Growth, Liquid Phase Epitaxy (LPE), Molecular Beam Epitaxy (MBE) and Vapor Phase Epitaxy (VPE). VPE can be broken down into two types, Chemical Vapor Deposition (CVD), such as Metal-Organic Chemical Vapor Deposition (MOCVD), and Physical Vapor Deposition (PVD), such as isothermal VPE (ISOVPE). There are variations and combinations of some of these growth methods, for example, a combination of

246 Handbookof Infrared Detection Technologies MBE and MOCVD called MOMBE is being used. 47'48 Another crystal growth method, similar to MBE, is called Atomic Layer Epitaxy (ALE). 49-52 7.4.1

Phase diagrams

Knowledge of the phase diagram of the various MCT compositions, as a function of temperature, pressure and x-value, is essential for developing a crystal growth method. The phase diagram for MCT has been measured and models have been developed that agree very well with experimental results. 5's3-ss See Figure 7.7 for examples of phase diagrams for MCT. Figure 7.7A shows the pseudobinary phase diagram useful for bulk MCT growth and Figure 7.7B shows the ternary phase diagram useful for LPE MCT growth. Schmit and Speerschneider summarized the initial phase diagram experiments as a function of temperature and pressure versus x-value, s~ Steininger collected additional results and developed a set of formulas to fit the phase diagram as a function of temperature and x-value, s7 Later, Steininger completed an extensive measurement of the phase diagram as a function of temperature and mercury pressure for various cases with varying ratios of the metal/non-metal, y, defined by, (Hgl_xCdx)y Tel_y. s8 Phase diagram at the Te corner, i.e. versus temperature and for y-values in the 0.5-1.0 range, have been measured s9-61 (see Gibbs Triangle). Similarly, phase diagrams at the Hg corner have been measured. 62'63 Both the Te corner and the Hg corner data are important for LPE growth methods. Models for phase diagrams have been developed to aid in the understanding and predicting of MCT crystal growth processes. Two models developed are the regular associated solution (RAS) model 64'6s and the generalized associated solution (GAS) model, s3 The RAS model results agreed very well with the liquidus data in the Te and Hg corners and with the solidus data in the Te

Te

Corner

~2

Solid

HgTe

Xl

x2 X-Value

CdTe

A. Pseudobinary Phase Diagrams Te concentration = 0.5

Hg Corner

T = Constant Value

Cd Corner

B. Ternary Phase Diagrams

Figure 7.7 Phase diagrams.

MCT properties, growth methods and characterization

247

corner. 5 The GAS model has shown excellent agreement with data for the phase diagram of HgTe-CdTe-Te. 53 7.4.2 Bulk growth

There are several methods for MCT crystal growth classified as bulk growth. These crystal growth methods include: 1. quench and anneal or solid state recrystallization (SSR), 2. traveling heater method (THM) or traveling solvent method, 3. Bridgman, 4. Slush method, 5. Zone Melting. 6. Czochralski. These methods are discussed in detail by Micklethweite. 5 MCT was initially fabricated by a quench and anneal method. 66-68 In this method, the three high purity elements are cleaned and loaded into a thick walled, small diameter quartz ampoule. The ampoule is evacuated, sealed and placed into a furnace. The furnace temperature is increased to approximately 950~ to melt the mixture in the ampoule. Mixing is accomplished by rocking the furnace. While still in the liquid state, the MCT ampoule is removed from the furnace and the MCT mixture is rapidly quenched to produce a uniform composition. Finally, the ampoule is exposed to a uniform annealing temperature for an extended period to allow crystal growth. This method produced a polycrystalline ingot from which the largest crystals are removed, often referred to as a 'crystal mining operation'. These small crystals were further annealed in Hg-vapor to convert them to n-type for IR detector fabrication. The quench and anneal method for MCT crystal growth has several disadvantages. The high melt temperature causes a very high Hg-vapor pressure and there is the danger of the ampoule exploding. Also, the non-equilibrium solidification, results in composition non-uniformities. Furthermore, a very long annealing time is required and only small crystals are produced. 5 An improved bulk MCT growth method was demonstrated by Triboulet called the traveling heater method (THM). 69 In the THM method a molten solvent zone is slowly moved through a solid homogeneous ingot of MCT by either moving the ampoule, which contains the ingot, or by moving the heater. Through convection and diffusion, the solid material is dissolved at the high temperature interface and deposited at the low temperature interface of the zone. Crystal growth occurs at a lower temperature than the quench and anneal MCT growth method, in the 500~176 range. THM has a growth rate of approximately one micrometer per minute. Ingots with diameters up to 3 cm and up to 8 cm in length have been grown. Radial variation in the x-value is better than + / - 0.002 and longitudinal variation better than + / - 0 . 0 2 . 7 o Another important MCT crystal growth method is called the slush method. 71 In this method, the MCT mixture is melted in the top portion of the furnace where

248 Handbookof Infrared Detection Technologies

Figure 7.8 Traveling heater method.

there is a constant temperature along the material. Then the material is quickly lowered to a second position in the furnace where there is a temperature gradient. The bottom portion of the melt is quickly frozen while the top part is still in the liquid state. A temperature gradient of approximately l OK/cm is maintained along the material about the solid/liquid interface. After approximately 40 days the furnace is turned off and the ampoule is cooled to room temperature. High quality single crystals have been produced by this method, which was adapted for detector grade MCT production. 5 7.4.3 Epitaxial growth

Epitaxial growth methods provide lower growth temperatures compared to bulk growth methods and provide crystals with uniform composition. Epitaxial growth methods have shorter growth times and provide larger crystals. With epitaxial growth methods, multilayered device structures are possible. Epitaxial growth, however, requires a suitable substrate. These substrates must meet very stringent requirements to be classified as being of IR detector quality. Substrates for epitaxial growth

Properties for a substrate for epitaxial growth include that it should: 1. be a very close lattice match to the MCT composition being grown, 2. be stable at the growth temperature, 3. not introduce impurities into the epilayer,

MCT properties, growth methods and characterization 2 4 9

4. be available in large size wafers, 5. be transparent in the IR spectral band of interest (unless it will be removed from the epilayer). CdTe was initially used as a substrate for epitaxial growth of MCT although it has approximately a 2% lattice mismatch for MCT. However, by adding 4% of ZnTe to the CdTe, a lattice matched substrate could be fabricated for MCT epitaxial growth. These substrates are transparent in the MWIR and LWIR spectral bands allowing the fabrication of backside illumination photodiode arrays with high q u a n t u m efficiency. The main disadvantage of these CdZnTe substrates is that growth methods to produce large area, high quality substrates have not been developed. Also, CdZnTe substrates have low mechanical strength. Another issue with these substrates is the poor thermal expansion match to silicon. Since most IR FPAs are built by interfacing the MCT detector array to a silicon readout chip, a good thermal expansion match between the detector and the readout provides for long term thermal cycle reliability. To provide a better substrate for MCT IR FPA production, several alternate substrates have been developed. A summary of the alternate substrate development has been given by Triboulet et al. 72 Alternate substrates that have been evaluated are sapphire, 73 GaAs, 74 InSb, 75 Si, 76 and Ge. 77 In most cases, a buffer layer must be grown on the alternate substrate before the MCT epilayer is grown. Of these alternate substrates, sapphire and Si are most widely used. Sapphire is used for the growth of SWIR and MWIR, backside illuminated IR FPAs since sapphire is transparent in the range from UV to approximately 6 l~m. A CdTe buffer layer is first grown on the sapphire substrate followed by the MCT layer. The CdTe buffer layer provides lattice matching between the MCT layer and the sapphire substrate. Silicon substrates have several advantages for MCT epilayer growth. Silicon is readily available in large sizes (i.e. up to eight inches in diameter). Silicon wafers are mechanically strong and of relatively low cost compared to CdZnTe and are transparent in the LWIR spectral region. Also, since large IR FPAs are bonded to silicon readout integrated circuits, there is no thermal coefficient of expansion mismatch between the readout and the silicon substrate. The disadvantages so far with silicon substrates are its high defect density (in the 106/cm 2 range) and the determination of a suitable buffer layer. Improvements in the silicon substrates for MCT growth are being made. Epitaxial MCT crystal growth methods can be divided into three types, liquid phase epitaxy (LPE), vapor phase epitaxy (VPE), and molecular beam epitaxy (MBE). LPE is currently the most widely used method to supply MCT crystals for IR detector production. However, MBE and VPE growth techniques are developing rapidly and IR FPAs have been fabricated from materials grown by these methods. /PE In the LPE crystal growth method, the MCT composition is dissolved in a solvent and the liquid mixture is placed in contact with a substrate. When the solvent is cooled, a solid MCT crystal layer forms on the substrate. LPE crystal growth

250 Handbookof Infrared Detection Technologies

Figure 7.9 MCT LPE growth methods.

methods are classified by the type of apparatus used to place the liquid into contact with the substrate and by the type of solvent. Methods classified by growth apparatus are; slider, dipping and tipping. (see Figure 7.9). All three methods are used for production of MCT layers for IR detector fabrication. LPE growth methods classified by solvent are, Te-rich, Hg-rich, and HgTe-rich. Terich and Hg-rich are the solvents used for producing MCT material for IR detectors. Layers grown with Te-rich melts are p-type because of the Hg vacancies induced during the growth cycle. These layers can be converted to ntype by annealing them in Hg vapor. Layers grown from Hg-rich melts are usually n-type. Sliders for LPE growth made from high purity graphite consist of a substrate holder and a charge holder (see Figure 7.9). A wiper is also provided in the slider to remove excess melt when the epilayer growth is completed. The slider is mounted into a tube inside a furnace. The tube can be closed or an open tube can be used with a constant flow of inert gas, typically hydrogen. The substrate sits in a well on the slider and charges are contained in compartments that can be moved over the substrate by an external mechanism. Advantages with this method are that the substrate can be cleaned by an initial melt-back of the substrate and multilayer structures can be grown. After melt-back of the substrate surface, the slider is lowered in temperature and a MCT layer forms on the substrate, which acts as a seed crystal. The properties of the substrate such as lattice matching, crystal orientation, defect density and purity are very important. Generally substrates are CdTe

MCT properties, growth methods and characterization

2 51

single crystals or lattice matched CdZnTe single crystals. However, growth on sapphire has been reported. 73 A buffer layer of CdTe is first grown on the sapphire before LPE growth of MCT. LPE epilayer growth using a slider and a Te-rich solution typically will proceed by ramping the temperature down at a rate of 0 . 2 - 0 . 3 degrees per minute. For a 40 minute growth period, a 2 0 - 3 0 lam thick epilayer can be produced. This does not include preparation time. Composition uniformities for x-value of less t h a n 0.002 have been demonstrated. These epilayers are p-type (1 x 1017) as grown and can be converted to n-type (mid 1 x 1015) by annealing in Hg vapor. LPE growth by using a slider has some disadvantages. Hg loss during growth can occur unless a source of Hg is provided. For Te-rich growths, the temperature is near 500~ and the equilibrium pressure of rig is 0.3 atm. If this is not controlled, the loss of Hg from the system will cause a variation in composition through the epilayer and will also lead to poor repeatability in producing epilayers. Several approaches have been explored to reduce Hg loss. One technique is to use a secondary well in the slider that contains a Hg source in the form of HgTe located close to the melt well. 78'79 In another technique, hydrogen gas is passed over an elemental Hg source prior to passing over the melt well. 8~ A third method is to introduce excess HgTe into the melt. 81 The dipping method is shown in Figure 7.9. In this method a Te-rich or Hgrich solution is maintained at the bottom of a vertically mounted quartz tube, which is mounted inside a furnace. A substrate holder is moved into the solution and the furnace is cooled to initiate crystal growth. Pulling the substrate holder out of the solution halts epi-layer growth. LPE by the dipping method can be separated into two methods, infinite melt and finite melt. An infinite melt (approximately 2kg) is used for Hg-rich solvents in order to prevent melt depletion due to the low solubility of Cd in Hg-rich solutions. Composition uniformity of x
252

Handbookof Infrared Detection Technologies

VPE There are basically two types of VPE, Physical Vapor Deposition (PVD) and Chemical Vapor Deposition (CVD). In the PVD process, a MCT source is vaporized by a high temperature at one region of an ampoule transported to a lower temperature region and deposited onto a substrate. In the CVD process, the sources are volatile chemicals that are transported to a region of a growth chamber where they react and deposit the MCT compound onto the substrate. The primary CVD method being developed is Metalorganic Chemical Vapor Deposition (MOCVD) in which some of the sources are metalorganics that are transported by hydrogen to the reaction chamber. Figure 7.10 shows a diagram of the MOCVD system. For example, dimethyl cadmium, (CH3)aCd, with diethyl tellurium, (C2H5)2Te and Hg vapor have been used in an open-tube hydrogen flow system. 85-8 7 In the PVD growth method, a source and substrate are placed in an ampoule and then the ampoule is evacuated and sealed, ss The source and substrate are very close together, usually separated with a thin spacer ring. A review of the isothermal VPE method (ISOVPE) has been presented by Bernardi. s9 For example, the ampoule containing the M C T source and the substrate is heated under isothermal conditions, the evaporated MCT source material deposits onto the substrate (CdTe wafer) and mixes by diffusion with the CdTe substrate to form a thin layer of MCT on the substrate. IR detectors have been fabricated from MCT grown by ISOVPE. 9~ CVD

MOCVD was first investigated by Manasevit. 91 The feasibility of using MOCVD for II-VI compound systems was demonstrated in 1969. 92 Pyrolytic MOCVD of MCT has been demonstrated using diethyl telluride (DETe), dimethyl cadmium (DMCd) and Hg. 85-87 The combination of two processes are involved in the formation of HgCdTe: 1. the CdTe formation from DMCd and DETE at the substrate, and 2. the formation of HgTe from DETe and Hg at the substrate. In MOCVD, the composition of MCT decreases as substrate temperature increases, as DMCd partial pressure becomes higher and as Hg partial pressures becomes lower. Two major difficulties with conventional MOCVD are control of epilayer composition and achievement of adequate lateral uniformity. Growth is very sensitive to local conditions of temperature, partial pressures and flow patterns making composition difficult to control. The problems of lateral uniformity and composition control have been addressed by Irvine by using a technique called Interdiffused Multilayer Process (IMP). 9s In IMP, very thin layers (0.1-0.2 lam) of HgTe and CdTe are deposited sequentially. These layers diffuse together during growth and the resulting epilayer has a composition which depends on the ratio of the thicknesses of the HgTe and CdTe layers. 94

MCT properties, growth methods and characterization 253

Figure 7.10 MOCVD system.

Low temperature MOCVD processes are being developed because of the need for sharp interfaces for heterojunction fabrication. The stability of DETe determines the lowest temperature at which growth is possible. Consequently, the various approaches to achieving lower temperature deposition have concentrated on decreasing the stability of the tellurium alkyl. The most direct approach involves the use of tellurium alkyls that are less stable than DETe. Another method involves pre-cracking the DETe in a hot zone upstream of the cooler growth area. In a third approach, photoassisted MOCVD, the organometallic species are partially broken down by the energy from ultraviolet photons produced by a broadband Hg-arc lamp. Several investigators have grown HgTe, CdTe, and MCT from alternate tellurium alkyls, including, di-isopropyltelluride (DiPTe), 95 di-tertiarybutyltelluride (DtBTe), 96 dimethyl-ditelluride (DMDTe), 97 and dihydrotellurophene (DHTeP). 9s By using increasingly unstable alkyls, it is possible to decrease the deposition temperature substantially from 400~ using DETe to 2 30~ using DtBTe. The tradeoff is that the vapor pressures of the alkyls also decrease with decreasing stability. For DMDTe, DHTeP and DtBTe, the growth rate is no longer limited by the substrate temperature, but is limited by the amount of tellurium alkyl that can be transported to the growth zone. Another problem with the alternate alkyls is the difficulty of synthesizing and purifying low vapor pressure materials. Investigations of low temperature MOCVD using pre-cracking of the tellurium alkyl have been reported. 99.1~176 Growth rates of 1-2 ~m per hour were reported for HgTe and HgCdTe at temperatures as low as 225~ by first passing the tellurium and mercury alkyls (DETe and DMHg) through a pre-cracking tube which is heated to between 400 ~ and 600~ The material grown by this method was n-type with room temperature carrier concentrations of 2-3 x 1017 cm-3 and mobilities 1-3 x 104 cm2/Vs. In the low temperature, photo-assisted MOCVD (PA-MOCVD) approach, ultraviolet photons must be incident on the reaction chamber to dissociate the

254 Handbookof Infrared Detection Technologies organometallic species. In order to ensure that the ultraviolet radiation has a clear path to the substrate during growth, the reactor wall above the substrate holder is continually flushed with pure carrier gas. This gas stream forms a barrier between the organometallics and the w a r m wall preventing deposits to m a i n t a i n a clear window. Several investigators have grown good quality CdTe, HgTe and HgCdTe epilayers using PA-MOCVD at temperatures ranging from 200 to 3 0 0 ~ 101-1()3 These systems have used high pressure mercury grid lamps as the sources of ultraviolet radiation and DETe, DMCd, and Hg as reactant species. Reported growth rates were low between O. 5 and 2.0 pm per hour. Typically, MOCVD is grown on CdZnTe oriented in (111)B or (211)B planes. MCT layers on (111)B oriented substrates exhibit pyramidal hillocks and MCT layers on (211)B oriented substrates exhibit void defects. Both of these defects can cause leaky or shorted photodiodes. Recently the MOCVD growth of MCT on (5:52) oriented CdZnTe has been reported. 1~ MCT layers grown on the CdZnTe (552) substrates have a surface morphology very different from MCT grown on (211)B substrates. The MCT layers grown on (552) substrates exhibited no void defects but had small hillocks (order of 40 ~m size) with densities in the 1 0 - 5 0 cm -2 range. These hillocks were determined to be flat and less than 0.75 Bm in height. Etch pit densities were measured in the 6• 104 to 5 • 1()~ cm -2 range. Other properties of the layers grown on (552) CdZnTe were comparable to layers grown on (211)B substrates. MBE

In MBE, molecular beams of the elements impinge on the heated substrate and crystallize into a MCT epilayer (see schematic, Figure 7.11). An ultra-high v a c u u m system is used for MBE crystal growth and Knudsen-type effusion cell are used to create the molecular beams. Since the sticking coefficient of Hg at the growth temperature is low, a specially designed oven is used for the Hg source. 1~176 The shutters in front of the molecular beam sources allow for the rapid change of composition or for controlled doping of the epilayers. Uniformity of the epilayer composition depends on the uniformity of the molecular beam

Figure 7.1 1 MBE s!tslem diagram.

MCTproperties. growth methods and characterization 2 5 5 across the substrate. Several investigators have d e m o n s t r a t e d single layer and multilayer MCT structures, lO7-1 o9 An i m p o r t a n t a d v a n t a g e of MBE is the capability to produce sharp interfaces for multilayered IR devices such as an IR FPA t h a t operate in two or three different spectral bands. This is due to the low g r o w t h t e m p e r a t u r e and the ability to rapidly s h u t t e r the sources. Since an ultra-high v a c u u m e n v i r o n m e n t is used for the g r o w t h chamber, several in situ analysis tools can be used to m o n i t o r the epilayer g r o w t h process and m e a s u r e the properties of the g r o w n layers. 11~ There are some disadvantages to MBE, the g r o w t h c h a m b e r is complex and expensive, precise control of the g r o w t h p a r a m e t e r s is essential over the entire g r o w t h period, t e m p e r a t u r e variations and beam flux variations across the substrate m u s t be kept very small, the g r o w t h rate is low and there are stringent r e q u i r e m e n t s on the substrate. A review of MBE results for detector quality MCT has been given by Wu. 113 Device quality material was d e m o n s t r a t e d for IR detectors operating in the SWIR, the MWIR and the LWIR. These layers were g r o w n on (211) CdZnTe substrate 2.5 c m x 2 . 5 cm in the t e m p e r a t u r e range 1 6 0 - 1 9 0 ~ 1 6 2Growth rates r a n g e d from 1-2 pm per hour, indium was used for donor doping and arsenic was used for acceptor doping. Uniformity m e a s u r e m e n t s on a sample 2.5 c m x 2 . 5 cm in area and 7.2 pm in thickness for an x-value of 0 . 2 5 9 had a s t a n d a r d deviation of 0 . 0 0 3 . For In doped layers, a carrier c o n c e n t r a t i o n of 2 x l O a S c m -3 was m e a s u r e d at 7 7 K with a mobility of 8 x l O 4 c m 2 / V s and a carrier lifetime of 1.O !as, comparable to LPE In doped layers. The In doping efficiency was determined to be approximately 100% for carrier c o n c e n t r a t i o n s less t h a n 2 x 10 ~8 cm-3. For the acceptor dopant, arsenic, a c a d m i u m arsenide c o m p o u n d and atomic tellurium were used to e n h a n c e the sticking coefficient of Hg and minimize Hg vacancies w h i c h inhibit the As from going to Hg vacancies instead of the group VI sublattice. ~~4. ~~s The As doping efficiency was estimated to be 50% and acceptor carrier c o n c e n t r a t i o n s over the range 10 ~6cm - 3 10 ~8 c m - 3 were demonstrated. MBE growth on silicon

Large area substrates are required to allow low cost production of MCT. High quality CdZnTe substrates are currently limited to approximately 30 cm 2 in area. Silicon is an alternate substrate that is low cost and available in areas from 81 cm 2 (four-inch wafers) to 182 cm 2 (six-inch wafers). The Si substrate also m a t c h e s the t h e r m a l expansion coefficient for a detector a r r a y indium b u m p bonded to a silicon readout resulting to lower strain induced during cooling of the FPA. la6 Several groups have d e m o n s t r a t e d MCT g r o w t h on small silicon substrates. ~~7-a 19 Since silicon is not lattice m a t c h e d to MCT. a buffer layer is needed for low defect density detector layers. Initial approaches were to use a C d T e / C d Z n T e layer first deposited on the silicon wafer for the buffer layer. Dislocation densities for MCT on silicon are reported in the 10 6 cm -2 range, approximately an order of m a g n i t u d e greater t h a n high quality LPE MCT grown on CdZnTe. Newer approaches use a ZnTe initiation layer followed by a CdZnTe layer on Si.

256 Handbookof Infrared Detection Technologies Recently, MWIR MCT has been grown on four-inch diameter silicon wafers. 12~ Measurements on MWIR MCT layers grown by this method show X-ray rocking curve data in the 8 0 - 1 0 0 arc-sec, range and etch pit densities in the range of 1-7 x 106 cm -2. The x-value value variation from the center of the wafer to the edge was 2%. MOMBE

MOMBE is a variation in the MBE growth method in which metalorganics are used for the molecular beam sources. 121 MOMBE may provide greater source control using gas flow system than the convential thermal sources used in MBE. MOMBE sources used for MCT growth are DmCd. DipTe. and elemental Hg. Donor doping with iodine using ethyliodide 122 has been reported with doping concentrations in the 4 . 4 x 1 0 1 4 - 5 • -3 range. Acceptor doping with asernic using a CdAs source resulted in samples with carrier concentrations in the range 8 x 1 0 1 6 to 3• cm -3 and mobilities in the 1 2 0 - 3 0 0 c m 2 / V s range. 123

7.5 Material characterization methods There are two main goals for characterization of MCT material. One is to provide accurate data for the parameters of MCT in order to gain scientific understanding of the material and of changes in material properties due to variations in external parameters. The other goal is to have a material characterization method that is suitable for a production environment. This type of characterization must be non-destructive, fast, and be capable of evaluating an entire wafer of MCT. Properties that are of interest include composition uniformity across the wafer and into the wafer, wafer thickness variations, carrier concentration, mobility, defects, impurity distribution, and surface morphology. 7.5.1 Material composition An initial method used to determine MCT wafer composition, i.e. x-value, was to measure the density of the wafer. This could not be applied accurately to small epitaxial layers on thick substrates. MCT composition can be measured accurately for thin layers by optical means, such as IR transmission or secondary ion mass spectrometry (SIMS) analysis. IR transmission is a non-destructive method and can be used to screen material for processing into IR detectors. If there is no composition gradient t h r o u g h the material, the measurement of the transmission curve versus wavelength at room temperature can be related to the detector cutoff wavelength at its lower operating temperature. Typically, a Fourier transform infrared transmission (FTIR) measurement can be used to determine both composition and layer thickness. However, if there is a composition gradient, the transmission analysis must be modified to determine the MCT composition. Theoretically, the transmission of a MCT layer, Ta, is given by the equation:

MCT properties, growth methods and characterization

Ta -- (1 - R)2exp(-c~t)

1 - R2exp(-2oet)

2 57

(28)

where R is the reflectivity, ~ is the adsorption coefficient and t is the layer thickness. For MCT, a has been related to temperature, T, x-value and photon energy, E,

as37.124

(2 9)

ot = A e x p ( a E )

where a--3.267x

104

(l+x) ( T + 81.9)

(30)

and A = otoexp(-aEo)

( 31 )

c~o -- exp(53.6 l x - 18.88)

(32)

Eo = 1 . 8 3 8 x - 0 . 3 4 2 4 + 0.148x 2

(33)

and

and

For the case where there is a composition variation, the equations have been used to determine the variation. 125 The method used was to calculate the derivative of the transmission curve and determine the m i n i m u m and m a x i m u m x-values. These values were then compared to Electron Probe Microanalysis data for the same samples. For linear composition variations, the method of Hougen 126 has been used and compared to the cutoff of the photodiode spectral response. This method can be used to predict the cutoff wavelength of a detector made from the measured material. 127 SIMS is a destructive technique for determining composition and variations in composition and is performed on a test sample. In SIMS, samples are sputtered with an ion beam and the secondary ions are collected by type in a mass spectrometer. This is a highly sensitive technique and can be used to determine doping and impurity levels as well as the composition of the MCT. a28 For MBE and MOCVD growth processes, the composition of the MCT layer must be monitored and controlled in situ. A review of the monitoring techniques for in situ MCT growth has been given by Irvine and Bajaj. 129 Composition is monitored during growth by using spectroscopic ellipsometry (SE). 13~ Ellipsometry is a sensitive optical technique used to monitor surface cleaning

258

Handbook of Infrared Detection Technologies

Fig1~re 7.12 Ellipsonleter.

processes and epitaxial material growth processes. 131 In ellipsometry, ellipitically polarized light is reflected from the surface being studies and the state of polarization of the reflected light is analyzed 132 (see Figure 7.12). Ratios of the parallel, E~, and perpendicular, Et,, optical field components before and after reflection, rv and r~ are combined into the ratio of the complex reflectivity, p, given in terms of two ellipsometric angles ~ and A as p-

tan(~)exp(iA)

(34)

In equation (34), tan(~) represents the amount that the ratio of the parallel and perpendicular components change by the reflection from the sampled surface. The term A in equation (34) is the amount of change in the phase difference for the two components due to the reflection. Material parameters are related to p through optical properties of the materials. For example, for a bare substrate, the complex refractive index, ns is related to p by the equation,

/

ns -- n~tan~ / I

V

4psin2~ -

(p+ 1):

(35)

where n~ is the refractive index of the ambient medium and ~b is the angle of incidence. By varying the wavelength, for SE, a more accurate determination of the epitaxial layer can be made. In SE, light from a Xe arc lamp is reflected off the sample and sent through a rotation analyzer and then to a diffraction grating where the spectrum over the 2 8 0 0 - 7 6 0 0 A range is measured by a set of detectors. The results are compared to a database to determine the composition. In this technique, the real and imaginary ellipsometric data is plotted in 2D giving the layer thickness and composition in real time. SE is routinely used to monitor MBE growth, to determine growth rate, composition, and to provide feedback to control material growth. 133.134 The accuracy of using SE to determine composition has been demonstrated to be within 0 . 0 0 1 5 for x-values in the 0.205 to 0.230 range. 135 Theoretical accuracy is calculated to be 0.0008. Accuracy due to variations in the angle of incidence and in spectral wavelength

MCT properties, growth methods and characterization

2 59

shifts has been studied and methods to further improve the accuracy of SE for composition determination have been suggested. 136

7.5.2 Measurements of carrier concentration and mobility Carrier concentration and mobility are routinely measured by the Hall Effect. Although this is usually a destructive method, since Hall samples are fabricated, in some cases accurate data can be measured on bulk wafers. For a two-carrier semiconductor, i.e. including holes and electrons, and at low magnetic fields, the Hall Effect is described in terms of the Hall coefficient, Rn, by the following equation, 137

(36)

q( lxhP + l~en) 2

where #h is the hole mobility, #,. is the electron mobility, p is the p-type carrier concentration, n is the n-type carrier concentration and q is the charge on an electron. The resistivity, PH, is given by 1

Plq

q(lzen + #hP)

(3 7)

These equations are a function of temperature since the mobilities and the intrinsic carrier concentration are functions of temperature as described previously. Carrier concentrations and mobilities for a sample can be determined from m e a s u r e m e n t s of the Hall coefficient and resistivity as a function of temperature. From equations (36) and (37), for high p-type MCT, RH= 6 . 2 5 x 1 0 is p-~, for intrinsic, p=n and R n = - 6 . 2 5 x l O is n -~ and for n-type, R n = - 6 . 2 5 x l O 18 n -1. Also, using equations (36) and (37) for an n-type MCT sample and with the condition that me>>mh, then RH=-pn#,,. Furthermore, at the temperature where RH=O, the equation for the ratio, b, becomes b=#~/l.tn. Equations (3 6) and ( 3 7), however, give only approximate values when applied to MCT. 138 Anomalous Hall results have been measured. These results are attributed to surface layers, such as a thin n-type layer on a thick p-type material, or to variations in carrier type through the material (e.g. multiple layers of different carrier type). To fully model and understand MCT Hall data, data is collected as a function of both temperature and magnetic field. The resistivity and Hall Coefficient including magnetic field dependence and temperature dependence for multiple carriers is given by the equation ~39 -~ m

tl i q i l"t i

'=

pH -

m

tliqild.i

12

,

i=1 1 + B2ft 2

+

-BY-.

2

niqilA 2 B

1 1 + B2g~

(38)

260 Handbookof Infrared Detection Technologies and }

R, =

X-~m niqi13 7 ) ) z_,i=~ l + B-#r

72[ m "=1 1 ~ B-2;2] -+- Ei=l 1 + ~-i2~-ff

2

(39)

where m is the total number of carriers, B is the magnetic field, qi is - 1 for electrons and +] for holes and/~i is - 1 for electrons and +1 for holes. These equations have been successfully applied to a layer model for LPE MCT 140 with ptype and n-type layers. 138.141-14 3

7.6 Summary MCT has been developed over the past 44 years and is a mature IR detector material. Its optical, physical and electrical properties are well understood. However, the demand for more complex IR detector structures for multiple spectral applications, avalanche photodiodes and high operating temperature IR FPAs, requires controlled epitaxial MCT growth methods. Most of the current research on MCT is to develop these epitaxial methods such as MBE and MOCVD. In addition, improved and new material evaluation and characterization techniques are being developed.

References 1. D. Long and J. L. Schmit, Mercury-cadmium telluride and closely related alloys, in Semiconductors and Semimetals, Vol. 5, 1 7 5 - 2 5 5 , edited by R. K. Willardson and A. C. Beer, Academic Press, New York (19 70). 2. P. W. Kruse, The emergence of Hgl_xCd• as a modern infrared sensitive Material, in Semiconductors and Semimetals, Vol. 18, 1-20, edited by R. K. Willardson and Beer, Academic Press, New York (1981 ). 3. Peter Capper, editor, Mercury Cadmium Telluride Part A, in Properties of Narrow Gap Cadmium-based Compounds, emis data review series, no. 1 O, INSPEC, the Institute of Electrical Engineering, London (1994). 4. A. Rogalski (ed.) Infrared Photon Detectors, SPIE Optical Engineering Press, Bellingham, WA ( 1995). 5. W. F. H. Micklethweite, The crystal growth of mercury cadmium telluride, in Semiconductors and Semimetals, Vol. 18, 4 8 - 1 1 9 , edited by R. K. Willardson and Beer, Academic Press, New York ( 1981 ). 6. A. Rogalski, Heterostructure infrared photovoltaic detectors, Infrared Physicsand Technology, 4 1 . 2 1 3 - 2 3 8 (2000).

MCTproperties, growth methodsand characterization 2 61 7. M. H. Weiler, Magnetooptical Properties of Hgl-xCdxTe Alloys, in Semiconductors and Semimetals, Vol. 16, 119-191, edited by R. K. Willardson and Beer, Academic Press, New York ( 1981 ). 8. W. D. Lawson, S. Nielsen, E. H. Putley, and A. S. Young, J. Phys. Chem. Solids, 9 , 3 2 5 (1959). 9. M. A. Herman and M. Pessa, Hgl-xCdxTe-Hgl-yCdyTe ( 0 < / - x , y < / - 1 ) heterostructures" prooerties, epitaxy, and applications, J. Appl. Phys. ~7 (8), 2671-2694(1985). 10. P. Norton, HgCdTe Infrared Detectors, in 6 th International Symposium on Long Wavelength Infrared detectors and arrays: Physics and Applications, edited by S. S. Li, M. Z. Tidrow, S. D. Gunapala and H. C. Liu, Boston Mass. (1999). 11. R. Dornhaus, G. Nimtz, and B. Schlicht, Narrow-Gap Semiconductors, Springer Verlag, Berlin ( 1983). 12. W. M. Higgins, G. N. Pultz, R. G. Roy and R. A. Lancaster, Standard relationships in the properties of Hgl-xCdxTe, J. Vac. Sci. Technol. A7, 2 7 1 - 2 7 5 (1989). 13. R. Dornhaus and G. Nimtz, The properties and applications of the HglxCdxTe alloy system, in Solid State Physics, 78, 1-119, Springer Verlag, Berlin (1976). 14. W. W. Anderson, Absorption constant of Pbl-xSnxTe and Hgl-xCdxTe alloys, Infrared Phys, 20, 363- 372 (1980). 15. G. L. Hansen, J. L. Schmit, and T. N. Casselman, 1. Appl. Phys. 53, 7 0 9 9 7101(1982). 16. J. R. Lowney, D. G. Seiler, C. L. Litter and I. T. Yoon, Intrinsic carrier concentration of narrow gap mercury cadmium telluride, ]. Appl. Phys. 71, 1253-258(1992). 17. G. L. Hansen, J. L. Schmit, Calculation of the intrinsic carrier concentration in Hgl-xCdx Te ]. Appl. Phys. 54, 1 6 3 9 - 1 6 4 0 (1983). 18. P. Capper, A review of impurity behavior in bulk and epitaxial HglxCdxTe, J. Vac. Sci. Technol. B9, 1667-1681 ( 1991 ). 19. A. Denisov, V. M. Lakeenkov, and O. K. Jouravlev, Physico-chemical properties of Ga and In dopants during liquid phase epitaxy of CdxHgl-xTe, J. Electron. Mat. 2 7 , 6 4 8 - 6 5 0 (1998). 20. O. Dingrong, T. Wenguo, S. Jie, C. Junhao, and Z. Guozhen, Solid State Commun. ~ 6 , 8 1 3 (1985). 21. P. Koppel, K. E. Owens, R. E. Longshore, High mobility HgCdTe by post growth indium doping, SPIE Proceedings, 1106. Future Infrared detector Materials, 70-78(1989). 22. I. R. Haynes and W. Shockley, Phys. Rev. 8 1 , 8 3 5 (1951). 23. D. E. Lacklison and G. Duggan, Determination of the minority carrier mobility of n-type cadmium-mercury-telluride using the Haynes-Shockley method, J. Appl. Phys. 8 5( 12 ), 4 2 5 7 - 4 2 6 5 ( 1984). 24. Y. Shacham-Diamand and I. Kidron, Haynes-Shockley experiment on ntype HgCdTe, J. Appl. Phys. 26(4), 1 1 0 4 - 1 1 0 8 (1984).

262 Handbookof Infrared Detection Technoloqies 25. S. E. Schacham and E. Finkman, A new method for measuring ambipolar mobility and its implementation in p-type HgCdTe, J. App1. Phys. 57(4), 1 1 6 1 1164(1985). 26. J. P. Rosbeck, R. E. Starr, S. L. Price and K. J. Riley, J. Appl. Phys. 5], 6430 (1982). 2 7. J. R. Meyer, F. J. Bartoli, J. Vac. Sci. Technol. 2 l, 237 ( 1982). 28. G. A. Hoffman, F. J. Bartoli, J. R. Meyer, J. Cryst. Growth 72, 4 4 3 - 4 4 7 (1983). 29. J. R. Meyer, F. J. Bartoli, G. A. Hoffman, J. Appl. Phys. 57, 4 6 3 8 - 4 6 4 4 (1985). 30. J. R. Meyer, G. A. Hoffman, F. J. Bartoli, D. A. Arnold, S. Sivanathan and J. P. Faurie, Semicon. Sci. Technol. 8 , 8 0 5 - 8 2 3 (1991 ). 31. R. D. Graft, F. F. Carlson, J. H. Dinan, P. R. Boyd, R. E. Longshore, Surface and interface recombination in thin film HgCdTe photoconductors, J. Vac. Technol. A l , 1 6 9 6 - 1 6 9 9 (1983). 32. R. Fastow and Y. Nemirovsky, The excess carrier lifetime in p-type HgCdTe measured by photoconductive decay, J. App1. Phys. 66(4), 1 7 0 5 - 1 7 1 0 (1989). 33. D. L. Polla, S. P. Tobin, M. B. Reine, and A. K. Sood, Experimental determinination of minority-carrier lifetime and recombination mechanisms in p-type Hgl-xcdxTe, J. Appl. Phys. 52 5 1 8 2 - 5 1 9 4 . 34. T. N. Casselman and P. E. Petersen, A comparison of the dominant Auger transitions in p-type (Hg, Cd)Te, Solid state Commun. 33, 6 1 5 - 6 1 9 (1980). 35. P. E. Petersen, Auger recombination in mercury cadmium telluride, in Semiconductors and Semimetals, Vol. 18, 1 2 1 - 1 5 5 , edited by R. K. Willardson and Beer, Academic Press, New York ( 1981 ). 36. T. N. Casselman, Calculation of the Auger lifetime in p-type Hgl-xCdxTe, J. Appl. Phys. 52 8 4 8 - 8 5 4 ( 1 9 8 1 ) . 37. S. E. Schacham and E. Finkman, Recombination mechanisms in p-type HgCdTe: freezeout and background flux effects, J. Appl. Phys. 57(6), 2 0 0 1 - 2 0 0 9 (1985). 38. D. Long, Photovoltaic and photoconductive infrared detectors, in Topics in Applied Physics, Springer-verlag, Berlin, 1 9 . 1 1 0 - 1 1 2 ( 1977). 39. M. A. Kinch, M. J. Brau and A. Simmons, J. AppI. Phys. 44, 1649 ( 19 73 ). 40. J. S. Blakemore, Semiconductor Statistics. Pergamon, Oxford (1962). 41. A. B. Chen, and A. Sher. Semiconductor Alloys, Plenum, New York (1995). 42. S. Krishnamurthy and T. N. Casselman, A detail calculation of the Auger lifetime in p-type HgCdTe, J. Electron. Mater. 29(6), 8 2 8 - 8 3 1 (2000). 43. W. Shockley and W. T. Read, Jr., Phys. Rev. 87, 835 ( 1952). 44. R.N. Hall, Phys. Rev. 8 7 , 3 8 7 (1952). 45. H. Kressel. The effects of crystal defects on optoelectronic devices, in Semiconductors and Semimetals, Vol. 16, 1-52, edited by R. K. Willardson and Beer, Academic Press, New York (1981 ). 46. M.C. Chen, R. S. List, D. Chandra, M. J. Bevan, L. Colombo, and H. F. Schaake, Key performance-limiting defects in P-on-N HgCdTe LPE heterojunction infrared photodiodes, J. Electron. Mater. 25, 1 3 7 5 - 1 3 8 2 (1996).

MCT properties, growth methods and characterization

263

4 7. C. J. Summers, B. K. Wagner, R. G. Benz II, and D. Rajavel, Chemical beam epitaxy ofHgCdTe, Semicond. Sci. Technol. 6, C1 O-C14 (1991). 48. C. J. Summers, B. K. Wagner, R. G. Benz, and D. Rajavel, Chemical beam/ gas source epitaxial growth of HgCdTe, Tenth Anniversay Issue of Chinese Journal of Infrared and Millimeter Waves, 11, 368 ( 1992 ). 49. T. Suntola and J. Antson, Finnish Patent No. 52359 (1974) and U. S. Patent No. 4 0 5 8 4 3 0 (1977). 50. C. H. L. Goodman and M. V. Pessa, J. App1. Phys. 60, R65 (1986). 51. T. Suntola, Materials Science Reports 4(7), 261 (1989). 52. N. H. Karam and K. Linden, H. Ehsani, and I. B. Bhat, Uniform deposition of CdTe, HgTe, and HgCdTe by ALE and MOCVD, Proc. SPIE 2 0 2 1 , 4 6 - 5 5 ( 1993). 53. T. Tung, C. H. Su, P. K. Liao, and R. F. Brebrick, ]. Vac. Sci Technol. 21, 117 (1982). 54. R. F. Brebrick, Thermodynamic modeling of the Hg-Cd-Te systems, J. Cr!lst. Growth 86, 3 9 - 4 8 (1988). 55. J. C. Brice, A numerical description of the Cd-Hg-Te phase diagram, Proc. Cryst. Growth and Charact. 13, 3 9 - 6 1 ( 1986). 56. J. L. Schmit and C. J. Speerschneider, Infrared Phys. 8 , 2 4 7 (1968). 57. J. Steininger, J. Appl. Phys. I 1 (6), 2 7 ] 3 ( 19 70). 58. J. Steininger, J. Electron. Mater. 5(3), 299 (1976). 59. T. C. Harman, J. Electron. Mater. 8(2 ), 191 ( 1979 ). 60. T. C. Harman, J. Electron. Mater. 9, 945 (1980). 61. R. F. Brebrick and A. J. Strauss, J. Phys. Chem. Solids 2 6 , 9 8 9 (1965). 62. J. P. Schwartz, T. Tung, and R. F. Brebick, ]. Electrochem. Soc. 128, 438 (1981). 63. T. Tung, L. Golonka, and R. F. Brebick, J. Electrochem. Soc. 12g, 451 (1981). 64. A. S. Jordan, Metall. Trans. 1,2 39 {19 70). 65. A. S. Jordan, in Calculations of Phase Diagrams and Thermochemistry of Alloy Phases, edited by Y. A. Chang and J. F. Smith, The Metallurgical Society of AIME, Warrendale, p. 1 O0 ( 1979 ). 66. S. G. Parker andH. Kraus, U. S. Patent No. 3 468 363 (1969). 67. M. J. Brau, U. S. Patent No. 3 656 944 (1972). 68. K. F. Riley, PhD Thesis, Syracuse Univ. ( 19 77). 69. R. Triboulet, Rev. Phys. Appl. 12, 123 ( 1977). 70. R. Triboulet, T. Nguyen Duy, and A. Durand, The 1984 U. S. Workshop on the Physics and Chemistry of HgCeTe, Extended Abstracts, p. 43. San Diego, CA {1984). 71. T. C. Harman, J. Electron. Mater. 2( 1 ), 23{)(1972). 72. R. Triboulet and A. Tromson-Carli, D. Lorans and T. Nguyen Duy, Substrate issues for the growth of mercury cadmium telluride, J. Electron. Mater. 22(8), 8 2 7 - 8 3 4 (1993). 73. E. R. Gertner, W. E. Tennant, J. D. Blackwell and J. P. Rode, HgCdTe on sapphire ~ a new approach to infrared detector arrays, J. Cryst. Growth 72. 4 6 2 - 4 6 7 {198 7).

264 Handbookof Infrared Detection Technologies 74. L. J. Kozlowski, R. B. Bailey, S. C. Cabelli, D. E. Cooper, G. McComas, K. Vural and W. E. Tennant, 640x480 PACE HgCdTe FPA, Proc. SPIE 1 7 1 5 , 1 3 173(1992). 75. R. F. C. Farrow, G. R. jones, G. M. Williams, and I. M. Young, Appl. Phys. Lett. 1 9 , 9 5 4 (1981). 76. S. M. Johnson, J. A. Vigil, J. B. James, C. A. Cockrum, W. H. Konkel, M. H. Kalisher, R. F. Risser, T. Tung, W. J. Hamilton, W. L. Ahlgren and J. M. Myrosznyk, MOCVD grown CdZnTe/GaAs/Si substrates for large-area HgCdTe IRFPAs, ]. Electron. Mater. 22, 8 3 5 - 8 4 2 (1993). 77. J. P. Zanatta, P. Ferret, G. Theret, A. Million, M. Wolny, J. P. Chamonal and G. Destefanis, Heteroepitaxy of HgCdTe (211)B on Ge substrates by molecular beam epitaxy for infrared detectors, J. Electron. Mater. 27(6) 1998. 78. J. L. Schmidt, J. Crystal Growth 6 5, 2 4 9 - 2 61 ( 1983). 79. J. C. Tranchart, B. Latorre, C. Foucher and Y. Le gouge, J. Crystal Growth 7 2 , 4 6 8 - 4 7 3 (1985). 80. T. C. Harman, J. Electron. Mater. 1 O, 1 0 6 9 - 1 0 8 4 ( 1981 ). 81. K. Nagahama, R. Ohkata, K. Nishitani and T. Murotani, J. Electron. Mater. 1],67-80(1984). 82. C. A. Castro, A review of key trends in HgCdTe materials for IR focal plane arrays, Proc. SPIE 2 0 2 l, 2-9 ( 1993). 83. L. Colombo, G. H. Westphal, P. K. Liao, M. C. Chn, and H. F. Schaake, Infrared focal plane array producibility and related materials, Proc. SPIE 1 6 g ] , 33(1992). 84. G. Bostrup, K. L. hess, J. Ellisworth, D. Cooper and R. Haines, LPE HgCdTe on sapphire status and advancements, J. Electron. Mater. ]0(6), 5 6 0 - 5 6 5 (2001). 85. S. J. C. Irvine and J. B. Mullin, The growth by MOVPE and characterization of Cd(x) Hg(1-x)Te, I. Crystal Growth 5 5, l 07 (1981 ). 86. I. B. Bhat and S. K. Ghandhi, The growth of mercury cadmium telluride by organometallic vapor phase epitaxy, 1. Crystal Growth 75, 24 ] (1986). 87. W. E. Hoke, P. J. Lemonias and R. Traczewski, Metalorganic vapor deposition of CdTe and HgCdTe epitaxial films on InSb and GaAs substrates, Appl. Phys. Lett. 44(11), 1046(1984). 88. P. Becla, J. Lagowski, H. C. Gatos and H. Ruda, ]. Electrochem Soc. 11711173(1981). 89. S. Bernardi, iso-VPE growth of HgxCdl-xTe on hybrid substrates, Proc. SPIE2554, 15-24(1995). 90. S. H. Shin and J. G. Pasko, App1. Phys. Lett. 44,423 (1984). 91. H. M. Manasevit, Single crystal gallium arsenide on insulating substrates, App1. Phys. Lett 12(4), 156 (1968). 92. H. M. Manasevit and W. I. Simpson, The use of metal-organics in the preparation of semiconductor materials, J. Electrochem. Soc. 118(4), 644 ( 19 71 ). 93. S. J. C. Irvine, J. Tunnicliffe and J. B. Mullin, The growth of highly uniform cadmium mercury telluride by a new MOVPE technique, Mater. Lett. 2(4B), 305 (1984).

MCT properties, growth methods and characterization 265

94. H. J. Hyliands, J. Thompson, M. J. Bevan, K. T. Woodhouse and V. Vincent, Metal-organic chemical vapor deposition of mercury cadmium telluride epitaxial films, ]. Vac. Sci. fechnol. A4(4), 2217 ( 1986). 95. W. E. Hoke and P. J. Lemonias, Metalorganic growth of CdTe and HgCdTe epitaxial films at a reduced temperature using diisopropyltelluride, App1. Phys. Lett. 46(4), 398 (1985). 96. W. E. Hoke and P. J. Lemonias, Low-temperature metalorganic growth of CdTe and HgTe films using ditertiarybutlytelluride, Appl. Phys. Lett. 48(24), 1669(1986). 97. D. W. Kisker, M. L. Steigerwald and K. S. Jeffers, Low temperature growth of II-VI compounds, presented at the Third International Conference on Metalorganic Vapor Phase Epitaxy, University City, CA, 13-17 April 1986 (unpublished). 89. J. D. Parsons, L. S. Lichtmann, E. H. Cirlin and D. W. Brown, Characterization of a new tellurium source for low temperature MOVPE of mercury cadmium telluride by unassisted pyrolysis, presented at the Third International Conference on Metalorganic Vapor Phase Epitaxy, University City, CA, 13-17 April 1986 (unpublished). 99. C. H. Wang, P. Y. Lu and L. M. Williams, Epitaxial growth of HgTe by precracking metalorganic mercury and tellurime compounds, Appl. Phys. Lett. 48(16), 1085 (1986). 100. P. Y. Lu, C. H. Wang, and L. M. Williams, Epitaxial Hg(1-x)Cd(x)Te grown by low-temperature metalorganic chemical vapor deposition, Appl. Phys. Lett. 49(20), 1372 (1986) 101. S. J. C. Irvine, J. B. Mullin, and J. Tunnicliffe Photosensitisation: a stimulant for the low temprature growth of epitaxial HgTe, 1. Crystal Growth 68,

188(1984).

102. D. W. Kisker and R. D. Feldman, Photon assisted OMVPE growth of CdTe, 1. CrystalGrowth 7 2 , 1 0 2 (1985). 103. S. J. C. Irvine, J. Geiss, J. B. Mullin, G. W. Blackmore and O. D. Dosser, Photo-metal organic vapor phase epitaxy: a low temperature method for the growth of Cd(1-x)Hg(x)Te, 1. Vac. Sci. Technol. B3(5), 1456 (1985). 104. P. Mita, F. C. Case, H. L. Glass, V. M. Speziale, J. P. Flint, S. P. Tobin and P. W. Norton, HgCdTe growth on (552) oriented CdZnTe by metalorganic vapor phase epitaxy, 1. Electron. Mater. 30(6), 7 7 9 - 7 8 4 (2001). 105. J. P. Faurie and A. Million, 1. Crystal Growth 5 4 , 5 8 2 (1981). 106. J. P. Faurie, Developments and trends in MBE of II-VI Hg-based compounds, 1. Crystal Growth 8 1 , 4 8 3 - 4 8 8 (1987). 107. J. P. Faurie and A. Million, and G. Jacquier, Thin Solid Films 90, 107 (1982). 108. P. P. Chow, D. K. Greenlaw, and D. Johnson, 1. Vac. Sci. Technol. AI, 562 (1983). 109. J. P. Faurie and A. Million, and J. Piaguet, Appl. Phys. Lett. 41, 713 (1982). 110. S. I. C. Irvine and J. Bajaj, Recent progress with in situ monitoring of MCT growth, Proc. SPIE 2 2 7 4 , 2 4 - 3 6 (1994).

266

Handbook of Infrared Detection Technologies

111. J. E. Jensen, J. A. Roth, P. D. Brewer, G. L. Olson, J. J. Dubray, O. K. Wu, R. D. Rajavel and T. ]. de Lyon, Integrated multi-sensor control of II-VI MBE for growth of complex IR detector structures, I. Electron. Mater. 27(6), 4 9 4 - 4 9 9 (1998). 112. L. A. Almeida, ]. N. Johnson, ]. D. Benson, J. H. Dinan and B. Johs, Automated composition control of Hgl-xCdxTe during MBE, using in situ spectroscopic ellipsometry, 1. Electron. Mater. 27(6), 500-503 (1998). 113. O. K. Wu, Status ofHgCdTe MBE technology for IRFPA, Proc. SPIE 2 0 2 1 , 79-89(1993). 114. T.S. Lee, ]. Garland, C. H. Green, M. Sumstine, A. ]andeska, Y. Selamet, and S. Sivananthan, Correlation of Arsenic incorporation and its electrical activity in MBE HgCdTe, J. Electron. Mater. 29(6), 8 6 9 - 8 7 2 (2000). 115. H. F. Schaake, On the kinetics of the activation of arsenic as a p-type dopant in Hgl-xCdxTe, 1. Electron. Mater. 30(6), 7 8 9 - 7 9 1 (2001). 116. T. J. de Lyon, R. D. Rajavel, J. E. Jensen, O. K. Wu, S. M. Johnson, C. A. Cockrum and G. M. Venzor, J. Electron. Mater. 25, 1341 (1996). 117. T. ]. de Lyon, ]. E. Jensen, m. D. Gorowitz, C. A. Cockrum, S. M. Jonson and G. M. Venzor, J. Electron. Mater. 2 8 , 7 0 5 (1999). 118. N. K. Dhar, M. Zandian, ]. G. Pasko, ]. M. Arias and J. H. Dinan, Appl. Phys. Lett. 70, 1730 (199 7). 119. R. Sporken, M. D. Lange, S. Sivananthan and J. P. Faurie, Appl. Phys. Lett. 59,81(1991). 120. K. D. Maranowski, ]. M. Peterson, S. M. Johnson, J. B. Varesi, A. C. Childs, R. E. Bornfreund, A. A. Buell, W. A. Radford, T. J. de Lyon and ]. E. Jensen, MBE growth of HgCdTe on silicon substrates for large-format MWIR focal plane arrays, J. Electron. Mater. 30(6), 6 1 9 - 6 2 2 (2001). 121. R. G. Benz II, B. K. Wagner, D. Rajavel and C. ]. Summers, Chemical beam epitaxy of CdTe, HgTe and HgCdTe, J. Crystal Growth 111 7 2 5 - 7 2 9 (1991). 122. C. ]. Summers, B. K. Wagner and R. G. Benz II, Recent advances in the metalorganic molecular beam epitaxy of HgCdTe, Proc. SPIE 2 0 2 1 , 51-66 (1993). 123. L. H. Zhang, S. D. Pearson, W. Tong, B. K. Wagner, J. D. Benson and C. J. Summer, P-type As-doping of Hgl-xCdxTe grown by MOMBE, J. Electron. Mater. 27(6), 6 0 0 - 6 0 4 (1998). 124. E. Finkman and S. E. Schachman, J. Appl. Phys. 56, 2896 (1984). 125. M. Anandan, A. K, Kapoor, A. Bhaduri and A. V. R. Warrier, Optical characteristion of Hgl-xCdxTe bulk samples, Infrared Phys. 31 (5), 4 8 5 - 4 9 1 (1991). 126. C. A. Hougen, J. Appl. Phys. 66, 3763 ( 1989). 127. D. Rosenfield, V. Garber, V. Ariel and G. Bahir, Composition graded HgCdTe photodiodes: prediction of spectral response from transmission spectrum and the impact of grading, ]. Electron. Mater. 24(9), 1 3 2 1 - 1 3 2 8 ( 1995 ). 128. ]. Sheng, L. Wang and G. E. Lux, SIMS characterization of HgCdTe and related II-VI compounds, J. Electron. Mater. 25(8 ), 1165-1171 (1996). 129. S. ]. Irvine and ]. Bajaj, Recent progress with in situ monitoring of MCT growth, Proc. SPIE 2 2 7 4 , 2 4 - 3 6 (1994).

MCTproperties. growth methods and characterization 267 130. R. H. Hartley, M. A. Folkard, D. Carr, P. J. Orders, D. Reees, I. K. Varga, V. Kumar, G. Shen, T. A. Steele, H. Buses and J. B. Lee, J. Vac. Technol. B I O, 1410

(1992).

131. R. M. A. Azzam and N. M. Bashara, Ellipsometry and Polarized Light, North-Holland, Amsterdam ( 19 77). 132. D. R. Rhiger, Use of ellipsometry to characterize the surface of hgCdTe, J. Electron. Mater. 22(8), 8 8 7 - 8 9 8 (1993). 133. J. E. Jensen, J. A. Roth, P. D. Brewer, G. L. Olson, J. J. Dubray, O. K. Wu, R. D. Rajavel and T. J. deLyon, Integrated multi-sensor control of II-VI MBE growth of complex IR detector structures, J. Electron. Mater. 27(6), 4 9 4 - 4 9 9 (1998). 134. L. A. Alameida, J. N. Johnson, J. D. Benson, J. H. Dinan and B. Johs, Automated composition control of Hgl-xCdxTe during MBE, using in situ spectroscopic ellipsometry, J. Electron. Mater. 27(6), 5 0 0 - 5 0 3 (1998). 135. D. Edwall, J. Phillips, D. Lee and J. Arias, Composition control of long wavelength MBE HgCdTe using in-situ spectroscopic ellipsometry, J. Electron. Mater. 30(6), 6 4 3 - 6 4 5 (2001) 136. M. Daraselia, J. W. Garland, B. Johs, V. Nathan and S. Sivananthan, Improvement of the accuracy of the in-situ ellipsometric measurements of temperature and alloy composition for MBE grown HgCdTe LWIR/MWIR structures, J. Electron. Mater. 30(6), 6 3 7 - 6 4 2 (2001 ). 137. A. Melissinos, Experiments in Modern Physics, Academic Press, New York (1966). 138. L. F. Lou and W. H. Fry, Hall effect and resistivity in liquid-phaseepitaxial layers ofHgCdTe, J. Appl. Phys. 56(8), 2 2 5 3 - 2 2 6 7 (1984). 139. F. J. Blatt, Physics of Electronic Conduction in Solids, McGraw-Hill, New York (1970). 140. T.T.S. Wong, Masters Degree Thesis, MIT (1974). 141. P. Koppel and K. Owens, Transport properties of liquid-phase-epitaxial Hgl-xCdxTe n/p structures, J. Appl. Phys. 67( 11 ), 6 8 8 6 - 6 8 9 8 (1990). 142. S. P. Tobin, G. N. Pultz, E. E. Krueger, M. Kestigian, K.-K. Wong and P. W. Norton, Hall effect characterization of LPE HgCdTe P/n heterojunctions, J. Electron. Mater. 22(8), 9 0 7 - 9 1 4 ( 1993). 143. M. L. Young, J. Giess and J. S. Gough, Assessment of electrical inhomogeneity of undoped and doped Hgl-xCdxTe MOVPE (IMP) layers by variable magnetic field Hall profile measurements, J. Electron. Mater. 22(8), 9 1 5 - 9 2 1 (1993).

This Page Intentionally Left Blank

Chapter 8

HgCdTe 2D arrays-technology and performance limits lan M. Baker

8.1 Introduction The concurrent development of infrared focal plane arrays, cryogenics, electronics and display technology by companies around the world, has made available a wide range of infrared cameras and systems-to meet an equally wide range of applications. Often the focal plane array is the discriminator in a system defining its resolution, sensitivity and imaging quality. For this reason there is enormous interest in the development of two-dimensional infrared arrays. HgCdTe is usually the infrared sensitive material of choice due to its high intrinsic sensitivity, moderate cooling requirements and spectral versatility. HgCdTe 2D arrays and processes vary significantly from one company to another but have the following common features. Firstly they are all based on arrays of photodiodes and secondly they are all mass-connected to a silicon integrated circuit, which performs the function of scanning the array. There are many possible technologies for producing 2D focal plane arrays in HgCdTe and there is a tendency for different manufacturing centres to adopt their own unique processes. The literature is therefore quite complex. One of the aims of this chapter is to try to explain the underlying technology and performance limits of 2D focal plane arrays and then describe how manufacturers have approached the challenge. Usually the specifications for infrared arrays contain fairly specialised parameters and so the various measurement techniques and specification parameters are also described. The chapter finishes with a description of the research and development paths that centres throughout the world are using to advance towards so-called GENII! detectors. It may seem that the infrared community has been slow to take up 2D focal plane arrays for thermal imaging, following in the footsteps of the visible imaging community. However first-generation infrared cameras for thermal imaging have a well established market. These are based on optical

270 Handbook of Infrared Detection Technologies scanners and small linear arrays of photoconductive HgCdTe elements and have set a benchmark for display picture points and cost. The evolution of 2D focal plane arrays has depended on matching the spatial resolution (number of elements in the array) and cost of existing systems, which until recently has been quite difficult. There are several reasons why infrared arrays lag well behind visible imagers. The wavelength of infrared radiation presents a physical limitation to the pixel size and it is not possible to use the very small pixel sizes commonly found in visible imaging arrays. Infrared arrays then tend to be physically large and, together with the extra complexity of the manufacturing process, relatively expensive. Volumes are also too low to make much impact on price. One of the aims of this chapter is to describe the work being conducted in many centres to develop very low cost technology and cameras with a low cost of ownership. Hgl_xCdxTe is the most widely used material for high performance infrared detectors at present. By changing the composition x, the detector spectral response can be made to cover the range from 1 Bm to beyond 17 Bm. The advantages of this system arise from a number of features, notably: close lattice matching, high optical absorption coefficient, low trap density, high electron mobility and readily available doping techniques. These advantages mean that background limited performance can be achieved at relatively high operating temperatures. HgCdTe continues to be developed as the material of choice for high performance long wavelength (8-12 lam) arrays and has an established market at the medium (3-5 ~tm) and short wavelength ( 1-3 lam) ranges. 8.1.1 Historical perspective

In the mid-seventies, work started on second-generation infrared detectors in the two commonly used atmospheric windows at 3-5 and 8 - 1 4 ~m. At that time it was seen that many infrared applications in future would need higher radiometric performance and/or higher spatial resolution than could be achieved with first generation photoconductive infrared detectors. The main limitation of photoconductive detectors is that they cannot easily be multiplexed on the focal plane. In consequence, there needs to be a separate electrical connection and amplifier for each detector element, which makes cryogenic encapsulations expensive for more than a few hundred elements. Also, the relatively high power dissipation on the focal plane provides another limitation for the larger array sizes. Photovoltaic arrays, on the other hand, consume very little power and can be easily multiplexed using an on-focal plane silicon chip. They are therefore well suited to long linear and two-dimensional infrared arrays. Systems based upon such focal planes can be made smaller, lighter, with lower power consumption and can result in much higher performance than systems based on first generation detectors. During the past two decades the development and industrialisation of HgCdTe based 2D arrays has taken place in centres throughout the world and a variety of viable material growth processes and detector technologies are now in the

HgCdTe 2D arrays - - technology and performance limits

2 71

market place. The field is a rapidly growing one as manufacturers continually push performance and cost barriers. However, it is not straightforward to grow high quality HgCdTe material and fabricate noise-flee arrays due to the crystals' low binding energies and the electrically active nature of defects. So the maturation of products has been relatively slower than in other material systems. Another limitation is the silicon multiplexer (or Readout Integrated CircuitROIC). The evolution of the size and pixel density has been constrained by the availability of silicon processes and chip sizes. Silicon critical dimensions have become finer and very large scale integration (VLSI) processes have become established progressively over the past 20 years. The infrared industry has then been able to design larger and larger arrays with better sensitivity and functionality. This trend is illustrated in Figure 8.1. In fact the role of the silicon multiplexer in determining the performance of the 2D focal plane array cannot be overemphasised and Section 8.4 is devoted to this area.

8.2 Applications and sensor design Currently the most important market for HgCdTe 2D arrays is in thermal imaging in the medium waveband (3-5 pm). Mid-performance thermal imagers tend to use sensors in the so-called half-TV formats of 3 2 0 x 2 4 0 or 3 8 4 x 2 8 8 and these imaging systems are finding use in many defence, security and industrial fields. High performance imagers use sensors up to 6 4 0 • Terminal guidance applications tend to use smaller arrays because of the data processing overhead in the signal processing electronics. At the other extreme, some very large arrays have been manufactured mainly for astronomy. Short wavelength 2D sensors are needed mainly for active systems where the scene is illuminated by an infrared laser, usually at the eyesafe wavelength of 1.54 lam.

Figure 8.1 Evolution of array size as a function of time. Data courtesy of Rockwell~Boeing.

272

Handbook of Infrared Detection Technologies

At longer wavelengths ( 8 - 1 2 ~m) 2D arrays need to compete with established scanned systems using long linear HgCdTe arrays and so most of the applications use relatively small arrays for specific guidance applications. However, manufacturers are now supplying 2D LW arrays for an increasing n u m b e r of imaging applications where the better sensitivity and faster flame rate offers operational advantages. It is important to note that the technology is often dictated by the application, and often quality and cost need to be carefully considered. For instance, the best thermal imaging systems need detectors with the highest possible sensitivity and will call for the best crystal growth techniques and processing technology. It may not be economically viable to use this technology for high volume thermal imaging products or very large focal plane arrays. Lower cost wafer scale technology is emerging but the challenge is to match the defect levels and excess noise of the best technologies. Often the detector is the performance-limiting component in the system, and it is necessary to use detectors with a sensitivity limited only by the r a n d o m rate of arrival of photons from the scene (so-called background limited detectors-BLIP detectors). In a narrow gap semiconductor it is necessary to cool to achieve BLIP performance so that thermally generated leakage currents are frozen out. For thermal imaging applications the commonest means of cooling is a cryocooler based on the Stirling cycle which can achieve temperatures as low as 55 K but is normally regulated to a temperature in the range 8 0 - 1 4 0 K. Figure 8.2 shows a typical integrated detector-cooler assembly (IDCA). The cut-away in Figure 8.2 shows the front end of the detector in more detail. The 2D focal plane array is bonded directly on the cold finger of the cooler. To ensure that only photons from the scene reach the detector, a cold stop or radiation shield is bonded to the cold finger. The cold stop will have internal baffles and special coatings to stop internal reflections and is a vital component to achieve BLIP operation of the array. The detector will have an optical pass-band filter that controls the upper and lower wavelengths. This will vary with the application but typical bands are 3.2-4.2 ~m or 3.7-4.9 um. Two methods are commonly used. The filter can be incorporated in the cold stop, and is therefore cooled, producing a very efficient block to stray radiation, say from the dewar itself. The

Figure 8.2 Typical integrated detector-cooler assembly ( IDCA ). Courtes!l of BAE SYSTEMS Infrared Ltd.

HgCdTe 2D arra!ts- technology and performance limits

2 73

disadvantage is the extra thermal load on the cold finger and extra optical surfaces. Almost as good performance can be achieved with filters on the front window and a well-designed cold stop. External to the detector shown in Figure 8.2 will be various electronics boards to control the cooler temperature and process the signals from the focal plane. HgCdTe arrays are not uniform enough to image directly and so each pixel is individually corrected for offset and gain in a process called non-uniformity correction (NUC). The aim is to completely eliminate spatial noise in the image and not add any additional temporal noise. Manufacturers have developed sophisticated signal processing h a r d w a r e and software to achieve this. One of the points to emphasise is that the focal plane array is only one of a n u m b e r of components that need to be optimised to preserve the BLIP performance. The optical design, cryogenics and signal processing are critical to m a i n t a i n the sensitivity of the focal plane array. The precision engineering involved in designing IDCAs, particularly for survival in high shock and vibration environments, is one of the reasons that the m a n u f a c t u r i n g cost of IDCA detectors is still relatively high. An alternative means of cooling is the Joule-Thompson cooler, which relies on a bottle of compressed gas. Figure 8.3 illustrates a typical product used in a missile guidance head. HgCdTe arrays are well suited to this type of application because they do not need to be cooled as deeply as other materials, such as: indium antimonide, and argon (90 K terminal operating temperature) can be used as a working gas resulting in cooldown times of a few seconds. Focal plane arrays for this application need to be thermo-mechanically robust to withstand the shock of rapid cooling. For specialised applications, thermo-electric cooling can be used as illustrated in Figure 8.4. State of the art coolers can achieve temperatures as low as 180 K but in general they are not as efficient as Stirling cryocoolers and the cooling

Figure 8.3 Typical fast cooldown detector using Jouh'-Thonlpson cooling. Courtes!l of BAE SYSTEMS Infrared Ltd.

274

Handbookof Infrared Detection Technologies

Figure 8.4 Typical thermo-electrically cooled detector. Courtesy of B AE SYSTEMS Infrared Ltd.

performance can be poor in high ambient temperatures. A temperature of 180 K is not quite cold enough for BLIP thermal imaging applications using current detectors. However, for lower sensitivity applications, thermo-electric cooling is sometimes used and its main strengths are much better long term reliability, greater shock resistance and smaller space-volume. Currently, infrared cameras based on half-TV of full-TV array formats and IDCA encapsulations are in production and are producing images with sensitivities in the region of 1 0 - 2 0 mK. Figure 8.5 shows a representative image from a state of the art HgCdTe 2D array.

8.3 Comparison of HgCdTe with other 2D array materials HgCdTe arrays have competition from lead salt detectors, indium antimonide (InSb) and metal silicide Schottky detectors in the MW band and multiple quantum well detectors in the LW. InSb based detectors have a large share of the MW thermal imaging market, particularly in the US, but less so in Europe and produce uniform and sensitive detectors. Array sizes up to about 640• 512 are available commercially. The advantages of InSb are that it is a binary compound (and hence compositional uniformity is not an issue) and that it is a III-V material. It is thought to be easier to grow and process III-V materials (less susceptible to damage) than II-VI materials such as HgCdTe. In addition, n-type InSb has a very large mobility, which reduces the series resistance effects in large arrays. Its main disadvantage is that the trap densities are typically 10 times higher than HgCdTe and this reduces its performance at temperatures above 80 K. In fact, HgCdTe can operate at temperatures up to 80 K higher than InSb due to this and the fact that HgCdTe can be tuned to exactly match the wavelength of interest as opposed to the fixed 5.5 um cut-off of InSb. In comparison with HgCdTe the sensitivity is very similar but InSb has potential limitations in imaging quality. HgCdTe is produced in thin layers and each pixel can be well defined leading in principle to much better image blurring effects. Most InSb manufacturers use bulk grown InSb and a

HgCdTe 2I) arra!ts - - technolog!t and perfornumce limits

275

Figure 8.5 M W infrared image from a 3;q4 x 2 8 8 element arra!l (OSPREY) and the Kenis system. Courtes!l of Kentron Ltd.

planar device structure. This can lend itself to crosstalk and blooming effects and restricts the smallest practical pixel size. Also there is the potential for image lag problems due to carrier trapping in the front surface of the InSb material. Lead salt detectors were developed for imaging systems until about the mid 70s when the work was stopped partly because the large dielectric constant made them unsuitable for photoconductors. For photodiodes, however, this can be an advantage and more recent work has investigated producing photodiodes on silicon substrates. The advantage of lead salt detectors is the potentially low cost due to the compatibility with growth on silicon for large arrays. The cut-off wavelength is tuneable across the MWIR and LWIR bands and the Auger generation rates are lower compared to the InSb and HgCdTe materials. The disadvantage at present is higher thermal leakage and higher 1/f noise that makes them inferior to InSb and HgCdTe on the grounds of sensitivity alone. Metal silicide Schottky detectors are formed directly on the silicon readout circuit and are available in large formats (1040• 1040) limited only by the silicon technology. They have the advantages of very good uniformity and excellent stability over long time periods and high storage temperatures. Their main disadvantage is that they have a relatively low quantum efficiency which drops with wavelength towards the cut-off and this means that the arrays tend to need deep cooling. Current development work is aimed at improving the quantum efficiency and extending the cut-off wavelength. Multiple quantum well (MQW) detectors use a well developed, robust III-V technology (usually A1GaAs/GaAs), which has already been developed for high

2 76

Handbook of Infrared Detection Technologies

speed electronics, to produce high quality detector arrays mainly in the LWIR band. Unlike other detector technologies MOWs are not sensitive to a broad range of wavelengths. They are essentially tuned to a specific wavelength as defined by the intersubband absorption and so normally a peak wavelength and spectral width is quoted. Another key difference is that the absorption depends on the angle and plane of polarisation of the photon and this leads to detector designs with optical structures to reflect and diffract the incoming flux. The main benefit is that low frequency noise (]/f noise) is generally negligible in MOW arrays in contrast to HgCdTe arrays where low 1/f noise demands very good processing. Also large area. 2D, arrays benefit from the lower cost and fewer defect levels associated with III-V materials. The main disadvantages are the low cooling temperatures needed which can lead to reduced cooling engine life and the limited sensitivity due to the lower quantum efficiency-gain product. The dark current depends on the detailed band structure used in the MQW stack. The band structure and the optical grating design is under intense development at various centres to raise the operating temperature of LW arrays closer to that of HgCdTe. However, the ultimate sensitivity of HgCdTe is better and has made this the material of choice for future high performance 2D 1 arrays.

8.4 Multiplexers for HgCdTe 2D arrays Both charge coupled devices (CCDs) and standard foundry CMOS devices have been used for multiplexing infrared 2D arrays but CMOS is now the preferred choice. The function of the multiplexer is to integrate the photocurrent from the photodiode array, perform a limited amount of signal processing and sequentially read out the signals. Typically, the photocurrent is accumulated on capacitors in a period called the integration time (or stare time) and then sequentially scanned out in a period called the readout time. Most manufacturers produce their own multiplexer designs because these often have to be tailored to the application. One of the advantages of CMOS to the infrared industry is that CMOS circuits will operate well at low temperatures. Typically long wavelength IR detectors will require cooling to 7 5 K or less. At these temperatures almost all aspects of the CMOS circuit operation improve, in particular, metal track resistance, MOSFET gm and white noise and the design of a circuit to operate at low temperature is essentially the same as a normal IC design. The trend in state of the art CMOS processes has been to shrink the critical dimensions (towards 0.1 Bin) and reduce the rail voltages (to 1.7 V or lower). In multiplexers designed for infrared applications, a low rail voltage can seriously restrict the dynamic range of the device and hence its sensitivity. Most CMOS designs for infrared applications still use ().5-().8 l.tm design rules, to take advantage of the 5-7 V supply rails, respectively. Other considerations are noise and uniformity. Modern processes are not necessarily developed with low noise and high analogue uniformity in mind. The design engineer can only rely on the MOSFET size to produce devices with low ]/f noise and high uniformity.

HgCdTe 2I) arrays - - technolog!l and performance limits

2 77

For this reason it is not always possible to exploit the density of processes to provide smaller and smaller pixels. These constraints have tended to slow up the adoption of leading edge processes for infrared. Section 8.6.1 explains the dependence of the sensitivity of the detector on the integration capacitance value. For background limited detectors the NETD (Noise Equivalent Temperature Difference)is shown to have a very simple relationship with the charge capacity of the multiplexer. It is inversely dependent on the square-root of the number of integrated photoelectrons. High sensitivity can therefore only be achieved with a high pixel capacitance. Normally the capacitance is created using a gate oxide capacitor and in 5 V tolerant processes, capacitance densities as high as 3 fF/ktm 2 are typical. Nevertheless, it is difficult to achieve capacitances of greater than 1 pF in pixels of around 25 ~m square and so the NETD is restricted to about 1 () mK per flame. It is important to note that this NETD limitation has nothing to do with the HgCdTe but only depends on the multiplexer design, emphasising the importance of the multiplexer on 2D array performance. Some of the key issues in multiplexer design for HgCdTe arrays are discussed below and particularly the aspects that affect the sensitivity and imaging performance of the array.

8.4.1 Photocurrent injection techniques There are a number of methods for injecting photocurrent from HgCdTe array to a silicon multiplexer. The choice depends on the application and good comprehensive summaries are presented by references 2. 3 and 4. The main techniques are described here.

Direct injection Injection of photocurrent from an infrared photodiode has been the subject of considerable interest in the infrared community because good injection efficiency is a prerequisite for achieving good signal to noise. In principle a perfect diode will inject photocurrent into a common gate MOSFET with 1 ()()% efficiency. This is called direct injection and simply requires one small MOSFET to hold the diode biased at close to zero bias and a capacitor on the drain to perform the integration. In reality, infrared diodes do not have perfect current-voltage characteristics because of the presence of leakage currents in the narrow band gap material. In effect this produces a shunt resistance and some photocurrent is lost to the shunt resistance rather than being injected into the multiplexer. Achieving a high shunt resistance has been a driver in the development of most infrared technologies. Very roughly there is a distinction between medium wavelength (MW) arrays with cut-offs up to 6~m and long wavelength (LW) arrays. In MW arrays there is usually no problem achieving high injection efficiencies because the photodiodes are near ideal. Near 1()()% injection efficiency have been measured at photon flux levels equivalent to at least F8 and 300 K backgrounds. In the long waveband there will be a limit to the upper cut-off wavelength and/or flux levels that routinely allow high injection

278 Handbookof Infrared Detection Technologies efficiencies using direct injection. Nevertheless the majority of multiplexers depend on direct injection because of the simplicity and low power consumption. Active circuits for injection

There are several situations where an active circuit is needed to e n h a n c e the injection efficiency. The first is where the wavelength is so long that diodes with adequate dynamic impedance are technologically challenging. The second is where the flux is so low that the input time constant of the diode and direct inject MOSFET is too high to give adequate temporal response. In both of these cases the aim is to produce a circuit with lower input impedance. Many circuits have been proposed but two circuits are in common use. The first is often called buffered direct injection, because the aim is to feed back a negative-amplified signal to the gate of the direct inject MOSFET so as to reduce the input impedance. The second is the so-called capacitive transimpedance amplifier (CTIA). Here the integration capacitor is in the feedback of an inverting amplifier. The CTIA is the most widely used technique because it has very favourable topology for compacting into a pixel unit cell and the integration capacitor can be made very small for high sensitivity. The disadvantage of both of these approaches is extra power consumption and extra sources of nonuniformity in the array. Nevertheless the CTIA circuit is a powerful way of overcoming temporal response and sensitivity problems in applications with low flux levels.

8.4.2 Scanning architectures The readout or scanning of the pixel voltage information in a CMOS array is done by sequentially X-Y addressing each pixel. The pixel will typically contain a column-addressed source follower. This will send the pixel voltage to a peripheral buffer where it is read out by a fast shift register. Often there are no active amplification circuits used since the voltages to be scanned out are of the order of volts. Many multiplexers offer optional functionality in the scanning such as a windowing function to permit more frequent scanning of a small area of the array, variable gain or variable numbers of parallel outputs (for faster readout times). In 2D arrays there are three basic scanning architectures: snapshot or blinking mode, fully staring mode and rolling readout. The difference between these is outlined as follows. In snapshot mode the integration period alternates with the readout. Clearly there is a duty cycle in this type of design that is dependant on the scan time and often multiple parallel outputs are used to minimise the readout time. This type of design is used for most applications, the exception being when the photon flux is very low. The integration capacitor is usually maximised to give the longest integration time and best signal to noise. It is also the preferred architecture in microscanned applications where the scene needs to be shifted between integration periods to improve the spatial resolution. When the flux is so low that the integration time approaches the available time to produce a flame there is some extra performance to be gained if the array

HgCdTe 2I) arra!lS - - technolog!l and performance limits

2 79

continues to integrate during the time it is being scanned. One way of doing this is to design a sample and hold function in the pixel so the previous frame of data can be stored. Some care needs to be taken to ensure that the signal to noise is not degraded by switching noise (KTC noise). The drawback of this approach is that the integration capacitor cannot be as large as the snapshot array because of the area taken up by the sample and hold circuit. Its ultimate sensitivity will therefore be less than the snapshot mode. Another way of achieving a fully staring architecture is the so-called rolling readout device. In a rolling readout design the integration capacitors in each row are reset after each row is read out so that the integration time is almost a full frame time. The rows have a unique integration period progressively shifted by a line time and this may not be acceptable in some systems. However the socalled latency (time from sensing to readout) is minimised in this design and is favoured in tracking applications. In m a n y multiplexer designs it is possible to have the option of snapshot or rolling readout because the pixel design is the same. 8.4.3 Future trends

Advanced detectors will critically depend on the future development of multiplexers. Challenges for the future include: arrays with larger physical size, smaller pixels, higher sensitivity, faster frame rate and even perhaps with retinatype processing to reduce the need for external signal processing. This will be a major challenge in view of the contradictory requirements for small pixels, more complex circuits and improved analogue performance. As mentioned there are strong arguments for remaining with relatively conservative CMOS processes. In addition the engineering costs associated with setting up a prototype run of wafers can be very much higher with leading-edge CMOS processes and the economic a r g u m e n t is not easy to make. However. there are technical reasons to eventually migrate to these processes. Firstly. extra signal processing speed can be obtained without increasing the power consumption: and secondly, the packing density of circuits (critical for small pixel sizes) is considerably eased by smaller metal track width. One of the first circuits to be enabled by denser CMOS is the on-chip analogue to digital converter (ADC). Putting the ADC on the focal plane reduces the susceptibility of analogue signals to electromagnetic interference and allows much higher output data rates. Another interest is the bi-spectral or multi-spectral array. Here the pixel design may need to include two or more duplicate sets of circuitry to cater for the different wavelengths and this can only be done with very dense metal tracking.

8.5 Theoretical foundations for HgCdTe array technology A good s u m m a r y of photodiode fundamentals and the properties of HgCdTe diodes in particular are included in reference 5 and the excellent reference list contained therein. The purpose of this section is to highlight the key issues

280 Handbook of Infrared Detection Technologies that control the thermal generation, leakage currents and q u a n t u m efficiency in detectors. From this analysis it is easier to understand the directions in which detector manufacturers have gone to make practical high performance detectors. 8.5.1 Thermal diffusion currents in HgCdTe

Expressions for the diffusion current in photovoltaic devices have been derived by Reine et al. 6 A well known parameter is R,,A, (zero bias resistance-optical area product) which is inversely proportional to the thermal generation current. A fundamental RoA expression is given in equation ( 1 ). RoA(n - side) -

(N.,/qn~)(kgr,./q.,.) 1/2

(])

where Na is the net donor concentration on the n-side, ni is the intrinsic carrier concentration, which dominates the temperature dependence, and rh and/~h are the minority carrier lifetime and mobility respectively. A similar expression describes the contribution of the p-side. In practice, the diffusion length, L, in normally doped HgCdTe is often larger than the 1()/.tm needed for effective infrared radiation absorption. In this case, the volume available for the generation of diffusion current is restricted, and a suitable modified expression is: RoA(n - side) - (kTr,,N,,)/(q2,,~t,,)

(2)

where t,, is the thickness of the n-side material. In the case of L being greater than t,,. the surfaces and contacts can act as sources of extra diffusion current if the surface recombination velocity is greater than the diffusion velocity. D/L. It is essential to ensure properly passivated surfaces and to employ a minority carrier barrier, such as a higher doped or wider band gap zone in front of the metal contact. The behaviour of equation (2) with doping, depends upon the dominant recombination process, i.e. radiative. Auger or Shockley-Read (S-R). Auger recombination in HgCdTe is a well-understood p h e n o m e n o n involving the interaction of three carriers. The Auger 1 lifetime in n-type material is due to the interaction of two electrons and a hole and is generally minimised by using a low carrier concentration on the n-side. In p-type material Auger recombination involves two holes and an electron and is referred to as Auger 7. The lifetime in ptype HgCdTe is reported by Lacklison et al.." Polla et al. s and Radford et al. ~ to show an inverse, linear dependence on doping, and this is attributed to the Shockley-Read process. S-R recombination is associated with the Hg vacancy and can be modelled by a strong donor level located .-~3() mV from the conduction band. which appears to be independent of composition in the .r=0.2- 0.3 range. The density of these donor S-R centres is proportional to the Hg vacancy concentration, but lower by a factor of .-.,2(). Mercury vacancies are often the dominant source of thermal diffusion current. In p-type

H~ICdTe 2I) arra!lS - - technolog!l and performance limits

281

material the lifetime is generally better (by up to an order) if Hg vacancies are replaced by acceptors (such as Cu. Ag. Na or Au). The use of extrinsic doping is an effective way to reduce thermal currents in homojunction devices. In order to engineer a detector with low thermal current (or high RoA) it is important to recognise that it is only necessary to use one side of the junction for collecting photocurrent: the other side. in principle, can be made with a wider bandgap, thereby minimising the diffusion current contribution. Devices with layers of different band gap are called heterostructures and the most common design is to use a wider band gap on the p-side to eliminate the dominant p-side contribution. 8.5.2 Leakage currents Interband tunnelling

Due to the very low effective mass of the electron in HgCdTe, direct band-to-band tunnelling can occur from filled states in the valence band, to empty states at the same energy in the conduction band. An expression for the current due to this process has been developed by Anderson. lc~ The tunnelling current increases very rapidly as the applied voltage or doping is increased, or the band gap decreased. It is the normal modern practice to use low doping on the n-side of the junction to minimise interband tunnelling under normal operating conditions.

Generation in the depletion region The thermal generation rate. ga,,p, within the depletion region via traps, is given by the usual Shockley-Read expression:

(3) where nl and Pl are the electron and hole concentrations which would be obtained if the Fermi energy was at the trap energy, and ~,,,, and ~p,, are the lifetimes in the strongly n-type and p-type regions. Normally. one of the terms in the denominator of equation (3) will dominate, and for the case of a trap at the intrinsic level, n~ and El=hi. giving #a,.p=ni/~. It is the weaker dependence on ni, and therefore on temperature, that distinguishes generation within the depletion layer from thermal diffusion current. Where the depletion region intercepts the surface, there is often enhanced generation due to the presence of a high density of interface states. This can be exacerbated if the surface passivation is not properly optimised to give a fiat band potential at the junction. In an extreme case, the surface on one side of the junction may become inverted, creating an extension of the depletion layer along the surface, and leading to high generation currents. A practical solution to this mechanism is to widen the band gap in the material where the junction intercepts the surface, so-called heteropassivation. This can be achieved by using thin film of CdTe together with a low temperature anneal, and this is the commonest passivation technique used by manufacturers.

282 Handbookof Infrared Detection Technoloqies

Impact ionisation The reverse-bias characteristics of homojunction arrays show some characteristic behaviour. Firstly, the product of p-side diffusion currents (including photocurrent) and reverse bias resistance is fairly insensitive to temperature and cut-off wavelength over a wide range (at least 4-11 I~ms), and secondly, the current increases much more slowly with reverse-bias than tunnelling models would predict. A model based upon an impact ionisation effect within the depletion layer gives a good fit to the practical observations. 11 The model proposes that leakage current arises because of the creation of extra electron-hole pairs within the depletion region due to impact ionisation by minority carrier electrons from the p-side. The leakage mechanism has been confirmed to have a linear relationship with optically injected minority carriers over a wide temperature range. 12 Calculations have been performed on the effect of impact ionisation on homojunction performance as a function of the doping levels. 1~ These predict that, to achieve a high dynamic resistance and therefore a high injection efficiency, the n-side doping must be very low. Routinely achieving low carrier concentrations ( < S x l O 1 4 c m -3) is an important aim for homojunction technology. However, there is some evidence that very low values may lead to instability and some manufacturers back-dope with indium to control levels at around 3-5e 14 cm -3. The carrier concentration in the p-region has a secondorder effect compared with that of the n-region, for the range of concentrations normally used. A weakly doped and uniform n-side is therefore essential for good junction characteristics in homojunction structures. 8.5.3 Photocurrent and quantum efficiency HgCdTe has an strong optical absorption coefficient and only thin layers are needed to produce high quantum efficiency. Typically in MW detectors the absorber need only be 4-5 l~m thick, and about twice this in LW detectors. Ideally the absorption should occur well within a diffusion length of the p-n junction to avoid signal loss due to recombination. A long carrier lifetime is nearly always observed in n-type material with low carrier concentration. Device engineers tend to favour using n-type absorbers for the best quantum efficiency and try to minimise the volume of the p-region for lower thermal leakage currents.

8.6 Technology of HgCdTe photovoltaic devices Tables 8.1 and 8.2 show a summary of some of the technologies being used around the world to produce 2D focal plane arrays. The list is by no means exhaustive and new technologies and products are being launched continuously. The activity in the US has been pushed towards heterostructures and doped detectors due to a high emphasis on reducing thermal leakage currents. In Europe the trend has been towards simple homojunctions in vacancy-doped material, which can give fewer defects and lower 1/f noise. The

HgCdTe 21) arra!lS - - technology and performance limits

283

Table 8.1 European Companies (in alphabetical order) active in 2D FPA manufacture Company

Technology

Photodiode type I waveband t

Selection of arrays t pitch }

AIM (AEG InfraredModules GmbH) [Reference 14]

Te-solution LPE on CdZnTe

l'lanar, n-p I SW and MW~

2 5 6 x 2 5 6 [4()pm) 384 x 288 [ 24 laml 64{) x 512124 t2m t

BAE SYSTEMS Infrared Ltd, Southampton, UK [Reference 15]

Te-solution LPE on CdZnTe MOVPE on GaAs or Si

Via-hole, n-p {SW, MW and LW) Heterodiode, N-n-P IMW and 13A'~

320 x 256 (3() pm} 384 x 288 (2() l~m} 64()x 512 (24 pro)

Sofradir, Grenoble, Fr [Reference 16]

Te-solution LPE on CdZnTe

Planar. n-p t SW, MW and I3,VI

128 x 128 (5() IJm} 320 x 2 56 { 3() pm }

Table 8.2 US Companies (in alphabetical order) active in 2D FPA manufacture Company

Technology

Photodiode type iwaveband~

Selection of arrays (pitch

BAE SYSTEMS, Lexington [Reference 17]

2-layer LPE on CdZnTe MOVPE on CdZnTe

Heterodiode. P-n {SW. MW and IAV1

2 5 6 x 2 56 (3()l.tm) 4 3 2 x 4 3 2 (301.tml

DRS Infrared Technologies, Dallas, TX [Reference 18]

Te-solution LPE on CdZnTe, Doped with Cu or Au

Via-hole, n-p (SW. MW and IAVt

64() x 512 ( 2 5 l-tmi

Raytheon Infrared Centre of Excellence (RIRCoe) and Hughes Research Labs (HRL) [Reference 19]

2-layer LPE on CdZnTe or Si MBE on CdZnTe or Si

Heterodiode. P-n (LW} Heterodiode. P-n (LWt

6 4 x 6 4 (61 pm) 128x 128 {401~m) 256x256{3()btm)

Rockwell/Boeing [Reference 20]

MBE on CdZnTe or Si Hg-rich LPE on MOVPE CdTe on sapphire

Planar (buried in) P-n (SW and MW} Planar (buried jn i n-p ISW and MW)

256 x 2 56 (40 l.tm} 32()x 24() ( 30 ~m) 640 x 480 {2 7 l-tm} 2048 x 2()48 ( 18 pm)

origin of this trend may have been a tendency for US programmes to have longer wavelength specifications, thereby placing a premium on minimising cooling. Another trend is for companies to move towards smaller pixel sizes and lower cost substrate materials as arrays become larger than about lOx 10 mm. In Tables 8.1 and 8.2 upper case P and N denote wider gap (longer wavelength)

284 Handbookof Infrared Detection Technologies layers e.g. P-n denotes a wider band gap p-type layer grown on a narrow gap ntype absorber layer. 8.6.1 Materials growth technology

Considerable progress has been made in the last two decades in the epitaxial growth of HgCdTe. Bulk growth methods are still used for providing good quality material for photoconductor arrays, but for photovoltaic arrays there are problems associated with crystal grain boundaries, which are electrically active, and cause lines of defects. Also there are limitations in the boule size, which makes it suitable for only small arrays. Several epitaxial growth techniques are in use today. Manufacturers will select a technique that suits their device technology and the type of detectors they are trying to make. For instance, high performance LW arrays will call for the best possible crystal quality, whereas, large area MW arrays can probably accept poorer quality material but must have large, uniform wafers. It is the aim of most manufacturers to produce high quality layers in large areas at low cost but this ideal has been elusive. At the present time the best structural quality material is grown using liquid phase epitaxy onto lattice-matched crystals of CdZnTe. Large area techniques are based on vapour phase epitaxy, VPE. Liquid phase epitaxy, LPE Liquid phase epitaxy (LPE) of HgCdTe at present provides the lowest crystal defect levels, and very good short and long-range uniformity. Capper et al. 21 has produced a detailed summary of the field. LPE layers are grown using an isothermal super saturation or programmed cooling technique, or some combination. A detailed knowledge of the solid-liquid-vapour phase relation is essential to control the growth, particularly in view of the high Hg pressure. Challenges include: the compositional uniformity through the layer, the surface morphology, the incorporation of dopants and the specifications for thickness, wavelength, etc. A common component leading to high structural quality is the use of lattice-matched substrates of CdZnTe. These are grown by a horizontal Bridgman process and can supply layers as large as 6 x 4 c m . The CdZnTe substrates must be of the highest quality and often this is a significant cost driver for the process. Two different technical approaches are used: growth from a Hg-rich solution and growth from a Te-rich solution. Advantages of the Hg-rich route include: excellent surface morphology, a low liquidus temperature, which makes cap layer growth more feasible, and the ease of incorporation of dopants. Also large melts can provide for very good compositional and thickness uniformity in large layers and give consistent growth characteristics over a long period of time. Growth from Te-rich solutions uses three techniques to wipe the melt onto the substrate: dipping, tipping and sliding. A sliding boat uses small melt volumes and is very flexible for changing composition, thickness and doping. Tipping and dipping can be scaled up easily and can provide thick, uniform layers but the large melts limit flexibility. Double layers are also more difficult to grow.

HgCdTe 21) arra!!s - - technolog!l and performance limits

28 5

Most manufacturers have taken their chosen growth system and tailored it to provide optimum material for their device technology. In particular, the use of dopants and the deliberate introduction of compositional grades is very specific to the device structure. A crucial figure of merit, however, is the dislocation count that controls the n u m b e r of defects in 2D arrays. Etch pit densities of 3 - 7 x 1 0 4 c m -2 are typically seen in the Te-rich sliding boat process, reproducing the substructure of the CdZnTe substrate. 22 The etch pits are associated with threading dislocations which appear to be normal or near normal to the layer surface. The substrate defect level can be as low as mid] 0 3 cm -2 in some horizontal Bridgman CdZnTe but this is not easy to reproduce. Similar defect densities are found in CdTeSe material but the impurity levels have proved difficult to control in the past. Device processing therefore must expect to cope with defect levels in the mid-1 ()4 cm-2 range for routine CdHgTe epilayers and this will set the ultimate limit on the n u m b e r of defects in HgCdTe 2D arrays. Nevertheless, LPE produces the best quality material at present and the challenge is to maintain the quality at reduced cost. The major obstacle to cost reduction is the substrate, and laboratories are turning towards a range of cheaper substrates including: CdZnTe or CdTe on GaAs/Si wafers. 2 More advanced device structures are mainly based on Hg-rich LPE. The Double-Layer Heterojunction, DLHJ. is a clever design using layers doped with slow diffusing impurities and compositional gradients to promote high q u a n t u m efficiency and low surface recombination. 24

Metal-Organic Vapour Phase Epitaxy, MOVPE

A summary of the state of the art for MOVPE technology has been produced by Irvine.2 s MOVPE growth depends on transporting the elements Cd and Te (and dopants In and As) at room temperature as volatile organometallics. They react, along with Hg vapour, in the hot gas stream above the substrate or catalytically on the substrate surface. The drive to lower temperatures and hence lower Hg equilibrium pressures has resulted in the adoption of the Te precursor diisopropyl telluride, which is used for growth in the 3 5()-4()() ~ C range. A key step in the success of this process is to separate the CdTe and HgTe growth so they can be independently optimised. This is called the IMP process for Interdiffused Multilayer Process. 26 IMP results in a stack of alternating CdTe and HgTe layers and relies on the fast interdiffusion coefficients in the pseudobinary to homogenise the structure at the growth temperature. Doping is straightforward using Group III metals (acceptors) and Group VII halogens (donors). For instance, ethyl iodide is used for iodine doping. There is a dual purpose in developing technologies such as MOVPE. Firstly, MOVPE is a tool for growing new device structures and secondly, it allows growth on low cost substrates. Certainly the elegance of this technique is that the composition, thicknesses and doping levels can be programmed, in principle to grow any required detector structure. For instance, multilayer, fully-doped heterostructures have been grown which demonstrate low thermal leakage current by a process called Auger exclusion. 2" Specialised device structures

286 Handbookof Infrared Detection Technolo#ies generally need material with good crystallinity and lattice matched substrates of CdZnTe are generally used. For 2D arrays, clusters of defects can be caused by macro defects in layers called hillocks, which are caused by preferred 111 growth, nucleated from a particle or polishing defect. Orientations 3-4 ~ off 1()() are used, primarily to reduce both the size and density of hillocks. Molecular Beam Epitaxy, MBE A summary of the state of the art for MBE technology has been prepared by Wu. 2~ MBE offers the lowest temperature growth under an ultra high vacuum environment, and in common with MOVPE in situ doping and control of the composition and interfacial profiles. These are essential for the growth of advanced and novel device structures. Typically, growth is carried out at 1 8 0 190~ on 211 CdZnTe substrates. Effusion cells of CdTe, Te and Hg are commonly used. Hg is incorporated in the film only by reacting with free Te thus the composition depends on the Te to CdTe flux ratio. The structural perfection depends strongly on the Hg to Te flux ratio and growth is usually restricted to a tight temperature range. Indium is the most widely used n-dopant and is well activated. P-type dopants are less conveniently incorporated in situ but manufacturers have devised a number of processes to force As on to the proper Te site. Again the Hg to Te ratio and growth temperature is crucial to achieve good activation. In general reproducibility seems more difficult to achieve than MOVPE. MBE structural problems centre mainly on pinholes or voids. Some very good EPD levels have been reported but in general the EPD levels are an order or more higher than the best quality LPE. Large area materials technology MBE and MOVPE are being used in the drive towards ever-larger focal plane arrays and much research effort has been focussed on the growth of HgCdTe directly on Si. The use of silicon as a substrate gives a perfect thermal expansion match to the silicon multiplexer and so provides a crucial enabling technology for very large arrays. Also silicon is mechanically strong, is available in large wafers at low cost and avoids the recurring problem of impurity out-diffusion in CdZnTe substrates. A 7 5 mm diameter silicon wafer can produce about twelve 64()x 512 arrays. The lattice constant mismatch can potentially provide high levels of misfit dislocations and this is an important figure of merit for the process. However. at shorter wavelengths, particularly 1- 3 l~m band. the effect of misfit dislocations is reduced enough to make these wafer scale processes viable, at the present time. Most of the MOVPE work on silicon has used a GaAs layer to buffer the 19% lattice mismatch between Si and HgCdTe. MBE growth on silicon substrates of four inches in diameter have been used to grow MW HgCdTe layers as described by Varesi. 29 The MBE work has used buffer layers of CdZnTe and CdTe to minimise the density of misfit dislocations. Another approach has been reviewed by Bostrup. ~ The process known as PACE uses a 7"3 mm diameter substrate of high quality sapphire which remains with the device throughout. A buffer layer of CdTe is deposited using an organometallic vapour phase epitaxy (OMVPE)

HgCdTe 2I) arrallS - - technology and l~erforntance limits

287

process to 8.5 Bm thickness followed by a 1 ()~tm thick HgCdTe layer grown by LPE. This process has been used to fabricate 1 ()24 • 1 ()24 arrays with 3.2 ~m cutoff 31 and 2048 • SW arrays of 16 cm ~ in area. 3_,

8.6.2 Junction forming techniques in homojunction arrays For n-p d e v i c e s , crystal growers obtain the d e s i r e d p-type level by controlling the density of a c c e p t o r - l i k e mercury vacancies within a c a r r i e r c o n c e n t r a t i o n range of, say, 1()~ ~-1017 cm-3. The photodiode junctions are c r e a t e d by locally typeconverting the material, by neutralising the Hg vacancies. It is not fully understood why HgCdTe adopts an n-type nature when Hg vacancies are eliminated but the effect has been exploited in a number of junction forming processes. Mercury can be introduced by thermal diffusion from a variety of sources but high temperature processes are not very compatible with HgCdTe at the device level. However. type conversion can be readily achieved by processes such as: ion beam milling as described by Blackman et al. 33 and ion implantation as described by Margalit et al.. 34 Kolodny and Kidron. 3s Bubulac et al. 3~'37 and Syz et al. 38 Kinch 39 also reports type-conversion using plasma-enhanced milling in the VIAP process and White 41~ has similar observations for H2/CH4 plasmas stressing the role of hydrogen in the process. The common feature is that the conversion depth is m u c h deeper than would be expected from the implantation range alone. The current knowledge on type conversion using ion beam milling is described by Baker 12 and the current knowledge for ion implantation is summarised by Bubulac and Viswanathan. 41 The explanation for the behaviour of HgCdTe under ion beam bombardment involves a n u m b e r of physical mechanisms. Firstly, the low binding energies, ionic bond nature and open lattice of HgCdTe make the liberation of free mercury at the surface and subsequent injection by the ion beam highly favourable. The injection mechanism probably involves a recoil implantation process. Once the Hg interstitial is injected, the mobility is apparently extremely high and there is some evidence that this is stimulated by the ion beam in a process related to the anomalously high diffusion rates of impurities often observed in SIMS analysis. Another factor is the movement of dislocations under the influence of the ion beam and possibly stimulated by strain fields. A n u m b e r of workers report that the n-type carrier concentration in the converted region is very low and in fact fast-diffusing impurities such as Group IB elements Cu. Ag. and Au. and Group IA elements Na, K. and El. which reside on the metal sub-lattice, are swept out of the n- region of the diode by the flux of Hg interstitials. This impurity sweep-out effect is serendipitous because it means that in ion-beam-generated junctions, the n-regions are very pure. free of S-R centres and weakly doped, creating an ideal structure for high performance detectors. The detailed atomic level processes taking place including the role of mercury interstitials, dislocations and ion bombardment in the junction forming process, are complex and not well understood in detail. Despite the complex physics involved, manufacturers have achieved good phenomenological control of the junction depth and n-dopant profiles with a variety of processes.

288 Handbook of Infrared Detection Techtlologies 8.6.3 Device structures

(a) Planar devices

The planar device structure illustrated in Figure 8.6 is the simplest device structure currently used. It is consistent with a number of junction forming processes, e.g. ion implantation, diffusion and ion milling. The matrix of junctions is mass connected to an underlying silicon multiplexer using indium bumps. The strength of the process is the simplicity and the compatibility with epitaxially grown materials. Tribulet et al. 42 describes a process based on high quality LPE material and ion implanted junctions, which is conducive to good quality detectors and volume production. In the simplest form, the process needs three masking stages, for the junction and pixel contact, and one for the contact to the p-side. The device is backside illuminated, i.e. it is illuminated through the substrate, and so this must be of high optical quality. Careful control of the junction geometry is needed to avoid crosstalk due to the diffusion of minority carriers into adjacent pixels. In general this type of structure has a poorer modulation transfer function (mtf) than other processes, especially in the case of small pixel sizes. The thermal mismatch between the HgCdTe/CdZnTe substrate combination and the silicon multiplexer is another important limitation in this device structure and can restrict the practical size of the array unless the substrate is thinned.

Figl~re 8.6 Schematic ofl~/anar indizml-blm~ped h!!brid.

HgCdTe 2I)arra!lS- technolog!l and performance limits

289

(b) Via-hole devices

Via-hole devices have been commercially produced since 1980 and are known as loophole devices 43 and VIAP devices 2~ By using a thin monolith of HgCdTe bonded rigidly to the silicon, the thermal expansion mismatch problem is overcome because the strain is taken up elastically. This makes the devices mechanically and electrically very robust. The loophole device is illustrated in Figure 8.7 following Baker et al. 22 This is a lateral collection device with a small central contact. The photosensitivity is by absorption in the n and p-regions and diffusion to the junction. The device structure inherently produces low crosstalk and so good mtf is achieved in small pixels. The technology permits very long monoliths to be used (over 16 mm length for 512 x 2 long linear arrays) without thermo-mechanical problems and very small pixels (down to 15 ~tm square). It has front surface illumination, and the contact obscuration is removed by only metallising one side of the via-hole. It does not have the substrate transmission issues associated with backside-illuminated structures. The process has two simple masking stages. The first defines a photoresist film with a matrix of holes of, say, 5 13m diameter. Using ion beam milling, the HgCdTe is eroded away in the holes until the contact pads are exposed. The holes are then backfilled with a conductor, to form the bridge between the walls of the hole and the underlying metal pad. The junction is formed around the hole during the ion beam milling process as described in Section 8.6.2. The second masking stage enables the p-side contact to be applied around the array. Figure 8.8 shows a photomicrograph of one corner of the hybrid ilustrating the membrane-like nature of the HgCdTe on the silicon. The HDVIAP process 39 results in a similar structure but uses a plasma etching stage to cut the via-hole and an ion implantation stage to create a stable junction and damage region near the contact. In order to achieve higher lifetimes and lower thermal currents, Cu is introduced at the LPE growth stage. This is swept out during the diode formation and resides selectively in the p-region, partially neutralising the S-1R centres associated with Hg vacancies and achieving dark currents approaching that of fully doped heterostructures.

Figure 8.7 Schematic t~floophoh' CdHgTe-silicon h!tbrid.

290 Handbook of Infrared Detection Technologies

Figure 8.8 Photomicrogragh of corn('r of 384 x 288 loolflloh' arra!t (()SPRtfY). Court('s!! of BAE SYSTt~MS Infrared Ltd. (a) O S P R E Y - - 384 x 28g. HgCdTe on C\IBS h!lbrid with 20 lxm pixels. (b) SEM detail of one corner of HgCdTe monolith.

In via-hole processes the cylindrical shape of the junction minimises the intersection with threading dislocations {see Section 8.6.1). This has been demonstrated by the difference in yields when the HDVIP device is extended by a planar junction. 39 The device is front side illuminated leading to near theoretical MTF performance.

(c) Double layer heterojunction (DLHJ)

Thermal leakage currents in HgCdTe devices tend to arise from the p-type side and most advanced technologies strive to improve the operating temperature by using the n-type side as the absorber. N-type HgCdTe is easier to control at low carrier concentrations and is relatively free of Shottky-Read centres that limit the lifetime in p-type material. The p-type side then is minimised in volume or uses a wider bandgap to reduce the thermal generation. There are a number of elegant device structures that have been reported to produce RoA values an order higher than that with homojunctions. The back illuminated mesa P-n heterojunction is the most widely used device. Figure 8.9 shows a schematic of the device. For heterostructures that are grown in s i t u , it is necessary to completely isolate the junction forming a so-called mesa device. This is usually performed using a chemical etch because of the electrical side effects of using dry processing techniques. This structure makes good use of the n-layer to give high fill factor. The absorber layer is doped with indium to around l x 1015 cm-3 or less and is grown either by the horizontal slider LPE process from a Te-rich solution 44-46 or by a vertical dipper LPE process from a Hg rich solution. 47 The p-type layer is doped with arsenic around 1-4 x 1017 c m - 3 and is grown by the vertical dipper LPE from a Hg rich solution. Arias et al. 4~ have

HgCdTe 21) arra!!s - - teclmolog!! and l~et:formance limits

291

Figzlre 8.9 Schematic of l)otibh, La!ler Heterojzlnction ( I)LHJ ).

reported a wide gap P+-n- technology grown by an MBE process. The alternative structure of n--on-wide gap p+ is reported by Tung. 4~ This device results in very low crosstalk because of the complete electrical isolation. Mesa heterostructures are effective for reducing thermal leakage currents in small photodiodes. They have, however, a technologically difficult passivation stage on the sidewalls of the mesa, which can lead to reverse bias leakage and uniformity problems if not optimum. Arias et al. ~ have demonstrated a clever diffused heterostructure, which overcomes the sidewall problem called double layer planar heterostructure (DLPH). The back-illuminated arsenic-implanted pn-N planar buried junction heterostructure was the first photodiode structure to exploit MBE. 51 The junction is formed by arsenic implantation into a three layer N-n-N film grown in sitll by MBE onto a CdZnTe substrate (see Figure 8.1 ()). The unique feature of this structure is that the junction is buried beneath the top wide gap n-type layer. The junction intercepts the surface in wider gap CMT thereby reducing the generation rates for any surface defects that may be present. The arsenic doped region must extend deep enough into the narrow gap absorber layer in order to collect photocarriers. This necessitates that the wide gap layer be thin on the order of O. 5 pm and that the compositional interdiffusion between the wide gap and absorber be minimal. Growth by MBE at low temperature (175 ~ satisfies these requirements. Another feature of this structure is the wide-gap n-type buffer layer between the absorber and the substrate. This buffer layer keeps carriers that are photogenerated in the absorber away from the filmsubstrate interface where there can be interface recombination. The DLPH process has been used to produce arrays of up to 1 0 2 4 • 1024 on 18.5 ~m pitch (HAWAII multiplexers) with high q u a n t u m efficiency, s2

292

Handbook of Infrared Detection Technologies

Figure 8.10 Schematic o fI)oHl~le La!ler Planar Heterostrz~ctzlre ( I)LPH ).

8.7 Measurements and figures of merit for 2D arrays Infrared detectors are assessed against a number of performance parameters. These can include, for instance: sensitivity, response time, noise spectrum, defect levels or manufacturing cost, and this data can be used to compare detectors from different technologies. In manufacturing the main need is to assess yield and compliance with the specification. This is because most infrared detectors are based on technologically challenging materials and can experience non-ideal behaviour, such as: excess noise, non-uniformity, pixel defects and other electrooptic effects that can mar imaging performance. Detector specifications are drawn up around certain adopted parameters aimed to meet the imaging requirements. This section describes the commonly used parameters and their measurement. For m a n y infrared components the accepted figure of merit is the detectivity or D* which is the signal-to-noise normalised for pixel area and noise bandwidth. D* is a rate dependent variable which applies well to non-multiplexed detectors and is the most useful parameter for comparing different technologies. When a silicon readout is used, the signal is usually sampled for a period called the integration time, which is less than the frame time. D* alone cannot be used to predict the detector performance in the system because there is now a duty cycle and a different bandwidth. A better figure of merit is Noise Equivalent Temperature Difference (NETD). This is not normalised to area or bandwidth and so it cannot easily be used to compare different technologies. However. it provides a direct measure of the sensitivity of a multiplexed detector and is an

H~ICdTe 2I) arrays -- technolog!! and performance limits

29 3

excellent figure of merit for specification purposes. This section describes how NETD is measured and the relationship between D* and N E T D . 8.7.1 NETD-theoretical calculation

For thermal imagers, it is necessary to measure small changes in temperature rather than infrared power. The thermal contrast (c) is defined as the fractional change in detector output per change in scene temperature 1 (aldet(Ts,,,,,,,.J)lo)) c - I,t,,t(T,,,,,,,,,,fno) -~~,i

(4)

where Idet is the detector output produced by absorption of infrared from the scene and depends on the scene temperature and the lno of the optics used. For the ideal spectral response and a scene temperature of 295K. the contrast is calculated to be 4.3% K-1 for a cut-off of 4.2 ~tm and this reduces to 1.9% K-l at 1 2 g m . It is instructive to calculate the NETD in terms of the number of electrons stored in the readout circuit tt,~t,,,.,,a. In a sampled imaging system, the output current from each detector is integrated on a storage capacitor in the readout circuit for a time r~,,,.,,. Some of the electrons stored will have been generated by photons arriving from the scene tlvj,,,t,,. The rest will have been generated by u n w a n t e d noise processes such as dark current from the detector, equation (3) can be used to calculate the mean change in the number stored by changing the scene temperature. dllphoto aT = Cnpho,,

(5)

Because the detector current is generated by random processes (e.g. random arrival of photons) there will be a variance in the number of electrons stored during a fixed stare time. For white noise, the variance in the number of electrons stored is v/nsto,.,,,t and hence the NETD can be deduced from equation ( 3 ). N E T D - n,,oi,,.,, k, d n /

ctlp~,,,t,, - Cv/n~t,,,.,,,t \ tlph,,t,,/

This gives a simple equation for the minimum background limited NETD when all the stored charges are generated by the scene. In addition it shows that this NETD does not depend on the q u a n t u m efficiency but only on the number of stored electrons (the q u a n t u m efficiency determines how long it takes to integrate the required number of electrons). For a silicon readout circuit, the m a x i m u m storage capacity is about 3 x 1 ()" electrons/pixel and this limits the NETD per frame to about 5 mK in the MWIR band and 1() mK in the LWIR band irrespective of the detector performance or temperature of operation.

294 Handbook of Infrared Detection Technologies 8.7.2 NETD-experimental measurement

The N E T D is a figure of merit, which is commonly applied to multiplexed arrays of detectors where the output is sampled. The N ETD of the detector can be used to predict the system performance simply by allowing for the additional system effects such as optics and signal processing. The N ETD for each pixel in the array is simply the noise divided by the thermal sensitivity (defined below) of the detector. Experimentally, the NETD can be calculated as follows. For every diode in the array, the thermal sensitivity. Vs. is derived from the difference between the mean values of the output voltage of the device viewing the ambient plate of the blackbody at temperature T and the corresponding voltage when the blackbody temperature is set Td warmer (typically 1 () K), i.e.

J( T,

) -

i.j

v/,./,,(r,

+ r,,.

(v K-')

(7)

where V,,,,t i'j is the measured output voltage for the pixel at co-ordinate (i.j). This will be a function of the blackbody temperature T and the stare time r,~t,,,.,,. At each temperature, Ns samples are acquired for each pixel in the array (where Ns is typically 100). The equivalent root-mean-square (rms)noise voltage (V,,,,i,,,) for each of the diodes is computed as the standard deviation of the Ns samples. ,.j V,,,,i.~e (T, r~t,,,.,,) --

~/~.~

~.j

(v,,,,,(r.

k= 1

_

99 (C/,,(T.

2

Ns - 1

The equivalent noise bandwidth for this derived noise voltage is set by the stare time, rst,,,.,,. and is given by l__k___.The noise equivalent temperature difference (NETD) for all the elements is computed from: -- T,,tar,,

NET17J(T,

r~t,~,.,,) -

VI/'i~"(T" r)

(9)

8.7.3 Relationship of NETD with other figures of merit

The responsivity and noise discussed above, determine the minimum infrared power that can be can detected above the noise level. The noise equivalent power or NEP can be calculated using. NEP-

Noise - -

R

(1())

where R is the responsivity in whatever output units are appropriate (e.g. V or A) per unit infrared power of signal being detected and N o i s e is the noise observed

HgCdTe 21) arrallS -- techtlolog1! and perfornlance limits 29 5 under the m e a s u r e m e n t conditions (bandwidth and f/no of cold shield). For m a n y detectors, the NEP is proportional to the square-root of both the detector area and the bandwidth so that a figure of merit known as the detectivity or D* can be used.

D* ---- vIA B W (cmx/-Hzz W-I'~ \ / NEP

(11)

where A is the detector area and B W is the bandwidth. Some caution should be exercised in using equation {1 1) to extrapolate the performance of detectors to widely different areas and bandwidths or to compare detectors measured under different conditions. 1/f noise will reduce the detectivity at low frequency. Similarly. noise generated at the perimeter of a device will reduce the detectivity for small area devices. From equation (4). the signal produced by a change in the scene temperature is

(12)

O~ldet = Cldot(TNcet,,.,fllo)aTs,.,,,,,.

By normalising the signal-to-noise ratio with respect to the detector area and the noise bandwidth, a figure of merit can be defined which compares their ability to detect changes in the scene temperature.

Idet(Tsc,,,,e, fno) ~/~__~~ M* - c , ~ : i ~ : [ - ~ ) --(x/Hzz

cm-' K-')

(13)

Because M* is a function of the fllO it is normally quoted for a 2x field of view. M* can be used, for example, to compare the potential system performance of detectors with different cut-offs. A frequently required relation is between N ETD and D* for a sampled detector. This may be required for example, to estimate the potential performance of a 2D array from measurements on a detector or to determine if in a multiplexed array the detector performance has been degraded by the multiplexer. In equation (6) the number of photo-generated electrons and the noise can be related to detector properties.

n,,h,,to --

n,,,,~,, =

rl

r

T,.,.,,,,,,)} r,,t,,,.,,A

-4-F~

i,lois,, Bx/~7 q

i,l,,i.~,, 1 q v/2 r',t,,,,'

(14)

(15)

Where the term in brackets in equation {14) is the detected photon flux under the array operating conditions. Combining equations {1 ()}. ( 1 1 ), {14) and ( 1 5 ) gives the required relationship.

296 Handbook of Infrared Detection Technoh~lies

D* = Ri 4fno 2 rl c~(;.,., T)NETDv/2r~.,,.,.A

(16)

8.8 HgCdTe 2D arrays for the 3-513m (MW) band In the MW band, the development of array technology has reached a stage when all the reported device processes will result in essentially background-limited performance for a wide range of systems. Differences will appear in the operating temperature and in imaging artefacts, such as. uniformity, crosstalk and blooming. Standard vacancy-doped HgCdTe diodes, such as planar or via-hole homojunctions, will be BLIP in F2 tlux levels at around 145 K. The standard HDVIP TM, which is extrinsically doped with many fewer vacancies, is BLIP to ,-~155 K. HgCdTe diodes fabricated on material with no vacancies will be BLIP to ~,175 K. In general, LPE on CdZnTe based processes will give low numbers of defects, set primarily by threading dislocations in the substrate material, and negligible 1/f noise. In material grown on GaAs or silicon the parameters that degrade are the numbers of defects and general 1/f noise, due to high densities of misfit dislocations. Nevertheless, even for these technologies, the performance of 2D arrays can be close to BLIP and the NETD defined by equation (6). Furthermore, the number of defects declines slowly with temperature so extra cooling can improve array defect levels. Following equation (6), the NETD depends on the number of integrated photoelectrons so, provided the integration time can be increased, the q u a n t u m efficiency is not necessarily an important parameter. Issues relating to imaging quality include the modulation transfer function (mtf). Fully defined MOVPE or MBE diode structures will have the best mtf performance but via-hole technologies (loophole and HDVIP ~;) closely approach them because of their concentric junction shape and this permits true resolution improvement in small pixels. DLHJ structures and planar structures rely on lateral collection to give very high fill-factor and high q u a n t u m efficiencies, but they have more blurred pixel edges. Blooming due to an intense optical highlight follows a similar argument.

8.9 HgCdTe 2D arrays for the 8 - 12 ~tm (LW) band 8.9.1 Array design issues

LW HgCdTe detectors were originally developed for long linear arrays and the processes can be extended to 2D arrays with one additional consideration. The earth plane of the array must carry the sum of the photocurrent from all the

HgCdTe 2I) arra!/s - - technolog!l attd performance limits

29 7

pixels in the array and if there is insufficient conductance, a voltage will appear between the centre and edge of the array. This is called 'substrate debiasing' and in the extreme case will inhibit injection in the centre diodes. In any event it provides a potential long-range crosstalk m e c h a n i s m and detector designers will try to minimise the resistance of the earth plane layer. In homojunction arrays the p-type common layer cannot be doped high enough and a special metal grid is needed as is shown in the DRS array in Figure 8.1 1. In other technologies, the resistance of the common layer must be minimised by using high doping and m a n y manufacturers prefer to use N-type common layers to achieve this. 8.9.2 Introduction to performance limitations in LW arrays

Figure 8.12 illustrates a typical current-voltage relation for a LW HgCdTe diode. The narrow bandgap nature of the device makes it difficult to achieve ideal diode behaviour and Figure 8.12 shows the main problems that need to be addressed. Ideally the device engineer wants very high dynamic resistance to allow good injection efficiency into the silicon buffer circuit and no excess leakage currents or noise. In practice, thermally generated leakage must be m u c h lower than the photocurrent to prevent excess white noise and this is usually ensured by cooling adequately, but processes vary in the degree of cooling necessary. Finally, the excess low frequency noise from a variety of potential sources must be minimised. In LW arrays it is usually necessary to operate HgCdTe diodes in reverse-bias to take advantage of a higher dynamic resistance. A high dynamic resistance ensures a good injection efficiency into the silicon multiplexer and also provides i m m u n i t y to long-range electrical crosstalk. In reverse-bias there

Figure 8,11 L W HDVIP |

Technologies Ltd.

array with metal grid for substrate debiasing control. Courtesy of DRS

298

Handbook of Infrared Detection Technologies Excess current due to high-bias leakage mechanisms

Slope due to shunt resistance mechanisms 20" 15A

c

10

Thermal current

C

,-

5

Photocurrent i

U 0

.s

0 h.J v!

Normal operating region

-5 -10

.

-20

0

20

.

.

40

.

60

.

80

100

1 .~0

Diode bias voltage (mV) Figure 8.12 T!lpical L W HgCdTe current-voltage characteristic.

are leakage currents that limit the maximum dynamic resistance that can be obtained in practice. The control of these leakage currents is therefore of crucial interest to the device engineer. In addition, for many applications, the sensor is effectively D.C. coupled to the infrared scene, especially in staring applications, and the importance of excess low frequency noise is emphasised. Although this is generally much lower in diodes than photoconductors, the use of reverse bias for enhanced injection efficiency introduces the possibility of 1/f noise associated with leakage currents and the modulation of the sensitive area by traps in channels. 8.9.3 Cause of defective elements in HgCdTe 2D arrays

Most manufacturers have developed processes that minimise general 1/f noise with appropriate choice of passivation technology and junction processes. The quality of the sensor then becomes limited by defects in the array. Defective diodes, when they arise, are usually characterised by low reverse-bias resistance and high 1/f noise. One of the problems of analysing the literature is that the actual device structure is critical in determining the importance of various leakage mechanisms. The scatter within databases is usually large and the noise depends strongly on the technology used. Many workers, such as: Szilagyi and Grimbergen, s~ Syllaios and Colombo, s4 Pelliciari and Baret, ss Johnson et al. s6 and Norton and Erwin, s7 have found that the reverse bias characteristics of HgCdTe diodes depend strongly on the density of dislocations intercepting the junction. This is an important observation for HgCdTe device technology, because of the ease with which dislocations can be

HgCdTe 21) arraJIs - - technolog!l and l~erformance limits

299

introduced, due to materials growth and strain in the device fabrication process. In HgCdTe processes almost any artefact that causes strain or physical damage will manifest itself as a defect. Johnson et al. s~, showed that dislocations can increase the g-r current linearly in p-on-n heterostructures, along with a corresponding increase in 1/f noise current density. Tunnel currents are also strongly affected, particularly at low temperatures. Diffusion current and quantum efficiency are only weakly affected. The electrically active nature of dislocations is well reported, for instance Tregilgas ss and Hirth and Ehrenreich. s The nature of defects in HgCdTe LW arrays has been studied in detail by Baker 12 in homojunction via-hole arrays (loophole arrays) made using high quality LPE material. A key observation is that diodes with etch pits on the junction have enhanced ]/f noise and degraded reverse-bias characteristics. However, dislocations had no effect on the diode performance unless they physically intercepted the p-n junction. Glancing incidence has no apparent effect, indicating that dislocations do not act as significant centres for higher thermal generation or alter impurity distributions in the nearby region by gettering. Layers of LPE can have threading dislocations that originate from the CdZnTe substrate and rise vertically through the layer. In arrays, interception of threading dislocations with the junction can cause strong leakage in reverse-bias, but this is a variable effect, possibly because these dislocations can become randomly decorated during growth. Consequently, the threading dislocation density in the substrate is very important for controlling defect levels in the array. A model has been proposed based on two observations: firstly the 1/f noise is dependent on both the thermal diffusion current and the photocurrent from the p-type side of the junction (supporting the work of Williams 6()) and secondly the I / V characteristics of noisy diodes show a fiat or slightly decreasing dynamic resistance with increasing reverse bias. The model proposes that dislocations cause a microscopic extension of the n-region into the p-region causing the collecting volume for photocurrent and thermal current on the p-side to be controlled by the extension. Any fluctuation in trap occupancy along the dislocation will cause the depletion region extension to fluctuate and apply a noise modulation to the current from the p-side. The link between dislocation trapping and ]/f spectra is well established, s3-s7 The model explains the dependence of noise on diffusion and photocurrent from the p-side. The variability in the database would result from the range of possible interception angles of the dislocation and the junction. The role of crystal imperfections in LW arrays is established, but the sensitivity may vary between device structures. More data is needed for the VPE processes because the nature of the misfit dislocations and the geometry of the absorber will influence the magnitude of the excess noise.

8.10 HgCdTe 2D arrays for the 1-3 ~m (SW) band Important applications in the SW (1-3 Bm) waveband include: thermal imaging (using nightglow), spectroscopy and active imaging using lasers. Competing

300 Handbook of Infrared Detection Technologies detectors are available but products such as InGaAs tend to have high noise beyond 1.7 ~m due to defects from lattice mismatch with the InP substrate. Most HgCdTe technologies can be extended to SW with little change to the processing. Carrier lifetimes would be expected to be dominated by radiative recombination but in fact S-R centres and technology-related limits probably apply in practice. SW is very m u c h more forgiving of crystal defects and so is amenable to detector technologies based on low-cost substrates. For imaging and spectroscopy, typical operating temperatures are around 2()() K and thermo-electric coolers are often offered as standard products. Some very large arrays have been produced in the SW band including the 2048• array 32 produced by Rockwell/Boeing for an astronomy application, shown in Figure 8.13. The arrays are 37 • 37 m m in size and are backside illuminated through a sapphire substrate. The process involves growing an MOVPE CdTe buffer layer onto highly polished sapphire and then growing p-type LPE onto this. The n-p diodes are formed by ion implantation. A growing area is active imaging using eyesafe 1.54 btm lasers. The shorter wavelength results in higher resolution for a given optical aperture size and is particularly beneficial in long-range sensor systems. The very short pulse nature of the laser can be used to give range information and is often used with gated detectors to produce imaging in a certain range zone (so-called laser shape profiling). When range information is produced it is called laser range profiling

Figure 8.13 2048 x 2048 SW arra!l for astronom!l. Courtes!t of Rockwell~Boeing.

HgCdTe 2I) arra!ls - - technolog!t and l~erJormance limits

301

and, in the limit, 3D imaging. At present the field is in its early development stage and workers are using relatively small arrays to perform initial research. Detectors for this field need to be very fast and use a wide. weakly doped absorber region to give low capacitance. This is a so called PiN diode where the 'i' stands for intrinsic. It also needs a very low series resistance to withstand the shock of the laser return and this lends itself to metal grids and heavily doped heterostructures. In most SW applications, the photon flux is low and there is considerable interest in providing some avalanche gain in the device. For low noise avalanching it is desirable to have the electron and hole ionisation rates very different and this occurs naturally in HgCdTe. So avalanche gain can readily be achieved for wavelengths above about 3 btm when the absorber region is p-type and electrons are the minority carrier. Typically a gain of 1 () is observed for around 6 V at a cut-off of 5 btm. However. there are technological problems in sustaining this voltage across narrow bandgap junctions. An alternative structure uses SW material (1.6 btm) and a resonant e n h a n c e m e n t of the hole impact ionisation rate when the bandgap equals the spin-orbit split-off energy. ~'1 Gains of 3 0 - 4 0 have been seen with voltages of 85-9() V. SW detectors and active imaging will see rapid growth in the next few years as sophisticated ROICs become available to decode range and intensity information.

8.11 Towards GEN III detectors The definition of a GEN III detector can differ between different nations but the general guideline is any detector that offers an imaging advantage over conventional first and second-generation systems. Common agreed examples are, for instance, megapixel arrays, dual colour or even multispectral arrays, higher operating temperature, fast readout rates, very low NETD due to pixel level signal compression and retina-level signal processing. Section 8.4.3 has described some of the silicon multiplexer developments that are crucial to GEN III detectors. Some of the other research areas are described here.

8.11.1 Two-colour array technology Two-colour arrays are used to discriminate objects of similar photon flux levels in a thermal image by resolving the spectral signature. A common example is to separate a sun glint from a hot thermally emitting object. The field is still in its early development stages but some good results have been achieved. In general, signal processing algorithms need simultaneous integration as a first priority. Two-colour arrays use two layers of HgCdTe with the longer wavelength layer underneath. The technological challenge is to make contact to the two layers without obscuring the sensitive area. Ideal devices, where the integration takes place simultaneously and the sensitive area is co-spatial, are practically difficult to make on normal pixel sizes, say less than 5()~m. Most dual-colour devices

302 Handbookof Infrared Detection Technologies using indium bumps need two contacts within each pixel and make contact to the underlying layer using an etching stage. The silicon needs to be custom designed because the flux levels in the two bands may be markedly different. The polarity of the input MOSFET and the gain within the silicon must be matched to the technology and application. Several successful arrays have been reported. Reine r reports a 6 4 • simultaneous MW/LW dual-band HgCdTe array on 75 ~m pitch. The array is fabricated from a four-layer P-n-N-P film grown ill situ by MOVPE on CdZnTe and mounted on a custom ROIC. This approach has been applied to the double layer planar heterostructure, DLPH, in a process called SUMIT for Simultaneous Unipolar Multispectral Integrated Technology. r The DLPH s t r u c t u r e can be turned into a two-colour device by having two absorbers of different bandgap separated by a wide bandgap barrier. An etch step is used to expose the lower, shorter wavelength absorber and form an arsenic doped junction. A pixel size of 40 ~m for each colour is used in 128 x 128 demonstrators for MW/MW arrays. A variant of this basic process has also been presented by AIM ~4 and they have achieved two-colours in a concentric arrangement in 1 9 2 x 192 and 2 5 6 • arrays with 56 Bm pitch, again using a full custom ROIC. Kinch 39 reports a via-hole type s t r u c t u r e in which each pixel has two viaholes. One is for connecting the longer wavelength array to the silicon in the normal way and the other is isolated from the longer w a v e l e n g t h material and connects the shorter wavelength to the silicon as is illustrated in the Figure 8.14. The spectral response of the two colours is shown in Figure 8.15 for t w o different doublets. The multiplexer is a standard 64()• 512 readout and each two-colour pixel occupies two silicon pixels i.e. 3()• 6() ~tm. The shrink potential of dual-colour technologies is limited by the need for two contacts per pixel. An a l t e r n a t i v e approach is to use a so-called bias-selectable detector. ~s'~ which needs only one bump and is compatible with small pixel sizes. The device uses a p-n-N-P sandwich structure with each p-n junction in different composition material. Provided the transistor action is suppressed by appropriate barriers, the wavelength can be selected by the bias polarity. The silicon circuit requires two input MOSFETS, a PMOS and NMOS, which are

Figure 8.14 MW/L W HI)VIP t:PA two-Colo.r Composite. Courtes!! of I)RS Technologies Lut.

HgCdTe 2I) arra!!s - - techtlolot,l!l and performance limits

3() 3

Figure 8.15 Spectral response Qf MW/MII' two-colonr HI)I'IP HgCdTe t:PAs. ('ollrtes!l o fl)RS Technologies.

switched when the bias is switched. This is an elegant technique but has the operational disadvantage of non-simultaneous integration. 8.11.2 H i g h e r o p e r a t i n g t e m p e r a t u r e (HOT) d e v i c e s t r u c t u r e s

The general principles for designing HOT HgCdTe detectors have been proposed over the past two decades, primarily as a result of research in the UK. Elliott ~7 has produced a good s u m m a r y of this work. The advantages afforded by HOT detectors are in the area of cost of ownership and portability. LW arrays operating at only a few tens of degrees higher, can significantly improve the mean-time-before-failure (mtbf) of a Stifling engine cooler. In the MW band, BLIP detectors on thermoelectric coolers would offer a much cheaper, smaller and more robust product, ideally suited to continuously operating security-type applications. The possibility of operating detectors at temperatures near room temperature was proposed by Elliott and Ashley. ~ The basic structure is a P-p-N illustrated in Figure 8.16. The active volume of the absorber is small relative to a minority carrier diffusion length and it is operated in strong non-equilibrium by reverse biasing the minority carrier contact to completely extract all of the intrinsically generated minority carriers. To preserve space-charge neutrality the majority carrier concentration drops to the background dopant concentration and Auger generation is effectively suppressed. Remaining components are associated with S-R centres and possibly injection from the contact and surface regions. Realisation of HOT LW arrays will depend on the suppression of high 1/f noise associated with the reverse bias operation and in MW arrays the suppression of competing recombination mechanisms. Nevertheless the device concept has enormous importance as a route towards high performance infrared detectors with minimal cooling.

304 Handbook of Infrared Detection Technolo~lies

Figlire 8.16 Schematic ~[H()T device structure.

8.'1"1.3 Retina level processing

In very large arrays operating at high frame rates, the downloading of signal data and the subsequent signal processing can be daunting and is called the data processing bottleneck problem. The human retina presents an example of how evolution has dealt with the problem. The eye performs a number of image processing operations in the 'z-plane' including in order: logarithmic photon sensing, spatial filtering, temporal filtering, motion sensitivity and data decomposition. An essential function is the Difference of Gaussian or DoG filter which is used for both edge and contrast enhancement. ~ Focal plane arrays with neuromorphic processing are under development at a number of centres (for instance, references 7() and 71 ). A good example is the MIRIADS programme (Miniature InfraRed Imaging Applications Development System). MIRIADS uses a neuromorphic FPA. with temporal high pass filtering, frame co-adding and a Difference of Gaussians operation to detect motion, enhance edges and reject ambient light levels. The current reported array is a 64x64. More information is presented by Baxter et al. "2

8.12 Conclusion and future trends Photovoltaic 2D HgCdTe detectors are serving many applications worldwide. Such detectors have supplanted photoconductive detectors in second-generation systems for military, commercial and scientific applications, offering improved temperature sensitivity, lower power consumption, weight and volume. The continuing development work within manufacturing centres, supported by the wider scientific community, is aimed at increasing the radiometric performance, reducing defects and reducing manufacturing costs. Medium term research is being directed at the best advanced materials and device structures for the next generation of HgCdTe arrays, which will combine state-of-the-art background

HgCdTe 2I) arraJIS - - technolo~l!l and pet:fi)rmance limits

~() 5

limited performance with low cost wafer scale processing. In future it is envisaged that HgCdTe detectors will be grown directly on silicon or even silicon multiplexers for very low cost detectors. Band-gap engineering will produce heterostructure detectors with much higher operating temperatures. This will enable background-limited operation in MW arrays at near room temperature and operation of LW detectors on thermoelectric coolers or low power Stirling engines. Device structures will be extended to produce bi-spectral and multispectral capability. The ultimate performance potential of HgCdTe will ensure that it is the material of choice for all high performance infrared systems. In MW arrays it will replace InSb due to the better imaging characteristics and higher operating temperature. Active imaging will drive a major growth in SW array technology. In medium performance LW imagers competition from uncooled detectors will create the motivation to drive costs down and it is to be expected that applications will widen, for instance, to include detectors for non-defence applications such as: medicine, driver and pilot aids and low cost imaging sensors for the security industry.

Acknowledgements The author wishes to express his gratitude to Mike Kinch of DRS Technologies, Kadri Vural and Jose Arias of Rockwell/Boeing and Marion Reine and coworkers at BAE SYSTEMS, Lexington, for supplying material for this chapter and valuable advice. Also the advice and support from my technical colleagues, particularly: Peter Capper, Chris Maxey. Chris Jones, Peter Knowles and Les Hipwood and my management here at BAE SYSTEMS Infrared Ltd, particularly Graham Hall. Thanks also to my wife, Lesley, for help with the English.

References 1. A. C. Goldberg, S. W. Kennerly, J. W. Little and H. K. Pollehn, Comparison of HgCdTe and QWIP Dual-band FPAs, Proc. SPIE 4 3 6 9 , "332 (2()() 1 ). 2. L. J. Kozlowski, Proc. SPIE 2 7 4 5 . 2 (1996). 3. L. J. Kozlowski, J. Montroy, K.Vural and W. E. Kleinhans, Proc. SPIE 3 4 3 6 , 162(1998). 4. L. J. Kozlowski and J. Luo, Proc. SPIE 3 3 6 0 (April 1998 ). 5. M. B. Reine, Chapter 12, Infrared Detectors and Emitters: Materials and Devices, Electronic Materials 8, published by Kluwer Academic (and references contained therein). 6. M. B. Reine, A. K. Sood, T. J. Tredwell et al. in Selniconductors atld Setnimetals, Eds. R. K.Willardson and A. C. Beer(Academic, New York), 18, Chap. 6 (1981 ). 7. D. E. Lacklison, P. Capper. et al.. Se~icot~d. Sci. Technol. (UK) 2, 33 ( 198 7). 8. P. L. Polla, R. L. Aggarwal, D. A. Nelson. J. F. Shanley and M. B. Reine, Appl. Phys. Lett. (USA) 4 ] , 941 ( 198 3 ).

306 Handbookof Infrared Detection Technologies 9. W. A. Radford, R. E. Kvaas and S. M. Johnson. Proc. IRIS Special Group in Infrared Materials, Menlo Park, USA ( 1986 ). 10. W. W. Anderson, Infrared Ph!ts. (UK) 20, 353 ( 198()). 11. C. T. Elliott, N. T. Gordon, R. S. Hall and G. J. Crimes, 1. Vac. Sci. Technol. A (USA) 8 (1990). 12. I. M. Baker and C. D. Maxey, 1. Elec. Mat., 30(6), 682 (2()()1). 13. I. M. Baker, G. J. Crimes. C. K. Ard, M. D. Jenner, J. E. Parsons, 1R. A. Ballingall and C. T. Elliott, 4th Int. Conf. on Advanced Infrared Detectors and Systems, lEE Conf. Pub. 3 2 1 , 78 (199()) 14. www.aim-ir.com 15. www.infrared-detectors.com 16. www.sofradir.com 17. www.iews.na.baesystems.com 18. www.drs.com 19. T. ]. de Lyon, ]. E. ]enson, M. D. Gordwitz et al., I. Electron. Mater. 28 705 (1999). 20. www.rsc.rockwell.com 21. P. Capper, T. Tung, L. Colombo, Part 1, Chapter 2, Narrow-gap II-IV Compounds for Optoelectronic atld Electro~nagnetic Applications, published by Chapman and Hall. 22. I. M. Baker, G. J. Crimes, ]. E. Parsons and E. S. O'Keefe, SPIE 2 2 6 9 , p636 (1994). 23. S. M. Johnson, J. A. Vigil, J. B. James et al., ], Electron. Mater. 22, 835 (1993). 24. K. J. Riley and A. H. Lockwood, Proc. SPIE 2 1 7 , 2 0 6 (1980). 25. S. J. C. Irvine, Part 1, Chapter 3, Narrow-gap II-IV Conlpounds for Optoelectronic and Electromagnetic Applications, published by Chapman and Hall. 26. J. Tunnicliffe, J. Irvine, S. Dosser and J. Mullin. ]. Cr!lst. Growth 68, 245 (1984). 2 7. C. D. Maxey, C. J. Jones, N. Metcalf et al., Proc. SPIE 3 1 2 2 , 453 (1996). 28. O. K. Wu, T. J. de Lyon, R. D. Rajavel and J. E. Jensen, Part 1, Chapter 4, Narrow-gap II-IV Compounds for Optoelectronic and Electromagnetic Applications, published by Chapman and Hall. 29. ]. B. Varesi, R. E. Bornfreund, A. C. Childs el al., ]. Electron. Mater. ] 0 ( 6 ) , 566(2001). 30. G. Bostrup, K. L. Hess, J. Ellsworth, D. Cooper and 1R. Haines, 1. Electron. Mater., 30(6), 560 (2001). 31. L. J. Kozlowski, K. Vural and D. E. Cooper, et al., Proc. SPIE 2 8 1 7 , 150 (1996). 32. K. Vural, L. J. Kozlowski and D. E. Cooper, et al., Proc. SPIE 3 6 9 8 , 24 (1999). 33. M. V. Blackman et al., Elec. Letts. (UK) 23, 9 78 ( 198 7). 34. S. Margalit, Y. Nemirovsky and I. Rotstein. ]. Appl. Phys. (USA) 50, 6386 (1979). 35. A. Kolodny and I. Kidron, IEEE Trans. Elec. Dev. (USA) ED-27, 3 7 (1980).

HgCdTe 2I) arra!ls-- technolog!l and performance limits

3()7

36. L. O. Bubulac, W. E. Tennant, R. A. Riedel and T. J. Magee, I. Vac. Sci. Technol. (USA) 2 1 , 2 5 1 (1982). 37. L. O. Bubulac and W. E. T e n n a n t , Appl. Ph!ls. Lett. (USA) 51, 35:3 (1987). 38. J. Syz, J. D. Beck, T. W. Orient and H. F. Schaake, J. Vac. Sci. Technol. (USA) AT, 396 (1989). 39. M . A . Kinch, Proc. S P I E 4 3 6 9 , 566 (2()()1 I. 40. J. White et al., J. Electron. Mater. 3 O( 6 ), 762 (2 ()() 1 ). 41. L. O. Bubulac and C. R. Viswanathan, J. Cr!lst. Growth (Netherlands) 123, 555(1992). 42. P. Tribulet, J.-P. Chatard, P. Costa and S. Paltrier, 1. Electron. Mater. 30(6), 574(2001). 43. I. M. Baker and R. A. Ballingall, SPIE Proc. 5 1 0 , 21 () ( 1985 ) 44. C. C. Wang, 1. Vac. Sci. Technol., B9.74() (1991 ). 45. G. N. Pulz, P. W. Norton, E. E. Krueger and M. B. Reine, ]. Vac. Sci. Technol., B9, 1724 (1991). 46. P. W. Norton, P. LoVecchio and G. N. Pultz et al., Proc. SPIE 2 2 2 8 , 73 (1994). 4 7. T. Tung, M. H. Kalisher and M. H. Stevens et al., Mater. Res. Soc Syrup. Proc. 9 0 , 3 2 1 (1987). 48. J. Arias, M. Zandian and J. G. Pasko et al., J. Appl. Phys. (USA) 69, 2143 (1991). 49. T. Tung, J. Cryst. Growth (Netherlands) 86, 161 (1988). 50. J. M. Arias, J. G. Pasko and M. Zandian et al., Appl. Phys. Letts. (USA) 62, 976(1993). 51. J. Bajaj, Proc. SPIE 3 9 4 8 . 4 2 (2()()()). 52. K. W. Hodapp, J. K. Hora and D. N. B. Hall et al., New Astronomy 1, 177 (1996). 53. A. Szilagyi and M. N. Grimbergen, J. Cryst. Growth (Netherlands) 8 6 , 9 1 2 (1988). 54. A. J. Syllaios and L. Colombo, Proc. IEDM Conf., IEEE, New York, p. 137 (1982). 55. B. Pelliciari and G. Baret, J. Appl. Phys. (USA) 62, 3986 (198 7). 56. S. M. Johnson, D. R. Rhiger, J. P. Rosberg, J. M. Peterson, S. M. Taylor and M. E. Boyd, ]. Vac. Sci. Technol. (USA)BIO, 1499 (1992). 57. P. W. Norton and A. P. Erwin, J. Vac. Sci. Technol. (USA} AT, 503 ( 1989 ). 58. J. H. Tregilgas, J. Vac. Sci. Technol. {USA} 2 1 , 2 0 8 (1982) 59. J. P. Hirth and H. Ehrenreich, ]. Vac. Sci. Techllol. (USA) A3, 367 (1985). 60. G. M. Williams et al., J. Electron. Mater., 22(8), 931 ( 1993 ). 61. T. J. de Lyon, B. Baumgratz and G. Chapman et al.. Proc. SPIE, 3 6 2 9 , 2 56 (1999). 62. M. B. Reine and A. Hairston, P. O'Dette et al., Proc. SPIE 3 3 7 9 , 200 (1998). 63. W. E. Tennant, M. Thomas and L. J. Kozlowski et al., J. Electron. Mater. 30(6), 5 9 0 ( 2 0 0 1 ) . 64. W. Cabanski, R. Brieter and R. Koch, et al., Proc. SPIE 4 3 6 9 , 547 (2001).

308 Handbookof Infrared Detection Technologies 65. T. N. Casselman, G. R. Chapman and K. Kosai et al., US Workshop on Physicsand Chemistry of MCTand other II-IVcolnpounds, Dallas, TX (Oct 1991 ). 66. R. D. Rajavel, D. M. Jamba and (). K. Wu et al., ]. Cr!lst. Growth, 1 7 5 / 1 7 6 , 653(1997). 67. C. T. Elliott, Chapter 11. Infrared Detectors and Etnitters: Materials and Devices, Electronic Materials 8, published by Kluwer Academic. 68. C.T. ElliottandT. Ashley, Elec. Lett. 2 1 , 4 5 1 (1985). 69. D. Mart, Vision, W. H. Freeman ( 1982 ). 70. M. Masie, P. McCarley and J. P. Curzan, Proc. SPIE 1 9 6 1 , 17 ( 199 3 ). 71. P. McCarley, Proc. SPIE, 3 6 9 8 , 716 (1999). 72. C. R. Baxter, M. A. Massie, P. L. McCarley and M. E. Couture, Proc. SPIE 4 3 6 9 , 129 (2001).

Chapter 9

Status of HgCdTe MBE technology T. J. de Lyon, R. D. Rajavel, J. A. Roth and J. E. Jensen

9.1 Introduction Since its initial synthesis and investigation in 1959. ~ the HgCdTe semiconductor alloy material system has been developed into one of the primary infrared detector materials for high-performance infrared focal-plane arrays (FPA) utilized in thermal imaging systems operating in the 3-SjJm and 8-121Jm spectral ranges. By virtue of its tunable direct energy bandgap that can be varied from semiconducting to semimetallic, its large optical absorption coefficient, and long minority carrier lifetime. HgCdTe has come to assume a central position in infrared detector materials research and development and infrared system manufacturing for over 4() years. In spite of this relatively long history of research activity, fabrication of high-performance multilayer heterojunction device structures in this materials system has only been realized during the past decade, with the adoption and refinement of sophisticated thin-film epitaxial growth techniques, such as molecular-beam epitaxy (MBE) and organometallic vapor phase epitaxy (OMVPE). capable of reliably synthesizing complex doping and composition profiles. While liquid-phase epitaxy {LPEI, a crystal growth technique first developed in the late 19 70s for HgCdTe, has proven itself as the workhorse epitaxial growth technology for manufacture of single-color, p-n heterojunction device structures, 2 the need to address mission requirements that can only be satisfied with the development of more complicated multilayer device structures has motivated an increased level of research investment in the MBE growth of HgCdTe alloys. Over the course of the past decade, significant progress has been achieved in HgCdTe MBE growth technology. ~.4 which has enabled the synthesis of complicated multilayer IR detector device structures. The development and integration of in situ sensors for monitoring and control of the sensitive HgCdTe MBE growth process, have significantly advanced the capabilities and

310

Handbook of Infrared Detection Technoloqies

consistency of the technology, culminating in demonstrations of excellent device perfomance in a wide variety of IR device architectures such as multispectral detectors, 5 near-IR avalanche photodiodes (APD), ~' high-performance MWIR detectors, 7 and megapixel arrays on Si substrates, s In this chapter, we will summarize the current status of HgCdTe MBE growth technology as practised in our Laboratory. The chapter will address the following topics: 1. a s u m m a r y of the 'hardware' aspects of MBE growth technology, including the role of process sensors (Section 9.2): 2. a discussion of the key parameters of the epitaxial growth process and associated thin-film material properties (Section 9.3). and 3. a discussion of current state-of-the-art results in device application areas that have been particularly well-served by the evolution of HgCdTe MBE technology (Section 9.4).

9.2 HgCdTe MBE equipment and process sensors 9.2.1 Vacuum equipment a n d sources

Cryopumps have found widespread use in MBE systems designed for the growth of HgCdTe alloys. A liquid-nitrogen-cooled trap is usually employed on the upstream side of the cryopump to condense mercury and prevent its transport into the cryopump, while the cooler internal surfaces of the cryopump are utilized to pump condensable species such as oxygen and hydrogen. Liquidnitrogen-trapped diffusion pumps that use mercury as the operating fluid are also in use, and these have provided extended trouble-free operation in our laboratory. The presence of large quantities of mercury in HgCdTe MBE systems precludes the use of ion pumps that are typically used to pump MBE chambers devoted to the growth of III-V semiconductors. In most MBE systems, titanium sublimation pumps are also present for removal of oxygen and water vapor. The Ti sublimation pump is generally employed prior to a long growth, or to assist in the pump down after a v a c u u m break, or to maintain low background pressure while effusion sources are outgassed. A large portion of the excess mercury resulting from the extreme overpressure used during growth is pumped by the liquid-nitrogen-cooled cryopanels surrounding the substrate and the source flange, t h r o u g h the freezing of Hg vapor onto the cold surface. It is therefore advantageous to maximize the surface area of the cryopanel. Growth chamber pressures around 1• -s torr. consisting primarily of the Hg background vapor, are typical during the growth of HgCdTe layers. In the absence of Hg flux, low background pressure levels of 1 to 5 • 10 -9 torr are achieved in the MBE chamber by maintaining a continuous flow of liquid nitrogen through the cryopanels during the course of a growth campaign that can last 6-9 months. As liquid mercury tends to pool at low

Statzls of HgCdTe ,\IBE technolog~t 311

points within the MBE system, internal chamber drains are designed to direct the mercury to a low-lying region, from which the accumulated liquid can be easily removed. In addition, the MBE system is baked at 15()-2()()~ to remove the remaining mercury from the cryopanels and adjoining areas to cooled collection areas, such as the traps located on the upstream side of the primary pump, prior to opening the MBE chamber for servicing. Until the early 1990s, a few groups conducted research on the growth of HgCdTe alloys by a variant of MBE referred to as metallorganic MBE (MOMBE), which employs metallorganic compounds as the source material. ~ At present, though, virtually all MBE growth of HgCdTe alloys is performed using source fluxes derived from the vaporization of elemental (Hg and Te) and binary (CdTe, for providing Cd) source materials. As in liquid-phase epitaxy, indium and arsenic are the preferred n- and p-type dopant elements, respectively. Commercially available dual-zone effusion cells are employed for the sublimation of Te and CdTe. The heated zone near the effusion cell's crucible orifice is maintained at a sufficiently high temperature to prevent condensation near the orifice, which can lead to clogging. A second independently-heated zone located near the bulk of the charge is maintained at a temperature sufficient for generation of the required flux at the substrate surface. Due to the high vapor pressure of Hg (41 () torr at 2()()~ at temperatures needed to bake the MBE system, Hg effusion sources are designed to store liquid mercury remotely from the MBE system. The remote location of the reservoir containing Hg facilitates refilling without having to break the v a c u u m within the main MBE system. This feature is particularly useful since large quantities of Hg are used during a growth campaign, which necessitates frequent refilling. The large external reservoir (typical c a p a c i t y = l - 2 litres) is connected to a heated zone that is located within the MBE system to supply the required Hg flux. Two different types of mercury sources have evolved over the years. In the first type, such as the one described by Million et al., ~ the Hg reservoir is supported on a platform that can be raised or lowered. The reservoir is connected to a crucible located inside the MBE system t h r o u g h flexible coupling such as a bellows. The height of the reservoir can be adjusted to supply, by gravity, the required a m o u n t of Hg to the crucible, which is heated to the desired temperature to provide the necessary Hg flux. The reservoir is lowered to drain the crucible during flux measurements of the other constituents or while the MBE system is being baked. In an alternate design, the reservoir is located remotely from the MBE system, but remains stationary. The reservoir is connected by rigid, heated tubing to a heated delivery nozzle located inside the MBE system. In this design, the entire reservoir is heated to about 130 ~ to provide the required flux. Since the reservoir is isolated from the rest of the MBE system by a valve, the flux can be rapidly turned on or off. If feedback regulation of the flux is desired, a separate electrically-driven valve controlled by the signal from a pressure sensor, such as a capacitance m a n o m e t e r mounted downstream of the control valve, can be used to precisely modulate or regulate the Hg flux.

312

Handbook of Infrared Detection Technoloqies

9.2.2 HgCdTe MBE process sensors

The epitaxial growth of device-quality HgCdTe by MBE is extremely sensitive to fundamental growth parameters such as source fluxes and substrate temperature. Materials properties such as alloy composition, p-type dopant density, and crystalline defect densities are strongly affected by variations in growth conditions. Because the MBE process 'window' for achieving a particular detector cutoff wavelength with a minimal areal density of crystallographic defects is substantially narrower for HgCdTe than for most other semiconductor systems currently being grown by MBE. many industrial, 11-13 military 14 and academic 15 groups involved in HgCdTe MBE research and development have actively pursued the development and refinement of in situ sensors for monitoring and control of the growth process. Without direct sensing of the MBE growth process, the ability to precisely and repeatably grow complex heterostructure device structures can be hampered by source flux instabilities (either long-term drifts or transients induced by shutter actuation), which can result in unintentional composition changes and defect generation. Morphological defect formation, ~~ HgCdTe alloy composition, and dopant incorporation ~7 can all be significantly affected by substrate temperature variations as small as 1 ~ and by source flux variations as small as (). 1%, which clearly illustrates the need for external sensing and control of the HgCdTe MBE growth process. Conventional MBE growth procedures have typically employed numerous calibration runs to fine-tune the effusion cell fluxes and substrate temperature, which is an extremely time-consuming and expensive approach. Moreover, this approach fails to address the need to maintain close tolerances continuously throughout the HgCdTe device growth process, the duration of which frequently exceeds 12 hours. It1 situ sensors have recently been developed for HgCdTe alloy composition, 11-14'1~ substrate temperature, 19 and source fluxes, 2~ enabling the growth process to be monitored, and in many cases controlled, using closed-loop feedback control paradigms. Advanced monitoring and control capabilities of this type can substantially increase the yield of manufactured wafers that meet target specifications for optical properties and crystallographic defects, and can dramatically shorten the time required for establishment and qualification of new growth processes. Our laboratory has developed a particular MBE system architecture to address the need for sensor-based control of the HgCdTe MBE growth process. The system configuration, illustrated in Figure 9.1. is based on the integration of three optical sensors with a conventional VG Semicon VS()H MBE system designed for growth on three-inch diameter substrates. A similar approach has been adopted to implement real-time control on a Riber Epineat system that is used for production of HgCdTe epitaxial material on five-inch diameter substrates at Raytheon Vision Systems. The three primary sensors used on these MBE systems are: (1) Absorption-edge Spectroscopy (ABES) for substrate temperature, (2) Spectroscopic Ellipsometry (SE) for alloy composition, and (3) Optical absorption Flux Monitoring (OFM) for direct monitoring of Group II source fluxes.

Status of HgCdTe MBE technolog!l

31 3

Figure 9.1 HgCdTe MBE system integrates sensors for closed-loop feedback control of substrate temperature. alloy composition, and Group Il flux.

The use of optical sensors has the advantage that there are no known adverse effects due to interactions of the light sources with the growing epitaxial film; moreover, they are easily incorporated into conventional MBE growth chamber designs t h r o u g h the provision of additional dedicated optical access ports. The operation of the ABES substrate temperature sensor is based on the method of absorption-edge spectroscopy, ~9.21 performed at normal incidence in reflection mode, to infer the temperature of a 2 • 2 cm Si 'witness' wafer that is conductively mounted onto the back side of the substrate mounting block. This measurement technique requires optical access to the backside of the substrate mounting block, which is provided in our MBE systems by a straight quartz light pipe installed on the single-axis substrate manipulator. The quartz light pipe, tapered at one end to enable it to pass between the elements of the substrate heater assembly, is positioned approximately 7 mm from the Si witness wafer. The refiectivity measurements required for the temperature determination are made with a Model NTM-I remote temperature sensor manufactured by CI Systems, Inc. 22 The ABES technique for substrate temperature measurement has a precision of • ~ and is essential for elimination of substrate temperature transients induced by opening/closing of shutters in front of hot effusion cells and those arising from emissivity variations due to coating of manipulator and substrate mounting block surfaces during growth. Such uncorrected transients can easily exceed 10 ~ and are only partially sensed by the substrate heater thermocouple, which is not in direct physical contact with the substrate mounting block. The advantage of ABES sensing of substrate temperature is demonstrated in Figure 9.2, which presents a comparison of substrate temperature control utilizing

314

Handbook of Infrared Detection Technolo~ties

Figure 9.2 Feedback control of substrate temperature based on ABES sensor eliminates large temperature variations induced by opening and closing of MBE effusion cell shutters.

conventional non-contact thermocouple feedback to that utilizing ABES feedback. During the first half of the experiment depicted in Figure 9.2, power to the substrate heater was controlled using a feedback loop in a commercial temperature controller that maintained a constant thermocouple temperature. The thermocouple, due to its insensitivity to changes in actual substrate temperature induced by opening and closing of MBE source shutters, performs poorly as a means of regulating substrate temperature. The second part of the figure illustrates how temperature transients can be eliminated t h r o u g h use of a feedback loop based on ABES sensor data. Substrate temperature changes, arising from shutter actuation or induced by emissivity variations due to accumulation over time of coatings on the substrate holder and surrounding manipulator surfaces, are completely eliminated. For substrate temperature regulation, we employ a cascade-PID control structure that is depicted in Figure 9.3. An outer control loop software PID algorithm compares the ABES sensor temperature (indicated as 'True Substrate Temperature' in Figure 9.3) to the desired substrate temperature, producing an updated thermocouple temperature setpoint that is sent to a conventional 'hardware' PID controller, which regulates the power to the substrate heater.

Status o[ HgCdTe ,\IBt:, technolog!!

31 5

Figure 9.3 Dual-cascade control loop structure for substrate temperature regulation. The outer control loop is based in the MBE control software, while the inner loop is implemented in a conventional hardware PlI) controller.

The software PID algorithm employed for outer loop control is of the following general form:

ar,,

r- K (t3rr,,S-

+

][

(rr,:, -

+ D

dTARES ~ES at

(1)

in which A T h t r represents the computed change in inner-loop heater thermocouple setpoint, K is the loop gain, I is the integral time constant, D is the derivative time constant, Tr,:fis the true substrate temperature setpoint, TAt31:,sis the instantaneous substrate temperature derived from the ABES sensor, and fl is a setpoint weighting factor included to improve setpoint-following performance. 23 This control algorithm has been implemented in internallydeveloped MBE computer control software to provide substrate temperature regulation with a precision of _+0.5 ~ throughout the growth of complicated multiplayer device structures. Spectroscopic ellipsometry has been successfully developed as a technique for the in situ determination of HgCdTe alloy composition during the MBE growth of IR device structures. Considerable research activity has been directed at overcoming a variety of challenging experimental difficulties with this sensor, including data corruption due to substrate wobble during sample rotation, 13 temperature-composition cross-coupling effects, 24 and lack of comprehensive and accurate dielectric function databases for HgCdTe under MBE growth conditions. Progress in resolving these issues is evident in recent reports demonstrating the utility of SE for real-time measurements of composition in

316

Handbook of Infrared Detection Technologies

Hgl_xCdxTe alloys with x ranging from ().2 to 0.5, which covers the range necessary for fabrication of detectors operating in the LWIR and MWIR spectral ranges for thermal imaging applications. 1~-~4.~s Spectroscopic ellipsometry data are typically acquired over a spectral range of 1.7-5 eV using J. A. Woollam Co. 2s M2000 ellipsometers. The MBE systems are equipped with low-wobble (___0.15 ~ sample manipulators and heated low-birefringence windows allowing for coating-flee optical access to the substrate at a 70 ~ angle of incidence with respect to the substrate normal. The SE sensor provides an updated measurement of the surface x-value of the growing HgCdTe film approximately every eight seconds after averaging the optical data over two substrate rotations. Application of the SE sensor to derive the x-value requires a dielectric function database, which can be acquired through growth of uniform-composition layers that are independently analyzed for alloy composition via post-growth IR transmission combined with secondary-ion mass spectrometry (SIMS). The dielectric function spectra are then fit with a superposition of harmonic oscillator functions to construct a composition- and temperature-dependent optical constant database. During an actual growth run. the database is used to derive epilayer composition by minimizing the error between the real-time SE dielectric function spectra and spectra generated from an optical model of the growth surface constructed using the aforementioned library of optical constants. Figure 9.4 illustrates the fidelity of the composition determination that can be achieved with SE. In this figure, we compare the x-value determined by SE in real-time during MBE growth with a postgrowth SIMS depth profile. The SE and SIMS data show excellent agreement over the entire layer profile, and it is clear that the SE sensor is capable of detecting quite small changes in alloy composition, such as those that can occur in nominally constant-composition device layers grown without feedback control. In practice, the instrumental noise level of the SE sensor is low enough to permit the detection of composition changes as small as Ax=O.O003, corresponding to flux changes of less than O. 1%. The run-to-run performance of the sensor, when compared against post-growth IR transmission measurements, is characterized by a 95% confidence limit for composition determination of Ax=+O.O015.11'12'26 The use of SE for HgCdTe composition determination is sufficiently well developed that it is routinely used for feedback control of HgCdTe composition in our laboratory and by several other groups. ~2'~4 A cascade-PID control algorithm, similar to that described above for substrate temperature (cf. eqn. (1)), is employed to achieve closed-loop feedback regulation of alloy composition. Inner-loop control of either the CdTe or Te effusion cell temperature can be implemented in this cascade-loop architecture to achieve continuous regulation of composition in layers with either constant or ramped composition. Monitoring of the flux of Group II species (Cd, Zn, and Hg) is achieved through an optical flux monitoring technique that utilizes the resonant optical absorption of light by the atomic species comprising the fluxes emitted from effusion cells in the MBE chamber. The sensing light is generated from either hollow-cathode or electrodeless discharge lamps for each of the elements and enters the MBE

Status of HgCdTe MBE technoloq!!

317

DEPTH (um) 0.26

10

8

6

4

2

I

I

I

I

I

0.25 Z

0 I.-

.(

0.24

u_ 0.23 l,J.l ._J

SIMS, .

-

0

~ 0.22 "13 (J

SE

0.21

0.20

I

0

-

I

I

I

I

I

I

50

1 O0

1 50

200

250

300

TIME (min) Figure 9.4 Comparison of HgCdTe allo!l composition determined from postgrowth SIMS depth profiling with values derived from SE measurements obtained durin# ,\ I B I:, #rowt h.

growth chamber t h r o u g h heated optical viewports. As depicted in Figure 9.1. the optical beams make a double pass through the MBE chamber, sampling the effusion cell fluxes at a position just above the substrate. A portion of the incoming signal beam is diverted into a reference path external to the MBE chamber in order to cancel signal level changes arising from lamp intensity fluctuations and photomultiplier detector drift. Both the signal and reference beams are chopped and lock-in amplifiers are used for synchronous detection. Due to the extremely high absorption cross-sections for the Group II elements zinc ( Z = 2 1 3 . 8 n m ) , cadmium ( Z = 2 2 8 . 8 n m ) a n d mercury ( Z = 2 5 3 . 6 n m ) , the OFM sensor can readily detect flux variations at the 1 x 1 ()-9 torr level, which is approximately O. 1% of the Cd flux typically employed during the growth of LWIR HgCdTe at a growth rate of 2 ~tm/hr. Sensing of low levels of Zn flux can be useful in applications such as the growth of q u a t e r n a r y HgCdZnTe alloys 27 in which the incorporated Zn concentration is at or below the 1% level. The OFM sensor for Cd is particularly useful for analysis of short-term and long-term anomalies in the flux e m a n a t i n g from the CdTe effusion cell such as MBE shutter transients 1~ or flux instabilities arising from cell depletion, and can be used to detect source flux leakage past closed shutters.

318

Handbookof Infrared Detection Technolo#ies

9.3 HgCdTe MBE growth process 9.3.1 Substrate preparation

The current preferred crystallographic orientation for epitaxial growth of HgCdTe device structures by MBE is (211 )B. due to difficulties with twin 2's and hillock 2~ formation encountered with alternative low-index orientations such as (100) and (111). Because the lattice mismatch between HgTe and CdTe is only O. 3 %, epitaxial growth of low-dislocation-density ( < 1 x 1 ()s cm-2) HgCdTe films can be readily achieved on Cdl_vZnvTe substrates with y=().()l-().()4. As discussed later in Section 9.4.3. MBE growth is also performed on Si(211) substrates for SWIR and MWIR device applications, which can tolerate the higher dislocation densities (> 1()r cm -2) present in HgCdTe films grown on such lattice-mismatched substrates. Prior to loading into the MBE growth reactor. CdZnTe substrates are conventionally subjected to a wet chemical cleaning procedure consisting of: ( 1 ) solvent degreasing, (2) etching in dilute I(). 1-1 vol.%) Br-methanol solution, (3) rinsing in methanol, and I4) rinsing in deionized water. Following this procedure, the substrates are blown dry with nitrogen, mounted onto substrate holders using a colloidal graphite paste, and loaded into an entry lock chamber for pumping down from atmospheric pressure. The surfaces of CdZnTe substrates prepared in this m a n n e r are generally depleted of Cd and Zn and are covered with a layer consisting of a mixture of elemental Te and Te oxide. ~r ~1 This surface layer is removed from the substrates prior to initiation of HgCdTe epitaxy by thermal cleaning in the MBE system. The evolution of the CdZnTe substrate surface during thermal cleaning in the MBE growth chamber can be monitored with SE ~2 due to the sensitivity of SE to oxides and other overlayers on II-VI compounds. ~' ~4 The overlayer thickness, plotted in Figure 9.5, can be extracted through use of a simple optical model of the CdZnTe surface that assumes a rough overlayer, consisting of a 5()%/5()% mixture of v a c u u m and CdZnTe. whose dielectric function is treated in the effective-medium approximation. Prior to any heating of the substrate, the SE m e a s u r e m e n t indicates that the wet-chemical-cleaned CdZnTe surface has an equivalent overlayer thickness of ~2() A. As the wafer is heated and the temperature rises past 2()()~ at t - 1 7 minutes, the thickness of the overlayer decreases dramatically, corresponding to the sublimation of excess elemental Te from the substrate surface. Further reduction in the thickness of the overlayer at t=25 minutes, due to the sublimation of volatile Te oxide, is achieved only by heating to ~3()0 ~ under a Te overpressure for ~ 1 () rains. Failure to eliminate the residual surface overlayer introduced by the e.v sitl~ wet chemical cleaning procedure generally results in epitaxial films with compromised crystalline quality. For this reason, SE has been found to be a quite useful analytical tool for monitoring the quality of CdZnTe substrate surfaces prior to HgCdTe nucleation, supplementing more traditional surface-sensitive MBE techniques, such as reflection high energy electron diffraction (RHEED). for assessment of substrate cleanliness.

Status of HgCdTe ,\IBE technolog!! 319

25 I

2O

I

I

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

..". . . . . . . . . . . . . . . . . . .

"..r . . . . . . . . . . . . . . . . . .

i,

~...................

i

! .................

i

i

i

i

-

i

::

::

::

~, . . . . . . . . . . . . . . . . . . .

i,. . . . . . . . . . . . . . . . . . . .

~.

t

v r L,I.I

z

"T"

b-~_1

0

15

_\'-'i

-" ................ i[

,~ ,4

-

l

' ..............

!

:::

---I-

;:. . . . . . . . . . . . . . . . . . . .

!

.....

i!

!

t

i

!::

!

I

,i~-~. . . . . . . . . . . . . . . . . . . .

::i

~

j

!

-

-:. . . . . . . . . . . . . . . . . .

i

""

~: . . . . . . . . . . . . . .

!

I

'~

::

i!

- ~ -%- i i,. . . . . . . . . . . . . . . . .

"

-

i ::

i !

i ::

i ::

i i

" "

.

/:

i

i

::

::

::

i

i

-

i i

:: i

i i

:: i

:: i

:: i

::

i

i

i

i

2oo

-I i

~,

il il

::/

250

"

-

"

-

300

150

,!

" 5 -

0

i

:

~

l

i '

10

::

~

20

::

i i

I

30 TIME

!

i i

i

40

i ~

: ~,

50

60

-

m'~ r--

1 O0

m

so

~o "On

" " -

rri 3~

~, -'1

10 ................ ~..................i....................!...................!....................i...................!.................. .

m

, o 70

(mins.)

Figure 9.5 Evohltion of SE-determined overlayer thickness on wet-chemical-etched CdZnTe substrates dziring pregrowth thermal treatment.

Initial p r e p a r a t i o n of oxide-free Si(211 ) substrates for II-VI h e t e r o e p i t a x y can either be achieved t h r o u g h : (11 HF-based ex sitz~ chemical t r e a t m e n t s to strip native oxides and leave a robust h y d r o g e n - p a s s i v a t e d surface that is subjected to l o w - t e m p e r a t u r e ( < 6()() ~ t h e r m a l t r e a t m e n t in the MBE system for desorption of a h y d r o g e n m o n o l a y e r . ~s. or 12) h i g h - t e m p e r a t u r e ( > 8 5 ( ) ~ thermal t r e a t m e n t to desorb volatile Si oxides. ~ After p r e p a r a t i o n of an oxide-flee Si surface by either of these techniques, a variety of distinct prescriptions have been reported for s u b s e q u e n t substrate p r e p a r a t i o n procedures. T e r m i n a t i o n of the Si(211) surface with As has been reported to be useful for improving the m e c h a n i c a l stability of the II-VI/Si interface ~- and for avoiding formation of a m o r p h o u s SiTe, on the Si substrate surface. N.K. Dhar et al. have reported ~ on the utility of utilizing a thin. a m o r p h o u s film of ZnTe deposited at ,~4() ~ onto such A s - t e r m i n a t e d Si surfaces for suppression of n u c l e a t i o n in a threed i m e n s i o n a l g r o w t h mode. Additional e x p e r i m e n t a l work at the University of Illinois at Chicago 4r has suggested the use of thin { 2 - 2 ( ) n m ) ZnTe films deposited at 2 2 0 ~ onto A s - t e r m i n a t e d Si substrates, followed by a n n e a l i n g at 3 8 0 ~ prior to g r o w t h of a thick CdTel 211 ) buffer layer.

320 Handbook of Infrared Detection Technologies 9.3.2 Growth conditions

As discussed in Section 9.2.1, MBE growth of HgCdTe generally utilizes Hg, CdTe, and Te as source materials. Typical beam equivalent pressures of these constituents during the growth of LWIR-HgCdTe films (.x-=().22) at the rate of 1.5 ~ m / h r are: Hg=8 x 10 -4 torr. CdTe=4 x 1 ()-- torr and T e 2 - 1 . 5 x 1 ()-~' torr. In general, there is a strong correlation between the vapor pressure of a constituent element and its sticking coefficient on the growth surface. Mercury, which has the highest vapor pressure of all group II and VI elements ( l x l ( ) - ~ torr at 300 K), has an extremely low sticking coefficient (~1/3()() at 180 ~ on a Testabilized surface) that also exhibits a strong temperature dependence. Consequently, HgCdTe growth is generally conducted in the 175-2()0~ temperature range, under excess Hg growth conditions. Under these conditions there is sufficient surface mobility of Hg, Cd and Te2 to ensure epitaxy. With the exception of the Hg-containing alloys, MBE growth of II-VI compounds is usually performed at 2 5 0 - 3 0 0 ~ In this temperature range, the re-evaporation rate of the excess constituent element is high enough to enable the epitaxial growth process to be self-regulating, resulting in a film that is stoichiometric despite using an excess flux of either the Group I! or Group V! constituents. At typical HgCdTe growth temperatures of ~,18"3 ~ the re-evaporation rate of Te is sufficiently low that the growth process is not self-regulating. Therefore, despite the presence of a large excess flux of Hg, there is a tendency to form Teprecipitates when the optimal growth temperature is exceeded. Hence the growth temperature and ratio of Hg-flux to Te-flux must be precisely controlled to optimize the quality of the HgCdTe layer and inhibit the formation of crystallographic defects, as discussed in further detail in Section 9.3.3. 9.3.3 Defects

Threading dislocation segments and macroscopic defects that are manifested as morphological defects on the surface of a HgCdTe film are the two most important classes of defects that affect the performance of IR detectors. Dislocation densities in thick HgCdTe films (thickness > 4 l a m ) c a n be readily measured by subjecting the film to a wet chemical etch that preferentially decorates threading segments intersecting the surface of the film through the formation of pits. The etched surface can be observed under an optical microscope to measure the density of these pits. 4~ Threading dislocation densities < 5• 1() ~ cm -2 are desirable for the satisfactory operation of LWIR diodes at 7 7 K. The dependence of diode performance on the dislocation density has been reported by Johnson et al. 42 The performance of MWIR diodes appears to be less sensitive to the dislocation density as attested by the performance of MWIR p-n diodes grown on Si substrates for which the dislocation density is generally in excess of l x l() ~' c m - 2 . 4~ The dislocation density in a device structure is primarily dependent on the degree of lattice-matching with the underlying substrate. Figure 9.6 shows the near-surface dislocation densities measured for a series of LWIR device structures.

Status of HgCdTe MBE technolog!l

~21

Figure 9.6 Dislocation densit!l measured for series of I.i'~'lR devices grown in our lahoratory. A vast majoritzl of the device structures have near-surfi~ce dislocation densities of about ~ • 10 s cm 2.

Typical dislocation densities for LWIR and MWIR device structures grown on CdZnTe substrates are 3 • 1 () 5 c m - 2 and 5 • 1 () ~ c m - 2, respectively. One of the most challenging aspects of HgCdTe device growth lies in controlling the macroscopic defect density. For a chosen Hg/Te flux ratio, the substrate temperature must be controlled within a range of approximately 5 ~ to realize the optimum quality (minimum defect density) in the film. At temperatures higher t h a n the optimum growth temperature, or when the Hg flux is deficient, a large density of morphological defects, known as voids, forms in the film. The term 'void' is used in the literature to describe a polycrystalline defect whose diameter ranges between 2 and 1() microns and which has an irregular shape. Voids can be seen when the surface of the HgCdTe film is observed under Nomarski contrast optical microscopy. This defect, illustrated in Figure 9.7, is a Te-rich polycrystalline region in the film. Cross-sectional microscopy and TEM studies have shown this defect to have a conical profile, resembling an ice c r e a m c o n e . 4 4 - 4 6 Void defects have been shown to cause shorted diodes, resulting in inoperable pixels. 47 The density of voids decreases as the growth temperature is lowered and approaches the optimum growth temperature for the chosen Hg/Te flux ratio. However, as the growth temperature is reduced below the optimum growth temperature, a different population of defect, the microvoid, is generated. These defects are smaller in diameter than conventional void defects and have a more regular shape. Microvoids are thought to form when the Hg flux is higher than the optimum value, or the substrate temperature is lower than the optimum growth temperature. Piquette et al. have reported 4s in detail the dependence of

322

Handbook of Infrared Detection Technologies

Figure 9.7 Cross-sectional TEM view of a HgCdTe.fihn showing the "conical'prqfile of a void defect. The image on the right illustrates the pol!tcr!lstalline region contained in the defect.

the density of both voids and m i c r o v o i d s on growth conditions. In our laboratory we find that, under optimal growth conditions, the total density of both voids and micro-voids is typically between 5()() and 1 5()() cm-2 for n-p-n heterostructures, such as dual-band detectors, whose thickness is 2() gm.

9.3.4 Doping The growth of n-type HgCdTe layers with the desired levels of In concentration and structural perfection is relatively straightforward. Typical mobilities for MWIR and LWIR HgCdTe alloys doped with In at 1-3 • 1 ()is cm-3 measured at 7 7 K for films grown in our laboratory are 50()()() and 80 ()()() cm2/Vs, respectively. Data on the reproducible growth of n-type HgCdTe films have also been reported by Rockwell. 4~ In contrast with the relative simplicity of n-type doping with In, the low sticking coefficient of As and the tendency for compensation of acceptor impurities due to the presence of Hg vacancies, has raised challenging issues for in situ p-type doping, s~ The ability to achieve good structural perfection with desired levels of hole concentration in in situ doped p-type layers is a prerequisite for the growth of high-performance n-p-n two-color device structures. Our laboratory has developed a unique process that enables growth of in situ doped ptype layers with excellent structural and electrical properties. 51 To assess the electrical properties of these in situ doped p-type layers, individual 4 - 6 l.tm thick MWIR-HgCdTe:As films were deposited on (211)B CdZnTe substrates. The samples were subjected to a post-growth Hg-anneal to remove any Hg-vacancies in the films, which are deep level acceptor impurities. Electrical contacts were made with indium and variable temperature Hall effect and resistivity

Status of HgCdTe ,\IBE technolog!!

32 3

m e a s u r e m e n t s were performed. The in s i t u doped MWIR-HgCdTe:As films exhibit classic p-type conductivity for hole concentrations from 5xl()l~' to 2 x l O ~scm -3. As reported previously, the electrical activity in these in s i t u arsenic-doped films is in the range of (~()-1()()% as determined from SIMS and Hall effect measurements. An activation energy of 7 meV was measured for the As acceptor from variable t e m p e r a t u r e Hall m e a s u r e m e n t s for HgCdTe films with x=0.30. The dependence of hole mobility on the hole concentration for As-doped MWIR HgCdTe layers is illustrated in Figure 9.8. As expected, the hole mobilities exhibit a trend of decreasing values with increasing hole c o n c e n t r a t i o n due to increased ionized and neutral impurity scattering, similar to that observed for n-type alloys. 82 The hole mobility is also sensitive to the band gap of the HgCdTe alloy, with mobility increasing with a decrease in the band gap. For this reason there is some scatter in the data s h o w n in Figure 9.8, since the alloy composition of the As-doped samples extends over the range x = 0 . 2 8 - 0 . 3 8 . Nevertheless, the trend of decreasing mobility with increasing hole concentrations is clear, as s h o w n by the dotted line. The r u n - t o - r u n variability in dopant incorporation for As and In are contrasted in Figure 9.9, which summarizes dopant concentrations as determined from SIMS m e a s u r e m e n t s for a set of LWIR two-color detectors. Typical doping c o n c e n t r a t i o n for the n-type absorbing layer is 1- 3 • 1 ()) s c m - 3, and that for the p-type layer is 1-1 5 x 1 ()~ - c m - 3. The c o n c e n t r a t i o n of As that is

"-.....

350 I

A

300

..

x = 0.28-0.38

..................... :.....m ....~....................... .

04

!

.

::;'...

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

.

.

77K

Hall meas,.

.

I .

E

(.1 250

200

.(2 O

E

O

"I"

.................. 9

r,. r..

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

i

9

....... i~7:'~

100 50

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

9

............... ~..............................................

9

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

: .................. : ...... i .... ! .... j=,.~

",...

".... ,

15o

:'-:-.-:.....:....!.... i "'!.. :

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

",,

9

m

m

ium

.

9

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

. .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

. . . . . . .

. . . . . .

, ,I

. . . .

"

9

"'I ........................... i.... .....................

"'..

im

9 . . . . .

:

:

"

-

:

,

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

i ,

:::,.

..

.

i

i

:-:,,-; ............. i .......... :.............. !-= ..... i ....... i ...........................

....=

,".~

1017 Net hole concentration

i

, ,

I

1018 a t 7 7 K ( c m "3)

Figure 9.8 Hole mobility as a function of hole concentration for in situ As-doped HgCdTe films with x values between 0.28 and O. 38. The general trend of decreasing mobilit!t with increased doping is shown b!! the dotted lines that serve to guide the eye.

324

Handbook of Infrared Detection Technolo~lies

Growth Run Index Figure 9.9 Control over As and In dopant concentrations, a s measured b!t SI,\IS. jor a series of LWIR multispectral device structures. T/re horizontal lines indicate t/re acceptable range for the dopant concentrations.

incorporated in the films shows more scatter than the In concentration due to the strong substrate temperature dependence of As incorporation.

9.4 Device applications 9.4.1 Multispectral HgCdTe infrared detectors The capabilities of infrared thermal imaging systems can be significantly enhanced through the use of multispectral sensors in place of traditional singlewavelength infrared sensors. By providing sensitivity to wavelength-dependent variations in emissivity and reflectance, detection in multiple spectral bands can aid in scene interpretation and signature recognition. Multispectral FPAs can also provide additional flexibility through the selection of a particular spectral band in which scene contrast is optimized. The next generation of multispectral devices will be based on a structure that utilizes stacked p-n junctions grown by MBE. This approach enables significant savings in weight, power, and optical system complexity compared with multispectral system approaches that depend either on dispersion of the optical signal across multiple detector arrays, or the use of a filter wheel to perform spectral decomposition in tandem with a single FPA. The stacked junction approach also guarantees image

Ntatus of HflCdTe AIBE technolo#!i

32

alignment at the level of the individual pixels from the various spectral bands, which can significantly simplify video image processing by eliminating the need to accurately align separate FPA fields-of-view. The standard device architecture for two-color detection grown by MBE in our laboratory was developed at Raytheon Vision Systems s~-ss in the early 1990s. The architecture is based on an n-p-n device structure, as illustrated in Figure 9.10. The structure, which consists of two back-to-back p+-n HgCdTe photodiodes sharing a common p-type layer, enables both detectors to be physically positioned in the same pixel location. The spectral sensitivities of the two bands, labelled 'Band 1' and 'Band 2' in Figure 9.1 (). are determined by the alloy compositions of the two n-type HgCdTe absorbing layers, which can be precisely controlled in the MBE growth process. Because the structure is designed for backside illumination through the CdZnTe substrate, the alloy composition of the Band 1 absorber layer is higher (and corresponding detector cutoff wavelength shorter) than that of the overlying Band 2 absorber. An essential precursor technology for demonstration of these N-p-n architecture devices was a process for in sitzi p-type doping with As. Development of this process was hampered by several materials growth issues raised by introduction of As into the crystal growth process. Control of substrate temperature, in particular, has been a significant issue with respect to control of p-type doping, due to the sensitivity of As incorporation to substrate temperature, s~''s7 In addition, the e l e c t r i c a l a c t i v i t y of As has been shownS'~ to diminish at concentrations above a p p r o x i m a t e l y 2 • 1() ~'~ cm-~. Increasing the atomic concentration of As to l()2~cm-~ has also been observed to have a deleterious effect on dislocation density in the p-layer and topmost n-layer ofn-p-n devices. 59 For these reasons, reproducible control of As dopant c o n c e n t r a t i o n in the 1 - 1 0 x 1 0 1 7 c m -~ range has been desirable. The ABES n o n - c o n t a c t temperature sensor, discussed in Section 9.2.2, has enabled us to achieved sufficiently tight control over substrate temperature, as manifested in good runto-run reproducibility of As concentration during the growth of two-color device structures. ~'~)A typical depth profile, obtained with oxygen bombardment SIMS analysis, of the dopant species in a multispectral device structure is presented in

Figure 9.10 n-p-n device architecture developed b!l Ra!ltheo~l i'ision Sjlstelns for IR detection in two spectral bands. The HgCdTe allo!l composition of the Band 1 and Band 2 ahsorbin~l la!lers determines the spectral bands sensed b!l the detector.

326

Handbook of Infrared Detection Tecilnolo#ie,,;

Figure 9.11. The In and As dopant c o n c e n t r a t i o n profiles are observed to be wellbehaved, w i t h o u t any evidence of segregation or m e m o r y effects. The b a c k g r o u n d dopant levels of In and As are m e a s u r e d to be at the SIMS detection limits of 5 • ~cm-~ and 5 • under oxygen and cesium b o m b a r d m e n t , respectively. HgCdTe epitaxial wafers intended for multispectral device applications can be processed into either sequential-mode or s i m u l t a n e o u s - m o d e detector arrays, as illustrated in Figure 9.12. The fabrication sequence for such multispectral arrays utilizes process steps that are identical to those used in the m a n u f a c t u r e of singlecolor arrays, with the exception of the mesa delineation. This is performed with reactive ion etching r a t h e r t h a n wet chemical etching in order to create nearly vertical sidewalls on the mesas, for m a x i m u m optical fill factor in the top junction. In the sequential-mode detectors illustrated in Figures 9.12 (a) and 9.12 (c), electrical contacts are provided to the top and bottom n-type photon-absorbing layers, while the middle p-type layer is left floating, resulting in a format that has

1019

!

l,,sl A !

E 0

Z

0 m

i<1:

1

017

1

01 6

.- . . . . . . . . .

:

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

.

!. . . . . . . .

_

IX: I--

Z uJ

(3 Z

0 (3

_ _t

zl"'l

,o,4 i i i i i iiiiiii i i 1015

i

. . . . . . . .- - . . . . .

.i 1113

,

0

,

,

,

,

5

,

,

,

,

,

10

i

, .......

,

,

I

~/,

1

15

DEPTH (pm) Figure 9.11 In and As dopant com'entratiotls me~lsllred in an n-p-n H#CdTe two-color device structure with oxygen bombardment SIMS anal!Isis. The apparent A,,; cotwet~tration ill the n-tJIpe absorbin# la!ler,'; is limited by the SIMS measurement back#roztmt.

Status of H[ICdTe ,\IBE technolo[l!l

SEQUENTIAL-MODE Top Absorber Contact

Contacl i

l

-

SIMULTANEOUS-MODE p-type layer contact o

Arra~oCnt~

Array Common /

CTtact

Top Absorber 9

n-type MWlR

p-type MWIR

p-type MWlR

n-typeMWIR

n-type MWlR

Substrate

Substrate

f

IR radiation

(a)

~2 7

I IR radiation (b)

Figure 9.12 ( a ) Sequential-mode two-color detector is fabricated with ohmic contacts to the top amt bottom nt!lpe absorbers, with the middle p-t!lpe la!ter h'ft [loatitul. (b) ,';imultatu, ous-mode two-color detector requires ohmic contact to each of the three device lallers, allowin[l both junctions to ~lem'rate photocurrent sintultaneousl!!. (c) SEM tnicro#raph Q[processed seqm'ntial-ntode detector pixels with one indimn bllttl 1)per 61 lath unit ('ell. (d) SE,\I nticro[iraph Oll~roce.~,~ed simultaneotts-t,ode dete(tor pixels with two indimn l)mnl~S per ~0 I~m unit cell.

only a single indium bump per unit cell. The detector is operated by applying a bias to the pair of back-to-back diodes, with the polarity of the voltage determining which of the two junctions is reverse-biased for collection of photogenerated carriers. Because the bias polarity must be switched in the sequential-mode device structure in order to select between the two wavelength bands, images from the two bands must be collected in sequence during different intervals of time. Addition of an ohmic contact to the central p-type layer.

328

Handbook of Infrared Detection Technolo~lies

illustrated in Figure 9.12 (b) and (d), results in a design with two indium bumps per unit cell. This allows the two photodiodes to be independently accessed and simultaneously reverse-biased and integrated for acquisition of simultaneous imagery in the two spectral bands. Fill factors in excess of 80% for the topmost junction can be achieved with the simultaneous device architecture illustrated in Figure 9.9 (d). A typical current-voltage characteristic for a sequential-mode MWIR multispectral detector is presented in Figure 9.1 3. The 80 K data, collected with an f/2 field-of-view and a 300 K background, exhibits the classical 'S'-shaped form that is representative of two back-to-back diodes. Diode breakdown voltages in excess of 0.5 V are routinely observed for MWI1R sequential-mode devices with cutoff wavelengths in the 3.'3-4.'3 l~m spectral range at 8() K. The dynamic resistance-area (RA) product, at modest reverse bias of ~1()() mV, is in excess of 106 f2 cm 2, sufficient to ensure near-unity injection efficiency when such detectors are coupled with charge integration capacitors on direct-injection

500xl 0

-12

10 o 80K f/2 FOV

400

ii

i

10

7

300 l.Z

"O 10 6

200

14.1

I:I:: n"

100

10 s

0

0

r0 . o

I

o

3

10

-100 -200

=

-0,4

-0.2

0.0

0.2

0,4

4

103

V O L T A G E (V) Figure 9.1 3 Current-voltage characteristic a~zd d!lnamic impedance-area (RA) product at 80 K tff HgCdTe sequential-mode two-color detector, measured at f 2./ieht-ol:viexv under a ~00 K back~Iround. I)!!namic RA products exceed 10 (' f2cm 2 for these detectors ~vith area equal to ~.72• 1 0 s cm2 and 80 K cutoff wavelengths of ~. 7 and 4. ~ ~tm.

Status of HgCdTe ,\IBE technolog!l

~29

Si readout circuitry. In order to make a meaningful assessment of zero-bias RA product, current-voltage characteristics must be collected on simultaneousmode devices that allow for independent biasing of the two junctions. Results have been reported 61 on such devices, with RoA values in the 0.5-1.3 • 106 f2 cm 2 range observed for detectors with 77 K cutoff wavelengths in the 3.74.4 Bm spectral range. The 80 K spectral response of an integrated multispectral detector is displayed in Figure 9.14. Excellent uniformity of detector cutoff wavelengths is typically observed across 6 4 x 6 4 or 128x 128 element arrays. For the 6 4 x 6 4 element detector array from which the device characteristic in Figure 9. ] 4 was derived, a cutoff wavelength of 3.71+_().()()7 ~tm for the lower-junction detector and 4.29_+0.005 l~m for the upper-junction detector were measured. Such excellent uniformity of detector cutoff derives from the uniformity of HgCdTe alloy composition for films grown by MBE. The lower (shorter cutoff wavelength) diode exhibits uniformly high spectral response to the shortest measured wavelength of 2 ~tm, which indicates that the minority carrier (hole) lifetime in the lower

121-"" =

O

1.0

L ,

! "

"'"

........................ _, . . . .

~ .,p.,,,,, ir

.

"

~

08

I""

"'"

.

i

" : " L . .

i i !

9

"'"

""

i l ............

;

I .............................

i ! i

..,=

i

-

~ .................................

i

'

................. ,f ........7 .................. i

- Iv -- omvl

~

; ............................

i i ~ ~

't

"'

I

.............. ~i............................

~

- - ................................... ~. . i1

00

9

,l

7 .....................

i

,,-I, o2

i

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

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

!""

V

! i //I

b

=+80m

-

V,

I .....

-

i

i

::

i

-

i ::

i i

i i

i !

-

-

i

Ii

::

i

.

i

i

i

i

_

~

.

-

~

.

2

3

4 WAVELENGTH

5

6

(pm)

Figure 9.14 HgCdTe two-color, sequential-mode infrared detector spectral response characteristic at 80 K. Vt, is the bias applied to the top absorber ohniic contact with respect to the array common contact to the bottom absorber. Under positive bias (Vb >t 0 V). the z~pper photodiode (longer cutoff wavelength) is reverse-biased and producing photocurrent. Under negative bias (V~,
330

Handbook of Infrared Detection Technologies

absorbing layers is high. The response of the upper diode is limited to a spectral band defined by the bandgaps of the two absorbing layer since the bottom absorber acts as a filter to absorb photons shorter than 3.71 ~m. The spectral curves also exhibit excellent spectral separation with crosstalk below 1% outside of the narrow crossover region between 3.4 and 3.9 gm. which ensures excellent spectral discrimination in thermal imaging applications. Because the nature of the HgCdTe MBE process facilitates the synthesis of arbitrary doping and composition profiles, two-color device structures can easily be grown with varying combinations of spectral bands to meet diverse mission requirements. Figure 9.15 exemplifies the variety of spectral response that can be readily engineered in MBE-grown multispectral device structures. Precise control of the cutoff wavelengths of the two spectral bands can be ensured through use of the SE sensor for control of the HgCdTe alloy composition of the two n-type absorbing layers in the device structure. The utility of the SE sensor for significantly improved run-to-run composition control is illustrated in Figure 9.16, which summarizes the variation in the narrower bandgap (Band 2) absorber 7 7 K cutoff wavelength from target wavelengths (in the range of 10-1112m) during growth campaigns conducted both with and without SEbased feedback control of alloy composition. The deviations plotted on the

Figure 9.15 Flexibility in the production of MWIR/MWIR. I\IIA'IR/LWIR. and LWIR/LWIR two-color detector structures can be easily achieved through control of HgCdTe allo!l composition during the MBE growth process.

Status of HgCdTe ,\IBE technolog!t

3 31

Figure 9.16 Deviation of projected 77 K detector cutoff wavelengths from target wavelengths in the 1011 IJm spectral band, based on postgrowth 300 K IR transmission measurements of MBE-grown wafers. The large scatter in cutoff wavelength of wafers in the unshaded portion of the graph is t!lpical of results obtained without SE regulation of alloy composition (x-value). Subsequent wqfers (shaded region) were grown with SE feedback control for allo!t composition regulation.

ordinate are based on the difference between post-growth 3OOK FTIR transmission measurements of cutoff wavelength and the target wavelength for the particular growth, followed by conversion of the 3OOK wavelength difference into an equivalent wavelength difference at 7 7 K . Prior to the introduction of SE-based feedback control of HgCdTe alloy composition, the 77 K cutoff wavelength error is characterized by a standard deviation of 0.5 ~m. Adoption of a feedback control operating mode has considerably reduced standard deviation by roughly a factor of four, which substantially increased the yield of wafers that meet the stringent requirements on cutoff wavelength. 9.4.2 Near-infrared avalanche photodiodes

Low-noise, high-speed laser detectors sensitive to eye-safe laser wavelengths at 1.5 ~tm are critical components of next-generation laser rangefinders and laser radar systems, which have been enabled by the parallel development of smallfootprint, high-power lasers operating in this wavelength range. For accurate determination of target range with reasonable laser power, a detector/amplifier combination must provide high sensitivity at high bandwidth. For range determination with an accuracy of 1 m, the round-trip laser pulse time must be

332

Handbookof Infrared Detection Technologies

determined to an accuracy of,-~ 7 ns, which places an upper limit on the detector rise time of approximately 10 ns. For ranges on the order of 2 km and a typical optical return pulse on the order of a few nanowatts in a pulse width of 10 ns, approximately 78 photons at 1.5 5 lam wavelength must be detected. 62 Detection of such low intensity return signals requires low-noise amplification of the signal in the detector. HgCdTe, historically of interest primarily as a cryogenic photodetector material for thermal imaging applications at longer wavelengths, has recently been investigated as an alternative to III-V photodetector materials, such as GaInAsP, for high-sensitivity detection in the near-infrared. To support this need for complex multilayer heterostructure HgCdTe APDs, MBE growth techniques have been developed for deposition of high crystalline quality Hg l_xCdxTe alloys with ,x~> O. 6. Insertion of APDs in place of unity-gain PIN detectors can significantly improve the signal-to-noise performance of laser ranging systems, which translates into either improved range or decreased laser power requirements. 63 The noise characteristics of the APD are governed by two distinct sources: (1) shot noise associated with both the diode dark current and the opticallygenerated current, and (2) excess noise deriving from the statistical nature of the multiplication process. The shot noise component arising from the detector dark current is controlled by extrinsic sources such as electrically active semiconductor crystal defects, either in the bulk portion of the detector or at passivated surfaces of the device. The excess noise is determined by an intrinsic property of a particular semiconductor system, the ratio of the hole impact ionization coefficient (13) to the electron impact ionization coefficient (~), k= 13/0~. In comparison with HgCdTe-based APDs, commercially available APDs constructed in Si, Ge, or III-V materials such as GaInAsP have significant drawbacks when applied to the detection of eye-safe laser wavelengths near 1.5 ~tm. Si has near-ideal noise properties due to its non-unity k-value, but does not offer significant optical absorption at 1.5 lam. InGaAs, Ge and InP offer inferior noise performance at moderate frequencies (tens to hundreds of MHz) and limited gain-bandwidth product at higher frequencies (GHz and above) due to their near-unity k values. 63 HgCdTe alloys, because of their unique energy band structure, can be designed to provide low-excess-noise carrier multiplication. The advantage of HgCdTe alloys for low-excess-noise avalanche multiplication derives from a 'resonance' in hole impact ionization coefficient resulting from the material's band structure at a particular x-value around X - - ( ) . 7 3 . 64"6s At this alloy composition, the semiconductor's energy bandgap (E.q) and valence-band spin-orbit splitting (Ao) are equal, which induces a large ratio of hole-to-electron ionization coefficients, an effect which has also been previously observed in the GaAISb material system. 66 In order to take advantage of the flexibility of the HgCdTe alloy system to provide both efficient optical absorption of 1.5 5 pm radiation at x,~().6 and hole-dominant avalanche multiplication at x=().73, a separate absorption and multiplication-APD (SAM-APD) architecture has been adopted. With the addition of ohmic contact layers, the alloy composition of epitaxial layers in a single HgCdTe SAM-APD device structure can span a range as wide as

Status of HqCdTe ,\IBE technoloq!t

333

x = 0 . 2 - 0 . 8 . Because this range is considerably broader than typically encountered during the growth of conventional passive MWIR and LWIR devices, it has been necessary to utilize SE to perform in situ monitoring of alloy composition during the growth of HgCdTe APD device structures. Recent experiments 2~' indicate that SE has excellent sensitivity and run-to-run repeatability when applied as a sensor for surface HgCdTe composition during the growth of near-IR APD device structures. Figure 9.1 7 gives an indication of the noise floor of the SE-based composition sensing during monitoring of a wide bandgap, x = 0 . 5 9 6 layer, for which the standard deviation in derived x'-value is 0 . 0 0 0 1 7 during the course of the m e a s u r e m e n t in which the surface was held static under a stabilizing Hg flux. A deviation in .x'-value of this magnitude during epitaxial growth would correspond to a variation in CdTe flux of only ().()3%, which is substantially below what can be sensed by any other available technique. Precise regulation of HgCdTe alloy composition in the APD device structure is most critical in the avalanche gain layer. Although the range of x'-value over which hole impact ionization resonance occurs is not precisely known for HgCdTe, prior m e a s u r e m e n t s in the GaAISb system suggest that the range of A/Eq over which resonance occurs might be as small as _+().()2. This range would correspond to an .~*-value variation of _+().()1 in HgCdTe. The extent to which SE can be used to achieve this level of precision in run-to-run control of HgCdTe alloy composition has been assessed during the course of a campaign of more

0.602

i

I

I

!

]_

2-1962 STATIC SURFACE

0.600

rUJ 1.:::)1.1 0.598 ,_1 ><=: 0.596

....

" ~

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

q

x

I

0.594 0.592

"

....................... IX Std. De v. = 0.00017

365

I 370

I 375

I 380

i 385

390

T I M E (mins) Figure 9.1 7 The background noise h'vel associated with the SI:. determination of H[ICdTe allo!i composition is approx'imateh3 0 . 0 0 0 1 7 . In this anal!Isis, a static ( non-~lr~,wim.1) x=(). 5t)6 H~ICdTe stt(lace was held at constant temperature under a ~'otlslatll stabilizinq HgJlu.r to present a constant-composition sm~lace to the SE instrumentation.

334

Handbook of Infrared Detection Technologies

than twenty APD growths. In this study, we have determined that the run-to-run variation in the w-value determination by SE is approximately +().0()24, at a 95% confidence level (two standard deviations), when compared with postgrowth IR transmission measurements, as illustrated by the data in Figure 9.18, which compares SE and IR transmission-derived compositions determined for both APD materials (x-,~().6)and MWIR device layers (.r~().3)collected during separate growth campaigns conducted in the same MBE chamber using the same SE hardware. The errors in x-estimation for these epitaxial materials are approximately the same as previously reported for SE monitoring of MBE-grown LWlR HgCdTe.11.~2 Because the SE measurement is quite sensitive to incident beam angle, it is probable that run-to-run variation in substrate wobble presently limit the composition measurement precision to +().()()2. This level of repeatability in composition determination is more than adequate for precision growth of HgCdTe avalanche multiplication layers with the anticipated +().()1 requirement for composition precision. An added benefit of the SE technique for determination of HgCdTe composition is its extreme surface sensitivity, which facilitates its use as a composition profiling technique through collection of data during the growth of complex multilayer structures. Whereas postgrowth optical characterization techniques such as IR transmission are limited to determination of the composition of only the narrowest bandgap layer in a multilayer device structure, SE can access in real-time all growing layers for which optical dielectric functions are known. An example of the composition depth profiling capability of SE is presented in Figure 0.63 ,..0.62 .o ._m 0 . 6 1

"i

......

IAPDI

I .........

l ......

0.310 ' ' ' '

:' =

0.305

~:

0.300

.....

i'''''''''l','','J

. 10 V)

E

r O~

I--

Sss

~. 0.29s >6 0 . 2 9 0

0.58 0.57 " 0.57

, 0.59

, i 0.61 X, SE

(a)

o.oo i

r~

[_,o.oo24 (2o)

0.63

_

0.285 0.285

0.295

X, SE

0.305

(b)

Figure 9.18 (a) Plot of H qCdTe alloJt ('onu~osition determined from postgrowth JO0 K IR transmission measurements vs. that determined l?om in situ SE measurements during the ,\IBE growth of x.~ O. 6 epitaxial layers. (b) Plot of HgCdTe allo!! composition determined J?om postgrowth ]OOK IR transmission measurements vs. that determinedj?om in situ S E measurements during the ,\IB ti growth of AIWI R ( x~O. ~) epitaxial la#ers. The dashed lines in the graphs indicate the 9 ~','/,, cot(tidence limits ( 2c~ measurement error) for the estimation of allo!t composition with the SE measurement techniqm'.

Status of HgCdTe AIBE technolog!l

0.85

,

I

I

2-1983

0.80 m

I

3 35

0.75

_J

>, x

0.70

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

; .....................

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

.~.-.

I

I

0.65 0.60

I 300

400

500

600

TIME (mins) Figure 9.19 Spectroscopic ellipsometr!l enal~les real-time measurement of HgCdTe allo!l composition depth profiles during the MBE growth ofro ultila!ler epitaxial structures such as this API) device.

9.19, which contains a composition profile obtained during the growth of an APD device structure that was performed without feedback control of the effusion cells. The sensitivity of the SE measurement to minute features of the ,r profile is evident in the figure. For instance, a dip in .u can be seen at t - 3 7 ( ) minutes, arising from an undershoot of the flux from the CdTe effusion cell during the transition from x=().81 to ,r-().61 5. By adopting a feedback control method with the SE composition sensor controlling the CdTe and/or Te effusion cells, it is possible to improve composition regulation to+().()() 1 at all times. Implementation of feedback control of alloy composition and substrate temperature has resulted in significant improvements in the consistency of the crystalline quality of HgCdTe APD device structures. Dislocation densities for these high-x" epilayers were frequently in excess of 2x 1() 7 cm -2 when grown without sensors in 1998, while more recent growth campaigns utilizing sensors have yielded epilayers with densities consistently below 2 • -2. This reduction in dislocation density has a significant impact on APD dark current, whose associated shot noise can dominate the noise performance of the device, if sufficiently high. Excessive dark currents can negate the improvement in overall noise performance expected from the reduced excess noise component associated with the hole impact ionization resonance. ~'~Figure 9.2() illustrates the negative impact of dislocation density on diode dark current t h r o u g h presentation of a comparison of the distribution of dark current density measured across 5 • 5 mini-arrays for epitaxial material with either high or low dislocation density. The array-median dark current is reduced by over an order of magnitude in devices fabricated in epitaxial material with dislocation density below

336

Handbook of Infrared Detection Technologies

Figure 9 . 2 0 Tile dislocation detlsil!! in ,'tlBt:,-grown API) device materials has a sign(/icant impact on the distribution of dark current across arrajts orAl'I) devices. The l~lot shows tJipical dark clirrent distributions in APD arra!lS fllbricated from epitaxial material with either high ( > 2 x 1 (I: cm 2) or low ( < 2 x 10 s cm 21 dislocation densit!l. The diode size in each case is SO lira • ~0 l~m. Also displa!led is a spec([ication of dark current for commerciall!t availal~h, III- ~' (;ah~As l ~API) devices. ~'"

2• 105 cm -2 as compared to epitaxial material with dislocation density greater than 2 • 107 cm -2. It is also important to note that the median array dark current for these recent HgCdTe APDs is comparable to that specified for commerciallyavailable GaInAsP APDs. r which has enabled us to demonstrate APDs with noise performance comparable to that of GaInAsP APDs. The MBE-grown HgCdTe diode arrays are characterized by dark currents (normalized to gain) of less than 10 nA and noise-equivalent power (NEP) levels less than 2 n W . ~'2 MBE growth technology, besides allowing for the integration of sensors that can be used to precisely control the process, also offers the added benefit of excellent lateral uniformity in HgCdTe epitaxial material quality. The lateral uniformity of source and dopant fluxes over large substrate areas translates into excellent uniformity of device electrical characteristics. In Figure 9.21. we plot a histogram of the voltage required to attain a gain of 13 in a 1 () • 1 () mini-array of MBE-grown HgCdTe APDs. The diode voltages are distributed in a n a r r o w +_0.25 V range about 91.4V. Such a n a r r o w range of required bias across a photodetector array considerably simplifies the design of bias circuitry in associated readout chip electronics. The demonstration of excellent device uniformity and low NEP device performance are important milestones for MBEgrown APD epitaxial materials and offer encouragement for continued future investigation.

Status of HgCdTe ,\1BE technolog!l

337

Figure 9.21 APD arrays fabricatedJ)'on~ ,\IBt:,-~lrown epitaxial materials exhibit excellent un(formit!l of diode gain characteristics. The histogram.lot a 10 • 10 element arra!l of ~,01~m diodes illustrates the unilbrmit!l ~[ diode bias required to achieve a gain of 1 3.

9.4.3 High-performance MWIR detectors

Infrared FPAs designed for operation in space-based surveillance applications must currently be operated at temperatures below 12()K to attain adequate noise performance. 6~ Cooling to such low temperatures requires large cryoradiators. By boosting the FPA operating temperature above 145K, significant reductions in cooling subsystem size, weight, power and cost, can be realized. In order to address this need for an increase in FPA operating temperature, an approach based on reduction of the active volume of the detector and reduction of the concentration of defects contributing to dark current generation has been pursued. ~ By reducing the active volume of the detector material, either through reduction of the device thickness or crosssectional area, the deleterious effect of current leakage mechanisms that fundamentally limit device performance can be mitigated. The improvement in detector performance that can be achieved by reduction in absorbing layer thickness can be seen by examining the dependence of diode R o A product, which is commonly employed as a device figure-of-merit, on material parameters. The thermal ]ohnson-Nyquist noise associated with an If{ detector is inversely related to the diode R.A product. In a simple onedimensional analysis of diode diffusion currents, neglecting non-ideal currents arising from effects such as surface recombination and bulk generationrecombination, the RoA product is given by ~')

338 Handbookof Infrared Detection Technolo#ies

R,,A-

9

j*

(2)

where Na is the donor doping density on the n-side of the p-n junction, n~ is the intrinsic carrier concentration, ~p is the minority carrier (hole) lifetime, q is the electron charge, k is Boltzmann's constant, T is the detector operating temperature, and d is the absorbing layer thickness. The formula indicates that diode performance should theoretically scale inversely with d. The demands placed on the crystal growth process to achieve this ultimate performance limit include: (1) the ability to reproducibly control layer thicknesses, (2) the capability for growth of backside interfaces with low recombination velocity, and (3) the ability to control absorber alloy composition so that ni is minimized for a given detector cutoff wavelength specification. By virtue of its ability to construct complex doping and composition profiles, MBE is especially well-suited to fabrication of detector structures whose performance should closely approach the ideal limit expressed in equation (2). Theoretical calculations of detector performance as a function of absorbing layer thickness indicated that absorbing layer thicknesses in the 2-5 ~tm range, with no compositional grading, would likely prove optimum. Growth of such thin, ungraded HgCdTe epitaxial films by more established techniques, such as liquidphase epitaxy, would not be possible. To establish the utility of MBE for fabrication of thin, complex multiplayer MWIR detectors, the device structures illustrated in Figure 9.22 were grown with a 140 K cutoff wavelength target of 3.8 ~tm. The impact of variations in the absorbing layer thickness over the range of 3-7 lam have been examined, along with the effect of including a wide bandgap (x=0.45) buffer layer between substrate and absorbing layer. The absorbing layer doping ([In]=2• and cap layer doping ( [ A s ] = l • -3) were held fixed for all device structures. HgCdTe epitaxial wafers were processed

Figure 9.22 High-performance MWIR device structures: (A) no wide bandgap buffer layer/71~m-thick absorbing layer, (B) wide bandgap buffer la!ler/ 5 t~m-thick absorbing layer, (C) wide bandgap buffer layer/ 3 ~m-thick absorbing layer.

Status of HgCdTe MBE technolo~,l!l

] ]9

into mesa-architecture discrete diodes and diode mini-arrays using standard wet chemical etch processes and CdTe surface passivation. The electrical performance of the three device structures is summarized in Figure 9.23, which contains plots of the temperature dependence of the R o A product. All three device structures exhibit diffusion-limited performance down to 140K; below this temperature, test equipment impedance limitations dominate the measured data. Current transport mechanisms related to generation-recombination centers and trap-assisted tunneling do not appear to be significant above 140 K, attesting to the high crystalline quality of the MBEgrown HgCdTe. To assess the impact of structural variations on diode performance, m e a s u r e m e n t s of median 140 K R(~A values were taken on a total of 6() diodes populating two mini-arrays on each device wafer. The median RoA values are summarized in Table 9.1. It is evident from the values in the table that the insertion of a high x'-value buffer layer in Structures B and C to electrically isolate the substrate from the 10 9

''

' I ' '

'

'

' ' I

' .W ' l

9 9

l

~ "

~

!A

' ' I '

.................. $!..................... , ..................... !i. . . . . . . . . . . . . .~. .] . ! , . . . ~. . . ./ . .~. . . . .

10 7

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

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

'' i i

0

~,.,. 8 1U

9

i ~

!

*! .................

:

i

1 0 6 ........................................!........o~.. ..... ' ...................................

A

"

E

r,,,1 I

i%

10s

0

".......

~Z,',z.~. . . . . . . . .

L~

........................................~ ..............i.....................i..................... :]"................. q~ . .

E

1 0 4 ..................................~ ....................!.....................i.....................~.................

0 0

10

3

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

. .........

=9 10

2

~

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

~

Q b < 10

cm

S

A r e a = 40 p m x 40 p m

.................. ~ . . . ~ ..........................

id~

10 1

J

lO o

~

"A"

1-]

"B"

:

,,,i,,,i,,,i,,,i,,,

,,

2

~

0

4

6 1 O00/T

8

10

:

12

14

(l/K)

Figure 9.2J Temperature dependence of R~A product for 40 lira size diodes filbricated from Structure A, B. and C wafers with 140 K cutoff wavelengths in the J.4- J.8 ~m ran qe.

340

Handbook of Infrared Detection Technologies

Table 9.1 S u m m a r y of a r r a y - m e d i a n RoA product for Structure A, B, a n d C wafers. The RoA product for Structure A, corrected to 3.75 ~tm c u t o f f wavelength, has been e s t i m a t e d by assuming diffusion-limited performance, for w h i c h RoA xl/n~ z Structure

14() K cutoff

Median RoA ( 140 K )

A B

3 . 4 0 pm ( ~. 75 ) 3.75

C

3.75

3 x I ()~' f2 cm 2 ( ~ 2 x 1 ()-3) 1 . 2 x l ( ) ~' 2 . 3 x 1 ( ) ~'

absorbing layer appears to offer a substantial performance improvement relative to Structure A, which has no buffer. The improved performance associated with the buffer layer derives from a reduction in back-surface recombination velocity, which effectively increases the minority carrier lifetime. Inclusion of such performance-enhancing buffer layers can be readily achieved with MBE. In examining the results in Table 9.1, it is also evident that further improvement in the diode RoA product has been achieved through reduction in the absorbing layer thickness, from 5 pm in Structure B to 3 pm in Structure C. This performance enhancement is consistent with the diffusion-limited performance characteristic expressed in equation (2). Variation of the detector absorbing layer thickness, besides affecting the diode RoA product, can also significantly modify the shape of the detector spectral response characteristic. Figure 9.24 compares theoretical calculations of single-pass absorption and experimental diode spectral response data for absorbing layer thicknesses of 3, 5

I. 0 0.8 <

0.6

<

0.4-

.__~. .... .,__~,,~.,,._~r

":"%@ ~i\ ". I"--~,'

] . 0 ~ . ~ , I . _ _ I[

_

2

<

d=7 .urn ,,,~,,~I

i" .... '""-. ''z ~1 d:5 pm (,,B,,)I oz 0.8 , ).l F-'',, o_,. I.=~ . m (,,c,,)I '..I. a_ 0.6"' z

0.2 -I-r.Eo.u

'

i

IT=I 20K

fJ

0.0 2.5

I 3.0

I 3.5

I 4.0

WAVELENGTH ( u m )

(a)

4.5

o

0.4-

""

0 2 "

0.0

2.5

EXPERIMENTAL DIODEDATA T= 120K

I 3.0

I 3.5

1 4.0

4.5

WAVELENGTH (/~m)

(b)

Figure 9.24 (a) Calculated 120 K single-pass absorption spectra for x--.O. J ~ HgCdTe fihns with thickness of 3, 5, and 7 pro. (b) Experimental 120 K diode spectral response data for actual device structures A (absorber thickness, d,= 7 pro), B (d= 5 pm ), and C (d= ~ pm ).

Status of HgCdTe ,~,IBE technolo#!l

341

and 7 lam. The theoretical calculations in Figure 9.24(a) were performed using published values for the optical constants of HgCdTe at 120 K. 7~) Substantial softening of the spectral characteristic is evident in both the calculated and experimental diode spectra when the absorbing layer thickness is reduced to 3 ~m. Because such broadening of the diode turn-on characteristic can present difficulties in optical system design, the data suggest that an absorbing layer thickness of "5lum represents a reasonable compromise between between electrical and optical performance goals. The improvement in diode performance, relative to thick-base detectors grown by LPE, demonstrated to date using MBE for growth of these thin-base detectors translates to approximately a IOK increase in operating temperature. Further increases in operating temperature from improvements in device processing and lateral architecture are presently being pursued. 9.4.4 Large-format arrays on silicon substrates During the course of the past decade, one of the most significant emerging trends in HgCdTe FPA development has been the migration of HgCdTe epitaxial growth from lattice-matched CdZnTe substrates to Si substrates. Both economic and technological factors are motivating this transition in FPA platform from costly and fragile CdZnTe substrates to more mechanically robust Si substrates. FPA linear dimensions are presently constrained by limits in the size of state-of-theart CdZnTe substrates: formats larger than approximately 1 ()()()x 1 ()()() are not currently feasible because the FPA dimension would exceed the largest available substrate size. Even three-inch diameter Si substrates are insufficiently large to enable production of a single 2()48 • 2()48 format array with 2 5 ~tm unit cell size, thus necessitating the capabilities of production MBE systems suitable for handling of five-inch diameter Si substrates. Besides enabling larger array formats, processing of HgCdTe FPA materials on Si substrates also yields significant cost reduction benefits due to the increase in patternable area. As an example, a six-inch diameter Si substrate offers a factor of seven increase in available area over a conventional 3() cm 2 CdZnTe substrate. Additional benefits conferred by use of Si substrates include an automatic thermal expansion match to Si readout chips, compatibility with wafer handling~processing equipment developed for cost-efficient manufacturing in the Si microelectronics industry, and substantially lower cost. Because of these potential advantages, methods for epitaxial growth of HgCdTe detector array material on Si substrates has been vigorously pursued as a research topic for the past decade by several organizations, s'71-76 Because the lattice constants of Si and HgCdTe are significantly mismatched, it is undesirable to directly grow active HgCdTe device layers on Si since the generation of a high density of threading dislocation segments in active device layers would compromise device performance. Instead, it has been customary to first deposit a thick ( 5 - 1 0 microns thickness) buffer layer of CdTe or CdZnTe to allow dislocation annihilation processes to reduce the density of dislocation threading segments. Significant improvements during the past decade in the quality of

342 Handbookof Infrared Detection Technoloqies HgCdTe device structures grown by MBE on bulk CdZnTe(112) substrates has motivated efforts to deposit CdTe and HgCdTe on (112)-oriented Si substrates. The first report of direct II-VI heteroepitaxy on Si(112) made by Hughes Research Laboratories in 1995 ~s featured the use of a ZnTe initiation layer to deposit twin-free CdTe epilayers with X-ray rocking curve FWHM < 75 arc-sec and EPD of 2 x 106 cm -2. Subsequent investigations of the growth of CdTe on Si(112) have focussed on the effects of growth initiation procedure, growth interruptions for thermal annealing, and optimal growth temperature. "7-8~ Several research groups have reported CdTe/Si(ll2) EPD results in the 2-5 • 1 0 s c m - 2 range, which is approximately one order of magnitude larger than that of commercially available CdZnTe substrates at the present time. Development of processes for nucleation of low-defect-density CdTe( 1 ] 2 ) films on Si substrates has facilitated deposition of device-quality HgCdTe on Si( 112 ) by several laboratories. Threading dislocation densities in MWIR H g C d T e / C d T e / S i films are generally reported to be in the range of 5xl()~'cm --' to 2 • -2. approximately a factor of 10() above that of MWIR HgCdTe films deposited on lattice-matched CdZnTe substrates. LWIR HgCdTe films deposited on CdTe/Si substrates generally exhibit slightly lower EPD values around 2• 1() 6 c m - 2 . Postgrowth thermal cycle annealing at 49()~C has been reported to reduce the EPD of heteroepitaxial HgCdTe by approximately an order of magnitude, to values in the 2.6 to 11 x l 0 s c m -2 range, s l''s2 Scaling of the HgCdTe on Si growth process from a research and development effort on two-inch and three-inch diameter substrates at HRL Laboratories, LLC. to a manufacturing environment at Raytheon Vision S y s t e m s 71"72 using four-inch and five-inch diameter substrates required the development of techniques for control of substrate temperature using 'freely mounted' substrates. This approach, which limits the run-to-run variation in substrate temperature to +0.5~ allows for growth of the entire HgCdTe device structure on a CdTe buffer layer, starting from a bare Si substrate, without requiring removal of the substrate for dicing and remounting between the CdTe and HgCdTe growth steps. Development of techniques for control of substrate temperature on freely mounted substrates that can also be scaled to arbitrarily large substrates is an extremely important element of the HgCdTe/Si growth technology that permits future exploitation of larger Si substrates and production MBE equipment. In addition to uniformity of substrate temperature across large areas, uniformity of HgCdTe alloy composition and crystal quality across large substrates have emerged as significant technology challenges. These challenges have been addressed through optimization of the uniformity of source tluxes of Hg, Cd, and Te in the MBE growth chamber. Void defect densities, which are controlled primarily by substrate temperature and Hg flux. can be reduced to less than 2 0 0 c m -2 across an entire three-inch diameter Si substrate, with proper design of the Hg source delivery nozzle to ensure uniform flux across the rotating substrate area. Proper design of the Hg source nozzle also results in excellent composition uniformity across large substrates, as illustrated in Figure 9.25, which displays radial composition profiles derived from 3()()K IR transmission

Status of HgCdTe MBE technolog!l

0.40

-

' ' ' ' l ' ' ' ' l ' ' ' ' l ! ! i i

! i

0.38

.........

o o =

O

1! i

0.36

0.34

.

_

0.32

.

-

"

.

.

.

.

.

.

.

.

.

! i

! i

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

i

Ii i i

i

-

i

i

-

34 3

I

O,d HQ ,ource ! ax = 0"0016

1

.

i

-

i

~

-

i

i

i

-

i

i

i

-30

-20

-10

0

i i

i i

i

i

10

20

30

DISTANCE FROM CENTER (mm)

(a)

z 0 rj) O

0.240

]I I !

0.235

L

O.

L ...... E

'L

....

i ....

i ....

: .i Old Hg S o u r c e l l "

Ox=000 1, ' i

i

:

"

[ ....

! ....

iJ

..

i =

i

i

~ ' ~-~i~'v

-

i i

:

i T

!

~

.......

;

_1

I ,,~" 1

]

-1 1

1

=E O o

0.230

r i i i i i 1 I - . . . . . . . . . i .......... ~ .......... g .......... ~ .......... ~ ....... ~ 1

0

0.225

"~ T T i ~-i i / ~ . ......... II.......... 4I .......... ~I ....... I m p r o v e d Hg S o u r c e

-r

i

i

I,,,,,

i

i i i,,,,i,,,,

i

i

0

5

10

15

20

[ 0.220

G x = 0.0008

i

i

L

i

i

DISTANCE

:,

i i,,,,i,

FROM

i

i

-I

25

30

i ,,,i,,,,/

CENTER

j

(mm)

(b) Figure 9.2 ~ HgCdTe composition un([i~rmit!l has beneJited.lrom improvements in Hg source design that increase the lateral un(formit!l of H[I .flux. (a) Composition un(formit!! in x~,O. ~ ,\I~VIR la!lers is approximatel!l 0.0006 across Hlree-inch diameter Si sul~stratcs. ( b) Cotnposition un(formit!l in x,~0.2 :~ L WIR la!lers is approximatel!! O. O00g.

344 Handbook of Infrared Detection Technologies measurements on three-inch diameter Si substrates. For MWIR alloy composition, the standard deviation in x" is ().()()06, corresponding to an equivalent 7 7 K cutoff wavelength spread of less than+().01 ~m across the wafer. For the LWIR data in Figure 9.2 5(b), the .r-value standard deviation is 0.0()08, which translates to a 77 K cutoff wavelength spread of_+0.09 ~m. Much of the current research activity relating to device performance in this technology area is devoted to demonstration and assessment of large-format FPA performance in the MWIR spectral band, for comparison with detectors fabricated from InSb. Additional work seeks to address scaling issues for migration to even larger substrate size beyond four-inch diameter. A variety of array formats, including 128x 128 (4() ~m pitch), 6 4 0 • (2() ~m pitch), and 1 0 2 4 x 1024 (27 lam pitch) have recently been fabricated. Figure 9.26 illustrates the essential elements of the epitaxial layer structure and processed device architecture. The HgCdTe device layers are integrated with the Si substrate using intermediate buffer layers of ZnTe and CdTe that enable the HgCdTe epilayers to be grown with a dislocation density in the ( 1-5)x 1 ()~ cm -2 range. 71'72 Adoption of an all-MBE process for production of the HgCdTe/Si device structure results in considerable process simplification relative to alternative approaches. Starting with a bare Si substrate, the entire epitaxial device structure, including both CdTe buffer layer and HgCdTe device layers, can be grown in a single lowtemperature growth run in a single MBE chamber for highest throughput. Scaling of this process to accommodate even larger Si substrates is anticipated to be straightforward because of the ready availability of production-scale compound semiconductor MBE systems designed for growth on six-inch and larger diameter substrates. The processing of the mesa-architecture devices on HgCdTe/Si epiwafers, depicted in Figure 9.26, utilized standard processing techniques already well-developed for HgCdTe detectors on CdZnTe substrates. In the future, utilization of Si substrates should also significantly enhance the

Figure 9.26 Cross-section schematic illustrates the use of a ZnTe/CdTe buffer la!ter structure to integrate the MWIR HgCdTe p-n device onto a Si substrate. The SE,\! photo shows a set of 40 i~m-unil-cell pixels taken from a 128 x 128format HgCdTe-on-Sijbcal plane arra!! processed at Ra!ltheon Vision S!lstems from ,\IBEgrown epitaxial wafers.

Status of HgCdTeMBE technolog!! 345

throughput of such manufacturing processes by permitting increased use of automated wafer handling in place of expensive touch labor. A considerable amount of characterization of the electrical performance of MWIR HgCdTe/Si detectors with cutoff wavelengths in the 3.5-613m range has been conducted, ~'43 with the results consistently supporting the performance benefits of the technology. In this spectral range, the performance of HgCdTe detectors on Si substrates has been observed to be equivalent to that of devices grown on CdZnTe substrates for temperatures above about 120 K, as exemplified by the temperature-dependent R o A data in Figure 9.27 for devices with 120 K cutoff wavelength of 3.7 gm and 4.6 gm. A diffusion-limited current model is shown to offer an excellent fit to the experimental data in Figure 9.2 7. The performance of HgCdTe/Si detectors across the 3.5-6 l.tm spectral range has been systematically investigated by the MBE group at Raytheon Vision Systems. 8 Figure 9.28(a) contains a summary of the array-median R o A product from a variety of MWIR HgCdTe/Si wafers, processed at Raytheon Vision systems, plotted as a function of 140 K cutoff wavelength . The figure also contains a plot of a calculated performance limit, based on the assumption of minority carrier lifetime limited by radiative recombination

O

on Si (kco = 3 . 7 2 / a m ) I on CdZnTe (;~co = 3 . 9 7 / a m ) I - - - Diffusion Current Model I

0

- - - Diffusion Current Model

10 8

10 8 ",,' I II ~

on Si (~'co : 4 . 6 2 / a m ) on CclZnTe (~co : 4 . 6 8 / a m )

t .......

l"

' " ~' '~'~ ' '

.

i

10 6 IM

E (J i

E JE:

O v

s

'

0

10 6 N

10 4

E 0 i

10 4

E r 0

10 z

10 2

10 0

10 0 (~"

iI

]C)~< I 0 ' 2 c m -z s-' l

i

1 0 -2 4

8

1 ooo/-1- (1/K) (a)

12

1 0 -2

,l,,l,,,,l,,,,,,,lJ,,

4

8

12

1000/T (1/K) (b)

Figure 9.27 Comparison of the performance of ,~,IWIR HgCdTe detectors grown b!! MBE on latticemismatched Si substrates with that of devices grown on conventional lattice-matched CdZnTe substrates. The temperature dependence of the RoA product, measured at zero field-of-view, for MBE-grown detectors on CdZnTe and Si substrates is compared for HgCdTe/Si cutoff wavelengths of: (a) 3.72 ~m and (b) 4,62 ~m. In each plot, the experimental data are also compared with model calculations that assume the diode impedance is limited by diffusion currents from the n-side of the device.

346

Handbook of Infrared Detection Technologies

Figure 9.28 (a) Comparison plot of 140 K arra!t median R~IAproduct vs. cutoff wavelength for 40 ~m-pixel MWIR detectors processed at Raytheon Vision S!lstems from 3,IBE HgCdTe/Si. MBE HgCdTe/CdZnTe. and LPE HgCdTe/CdZnTe epitaxial materials. The perlormance of the devices is also compared to a trendline calculated under the assumption of minorit!l carrier lifetime limited b!l radiative recombination. (b) The temperature dependence of the 1R~Aproduct for HgCdTe/Si detectors ranging in size from ~0 ~m to 800 Bm with 140 K cutoff wavelength of 4.[q Bin.

processes. The performance of the HgCdTe/Si devices is seen to be within an order of magnitude of this radiative limit and more closely approaches this limit t h a n the historical devices grown by LPE on lattice-matched bulk CdZnTe substrates, a remarkable achievement for a comparatively new epitaxial technology. The temperature dependence of the RoA product, displayed in Figure 9.28(b), illustrates typical behavior of these HgCdTe/Si detectors, wherein the RoA product rises with falling temperature in a diffusion-limited m a n n e r down to approximately 125 K. Below 125 K, the characteristic slope decreases, suggesting the dominance of a generation-recombination current mechanism, possibly associated with the high HgCdTe/Si dislocation density, which is approximately 5 x 106 cm -2, more than an order of magnitude higher than for HgCdTe grown on lattice-matched CdZnTe. Fabrication of HgCdTe devices on Si substrates results in a slight noise penalty, relative to devices grown on CdZnTe substrates with LPE. Figure 9.29 compares noise plots, collected at 180 K, for an LPE-grown device with 4.8 lJm cutoff and an MBE-grown HgCdTe/Si device with 4.6 lum cutoff. The increased 1/f noise level in the HgCdTe/Si device, which is likely due to the high dislocation density, actually has only a minimal effect on FPA performance. To b e n c h m a r k the device performance of HgCdTe/Si arrays against industrystandard MWIR InSb arrays, detailed m e a s u r e m e n t s of HgCdTe/Si array operability have been conducted on 128 • 128 format FPAs with 40 ~m unit cell size. InSb FPAs, although lacking cutoff wavelength tunability and the ability to operate at high temperatures, are based on a more m a t u r e technology that has excellent sensitivity, uniformity, and operability. Figure 9.30 compares the distribution of responsivity for a HgCdTe/Si array with a cutoff wavelength of 5.7 ~m to an InSb array. The InSb array exhibits a 99.9% operability, while the HgCdTe/Si array has an operability of 98.0%, using an operability definition as the percentage of pixels

Ntatus of H~ICdTe AIBE technolo~l!l

34 7

Figure 9.29 Comparison plots of the noise spectral densit!! for ,\IBI" HgCdTe/Si (;.,.,,=4.6 I~m) and LPE HgCdTe/CdZnTe (;,,,,,=4.811m) devices, measured at 18OK. under reverse bias var!ling between 0 and - 5 0 inV.

Figure 9.30 Comparison of the histograms of detector responsivitlt for 128x 128 format HgCdTe/Si (),,,o=5.7 ptm) and InSb FPAs. The lower relative responsivit!l of the H~3CdTe/Si arra!t is due to the absence of an antireflection coating.

with responsivity in the range of ().7-1.4 times the array mean. While the InSb array exhibits slightly higher operability and uniformity, it is important to note that the HgCdTe/Si array was operated at a significantly higher temperature of 106 K vs. 89 K for the InSb array, a temperature increase that is highly desirable for IR system operation. The observed ~ 9 8 % operability of the HgCdTe/Si array

348

Handbookof Infrared Detection Technologies

appears to be consistent with a Poisson distribution of void defects, which are typically present at a level of about 1 ()0() cm -2 in these device layers, across the array of 40 ~m diodes. Since void defect densities below 2 0 0 c m -2 have been attained in a significant fraction of films grown on both three-inch and four-inch Si substrates, it is anticipated that reduction of the operability loss due to voidinduced device degradation will follow from further development of the production MBE growth process that allows for consistent growth of layers with void defect densities below 20()cm -2. Recent reports of HgCdTe/Si arrays with operability > 99% suggest that significant progress is being made in this area. 83 In addition to the impressive results on 128 • 128 format FPAs, high-quality imagery from a 6 4 0 • format HgCdTe/Si FPA has also been demonstrated. ~ The sensitivity and operability of this FPA compares favorably with state-of-theart InSb FPAs. The rapid progress demonstrated to date in MBE growth and processing of HgCdTe/Si epitaxial wafers indicates that economical production of arrays with formats significantly larger than currently feasible with CdZnTe or InSb substrates will be achieved in the near future.

Acknowledgements The authors programmatic, to list, at both summarized in

wish to acknowledge the extremely valuable technical, and managerial contributions of many colleagues, too numerous HRL Laboratories, LLC and Raytheon. The remarkable progress this chapter is due entirely to the excellence of their efforts.

References 1. W. D. Lawson, S. Nielsen, E. H. Putley and A. S. Young, J. Phys. Chem. Solids 9,325(1959). 2. T. Tung, J. Cryst. Growth 86, 161 ( 1988). 3. O. K. Wu, T. J. de Lyon, R. D. Rajavel and J. E. Jensen, in Narrow-gap II-VI Compounds for Optoelectronic and Electromagnetic Applications, edited by Peter Capper (Chapman & Hall, London, United Kingdom, 1997), pp. 9 7 - 1 3 0 . 4. J. M. Arias, in Narrow Gap Cadmium-based Compounds, edited by Peter Capper (INSPEC, London, United Kingdom, 1994), pp. 3 0 - 3 5 . 5. R. D. Rajavel, P. D. Brewer, D. M. Jamba, J. E. Jensen, C. LeBeau, G. L. Olson, J. A. Roth, W. S. Williamson, J. W. Bangs, P. Goetz, J. L. Johnson, E. A. Patten and J. A. Wilson, J. Cryst. Growth 2 1 4 / 2 1 5 , 1100 (2000). 6. T. J. de Lyon, B. Baumgratz, G. Chapman, E. Gordon, A. T. Hunter, M. Jack, J. E. Jensen, W. Johnson, B. Johs, K. Kosai, W. Larsen, G. L. Olson, M. Sen and B. Walker, SPIEProc. 3 6 2 9 , 256 (1999). 7. T. J. de Lyon, J. E. Jensen, I. Kasai, G. M. Venzor, K. Kosai, ]. B. de Bruin and W. L. Ahlgren, J. Electron. Mater. 31 220 (2002).

statz~s of HgCdTeMBE technology 349

8. J. B. Varesi, R. E. Bornfreund, A. C. Childs, W. A. Radford. K. D. Maranowski, J. M. Peterson, S. M. Johnson, L. M. Giegerich, T. J. de Lyon, and J. E. Jensen, J. Electron. Mater. 30, 566 (2001). 9. B. K. Wagner, D. Rajavel, R. G. Benz II and C. J. Summers, J. Vac. Sci. Technol. B9, 1656 (1991). 10. A. Million, L. DiCioccio, J. P. Gailliard and J. Piaguet, J. Vac. Sci. Technol. A6, 2813 (1988). 11. G. L. Olson, J. A. Roth, P. D. Brewer, 1R. D. Rajavel, D. M. Jamba, J. E. Jensen and B. Johs, J. Electron. Mater. 2 8 , 7 4 9 (1999). 12. D. D. Edwall, J. Phillips, D. Lee and J. Arias, J. Electron. Mater. 30, 643 (2001). 13. F. Aqariden, W. M. Duncan, H. D. Shih. L. A. Alrneida and M. J. Bevan, J. Electron. Mater. 2 8 , 7 5 6 (1999). 14. L. A. Almeida, J. N. Johnson, J. D. Benson, J. H. Dinan and B. Johs, J. Electron. Mater. 27, 50() (1998 ). 15. F. M. Daraselia, C. H. Grein, S. Rujirawat, B. Yang, S. Sivananthan, F. Aqariden and H. D. Shih, ]. Electro11. Mater. 2g, 74 3 (1999). 16. P. S. Wijewarnasuriya, M. Zandian, D. B. Young, J. Waldrop, D. D. Edwall, W. V. McLevige, D. Lee, J. Arias and A. I. D'Souza, ]. Electron. Mater. 28, 649 (1999). 17. S. Sivananthan, P. S. Wijewarnasuriya, F. Aqariden, H. R. Vydyanath, M. Zandian, D. D. Edwall and J. M. Arias, ]. Electron. Mater. 26, 621 (1997). 18. B. Johs, C. Herzinger, J. H. Dinan, A. Cornfeld. J. D. Benson, D. Doctor, G. Olson, I. Ferguson, M. Pelczynski, P. Chow, C. H. Kuo and S. Johnson, Thin Solid Films 3 1 3 - 3 1 4 , 4 9 0 (1998). 19. J. A. Roth, T. J. de Lyon and M. E. Adel, Mater. Res. Soc. S!jnlp. Proc. 3 2 4 (Pittsburgh, PA: Mater. Res. Sot., 1994), pp. 3 5 3 - 3 5 8 . 20. J. E. Jensen, J. A. Roth, P. D. Brewer, G. L. Olson, J. J. Dubray, O. K. Wu, R. D. Rajavel and T. J. de Lyon, J. Electron. Mater. 2 7 , 4 9 4 (1998). 21. E. W. Hellman and J. S. Harris, Jr., J. Cr!jst. Growth 81, 38 (1987). 22. CI Systems, Inc., 30961 West Agoura Rd., Suite 1 ()9, Westlake Village, CA 9 1 3 6 1 - 4 6 1 8 USA. 23. C. C. Hang, K. J. Astrom and W. K. Ho, IEE Proc. D: Control Theory and Applications 13 8, 111 (1991 ). 24. M. Daraselia, J. W. Garland, B. Johs, V. Nathan and S. Sivananthan, ]. Electron. Mater. 3 O, 637 (2 ()() 1 ). 25. J. A. Woollam Co. Inc., 645 M Street, Suite 1 ()2 Lincoln, NE 6 8 5 0 8 USA. 26. T. J. de Lyon, G. L. Olson, J. A. Roth, J. E. Jensen, A. T. Hunter, M. D. Jack and S. L. Bailey, J. Electron. Mater. 31 688 (2()()2). 2 7. S. M. Johnson, J. L. Johnson, W. J. Hamilton, D. B. Leonard, T. A. Strand, E. A. Patten, J. M. Peterson, J. H. Durham, V. K. Randall, T. J. de Lyon, J. E. Jensen and M. D. Gorwitz, ]. Electron. Mater. 29, 68()(2()()()). 28. R. D. Feldman, S. Nakahara, R. F. Austin, T. Boone, R. L. Opila, and A. S. Wynn, Appl. Ph~As. Lett. 51, 1 2 3 9 ( 1987). 29. A. Million, L. DiCioccio, J. P. Gailliard and J. Piaguet, J. Vac. Sci. Technol. A6, 2813 (1988).

350 Handbookof Infrared Detection Technologies 30. P. M. Amirtharaj and F. H. Pollak, Appl. Ph!ts. Lett. 4 5 , 7 8 9 (1984). 31. R. D. Feldman, 1R. L. Opila and P. M. Bridenbaugh, J. Vat'. Sci. Technol. A3, 1988(1985). 32. J. D. Benson, A. B. Cornfeld. M. Martinka, J. H. Dinan, B. Johs, P. He and J. A. Woollam, ]. Cryst. Growth 1 7 5 / 1 7 6 , 6 5 9 (1997). 33. D. R. Rhiger, J. Electron. Mater, 22, 887 ( 1992 ). 34. H. Arwin, D.E. A s p n e s a n d D . R. Rhiger, l. Appl. Ph!ls. 54, 7132 (1983). 35. T. J. de Lyon, D. Rajavel, S. M. Johnson and C. A. Cockrum, App1. Ph!ls. Lett. 66,2119(1995). 36. L. A. Almeida, Y. P. Chen, J. p. Faurie, S. Sivananthan, D. J. Smith and S.-C. Y. Tsen, I. Electron. Mater. 2 5, 14()2 i1996). 37. S.-C. Y. Tsen, D. J. Smith and S. S i v a n a n t h a n . Mater. Res. Sot'. S!llnp. Proc. 4 4 0 , (Pittsburgh, PA: Mater. Res. Sot'., 1997t. p. 265. 38. H.-Y. Wei, L. Salamanca-Riba and N. K. Dhar, Mater. Res. S!tmp. Proc. 4 8 7 , (Pittsburgh, PA: Mater. Res. Sot'., 1998). p. 6()7. 39. N. K. Dhar, C. E. C. Wood, A. Gray, H.-Y. Wei, L. Salamanca-Riba and J. H. Dinan, J. Vac. Sci. Technol. B I 4 , 2 3 6 6 (19961. 40. S. Rujirawat, D. J. Smith. J. p. Faurie, G. Neu, V. N a t h a n and S. S i v a n a n t h a n , J. Electron. Mater, 27, 1()47 (1998t. 41. P. Mackett, in Properties of Narro~v Gap Cadtniz~tn-based Conlpounds, edited by Peter Capper (INSPEC, London. United Kingdom. 1994 ~. pp. 1 8 8 - 1 9 1 . 42. S. M. Johnson, D. R. Rhiger, J. p. Rosbeck. J. M. Peterson, S. M. Taylor and M. E. Boyd, I. Vac. Sci. Technol. BIO, 1499 (1992). 43. T. J. de Lyon, R. D. Rajavel. J. A. Vigil. J. E. Jensen, O. K. Wu, C. A. Cockrurn, S. M. Johnson, G. M. Venzor. S. L. Bailey, I. Kasai. W. L. Ahlgren and M. S. Smith, I. Electron. Mater. 27, "55() I 1998 ). 44. L. Zhao, J. Speck, R. D. Rajavel, J. E. Jensen, D. Leonard, T. Strand and W. Hamilton, J. Electron Mater. 29, 732 (2()()()). 45. L. H. Zhang and C. J. Summers. ]. Electron. Mater. 27, 634 {1998). 46. D. Chandra, F. Aqariden, J. Frazier. S. Gutzler, T. ()rent and H. D. Shih, J. Electron. Mater. 29, 887 (200()). 47. P. S. Wijewarnasuriya, M. Zandian, D. B. Young, J. Waldrop. D. D. Edwall, W. V. McLevige, D. Lee, J. Arias and A. I. D'Souza. ]. Electron. Mater. 28, 649 (1999). 48. E. C. Piquette, M. Zandian, D. D. Edwall and J. M. Arias, J. Electro11. Mater. 3 0 , 6 2 7 (2001). 49. D. D. Edwall, M. Zandian, A. C. Chen and J. M. Arias, ]. Electrotl. Mater. 26. 493(1997). 50. P. S. Wijewarnasuriya, S. S. Yoo, J. P. Faurie and S. Sivananthan, J. Electron. Mater. 2~, 13()() (1996 j. 51. R. D. Rajavel, D. M. Jamba, O. K. Wu, J. A. Roth, P. D. Brewer, J. E. Jensen, C. A. Cockrum, G. M. Venzor and S. M. Johnson, ]. Electron. Mater. 28, 141 (){ 1996 ). 52. J. R. Meyer and F. J. Bartoli, J. Vat. Sol. Techllol. 2 1 , 2 3 7 (1982). 53. J. A. Wilson, E. A. Patten, G. 1R. Chapman, K. Kosai, B. Baumgratz, P. Goetz, S. J. Tighe, 1R. F. Risser, R. Herald, T. Tung and W. A. Terre, S'PIE Proc. 2 2 7 4 , 117 (1994).

Status of HgCdTe MBE technolog!l

3 51

54. R. D. Rajavel, D. M. Jamba, O. K. Wu, ]. E. ]ensen, J. A. Wilson, E. A. Patten, K. Kosai, P. Goetz, G. R. Chapman and W. A. Radford, J. Cr!tst. Growth 1 7 5 / 1 7 6 , 653(1997). 55. R. D. Rajavel, D. M. Jamba. J. E. Jensen. O. K. Wu, C. Lebeau, J. A. Wilson, E. Patten, K. Kosai, J. Johnson. J. Rosbeck. P. (;oetz and S. M. Johnson, J. Electron. Mater. 2 6 , 4 7 6 (1997). 56. A. C. Chen, M. Zandian, D. D. Edwall. R. E. De Wames, P. S. Wijewarnasuriya, J. M. Arias, S. Sivananthan. M. Berding and A. Sher. ]. Electron. Mater. 2 7, 595 ( 1998 ). 57. S. Sivananthan, P. S. Wijewarnasuriya. F. Aqariden, H. R. Vydyanath, M. Zandian, D. D. Edwall and J. M. Arias, J. Electrotl. ,\later. 26, 621 (1997). 58. A. C. Chen, M. Zandian, D. D. Edwall, R. E. DeWames, P. S. Wijewarnasuriya, J. M. Arias, S. Sivananthan, M. Berding and A. Sher, ]. Electron. Mater. 2 7 , 5 9 5 (1998). 59. T. J. de Lyon, J. A. Vigil, J. E. Jensen, O. K. Wu, J. L. Johnson, E. A. Patten, K. Kosai, G. Venzor, V. Lee, and S. M. Johnson, ]. Vac. Sci, Technol. B I 6 . 1321 (1998). 60. R. D. Rajavel, P. D. Brewer, D. M. Jamba. J. E. Jensen, C. LeBeau, G. L. Olson, J. A. Roth, W. S. Williamson, J. W. Bangs, P. Goetz, J. L. Johnson, E. A. Patten and J. A. Wilson, J. Cryst. Growth 2 1 4 / 2 1 5 , 11()()(2()()()). 61. R. D. Rajavel, D. M. Jamba, J. E. Jensen. (). K. Wu. P. D. Brewer, J. A. Wilson, J. L. Johnson, E. A. Patten, K. Kosai. J. T. Caulfield and P. M. Goetz, J. Electron. Mater. 2 7 , 7 4 7 ( 1998 ). 62. M. Jack, J. Asbrock, C. Anderson. S. Bailey, G. Chapman, E. Gordon, P. Herning, M. Kalisher, K. Kosai, V. Liguori, V. Randall, J. Rosbeck, S. Sen, P. Wetzel, M. Halmos, P. Trotta, A. Hunter, J. Jensen, T. de Lyon, W. Johnson, B. Walker, W. Trussel, A. Hutchinson and R. Balcerak, SPIE Proc. Vol. 4 4 5 4 . 198 (2001). 63. Andrew T. Hunter, SPIE Proc. 3 6 2 9 . 2 5(){1999). 64. C. Verie, F. Raymond, ]. Besson and T. Nguyen Duy, I. Cr!tst. Growth 59, 342(1982). 65. T. J. de Lyon, B. Baumgratz, G. Chapman, E. Gordon, A. T. Hunter, M. lack, J. E. Jensen, W. Johnson, B. ]ohs. K. Kosai. W. Larsen, G.L. Olson, M. Sen, B. Walker and O. K. Wu, J. Cryst. Growth 2 0 1 / 2 0 2 , 98()( 1999 ). 66. O. Hildebrand, W. Kuebart, K. W. Benz and M. H. Pilkuhn, IEEE J. (.)uantum Electron. QE-I 7 , 2 8 4 (1981 ). 67. Perkin-Elmer Optoelectronics InGaAs Avalanche Photodiode Type C30645E. 68. F. M. Roush, A. H. Kalma, I. Kasai, D. A. Estrada, T. J. de Lyon, J. E. Jensen, M. A. Kinch and J. E. Robinson, 2001 IEEE Aerospace Cotfference Proceeditlgs, 5, 2 1 3 7 - 2 1 4 8 (2001). 69. M. B. Reine, A. K. Sood and T. J. Tredwell. Sel~licondlu'tors atld Sel~lil~letals, 18, ed. T.K. Willardson and Albert C. Beer (New York: Academic Press, 1981 ), pp. 2 1 1 - 2 . 70. C. A. Hougen, J. Appl. Ph!is. 66, 3763 ( 1989 ).

352 Handbookof Infrared Detection Technologies 71. T. J. de Lyon, J. E. Jensen, M. D. Gorwitz, C. A. Cockrum, S. M. Johnson and G. M. Venzor, J. Electron. Mater. 2 8 , 7 0 5 (1999). 72. K. D. Maranowski, J. M. Peterson, S. M. Johnson, J. B. Varesi, A. C. Childs, R. E. Bornfreund, W. A. Radford, T. J. de Lyon and J. E. Jensen, ]. Electron. Mater. 3 0 , 6 1 9 (2001). 73. P. S. Wijewarnasuriya, M. Zandian, D. D. Edwall, W. V. McLevige, C. A. Chen, J. G. Pasko, G. Hildebrandt. A. C. Chen. J. M. Arias, A. I. D'Souza, S. Rujirawat and S. Sivananthan, ]. Electron. Mater. 27. "546 (1998). 74. R. Ashokan, N. K. Dhar, B. Yang, A. Akhiyat, T. S. Lee, S. Rujirawat, S. Yousufand S. Sivananthan, ]. Electrotl. Mater. 2 9 . 6 3 6 (2()()()). 75. N. K. Dhar, M. Zandian, J. G. Pasko. J. M. Arias and J. H. Dinan, Appl. Phys. Lett. 70, 1730 (1997). 76. A. Ajisawa, M. Kawano, M. Nomura. M. Miyoshi and N. Oda, NEC Res. and Develop. 3 9, 1 ( 1 9 9 8 ). 77. N. K. Dhar, C. E. C. Wood, A. Gray, H.-Y. Wei, L. Salamanca-Riba and J. H. Dinan, J. Vac. Sci. Technol. B I 4 , 2 366 (1996). 78. H.-Y. Wei, L. Salamanca-Riba and N. K. Dhar, Mat. Res. S!tmp. Proc. 4 8 7 , 607(1998). 79. S. Rujirawat, L. A. Almeida, Y. P. Chen, S. Sivananthan and D. I. Smith, Appl. Phys. Lett. 7 1 , 1 8 1 ( ) (1997). 80. S. Rujirawat, D. J. Smith, I. P. Faurie. G. Neu, V. Nathan and S. Sivananthan, ]. Electron. Mater. 2,7.1 ()47 ( 1998 ). 81. T. Sasaki and N. Oda, J. Appl. Phys. 78, 3121 ( 1995). 82. S. H. Shin, J. M. Arias, D. D. Edwall, M. Zandian, J. G. Pasko and R. E. DeWames, J. Vac. Sci. Technol. B 10, 1492 ( 1992 ). 83. J. B. Varesi, A. A. Buell, R. E. Bornfreund, W. A. Radford, J. M. Peterson, K. D. Maranowski, S. M. Johnson and D. F. King. J. Electrotl. Maa, r. 11 815 (2()()2).

Chapter 10

Silicon infrared focal plane arrays Masafumi Kimata

10.1 Introduction PtSi Schottky-barrier (SB) focal plane array (FPA) is the most popular Si-based quantum infrared FPA. The SB infrared FPA was proposed by Shepherd and Yang in 1973.1 Although the SB infrared FPA has attractive advantages such as a Si-compatible process, high responsivity uniformity, and low l / f noise, the usefulness of the SB technology was not recognized until commercialization of infrared cameras started with a PtSi SB infrared FPA in the late 198Os. Since the commercial infrared camera business started, many PtSi SB infrared FPAs have been developed by making the best use of Si-LSI technology. While PtSi SB detectors are applied to the short wavelength infrared (SWIR) and mid wavelength infrared (MWIR) spectral bands, long wavelength infrared (LWIR) imaging has already been demonstrated with Si-based heterojunction internal photoemission (HIP) FPAs. the photodetection mechanism of which is the same as that of the SB detector. Recent progress in micromachining technology has made it possible to fabricate structures that have very low thermal conductance, and this has thus improved the sensitivity of uncooled infrared FPAs. Although VOx microbolometer FPAs and hybrid ferroelectric FPAs are the usual alternatives in the uncooled technology arena, silicon-based technologies are being developed in order to reduce production cost and to raise productivity. This chapter reviews the Si-based infrared FPA technology, including the PtSi SB FPA, SiGe/Si HIP FPA, and some uncooled FPAs. Uncooled topics discussed in this chapter cover FPAs using pn junction diodes on a Silicon On Insulator (SOI), Si-based resistance bolometers, and thermopiles.

3 54 Handbook of Infrared Detection Technolo~lies

10.2 Cooled FPAs 10.2.1 Schottky-barrier FPAs

Operation of Schottky-barrier detector

Figure 10.1 (a) depicts the internal photoemission process of an electron for an n-type semiconductor. 2 Although the process for an n-type semiconductor will be explained in this section, the process for a p-type semiconductor can be understood in the same way as that for the n-type semiconductor. In the p-type semiconductor, the band in the semiconductor bends downward, and holes are involved in the photoemission process. During the internal photoemission, incident photons are absorbed in the metal, generating excited carriers. The excited carriers are then transported in the metal film until they reach the interface between the metal and the semiconductor, and are finally emitted into the semiconductor. The SB operates as an energy filter. Emission occurs only if the excited carriers have a component of momentum normal to the interface that corresponds to a kinetic energy equal to or greater than the barrier height ~t,. The quantum efficiency (QE) r/ is calculated by dividing the number of excited states which meet the momentum criterion by the total number of excited states, and is given by 3 ,7 - C~ ( b y -

4)!,) 2 hv

'

(1)

where h v i s the photon energy and CI is a constant called the q u a n t u m efficiency coefficient. The cutoff wavelength ;., for the internal photoemission is defined as the wavelength where OE becomes zero. This definition gives the formula for the cutoff wavelength in pm as ;.,1.24_ ,

(2)

where ~t, is measured in eV. In contrast to the nearly constant QE of intrinsic detectors, SB detectors operating in the internal photoemission mode exhibit a strong dependence of QE on the photon energy, especially near the cutoff wavelength. This feature is due to continuous energy states and isotropic m o m e n t u m distribution in the metal. The continuous energy state allow transitions that do not contribute to the emission, and the isotropic m o m e n t u m distribution reduces the n u m b e r of hot carriers with m o m e n t u m s that meet the m o m e n t u m criterion, especially near the cutoff wavelength. Equation ( 1 ) was derived without considering any scattering effects during the transport process. The scattering effect, however, plays an important role in internal photoemission. As shown in Figure 1().1 (b), there are several possible

Silicon infrared focal phme arra!ls

355

Figure 10. l Internal photoemission in Schottk!l-barrier detector. (a) illustrates an energ!l band diagram ex'plaining the internal photoemission process lbr an n-t!lpe semiconductor, and ( b ) shows possible scattering processes in the metal (2, reproduced b!t permission of Kluwer Academic Publishers).

scattering effects that need to be considered when precisely analyzing the characteristics of internal photoemission detectors. Wall scattering (2,3) and grain boundary scattering (6) are elastic and diffusive, and they redirect the excited electrons without energy loss, enhancing the chance of meeting the momentum criterion. When excited electrons collide with cold electrons (4), the energy loss is so great that emission is prevented.

356 Handbookof Infrared Detection Technologies Phonon scattering (5) is a semi-elastic scattering, and a small a m o u n t of energy is lost t h r o u g h this type of scattering, but several collisions are needed before thermalization. Of these scattering effects, wall scattering is the most important in internal photoemission. The OE of photoemissive detectors can be improved by making use of wall scattering. The first observation of QE being enhanced by the wall scattering effect was made by Cohen et al. ~ They found a remarkable increase in OE at the edge of an Au/n-Si SB diode and concluded that it was caused by the Au film being thinner at the perimeter because of shadows resulting from a poorly fitting metal evaporation mask. An experiment on PtSi SB detectors on p-type Si exhibited more than 1()-fold e n h a n c e m e n t in C~ when the thickness of the PtSi film was reduced from 7 8 to 9 nm. 4 Mooney and Silverman published a comprehensive model for internal photoemission, in which the energy loss by electron-phonon scattering and the effect of carriers removed by emission were incorporated in addition to elastic wall scattering, s Their model successfully explained a finite response below the extrapolated barrier height energy and a roll-off from a linear fit of the modified Fowler plot for higher photon energy, as follows. The former p h e n o m e n o n is related to energy loss by the electron-phonon scattering. While only a few phonon collisions suffice to thermalize the hot carrier for low excitation energy, at high energy, the carrier is less thermalized and more likely to be redirected into the escape direction. This makes the apparent extrapolated barrier height greater t h a n the actual barrier height. The roll-off p h e n o m e n o n in a higher energy region is caused by the reduction in the n u m b e r of available carriers by prior emission events when the thickness of the metal electrode becomes so thin that multiple wall scattering occurs. Generally, the dark current ID determines the highest operating temperature of infrared photon detectors. The current flowing through the barrier in Si-based SB detectors is dominated by a thermionic emission current. The SB detectors are generally operated under reverse bias in infrared FPAs. In the reverse-biased condition, barrier lowering due to the Schottky effect has to be taken into consideration. For a reverse bias greater than 3 kT/q, ~,

JD-- A*T2exp[ - q ( ~ ' -

'

(3)

where q is the magnitude of electronic charge. T is the temperature, k is the Boltzmann's constant, and A* is the effective Richardson constant. For holes in Si, A* is about 30 A/cmeK 2 in a moderate electric field range. A~l, is the magnitude of the barrier lowering by the Schottky effect, expressed a s 6 A4)~, -- ~/-4;~~(),

(4)

where E is the strength of the electric field in the semiconductor near the interface, ~ is the dielectric constant of the semiconductor, and ~o is the free

Silicon infrared focal plane arra!!s

357

space permittivity. Depending on the barrier lowering effect, the reverse current of the SB detector gradually increases as the reverse bias increases. It is worth noting that the barrier height obtained from the electric m e a s u r e m e n t is lower than that from the optical measurement. The typical measured difference is from 2() to 5() mV for PtSi SB detectors. Since the barrier height determined by the electrical m e a s u r e m e n t is not affected by phonon scattering and equals the actual barrier height, this difference is deduced to be the average energy loss by phonon scattering.

PtSi Schottky-barrier detector Many metals react with Si at comparatively low temperatures of one-third to one-half the melting point to form silicides.- Among the many silicides, PtSi is the only successful option for infrared FPAs at present. PtSi SB detectors on ptype Si generally have barrier heights of around ().2 eV. As this barrier height corresponds to a cutoff wavelength of about 6 pro, PtSi SB detectors cover the MWIR and SWIR spectral bands. PtSi SB detectors can be operated at around 80 K. PtSi has become one of the most important materials for Si-LSIs, and hence its reaction kinetics and electrical characteristics have been extensively studied. Pt is generally deposited by e-beam deposition or sputtering. For infrared FPA application, the e-beam deposition technique is preferable to that of sputtering because e-beam deposition is performed in an ultra-high v a c u u m and thus has the advantage of low contamination. Annealing is performed either in forming gas or in v a c u u m during or after the deposition. PtSi is formed in a self-aligned manner. Before the Pt deposition, windows are opened through a SiO_, insulation layer on the Si substrate to expose the bare Si surface in active detector areas. Pt deposited on inactive areas {on Si()21 remains unreacted during the silicide formation process, and the excess Pt is removed by aqua regia etching after silicidation is complete. A thin oxide layer, which is formed on the PtSi layer during this process, prevents dissolution of the silicide by the aqua regia. This self-aligned PtSi formation reduces the n u m b e r of mask steps in the FPA fabrication process. The initial reaction between platinum and Si produces the first metal rich phase, Pt2Si. Pt2Si formation occurs at reaction temperatures between 2()() and 500~ s This reaction continues until all deposited Pt is consumed. Above 300~ further reaction continues to form the second phase, PtSi. s PtSi formation is terminated when all the Pt,Si is converted into PtSi. PtSi is thermodynamically stable in contact with Si. and further annealing below 700~ causes no change. ~ The reactions of both phases proceed in a laterally uniform m a n n e r and exhibit t 12 time dependence, indicating that the reaction process is diffusion-limited. Measurements of the temperature dependence of the reaction rates show that the activation energies are 1.3 and 1.5 eV for Pt,Si and PtSi, respectively. 1~ Determination of the dominant diffusing species is an interesting subject, not only from the viewpoint of physics, but also from a practical perspective. Marker and tracer techniques are useful for investigating the dominant diffusing

358 Handbook of Infrared Detection Technologies

species, ~1 In the Pt-Si system, the dominant diffusing species is reported to be Pt for PtxSi formation 12.13 and Si for PtSi formation. 14 Oxygen is the major c o n t a m i n a n t in the silicide formation process. The presence of oxygen retards or interrupts the reaction in the PteSi and PtSi formation processes. 1~ If an excessive a m o u n t of oxygen is present during the reaction process, a SiO2 layer is formed between PtSi and unreacted Pt, ~s resulting in poor adhesion and poor electrical contact. However, in the PteSi formation process, Pt easily penetrates the thin native SiO2 that is generally found on a chemically-cleaned Si surface, and a clean PtSi/Si interface is formed some distance below the original surface. Therefore, the standard wet chemical treatment is adequate for cleaning before Pt deposition. Thinning of PtSi is indispensable for enhancing the wall scattering and obtaining higher OE, as discussed above. Pellegrini et al. reported the photoresponse of PtSi SB detectors on p-type Si with monolayer-level thickness.16 While the reaction with one to two monolayers is metastable, PtSi film thicker than three layers exhibits stable characteristics, which are consistent with those of the PtSi phase. PtSi SB detectors obtained by ordinary procedures generally exhibit excellent uniformity and reproducibility in electrical and optical characteristics. High uniformity is extremely important in thermal imaging because it is low-contrast imaging under high background conditions. Measured non-uniformity of less than 1% (~) is routinely obtained with large format PtSi SB FPAs even though they employ n a n o m e t e r level PtSi films. These superior characteristics are attributed to the reaction kinetics that create a single-phase final product and to the existence of a very clean interface, free from c o n t a m i n a n t s found on the initial Si surface. The photograph in Figure 1 ().2 shows the high uniformity of PtSi formation with an atomic-level fiat interface. It is a cross-sectional highresolution transmission electron microscope photograph of a 3 nm PtSi film formed on a p-type Si substrate by using the e-beam deposition technique. In addition to the responsivity non-uniformity, the 1/f noise sometimes limits the performance of infrared FPAs by degrading the non-uniformity correction.

Figure 10.2 High resolution transmission electron microscope picture of ~ tim PtSi fihn formed on p-t!lpe Si.

Silicon in l?ared.focalplane arra!lS

359

The noise from PtSi SB detectors is virtually white and no distinct 1/f noise is observed for frequencies down to 3 x 1 () s Hz. ~7 This low 1/f noise characteristic also ensures high quality imaging with PtSi SB FPAs. Figure 10.3 is an example of the spectral response of a PtSi SB detector on a ptype substrate. 18 A l t h o u g h SB PFAs are generally illuminated t h r o u g h the Si substrate, this example shows a result for a detector illuminated from the metal electrode side in order to compare the spectral response of the internal photoemission with that of the intrinsic response. While in the visible spectral range, the d o m i n a n t photodetection m e c h a n i s m is band-to-band transition by photons p e n e t r a t i n g t h r o u g h thin PtSi film into Si substrate, the internal

m

Front-illuminated PtSi/p-Si SB

B

4' S

B

m

..

B

m

10-1

m

u

>, o

~

> ffl c.-

f,

o Q.. m

rr

10-2

,,.,.

4"

010 ~.."

PtSi Thickness = 5 nm Reverse Bias = 10 V 10-3 0.1

~

o

1

~

~

9

~

10

Wavelength (l~m) Figure 10.3 Spectral response of l)ollt-illunlinawd PtSi Schollk!l-barrier detector (1,~'. reproduced b!l permission of Plenum Publishin~l Corporation).

360 Handbookof Infrared Detection Technologies photoemission is the only valid photodetection m e c h a n i s m in the spectral region beyond the wavelength corresponding to the band gap energy of Si. FPA architectures

PtSi is formed on Si substrates on which we can fabricate readout circuits with sufficient performance for large format FPAs. PtSi as a contact material for Si LSIs has been investigated for a long time, and it has been confirmed that introduction of PtSi into the Si LSI process has no detrimental effects on readout circuits. Though PtSi has a limited endurance to high-temperature heat treatments, we are able to design the process flow so as to form the PtSi film after all high-temperature treatments in the fabrication process of readout circuits have been completed. Therefore, monolithic PtSi SB FPAs can be manufactured by using processes fully compatible with the planar Si LSI technology used in current Si LSI fabrication plants. In addition to the monolithic structure, the process compatibility with Si LSIs offers m a n y other advantages such as low cost, high reliability, and stable mass production. Figure 10.4 shows a typical design of a PtSi SB FPA pixel along with potential diagrams, which explain its operation.-' The pixel shown in Figure 10.4 uses a CCD readout. The components in a pixel (a PtSi SB detector, a transfer gate, and a part of a vertical CCD (VCCD)) are monolithically integrated on an Si surface. Infrared rays reach the PtSi electrode through the Si substrate. In this back-illuminated structure, the Si substrate serves as a refraction-indexmatching layer for the silicide film, and thus reduces the reflection loss. Furthermore, absorption in the PtSi layer is increased by incorporating an optical cavity, which consists of an a l u m i n u m reflector and a dielectric film (Si02) between the reflector and PtSi film. The effect of the optical cavity is optimized by adjusting the thickness of the dielectric film to the wavelength range of interest. The example shown in Figure 1().4 is a self-integrator pixel, in which the signal charge is accumulated on the stray capacitance of the detector. In order to obtain practical sensitivities with low OE SB detectors, we have to operate SB FPAs in a full-flame integration mode. PtSi SB detectors on moderately-doped ptype Si have sufficiently high breakdown voltages and low dark current at around 80 K, which allow the self-integrator pixel to accumulate the signal and the dark charges generated during the integration time compatible with the standard TV flame rate. A typical design gives a m a x i m u m accumulation capacity of around 106 electrons for 2() to 3()/Jm square pixels at several volts of reverse bias. Although the example shows a pixel with CCD readout, the selfintegrator pixel can also be constructed with MOS switch readout. In the operation of the self-integrator pixel, the detector is reset to a reversebiased state by pulsing the surface-channel transfer gate at the beginning of integration (Figure ] 0 . 4 (c-l)), and then it is isolated from the readout circuit during integration (Figure 10.4 (c-2)). The photoemission of holes from PtSi into Si leaves excess electrons at the PtSi electrode, thus reducing the reverse bias on the SB detector. The accumulated excess signal electrons are read out to the VCCD channel at the next reset process.

Silicon infrared focal plane arra!ls

361

Figure 10.4 Pixel design and operation of PtSi Schottk!l-barrier infrared FPA. which uses CCD readout. (a) and (b) show a plane view and a cross-section of the A-A line. respectivel!t. (c-1) and (c-2) are the potential diagrams during the readout (reset) and integration states, respectivel!t (2. reproduced by permission of Kluwer Academic Publishers).

Self-integrator SB pixels have a built-in blooming control capability. Blooming is a signal charge which spreads to neighboring pixels when excess light is illuminated. This phenomenon is caused by diffusion of excess minority carriers generated in the substrate or signal charge overflow in CCDs, and is found in FPAs with intrinsic detectors, unless appropriate suppression methods are employed. In the self-integrator SB pixel, a strong illumination forward biases the detector and no further electrons are accumulated at the detector. This forward bias at the SB detector is too small for the pn-junction of the guard ring

362

Handbook of Infrared Detection Technologies

to inject electrons into the Si substrate. Therefore. blooming is perfectly suppressed in SB FPAs as long as the vertical CCD has a sufficient charge handling capacity. The most typical readout architecture used for PtSi SB FPAs is the interline transfer CCD (IL-CCD). l~ The IL-CCI) is constructed by arranging the selfintegrator pixels in a two-dimensional m a n n e r and then placing a horizontal CCD at the bottom of the pixel array. The signal charges in the vertical CCDs are transferred to the horizontal CCD row by row in each horizontal blanking period, and a charge-voltage conversion amplifier called a floating diffusion amplifier generates the output signal from the horizontal CCD. The IL-CCD readout architecture is widely used in commercial visible FPAs. This type of FPAs, however, needs a substantial area for the readout CCD and relevant components, leaving a limited area for the PtSi SB detector. Therefore it is important for low OE SB FPAs to enlarge the fill factor as much as possible. As a response to this problem, considerable efforts have been made to break t h r o u g h the fill factor limitation in the course of the development of the PtSi SB FPA. The Charge Sweep Device (CSD) 2~ is one of the most successful readout architectures that has been developed in order to improve the fill factor. Figure 10.5 shows the construction and operation of the CSD FPA. In the IL-CCD, signal charges accumulated in all pixels are simultaneously read out to vertical CCD and are transferred t h r o u g h the vertical CCD as in a bucket relay. In the CSD FPA, on the other hand, signal charges in a row of pixels are read out to vertical charge transfer devices in a horizontal period. The figure illustrates the timing when the third row is selected. The vertical charge transfer device called CSD can

Figure 10. 5 Construction and operation of Charge Sweep Device ( CSD) FPA. 2~

Silicon in fraredfi~calplane arra!ts

36

be compared to a drain with a large capacity. The signal charge is transferred through the CSD by pushing the potential barrier downward and it is then collected under the storage gates within the horizontal period. This operation allows the signal charge to spread over the vertical charge transfer device. Since the whole vertical channel of the CSD is used to store and transfer the signal charge from a single pixel, the charge handling capacity of the CSD is much larger than the charge storage capacity of the detector. Thus, by adopting highcapacity structures for detectors, the saturation levels can be enlarged even with very narrow vertical charge transfer devices. This large saturation level feature is especially important for infrared imaging because terrestrial infrared imaging is done under high background conditions. The operation of the CSD is based on the charge-coupled concept and inherently low noise. Using the conventional 0.8 lam Si LSI design rules, the CSD readout architecture enables us to design 20 lam square pixels with fill factors of around 7()% and saturation levels larger than 2 x 1 ()6 electrons.2 l Other pixel designs, shown in Figure 1 ().6. have also been proposed in order to improve the fill factor. One is the hybrid structure, which is generally used in compound semiconductor infrared FPAs. Although there are some difficulties in making the hybrid structure, it has the attractive advantage of providing independent optimization of the detector and multiplexer, as well as a large fill factor. In the hybrid detector structure shown in Figure 10.6 (a), individual Schottky electrodes are fabricated so close to each other that their depletion regions merge, and diode isolation is achieved only by a 2 lam oxide gap between the silicide electrodes without the guard ring (the self-guarded detector). 22 This structure further enlarges the effective detector area. A fill factor of 8()% was obtained for a 20 lam square pixel by employing this structure. 23 The other pixel design shown in Figure 1().6 (b)is that for the Direct Schottky Injection (DSI) FPA 24 proposed by Kosonocky et al. This structure substantially provides a 100% fill factor. The DSI focal plane array consists of a continuous silicide SB electrode (DSI surface) formed on one surface of a thinned (10-2 5 lam) Si substrate with the CCD readout registers formed on the other side. During operation, the Si substrate is depleted between the DSI surface and charge collecting elements of the readout structure. Hot holes injected from the DSI surface by internal photoemission drift along the electric field line toward the collecting elements. A cross talk of less than 2()% was reported with a 50 lam square DSI pixel. 24 High-resolution PtSi FPAs

The first two-dimensional PtSi SB FPA with 2 5 x 5 0 pixels was reported in 1978. 2~ Since then, the array size has been doubled every 18 months on average, as shown in Figure ] 0.7. This rate of evolution is the same as that of dynamic random access memory (DRAM), which is the most typical product using Si. State-of-the-art technology has reached a level that makes it possible to manufacture high-resolution PtSi SB FPAs with over one million pixels. During the evolution of the PtSi SB FPA, pixel size has been reduced from 1 6 0 x 80 lam 2 to 17 x 17 ILtm2. In spite of this 50-fold reduction in the pixel size, noise equivalent

364

Handbook of Infrared Detection Technologies

Si Multiplexer ,."

% %

,4

".. %

.."

,,. %

,,-

-., "..

.,,

% %

.,,

".. %

,4

% %

,,"

% %

.,.

p

%

%

%

%.

%

%

AR Coat

%

%

.s

,.-

%

,,~

%

,,,

%

,4

%

,s

%

.~,

%

s"

%

::ilnd~um,Bump: AI Reflector

',ilndium Bump, ,

% %

%

%

%

%

%

%

~tSi SiO2

%

%

%

%

%

%

%

%

%

%

%

%,t

p-Si Substrate/

IR

(a)

Charge-Collecting Electrode Gate

Charge-Collecting Transfer Electrode

\

[N'~----\,'I

Barrier

['~

~4~>~;~ ~;i. . . . . . . . . . . . . . . S iO2 ,,.,u~~, ' U'~CCD ] n-Well ~l~n-Wel ~ J] ) 25 ILtm ~ '

PtSi I

N~.

t

~%

~~

t

s

"

'

"~%

~)

~

AR Coat

t

IR

t

t

is

,'

Overflow Drain

p'-Si

t

) ..~VBs

(b) Figure 10.6 Pixel structures of hybrid Schottk!t-barrier FPA with self-guarded detector 2~ (a) and Direct Schottky Injection (DSI) FPA 24 (b) (Reproduced b!t permission of SPIE ).

temperature difference (NETD) of about 1 K in earlier PtSi SB SB FPAs has been improved to below 0.1 K in current advanced PtSi SB FPAs. Table 10.1 summarizes the specifications and performances of typical highresolution PtSi SB FPAs that have full TV resolution. Current PtSi SB FPAs are manufactured on 150 mm Si wafers using process technologies with around 1 lam design rules. Except for a 6 4 0 • 480 element hybrid FPA, all other FPAs in the table have a monolithic structure.

Silicon infrared focal plane arraw

365

1010 --

f

c~ 108

16M -

106

DRAM

/

"'I

"

J

/

I,"

J

1968x1968

/ J ~ 1040x1040

-

64K ~

--/

-

I

"41

~512x512

D-256x25~6 PtSi SB FPA

104

J

-

102

72

i

~ 25x50 I

I

76

I

I

80

I

84

i

I

88

Year

I

I

I

92

I

I

96

2000

Figure 10. 7 Evolution of the spatial resolution of PtSi Schottk!t-barrier f:PAs.

Table 10.1 High-resolution PtSi Schottky-barrier infrared FPAs Array size

Readout

Pixel size.

F.F.

Saturation

NETI) ( F / # )

Ref.

512x512 512x488 512x 512 640x486 640 x48() 640x488 64()x48() l()40x 1()4() 512x512 656x492 811 x5()8 801x512 537x5()5 1968 x 1968

CSD ILCCD LACA ILCCD MOS ILCCD HB/MOS CSD CSD ILCCD ILCCD CSD ILCCD ILCCD

26x2() pro-' 3 1 . 5 x 2 5 Bin-' 3()x 3()pro 2 2 5 x 2 5 lJm-' 24 x 24 ~un 2 21 x21 ~Ill 2 2()x2()~m 2 17x17~m 2 26x2()Mm 2 26.Sx26.5~m 2 18x21 ~m-' 1 7x2() lain-' 1 5 . 2 x 1 1 . 8 l.tm~ 3()x 3() l.tm-'

39% 36'Ii, 54% 54% 38% 4()% 8()% 53% 71% 46% 38% 61% 32% -

1.3xl()"e "5.5• l()Se 4.()• l()Se 5.5xl()Se 1 .Sx 1 ()" e 5.()x 1() s e 7.Sxl()Se 1.6xl()"e 2.9xl()"e8.()x l()Se 7. Sxl()Se ~ 2.1 x l()"e 2.Sxl()Se -

().()7 K (1 ().()7 K (1 ().I()K (1 ().I()K (2 ().()6 K ( 1 ().I()K (1 ().I()K (2 ().I()K (1 ().()3K(1 ().()6 K ( 1 ().()6 K (1 ().()4K ( 1 ().13K(1

2() 26 27 28 29 3()

2) 8) 81 8j ()) ()i ()I 2) 21 8) 2) 2) 2)

23 31 32 33 34 35 36 37

E E" Fill Factor: CSD: Charge Sweep I)evice: II,CCI): Interline Transfer CCI): I,ACA: lAne-Addressed Charge-Accumulation: MOS: Metal Oxide Semiconductor: HB: Hybrid.

For infrared FPAs operating inversely

proportional

electrons.

This means

in t h e s h o t - n o i s e - l i m i t e d

to t h e s q u a r e - r o o t that higher

sensitivities require

As can be seen from the table, the saturation l x]0 6 electrons. saturation

Even

for P t S i

condition,

of the number

SB F P A s

t h e N E T D is

of background

signal

larger saturation

levels.

l e v e l s o f all t h e I L - C C D s a r e l e s s t h a n with

level of the IL-CCD readout architecture

relatively

low

OE, t h e

limits the sensitivity.

low

366

Handbook of Infrared Detection Techtmlogies

Compared with IL-CCDs, the CSD FPAs included in the table have larger fill factors and larger saturation levels despite their smaller pixel sizes. The earliest 512 • 512 element CSD FPA, developed in 198 7. was made with design rules of 2 Bm and has a fill factor of 39% for a 2 6 • 2() IAm-' pixel. 2~J The merit of the CSD readout architecture becomes more prominent as the design rules are reduced, and in 1992 the fill factor for the same pixel size was improved to 71% by using a 1.2 btm fine pattern process. ~-'. Figure 1 ().8 shows photographs of the chip and pixel of the 71% fill factor FPA. The width of the A1 vertical scanning line is 1.2 Bm and the channel width of the CSD is 2 ~tm. The current CSD process using 0.8 l.tm design rules (the effective width of the guard ring is determined by the accuracy of mask alignment and is much smaller than the design rules) makes it possible to manufacture a high sensitivity FPA with a 78% fill factor. Figure 10.9 (a) demonstrates the high uniformity of the PtSi SB FPA. This is an uncorrected thermal image with an 8()1 • element PtSi SB FPA. ~s The measured responsivity non-uniformity was ().3% rms. A two-point corrected image with the same FPA is also shown in Figure 1().9 (b). The residual fixed pattern noise in the corrected image is less than the random shot noise. Figure 10. ] O is the first infrared image with a mega-pixel PtSi FPA. ~1 This FPA has four output ports, each of which operates at a 1 () MHz pixel rate. The image from this mega-pixel FPA is displayed on an HDTV monitor at a frame rate of 30 Hz. Several PtSi linear FPAs have been developed for applications such as spaceborne radiometers and spectrometers. Table 1().2 shows typical linear FPAs developed for the spaceborne SWIR remote-sensing application using PtSi SB technology. The 4 ( ) 9 6 - e l e m e n t x 4 - b a n d FPA ~s was developed for the Short Wavelength Infrared Band Optical Sensors (OPS-SWIR) of the Japanese Earth Resources Satellite-1 (JERS-1), which was launched in 1992, and the 2 1 0 0 - e l e m e n t x 6 - b a n d FPA ~ was developed for the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER)of the Earth Observing

Figure 10.S Chip (a) and pixel (b) photogral~h.~ of:, 12 • ~ 12 element l)tSi Schottk!l-barrier CSI) FPA with 71%fill/actor.

Silicon infrared focalphlne arra!ls

367

Figure 10. 9 Exanlples of thernlal inlages with gO 1 • ~ 12 eh'nlent PtNi Scilottk!l-l~arrier CSI) t:I)A. ( a ) is an Imcorrected inlage, and ( b ) is a two-point ('orre('ted image.

System-AM1 (EOS-AM1). which was launched in 1999. These spaceborne FPAs integrate multiple bands on a single chip. The multiband chip design reduces the size and power consumption of Stirling cycle coolers, which cools the FPAs down to around 80K, and simplifies the optical design. If we did not make use of mature Si LSI manufacturing technology, multiband integration would be impossible because otherwise multiband integration would give us a poor production yield. The QE of these PtSi linear FPAs is around 4% and is almost constant from 1.6 to 2.4 btm. Responsivity non-uniformity obtained from a broad spectral band measurement {without the filter) was 4.5% p-p for a single band of the 34 mm long ASTER FPA. The CCD used in these FPAs is optimized for lowtemperature and low-signal level operation. The transfer inefficiency of the CCD of ASTER was below 2.5• l()-s per transfer at 77K in a signal range down to 4 electrons. The measured non-linearity was less than 1 % . In a Co ~'cJ 1• gamma-ray irradiation test, no detectable degradations were observed up to 5 x 103 rad, and it was proved that the radiation hardness of the FPAs is sufficient for low-altitude satellite applications. A special packaging t e c h n o l o g y , shown in Figure 1(). 11, has also been developed to ensure high planarity of the focal plane at 80 K and heat cycle endurance between 6():C and 8() K. Other silicide detectors and cutoff extension techniques Besides PtSi. there are four silicides that have been used for SB infrared detectors so far' Pd,Si 4().411rSi 42-44 Co,Si 4s and Nigi. 4~' The first solid-state SB infrared FPA was demonstrated with Pd,Si detectors on p - t y p e Si. 4{) The barrier height of the Pd2Si detector is ().34 eV and the cutoff wavelength is 3.5 l.tm. The PdeSi detector cannot be used for thermal imaging because of its poor thermal sensitivity. The main interest of this detector is for applications of spaceborne remote-sensing in the 1-3 lam spectral band. 41 The advantage of the PdeSi detector is its higher operating temperature. The operating temperature is around 12()K, which is compatible with the current satellite passive-cooling technology. CoSi2 and NiSi detectors have also been .

,

,

368

Handbook of Infrared Detection Technologies

Figure 10.10 Thermal image with 104(I • 1040 element PtSi Schottk!l-barrier CSD FPA.

developed for remote-sensing applications in the short wavelength infrared band. Optical barrier heights of CoSi2 and NiSi detectors on p-type Si were reported to be 0.44 eV and 0.4() eV. respectively. 4 s.4~, By 1980, IrSi had been found to be the silicide with the highest electrical barrier on n-type Sl, 47 and it was expected to extend the cutoff wavelength for IrSi detectors on p-type Si. Since Ir can be deposited by e-beam deposition or by sputtering, and the reaction temperature is between 350 and 600~ IrSi detectors can be readily integrated with m i n i m u m modifications into the PtSi SB FPA fabrication process. In 1982, Pellegrini et al. reported the first IrSi SB infrared detectors with a cutoff wavelength of 8.2 btm. 42 A longer cutoff wavelength of 10.7 ~m has been reported for IrSi SB detectors as a result of using oxygen-free in situ v a c u u m annealing. 44

Silicon infrared focal phme arra!ls

369

Table 10.2 PtSi Schottky-barrier infrared FPAs for spaceborne remote-sensing applications {69 reproduced by permission of SPIE)

Satellite/Radiometer

]ERS-1/()PS-SWIR

E()S-A~I1/ASTER

Photograph

Number of Detectors Number of Bands

Detector Size Pixel Interval Chip Size Readout Process Package

21 ()() pixels/band 6 bands/chip ( 1.600-1.7()() lain, 2.145-2.185 lain, 2.185-2.225 lam, 2.2 ~5-2.285 gm, 2.295-2.365 jam. 2.36()-2.4 3() l.tml 2{) t-tinICT)• 1 7 ~amtAT) 1() lam iCT)x 1() l.tmI,;\T~ 16.5 btm {CTIx 3 3 tam {AT) 1() jam (CT)x 2() lain I:VI'I 48.() turn x 1(). 5 mm 49.4 mm• 7.()mm 4-phase BCCI), 2 CCI)s band, 2 outputs/band 3 t.tmdesign rules. 2 Polv/2 AI SiC-Ale() ~-SiC SiC-AIN-SiC

4096 pixels/band 4 bands/chip ( 1.6()-1.71 Vtm. 2.() 1-2.12 ~tm, 2.13-2.2 5 l-ml. 2.2 7-2.40 l.tml

A l t h o u g h infrared i m a g i n g with large f o r m a t IrSi SB FPAs has a l r e a d y been d e m o n s t r a t e d , 4s'49 t h e r e are two serious difficulties in IrSi f o r m a t i o n . 43 One is related to the r e a c t i o n kinetics of the Ir-Si system. Unlike PtSi f o r m a t i o n , Si is the m a j o r diffusing species d u r i n g the w h o l e r e a c t i o n process, and the c o n t a m i n a n t s at the original Si surface r e m a i n on the IrSi/Si interface and t h u s d e g r a d e diode c h a r a c t e r i s t i c s . The o t h e r difficulty is related to the p h a s e control. Generally, at least t h r e e phases, IrSi, IrSi• I x - 1 . 5 - 1 . 7 5 t. and IrSi 3, w e r e detected in r e a c t e d films and it is difficult to obtain a s i n g l e - p h a s e film using c o n v e n t i o n a l silicide f o r m a t i o n processes. In order to o v e r c o m e these difficulties in the IrSi f o r m a t i o n , Pt-Ir silicidesS() s~ and MBE codepositionS2s~ h a v e been proposed. While T s a u r et al. formed Pt-Ir silicide by s e q u e n t i a l e - b e a m deposition of Pt and Ir followed by f u r n a c e a n n e a l i n g , L a h n o r et al. insisted t h a t the most successful process for Ir-Si silicide f o r m a t i o n was Ir presilicidation followed by Pt silicidation d u r i n g deposition. H i g h e r OE coefficients w e r e observed in both these Ir-Si silicide SB detectors w h e n c o m p a r e d w i t h those of Ir-only SB detectors. A possible s t r u c t u r e for these Ir-Si silicide detectors m i g h t be a t w o - p h a s e m i x t u r e of Pt and Ir silicides forming a parallel SB. By using MBE codeposition t e c h n o l o g y , a c c u r a t e p h a s e control is possible, a n d a s i n g l e - p h a s e IrSi~, w h i c h is a potential p h a s e for LWIR application, c a n be formed at relatively low t e m p e r a t u r e s (6()()-8()()~ A l t h o u g h the b a r r i e r h e i g h t is p r i m a r i l y d e t e r m i n e d by the c o m b i n a t i o n of the m e t a l electrode a n d s e m i c o n d u c t o r , b a r r i e r l o w e r i n g c a u s e d by a h i g h e r e x t e r n a l electric field n e a r the interface is expected from the Schottky effect. I n c r e a s i n g the i m p u r i t y c o n c e n t r a t i o n n e a r the interface e n h a n c e s the Schottky effect by r e d u c i n g the depletion layer width w i t h o u t i n c r e a s i n g the reverse-bias voltage. Early w o r k to h a r n e s s the Schottky effect to extend the cutoff w a v e l e n g t h used

370

Handbook of Infrared Detection Technologies

Figzire 10.11 Packaging for long linear PISi Schottk!l-barrier FPA for ASTER. ~

low-energy implantation of thallium (T1)s4 and boron (B) ss in PtSi SB detectors. The ion-implantation technique can easily be incorporated in the standard PtSi SB FPA fabrication process. This technique, however, has a limitation caused by the existence of a narrow potential spike that substantially reduces the collection of photoexcited carriers. Further ion-implantation beyond the limitation only raises the dark current, and is no longer effective in lowering the optical barrier height. Lin et al. found that this detrimental potential spike is caused by the broad depth profile of the implanted impurities, and that a very narrow doping spike layer eliminates this potential spike, s~ Their estimation shows that an nm-order doping spike with an impurity concentration level of 1()2~ cm-3 may make it possible to create PtSi SB detectors with cutoff wavelengths suitable for LWIR imaging. The critical parameter in this doping spike structure is the width of the doping spike layer. To suppress the diffusion of the dopant, Lin et al. employed a low-temperature ( < 500~ Si molecular beam epitaxy (MBE) technique with an elementary B dopant source, and achieved a cutoff wavelength of 22 ~tm for a PtSi SB by incorporating a 1 nm p+ doping spike with a concentration of 2 x 10 2o c m - 3. s6 A recent development in MBE technology has also made it possible to grow high-quality strained SiGe films on Si substrates, which has given us another option for extending the cutoff wavelength. The band gap energy of the strained

Silicon infrared_focalplane arra!ts

3 71

SiGe is smaller than that of Si, and can be tailored by changing the composition. SB detectors on epitaxially-grown p-type SiGe have lower barrier heights than those on p-type Si substrate, even if the same metal electrodes are used, because the major band offset between SiGe and Si appears at the valence band. Kanaya et al. reported the first PtSi/p-SiGe SB detector using MBE technology, s7 Xiao et al. employed rapid thermal CVD as a SiGe epitaxy technique, and obtained a cutoff wavelength of 8.3 l~m for a PtSi/p-Si(~.~sGe(~.l s de tector.s~ They reported that the reaction of Pt with Ge prevents diodes from reducing the barrier height. Therefore, they inserted a thin sacrifice Si layer between the SiGe and deposited Pt layers, which is consumed during the PtSi formation process. Since even a very thin excess Si layer of nm-order thickness is sufficient to form an energy barrier and impede photoemission, the thickness of the sacrifice layer must be rigorously controlled in PtSi/p-SiGe SB detectors. Figure 10.12 compares the spectral response of detectors having extended cutoff wavelengths with that of a PtSi SB detector. 2

10.2.2 Heterojunction internal p h o t o e m i s s i o n FPAs Operation of heterojunction internal photoemission (HIP) detector The valence band discontinuity between SiGe and Si can also be used as an energy barrier for internal photoemission. The idea of utilizing the internal photoemission of heterojunction diodes for infrared detection was proposed by 10-1

' ' 9

10-2

i 0

' '

I

'

'

'

'

I

'

'

'

'

I

'

'

'

- o ~

,......

10.3

", " " ' , ".

"

" g spike

10-4 i e/Si

10-5

0

, ,

, ,

I

5

,

, ,

,

I

,

,

10 Wavelength (l~m)

, ,

I

15

,

, ,

I

2O

Figure 10.12 Comparison of spectral response of Schottk!t-barrier detectors having extended cutoff wavelength, with that of PtSi detector (2. reproduced b!t permission of Kluwer Academic Publishers).

372

Handbookof Infrared Detection Technologies

Shepherd et al. in 1971. s9 Although this technology was very attractive because it could be used to continuously control the barrier height and improve the q u a n t u m efficiency, it took quite a long time to demonstrate the operation of the first HIP infrared detector because of poor epitaxy technology. Developments in MBE technology made it possible to grow high quality SiGe thin films onto Si substrates and fabricate SiGe HIP infrared detectors. SiGe HIP detectors are constructed by replacing the silicide electrode of SB detectors with degenerately doped SiGe. Figure 1 (). 1 3 illustrates the structure and energy band diagram of the SiGe HIP detector. Since major band offset between SiGe and Si appears at the valence band. an energy barrier with barrier height ~t, is established when we choose p-type materials. Photon absorption mechanisms in the SiGe layer are free-carrier absorption and intra-valence band transitions. Assuming that the density of the state of the valence band is proportional to the square-root of the energy, the internal QE can be expressed as

60

1

(hv-ck;,) 2

- 8E~: (1~)~/2 (E,: + ~,,)~/2.

(s)

where E;: is the Fermi energy. Equation (5) is valid when the detector is considered to be a narrow band emitter (hv>>E;.,). On the other hand, if hv <<E~:, the internal OE is given by ~'1 ,7 -

8E~/~(E,:

1

(by-el)l,) 2 + ~,),/,_

(6)

1,,~

Since photons with energy hv <<E;: can only excite holes populating states from EF,-hv to EF, in the degenerately doped SiGe layer, equation (6) has a similar

(a)

(b)

Figure 10.13 Structure and operation of Si(;e/Si heterojunclion internal photoemission (HIP) infrared detector.

Silicon il(fran, d ji~cal plane arra!ls

~7

photon energy dependence of internal OE to the SB detector. While the Fermi energy for silicides is a few eV, for SiGe it is of the order of a few tenths to a few hundredths of meV. Therefore, the internal q u a n t u m efficiency is expected to be significantly greater for HIP detectors than for silicide SB detectors. Meanwhile, the absorption coefficient of SiGe is smaller than that of silicide SB detectors because of the lower density of free carriers in SiGe. The Fermi level of degenerately doped SiGe is located in the valence band of the SiGe layer, and it depends on the doping concentration. Therefore. the cutoff wavelength of the HIP detector can be tuned not only by the compositions, but also by the impurity concentration of the SiGe film. SiGe heterolunction internal photoemission detectors Lin et al. demonstrated the operation of the first SiGe HIP detector in 199(). 62

They observed 0.8% q u a n t u m efficiency at l ( ) p m with a 40 nm Sicj.sGe~.2 emitter. Tsaur et al. also reported that the cutoff wavelength was extended to 25 pm using an Sio.7sGe{~.22 emitter with a boron doping concentration of 4 x 1 ()2~ c m - ~.6~ Using elemental B as a dopant source of MBE. Lin et al. lowered the growth temperature to 35():C and improved the junction quality. They fabricated SiGe HIP detectors with higher q u a n t u m efficiencies of 5-().1% at wavelengths ranging from 2 to 12 pm by using this low-temperature MBE technique. ~'~ The optimum thickness of SiGe film is reported to be 2() nm. ~'~ which is an order of magnitude thicker than that of the optimum metal electrode for SB detectors. This difference in optimum emitter thickness is attributed to the smaller absorption of the HIP detector. Tsaur et al. discussed the improvement of the q u a n t u m efficiency of the HIP detector with a metal overlayer and double-heterojunction structure. ~'~ Park et al. proposed a stacked GexSi~_~/Si HIP detector, with which both the optical absorption and internal q u a n t u m efficiency can be optimised. ~'4 Their detectors consist of several layers of 5 nm or 2.5 nm thick p+-Si~.rGe~. ~ that are separated by 30 nm thick undoped Si layers. The stacked SiGe HIP detector exhibited a high q u a n t u m efficiency ofabout 2 % a t 1 5 pm with a cutoff wavelength of 2(). 5 pm. Figure 10.14 shows the typical spectral response of SiGe HIP detectors. Unlike SB detectors, SiGe HIP detectors exhibit spectral responses that increase initially with increasing wavelength and then decrease monotonically to zero at the cutoff wavelength. The characteristic of the spectral response in the shorter wavelength region is attributed to wavelength-dependent free carrier absorption of infrared rays in the degenerately doped SiGe layers. SiGe HIP FPAs

The first SiGe HIP LWIR FPA was a 4()()x 4(){) element array with a CCD readout developed by Tsaur et al. ~ They used an SiGe film with a g e r m a n i u m concentration of 0.44 and obtained a cutoff wavelength of 9.3 pro. Although their FPA exhibited an excellent uniformity of responsivity, the CCD readout is not adequate for SiGe HIP LWIR FPAs because the transfer efficiency of CCDs is seriously degraded at operation temperatures of 4()-5() K.

3 74

Handbook of Infrared Detection Technologies

10-1

-

Ge0.33Si0.67 K)

__•(40

>,,

o ~

.

c0 1 0 - 2 _ n

Ge0.22Si0.78 K)

m 10_3 E 23 t-"

23

0

Geo.42Sio.58 (50 K)

10-4-10-5 -10-6

0

I

2

I

4

!

6

I

!

I

10 12 8 Wavelength (pm)

I

14

I

16

18

Figure l O. 14 Spectral response of SiGe/Si HIP irffrared detector (~,o. reproduced b!l permission of SPIE ).

Wada et al. applied a MOS readout architecture to a 512 • 512 element GeSi HIP LWIR FPA with a cutoff wavelength of 1().713m. ~s. Figure 10.1 5 depicts their pixel design, showing the cross-section and circuit of a pixel. The pixel contains a source-follower amplifier and a storage capacitor (4 transistors and 1 capacitor). Although the Si substrate has small absorption in the LWIR spectral region, a back-illumination design was adopted in order to obtain a sufficient area for the storage capacitor. The storage capacitor is composed of the a l u m i n u m reflector and the other electrode connected with the drain of the transfer gate. The FPA was fabricated using a 0.8 13m single polysilicon and double a l u m i n u m NMOS process technology. Figure 10.16 shows an SEM photograph of pixels before the second-level alumint:m formation. The pixel size btm 2 and the fill factor is 59%. The FPA has a single output and is is 3 4 • operated at a flame rate of 30 Hz. The measured NETD at 300 K background is 0.08 K with f/2 optics when operated at 4 3 K. An example of a thermal image with the 512 • 512 element GeSi HIP LWIR FPA is shown in Figure 1 (). 17.

10.3

Uncooled

FPAs

10.3.1 Silicon on insulator (SOl) diode FPAs

Electrical characteristics of semiconductor devices have temperature dependence, and such devices can be used for temperature sensing. The pnjunction diode is a typical example of a semiconductor temperature sensor. Assuming an ideal diffusion-limited characteristic and a sufficiently large forward-bias voltage Vf, the current If flowing in the diode is given b y 66

Silicon infrared focal plane arra!ls

37 5

Figure 10.1 5 Pixel of SiGe HIP FPA with ,\lOS readout. (a) is the cross-sectional structure, and (b) is the circuit diagram for a pixel (6 ~. reproduced b!l permission of SPIl:, ).

Figure 10.16 Pixel photograph of ~,12 x ~ 12 element Si(;e HIP FPA with MOS readout beJore second level metalization (~'9. reproduced b!! permission ql'SPlI~ ).

(7)

If-S,d~exp(~)

Is-CT(3+•

-kTj,

(8)

where Sa is the junction area. 1~ is the saturation current. E~ is the band gap energy, y is a constant determined by the temperature dependence of the diffusion constant and carrier lifetime, and C is a temperature-independent

376

Handbook of Infrared Detection Technologies

Figure 10.17 Example of thermal image with ~ 12 • ~ 12 element SiGe HIP FPA. r' ~

constant. Figure 10.18 shows current-voltage characteristics of a diode for two different temperatures, derived from equations (7) and (8). If the diode is driven at a constant current mode, a forward voltage difference of AVr between two temperatures is obtained. Referring to the figure, we understand that A Vr is determined by temperature dependences of ],~ and gradient q / k T . Because the dominant temperature dependence of J,~ is derived from the exponential term in equation (8), AVf is expected to have only a small sensitivity to fluctuations of the production process. This feature assures stable mass-production of SOI diode uncooled FPAs. Figure 10.19 shows a pixel structure with a pn-junction diode temperature sensor, 67-69 which was proposed by Ishikawa et al. In their structure, pnjunction diodes on an SOI wafer are used as a temperature sensor. By using an SOI wafer, they were able to obtain a freestanding structure that contains singlecrystal diodes having lower 1/f noise than amorphous or poly-crystalline devices. Since the diode temperature sensor, support legs, horizontal addressing line, and vertical signal line have to be placed on the same level, only a small fraction of the area can be allocated to the freestanding effective infrared absorbing area if a single-level structure is employed. In order to overcome this problem, they attached an infrared absorbing structure, which covers almost the entire pixel area, to the freestanding diode sensor by using several pillars. The infrared absorbing structure consists of a thin infrared absorbing metal and a dielectric film for maintaining the shape of this thin metal. An interference absorber is formed with this infrared absorbing structure and the reflector deposited on the lower level. While conventional two-level microbolometer pixels have fill factors of around 6()%, a 9()% fill factor is feasible with this SOI diode pixel.

Silicon infrared focal plane arrays

377

TI> T2

Gradient= "

/ Js ( r2)

i

i

I

i

I

I

I

I

vt Figure 10.18 Current-voltage characteristics of forward-biased pn-junction diode, showing operation as temperature sensor.

The micromachining process for the SOI diode uncooled FPA detector is as follows. After completion of CMOS circuits and temperature-sensing diodes, etching holes are formed so as to access the bulk Si layer (Figure 10.20 (a)). Then, an amorphous Si sacrificial layer is deposited and patterned (Figure 10.20 (b)). In the next step, the multi|ayer infrared absorbing structure is formed (Figure 10.20 (c)). Finally, the amorphous Si layer and bulk Si under the temperature-sensing diodes are removed by a single etching process (Figure 10.20 (d)). The buried oxide of the SOl substrate plays the role of an etching stopper during the final etching process. Although the final release process is not a standard Si process, all other processes can be performed on a conventional SiLSI production line. The temperature-sensing diodes are active when they are forward-biased in the SOI diode FPA. The forward-biased design makes it possible to use the temperature-sensing diodes as switching devices and eliminates selection MOS switches in the pixel. Figure 10.21 shows a block diagram of the SOI diode uncooled FPA. The pixel diodes are biased row by row during a horizontal period,

378

Handbook of Infrared Detection Technoloqies

(a) Figure 10.19 Pixel structure of SOI diode uncooled FPA. ( a ) shows a cross-section of the whole structure, and (b) shows the detailed structure of an SOl diode temperature sensor (r,~. reproduced by permission of SPIE ).

and signal amplification and bandwidth reduction are performed in column gate modulation integrators. A single diode driven in the constant current mode generates a differential output voltage of 1-2 mV/K. Since this output voltage is not sufficient for FPA readout circuits, it is increased by connecting several diodes serially. The serial connection also improves the production yield because the pixel is defective only when all diodes in a pixel have poor reverse-biased characteristics. A 3 2 0 x 2 4 0 element uncooled infrared FPA has been developed using SOI diode technology. The pixel size is 40• 40 lam 2. and the chip measures 17 x 17 mm 2. Figure 10.22 shows the chip and the pixel photographs. The width of the slit between adjacent infrared absorbing structures is 2 Ftm, and the resulting fill factor is 90%. The measured thermal conductance of 1.1 • 1 0 - 7 W/K is the same level as those of advanced VOx microbolometer FPAs.

5;ilicon inlraredJbcal plane arra!ls

379

(a) CMOS, SOl Diode and Etching Hole-1 Fabrication

(b) a-Si Sacrifice Film Deposition

(c) IR Absorber and Etching Hole-2 Fabrication

Sacrifice Layer and Si Substrate (d) Etching

Figure 10. 2 0 Micromachining process fi~r SO1 diode mwooh'd F PA. The process proceeds from ( a ) to ( d ). r,,~

Eight-series diodes yield a temperature coefficient of 9.7 mV/K. High uniformity is one of the most important features of the SOl diode uncooled FPA. Non-uniformity of forward-bias voltage and its temperature coefficient across an 200 mm wafer are only 0.13% and 0.62%, respectively. Reflecting the high uniformity of the diode characteristics, a low responsivity non-uniformity of less than 2% has been achieved with 3 2 0 • element FPAs. The total FPA noise and the NETD with f/1 optics are reported to be ll01uVrms and 0.12K, respectively. Table 10.3 summarizes the specifications and performances of the 3 2 0 x 2 4 0 element SOI diode uncooled infrared FPA. 6~ Figure 10.23 shows examples of thermal images with the SO! diode uncooled FPA. 70.3.2 Si-based resistance bolometer FPAs

The results of the VOx resistance bolometer uncooled FPA reported in 19927(~ are so impressive that most of the companies that entered the uncooled FPA business later on have chosen VOx. However, it is not the only material that can be used for resistance bolometers. Si-based resistance bolometers including amorphous Si (a-Si) and single-crystal Si are being studied for use in uncooled FPAs. Early work on a-Si was done by a group at the Defense Science and Technology Organisation in Australia. 71 One of the issues in applying a-Si to uncooled FPAs

380

Handbook of Infrared Detection Technologies

A

~v

2E

(9 n" .C

~ 1~SOI

Diode

O3

2F__. 2F_.. 2E A

o 1=

> I

Current Source & Gate Modulation Integrator T

T

ut

Sample and Hold Circuit i

I

I

Horizontal Shift Register

I Preamp

Figure 10.21 Block diagram of SOI diode uncooled FPA. 68

is large 1/f noise. They prepared a-Si with sputtering and plasma-enhanced chemical vapor deposition (PECVD), and compared the noise characteristics of films with a similar resistivity. The level of 1/f noise is highly dependent on the film preparation method, and a-Si deposited with PECVD exhibited a very high value of 1/f noise, as shown in Figure 10.2 4. They deduced that the origin of the higher noise in PECVD a-Si is the higher hydrogen content, which results in an increase of metastable Si-H bonds. In addition, they also found that a-Si prepared by PECVD commonly has a random telegraph signal (RTS) noise. The RTS noise is a characteristic of current transported along a narrow filament through a barrier of wide-band-gap material. When thermal or UV loads are applied to a-Si, it exhibits instability behavior, which is caused by a high impurity level in the material. 72 Tissott et al. overcame this problem by mastering the kinetic cooling rate to ensure thermodynamic equilibrium below equilibrium temperature. Their films have an equilibrium temperature of 180~ which is much higher than those of standard a-Si films (around 90~ and they exhibited a high stability during a 1000 hour high temperature 12 5~ storage test. 72

Silicon infrared focal plane arrays

Figure 10.22 Chip ( a ) and pixel ( b ) photographs of ~20

Table 10.]

•

381

2 4 0 element SOl diode uncooled F PA. ~'~

Specifications and p e r f o r m a n c e s of SO! diode u n c o o l e d FPA

Array Size Temperature Sensor Pixel Size Fill Factor Chip Size

Spectral Band On-chip Amplifier Si P r o c e s s

Micromachining Process Thermal Conductance Thermal Time Constant Responsivity Noise NETD

Resp. Nonuniformity

32() x 24()

Forward-biased SOl Diodes (x8 I 4() p m x 4() p m 9()% 17 mm • 17 mm 8 - 1 2 pm (;ate Modulation Integrator 1 Poly-Si a n d 2 Metal O. 5 l~m CMOS Single Step Surface and Bulk Process 1 • 1 () r W / K 17 msec 9 ~() I.tV/K ~a f/1, 3()() K 11 () ~," rms (). 12 K ,a f / 1 . 3 ( ) ( ) K < 2 % t ~)

Amorphous silicon has temperature coefficient of resistances (TCRs) from 1.8%/K to 5.5%/K, depending on the growth techniques. Figure 10.25 shows an example of the relationship between the TCR and resistivity. 7 ~ Since it is difficult to control quality a-Si films with higher resistivities, highly-doped a-Si films with TCRs around 2.5%/K are generally used as bolometer materials. Successful demonstrations of a-Si resistance bolometer uncooled FPAs were given by LET172-74 in France and Raytheon' ~" ~'in the United States. Figure 1 0 . 2 6 shows the structure of the LETI pixel. 7~ Their microbridge comprises a thin layer (0.1 ~m) of doped a-Si with no extra supporting layer or membrane. Thin TiN electrodes on the a-Si bolometer layer and an aluminum reflector on the readout circuit constitute an interference infrared absorbing

382

Handbook of Infrared Detection Technologies

Figure 10.2 $ Examples of thermal images with ]20 x 240 element SOl diode uncooled FPA. White is hot fl~r (a). and black is hot for (b).

10-5 10-6 N

T 04 10-7 >

m

L=.

o

o 13_ o

10 -8

........... off-

Sputtered

.m_ 10 -9 o

z

10-10 10-11

10

I

I

I

I

I

i i a I

100 Frequency (Hz)

I

I

I

i

i I a L

1000

Figure 10.24 1 ff noise of amorphous Si ( a-Si ) resistive bolometers prepared b!! plasma-enhanced chemical vapor deposition ( PECVD) and sputterinfl techniqm',~ ( - i. reprodm'ed h!t permission of SPIE ).

structure. In order to obtain the m a x i m u m absorption at 1 () l.tm. the height of the microbridge from the reflector and the sheet resistance of the electrodes are adjusted to be 2.5 l.tm and 3 77 f2/square, respectively. The LETI group employs a micromachining process with a thick polyimide sacrificial layer. 74 Thanks to the very low thermal mass of this thin microbridge structure, a higher flame rate operation than the standard TV frame rate is possible with a thermal time constant of a few ms. After they confirmed the feasibility of their technology with a 50 pm pixel pitch 2 5 6 • element FPA. :~ they developed 32()• element FPAs with pixel pitches of 45 l.tm 72 and 35 ~ m . 74 The 35 pm pixel FPA has

Silicon infrared focalphme arra!ls 10 . . . . . . . . .... . . . .

i

i

•

J_

o4

nO

....

1

,,

100

i

i

I Iii

I

I

I

I

I

II

I

T - G - - I - F F n T F - - r - 3-1 -i-r-n *I- - - I - -I - I- I-I-I*I* I- - - - 'I- -- -I-- -t --I--I--H-I"

I I + -- --I-I I I

I

L _ J_

v

!

]8 ]

J_LLUJ.

L _ _ _

I

L_

J_

I I I III I I -I --I--I-I-14- I- -- -- -- 4- -- -4-I I I I II I I I I

I

I IIII

I

I

I I II~~Jw[

J_LLUJ. 4

I

I

I I I I I

--I--1--1-14

I I I I I I

_1 _ L L U • L _ _ _~,I,,,,~J~Z.[ _ L L U .!.

t i i t~L..I,,.I,,.~r s _ _i ~ I ~ ~ _ tJ • L

I

I

i i t t t tit L _ J _ J _i_ L U

I I I

I I I

I I I

I I I

I I I I I I I

I I I I I I f

I I I I I I I

I I I I I I I

T - -I-

I

I

I

I II

I

I I I

I I i

I I II I I II I I III

I I

I I I I I I I

I I I I I I I

I I I I I I I

I I I I I I

-I - I - l - n

I I I I I I I

I

I

I

T I- - - - I" - - I -

II II II II II II III

1000

I

I

I

I I I

I

I I I

I I I

I I I I

I I I I

I I I I I I I

I I I I I I I

I I I I I I !

I I I I I I I I I I I I if!

I

I -I-I- n

I I I I

I I I I

-r

I I I I I I

100(

)0

Resistivity (f~cm) Figure 10.25 Relationship between temperature coefficient of resistance (TCR) and resistivit!l for a-Si resistive bolometer ( 7 ~. reproduced b!l permission o fSl'lE).

achieved an NETD of 36 mK (f/l) by using an enhanced technology, in which 60% reduction in the 1/f noise has been accomplished by a thicker a-Si film and the thermal resistance has been increased to 4.2 • 1 ()" K/W. Raytheon reported a 16()x ] 2() element a-Si resistance bolometer uncooled infrared FPA with a pixel size of 46.8~tm square.' ~" r The high a-Si pixel resistance of 30 Mr2 of this FPA allows the use of a DC detector bias. The high pixel resistance and the switched capacitor integrating amplifier in each pixel make it possible to realize full-frame integration operation. Although the single-crystal Si has limited TCRs of (). 5-(). 7%/K, it is attractive because it has the same features as those of the SOl diode uncooled FPA. A group at Middle East Technical University in Turkey reported some results on a singlecrystal Si resistive microbolometer uncooled infrared F P A . ' " ' s Figure 1().27 shows the pixel structure of their FPA. The whole structure is fabricated using a standard CMOS process technology in an Si foundry, and the only additional process is the final release of the freestanding structure. Anisotropic etching of the bulk Si is done with Tetramethyl A m m o n i u m Hydroxide (TMAH) solution. 10.3.3 Thermopile FPAs The thermocouple is a temperature sensor that consists of two dissimilar metals or semiconductors connected in series. As the temperature of this junction varies, the electromotive force developed at the output terminal varies. Although single thermocouples have poor sensitivity, the sensitivity can be improved by

384

Handbook of Infrared Detection Technologies

Figure 1 0 . 2 6 Pi.vel structure ofLETI a-Si re.~istive l~ololm'ter t:PA ( - ; reproduced b!l permission of SI~IE ).

connecting thermocouples serially. Serially-connected thermocouples are called a thermopile. There are many combinations of materials for thermopiles, including materials that are used in current Si LSI technology. In a thermopile FPA pixel, the temperature difference between the hot and cold junctions is detected. The cold junctions are located on the heat sink (the substrate) and are used as a temperature reference. Therefore, the temperature drift of the FPA is automatically compensated without temperature stabilizer. The serial connecting design of multi-thermocouples, however, needs many interconnections between the freestanding structure and substrate, limiting the thermal isolation in the pixel. This results in lower sensitivity, compared with other types of uncooled FPAs. Because of these features of thermopile FPAs, a major effort is focused on developing low-cost moderate-sensitivity uncooled FPAs. Kanno et al. reported a 128• 128 element thermopile uncooled infrared FPA with a CCD readout. 7~ Their thermopile consists of 32 pairs of p-type and n-type polysilicon thin films deposited on SiO2 diaphragms. An NETD of 0.5 K was achieved with f/1 optics. Other efforts to develop low-cost thermopile uncooled FPAs have been made at the Physical Electronics Laboratory of ETH Zurich. They have developed the

Silicon infrared focal plane arra!ts

38 5

Figure 10.2 7 Microbolometer pixel using n-t!lpe sin~lle-cr!lstal Si (77. reproduced b!l permission of SPIE ).

Figure 10.28 Cross-sectional structure of polJisilicon/aluminum thermopile uncooled FPA (m~, reproduced by permission of Editions Frontieres ).

pixel structure shown in Figure 1().28. s~ Thermal isolation is performed by a bulk micromachining process from the backside of the wafer. All pixels on the FPA are located on a single membrane that is made of dielectric films. Neighboring pixels are separated by electroplated gold lines formed using a process service for tape-automated bonding. They chose a combination of n-type polysilicon and a l u m i n u m for their thermopile, which is available from a standard CMOS foundry service. They have developed two uncooled FPAs with array sizes of 10• 10 ~ and 16• 16. ~ Figure 10.29 shows an example of a thermal image using the 16 • 16 element thermopile uncooled infrared FPA.

386

Handbook of Infrared Detection Technologies

Figure 10.29 Example of thermal image with 16 • 16 element thermopile uncooled FPA (~I. reproduced b!t permission of lEEE ).

A group at the University of Michigan have also been developing thermopile uncooled infrared FPAs. They employed a combined front-undercut and backetching micromachining process technology. ~2 They demonstrated the operation of a 3 2 • element FPA with 32 n-type and p-type polysilicon thermocouples in each pixel.

10.4 Summary In this chapter, we have discussed Si-based infrared FPA technology. Although Si-based infrared FPA technology does not offer the highest sensitivity, it has many attractive features such as low cost, high productivity, and large-scale integration. A wide spectral range from SWIR to LWIR can be covered with Si-based photoemissive detectors. PtSi SB FPAs have led to the commercialization of cooled infrared cameras, which are being used in many commercial applications. High quality LWIR images have already been demonstrated with a full TV resolution GeSi HIP FPA. Uncooled FPAs are expected to be used for low-end applications to which cooled infrared systems cannot be applied. Such applications require low-cost high-volume production of infrared FPAs, and Si-based technology is the best choice for manufacturing such uncooled FPAs. Si-based uncooled infrared technologies have already reached a level suitable for many commercial applications, and will open up a huge new infrared market.

Silicon infraredjbcalplane arra!ls 387

References 1. F. D. Shepherd and A. C. Yang, Silicon Schottky retinas for infrared imaging, Tech. Digest IEDM, 31 ()- 31 3 ( 19 73 ). 2. M. Kimata, Metal silicide Schottky infrared detector arrays, in Infrared Detectors and Emitters: Materals atut Devices (ed. P. Capper and C. T. Elliott), Chap. 4, pp. 77-98, Kluwer Academic Publishers, Dordrecht {2(}(}1 ). 3. J. Cohen, J. Vilms and R. J. Archer, Investigation of semiconductor Schottky-barriers for optical detection and cathodic emission, Final Report, AFCRL-68-0651, Air Force Cambridge Research Lab., Boston ( 1968 ). 4. M. Kimata, M. Denda, T. Fukumoto, N. Tsubouchi, S. Uematsu, H. Shibata, T. Higuchi, T. Saeki, R. Tsunoda and T. Kanno, Platinum silicide Schottkybarrier IR-CCD image sensors, Jim. ]. Appl. Ph!ls., 21 {Suppl. 21-1 ), 2 3 1 - 2 3 5 (1982). 5. J. M. Mooney and J. Silverman, The theory of hot-electron photoemission in Schottky-barrier IR detectors, IEEE Tratls. Electron Devices, ED-32, 3 3 - 3 9 (1985}. 6. S. M. Sze, Metal-semiconductor devices, in Physics of Setniconductor Devices. Chap. 6, pp. 3 6 3 - 4 2 4 . John Wiley & Sons, Inc., New York {1969 ). 7. J. P. Gambino and E. G. Colgan, Silicides, in Encyclopedia of Applied Physics, Update 2, pp. 2 0 1 - 2 4 8 , Wiley-VCH, Weinheim {1999 }. 8. M.-A. Nicolet and S. S. Lau, Formation and characterization of transitionmetal silicides, in VLSI Electronics Microstrz,'ture Science, 6, (ed. N. G. Einspruch and G. B. Larrabee), Chap. 6, pp. 3 2 9 - 4 6 4 , Academic Press, New York {1983). 9. A. Prabhakar, T. C. McGill and M.-A. Nicolet, Platinum diffusion into silicon from PtSi, Appl. Phys. Lett. 43, 1118-112{) ( 1983 ). 10. C. A. Crider, J. M. Poate, J. E. Rowe and T. T. Sheng, Platinum silcide formation under ultrahigh v a c u u m and controlled impurity ambients, ]. App1. Phys. ~2, 2 8 6 0 - 2 8 6 8 {1981). 11. K. N. Tu and J. W. Mayer, Depth profiling techniques, in Thin Film Interdiffusion and Reactions, (ed. J. M. Poate, K. N. Tu and J. W. Mayer}, Chap. 6, pp. 1 1 9 - 1 6 0 , John Wiley & Sons, New York {19 78 }. 12. R. Pretorius, C. L. Ramiller and M.-A. Nicolet, Marker studies of silicide formation, silicon self-diffusion and silicon epitaxy using radioactive silicon and Rutherford backscattering, Nuclear Instruments and Methods 149, 6 2 9 - 6 3 3 (1978). 13. K. Affolter, X.-A. Zhao and M.-A. Nicolet, Transition-metal silicides formed by ion mixing and by thermal annealing: Which species moves? J. Appl. Phys. 58, 3 0 8 7 - 3 0 9 3 (1985). 14. J. E. McLeod, M. A. E. Wandt. R. Pretorius and C. M. Comrie, Marker and radioactive silicon tracer studies of PtSi formation. ]. Appl. Phys. 72, 22 3 2 - 2 2 4 1 (1992). 15. R. J. Blattner, C. A. Evans, S.S. Lau, J. W. Mayer and B. M. Ullrich, Effect of oxidizing ambients on platinum silicide formation, ]. Electrochem. Soc.: Solid-State Science and Technology 122, 1 7 3 2 - 1 7 3 6 ( 19 75 }.

388 Handbookof Infrared Detection Technolo#ies 16. P. W. Pellegrini, A. Golubovic and C. E. Ludington, A comparison of iridium silicide and platinum silicide photodiodes, Proc. SPIE 782, 9 3 - 9 8 (1987). 17. J. M. Mooney, Excess low-frequency noise in PtSi on p-type Si Schottky diodes, IEEE Trans. Electron Devices 38, 16()-166 ( 1991 ). 18. M. Kimata, M. Denda, S. Iwade, N. Yutani and N. Tsubouchi, A wide spectral band photodetector with PtSi/p-Si Schottky-barrier, Int. J. Infrared and M M Waves 6, 1 0 3 1 - 1 0 4 1 ( 1985 ). 19. A. J. P. Theuwissen, Solid-state imaging at a glance, in Solid-State Imaging with Charge-Coupled Devices, Chap. 4, pp. 1()9-13(), Kluwer Academic Publishers, Dordrecht (1995). 20. M. Kimata, M. Denda, N. Yutani, S. Iwade and N. Tsubouchi, A 512 x ,312 element PtSi Schottky-barrier infrared image sensor, IEEE JSSC SC-22, 1 1 2 4 1129(1987). 21. M. Kimata T. Ozeki N. Tsubouchi and S. Ito, PtSi Schottky-barrier infrared focal plane arrays, Proc. SPIE, 3 505, 2 - 1 2 ( 1998 ). 22. F. D. Shepherd, Recent advances in platinum silicide infrared focal plane arrays, Tech. Digest IEDM 3 7 0 - 3 7 3 (1984). 23. J. L. Gates, W. G. Connelly, T. D. Franklin, R. E. Mills, F. W. Price and T. Y. Wittwer, 488 x 640-element hybrid platinum silicide Schottky focal plane array, Proc. gPIE 1 ~40, 2 6 2 - 2 73 (1991). 24. W. F. Kosonocky, T. S. Villani, F. V. Shallcross, G. M. Meray and J. J. O'Neil, A Schottky-barrier image sensor with 1()()% fill factor, Proc. SPIE 1 3 0 8 , 7()-80 (1990). 25. B. Capone, L. Skolnik, R. Taylor, F. Shepherd, S. Roosild, W. Ewing, W. Kosonocky and E. Kohn, Evaluation of a Schottky IRCCD staring mosaic focal plane, Proc. SPIE 1 5 6 , 1 2 0 - 1 3 1 (1978). 26. H. Elabd, Y. Abedini, W. Shieh, J. Kim, M. Shih, J. Chiu, F. Nicol, W. Petro, J. Lehan, M. Duron, M. Manderson. S. Otto, C. Diaz, S. Lam, H. Balopole, P. Coyle, P. Cheng and R. Marin, 4 8 8 x 5 1 2 - and 2 4 4 x 2 5 6 - e l e m e n t monolithic PtSi Schottky IR focal plane array, presented at SPIE S!ll~lp., No. 1107-29 (1989). 2 7. W.-L. Wang, R. Winzenread, B. Nguyen and J. J. Murrin, High fill factor 5 1 2 x 5 1 2 PtSi focal plane array, Proc. SPIE 1 1 6 1 , 7 9 - 9 5 (1989). 28. E. T. Nelson, K. Y. Wong, S. Yoshizumi, D. Rockafellow, W. DesJadin, M. Elzinga, J. P. Lavine, T. J. Tredwell, R. P. Khosla, P. Sorie, B. Howe, S. Brickman and S. Refermat, Wide field of view PtSi infrared focal plane array, Proc. SPIE 1 3 0 8 , 3 6 - 4 3 (1990). 29. D. J. Sauer, F. L. Hsueh, F. V. Shallcross, G. M. Meray and T. S. Villani, A 640• PtSi IR sensor with low-noise MOS X-Y addressable multiplexer, Proc. SPIE 1 3 0 8 , 8 1 - 8 7 (1990). 30. K. Konuma, S. Tohyama, A. Tanabe, K. Masubuchi, N. Teranishi, T. Saito and T. Muramatsu, A 6 4 8 x 4 8 7 pixel Schottky-barrier infrared CCD image sensor, Tech. Digest ISSCC 2 1 6 - 2 1 7 ( 1991 ). 31. N. Yutani, H. Yagi, M. Kimata, J. Nakanishi, S. Nagayoshi and N. Tsubouchi, 1 0 4 0 x 1040 element PtSi Schottky-barrier IR image sensor, Tech. Digest IEDM 1 7 5 - 1 7 8 (1991 ).

Silicon infrared focalphme arraJls 389

32. H. Yagi, N. Yutani, S. Nagayoshi. J. Nakanishi, M. Kimata and N. Tsubouchi, Improved 512 • 512 IRCSD with large fill factor and high saturation level, Proc. SPIE 1 6 8 5 , 3 7 - 4 7 (1992). 33. Y. S. Abedini, O. R. Barrett, J. S. Kim. D. D. Wen and S. S. Yeung, 656 • 492-element platinum silicide infrared charge-coupled-device focal-plane array, Proc. SPIE 2 0 2 0 , 36-4() ( 1993 ). 34. M. Shoda, K. Akagawa and T. Kazama, A 41 ()K pixel PtSi Schottky-barrier infrared CCD image sensor, Proc. SPIE 2 7 4 4 , 2 3-32 (1996). 35. H. Yagi, T. Shiraishi, K. Endo, Y. Kosasayama, M. Kimata and T. Ozeki, High-performance 801 • 512-element PtSi Schottky-barrier infrared image sensosr, Tech. Digest 1 5 th Sellsor S!lzllnp. (it1 ]apall ), 155-16() ( 1997 ). 36. M. Shoda, K. Akagawa and A. Komai. New high-performance PtSi IRCCD and its electrical shutter operation, Proc. SPIE 3 1 2 2 , 3 9 9 - 4 0 8 ( 1 9 9 7 ) . 37. C. Kauffman, Emergence of tactical, framing infrared reconnaissance, Proc, SPIE 3 4 3 1 , 1 3 0 - 1 4 2 ( 1998 ). 38. M. Denda, M. Kimata, S. Iwade, N. Yutani, T. Kondo and N. Tsubouchi, Schottky-barrier infrared linear image sensor with 4-band • 4()96-element, IEEE Trans. Electron Devices 3 8, 1 1 4 5 - 1 1 5 1 ( 1991 ). 39. M. Ueno, T. Shiraishi, M. Kawai, Y. Yonada, M. Kimata and M. Nunoshita, PtSi Schottky-barrier infrared focal plane array for ASTER/SWIR, Proc. SPIE 2 5 5 3 , 5 6 - 6 5 (1995). 40. E. S. Kohn, S. A. Roosild, F. D. Shepherd and A. C. Yang, Infrared imaging with monolithic CCD-addressed Schottky-barrier detector arrays: Theoretical and experimental results, Proc. hit. Application of CCD's, 5 9 - 6 9 (1975). 41. H. Elabd, T. Villani and W. Kosonocky, Palladium-silicide Schottky-barrier IR-CCD for SWIR applications at intermediate temperatures, IEEE Electron Device Lett. EDL-3, 8 9 - 9 0 ( 1982 ). 42. P. W. Pellegrini, A. Golubovic, C. E. Ludington and M. M. Weeks, IrSi Schottky barrier diodes for infrared detection, Tech. Digest IEDM 157-16() (1982). 43. P. W. Pellegrini, A. Golubovic and C. E. Ludington, A comparison of iridium silicide and platinum silicide photodiodes. Proc. SPIE 782. 9 3 - 9 8 (1987). 44. B.-Y. Tsaur, M. M. Weeks, R. Trubiano. P. W. Pellegrini and T.-R. Yew, IrSi Schottky-barrier infrared detectors with 1 ()-lam cutoff wavelength, IEEE Electrotl Device Lett. 9, 6 5 0 - 6 5 3 ( 1988 ). 45. J. Kurianski, J. Vermeiren, C. Claeys. W. Stessens, K. Maex and R. De Keersmaecker, Development and evaluation of CoSi2 Schottky barrier infrared detectors, Proc. SPIE 1 1 5 7, 1 4 5 - 1 5 2 ( 1 9 8 9 ). 46. J. Kurianski, J. Van Dammer, J. Vermeiren. M. Maex and C. Claeys, Nickel silicide Schottky barrier detectors for short wavelength infrared applications, Proc. SPIE 1 3 0 8 , 2 7 - 3 4 (199() ). 47. I. Ohdomari. K. N. Tu, F. M. d'Heurle, T. S. Kuan, and S. Petersson, Schottky-barrier height of iridium silicide. Appl. Ph!ls. Lett. 33, 1()28-1()3()

(1978).

390 Handbookof Infrared Detection Technologies 48. N. Yutani, M. Kimata, M. Denda. S. Iwade and N. Tsubouchi, IrSi Schottky-barrier infrared image sensor, Tech. Digest IED,V1124-127 (1987). 49. B.-Y. Tsaur, M. J. McNutt, R. A. Bredthaue and IR. B. Mattson, 128x 128element IrSi Schottky-barrier focal plane arrays for long-wavelength infrared imaging, IEEE ElectronDevice Lett. 1 0 . 3 6 1 - 3 6 3 (1989). 50. B.-Y. Tsaur, M. M. Weeks and P. W. Pellegrini. Pt-Ir silicide Schottkybarrier IR detectors, IEEE Electron Device Lett. 9, 1 ()()-102 (1988). 51. P. Lahnor, D. Worle and M. Schulz. Tailoring of the Schottky barrier height of Pt-Ir mixed silicide infrared detectors. Appl. Ph!ls. A A64, 1 0 1 - 1 0 8 (1997). 52. T. L. Lin and J. M. Iannelli, Fabrication of IrSi3/p-Si Schottky-barrier diodes by a molecular beam epitaxy technique, Appl. Ph!ls. Lett. 56, 2013-2() 15 (1990). 53. D. A. Lange, G. A. Gibson and C. M. Falco. MBE-codeposited iridium silicide films on Si(100) and Si(111 ), Proc. SPIE 2 0 2 1 , 6 7 - 7 7 ( 1993 ). 54. P. Pellegrini, M. Weeks and C. Ludington, New 6.'5 lam photodiodes for Schottky barrier array applications, Proc. SPIE 31 1 , 2 4 - 2 9 ( 1981 ). 55. N. Tsubouchi, M. Kimata, M. Denda, M. Yamawaki, N. Yutani and S. Uematsu, Photoresponse improvement of PtSi-Si Schottky-barrier infrared detectors by ion-implantation, Tech. Digest 12th European Solid-State Device Research Conference, 1 6 9 - 1 7 1 (1982). 56. T. L. Lin, J. P. Park, S. D. Gunapala, E. W. Jones and H. M. del. Castillo, Long-wavelength infrared doping-spike PtSi detectors fabricated by molecular beam epitaxy, Proc. SPIE 2 0 2 0 , 3()-35 ( 199 3 ). 57. H. Kanaya, F. Hasegawa, E. Yamaka, T. Moriyama and M. Nakajima, Reduction of barrier height of silicide/p-Sil_xGex contact for application in an infrared image sensor, ]pn. J. Appl. Ph!ls. 28.1,544-L546 (1989). 58. X. Xiao, J. C. Sturm, S. R. Parihar. S. A. Lyon. D. Meyerhofer and S. Palfrey, Silicide/Sil_xGe• Schottky-barrier long-wavelength infrared detectors, Tech. DigestlEDM, 1 2 5 - 1 2 8 (1992). 59. F. D. Shepherd, V. E. Vickers and A. C. Yang, Schottky barrier photodiode with a degenerate semiconductor active region. Urlited States Patent. 3,6()3,847 (1971). 60. B.-Y. Tsaur, C. K. Chen and S. A. Marino, Long-wavelength GexSil_x/Si heterojunction infrared detectors and focal plane array, Proc. SPIE 1 540, 58()595(1991). 61. T. L Lin, J. S. Park, S. D. Gunapala, E. W Jones and H. M. De Castillo, Photoresponse model fo Sil_xGex/Si heterojunction internal photoemission infrared detector, IEEE Electroll Device Lett. 1 5.1 () 3-1 ()5 (1994). 62. T. L. Lin, A. Ksendzov, E. W. Dejewski, 12,.W. Fathauer, T. N. Krabach and J. Maserjian, A novel Si-based LWIR detector: The SiGe/Si heterojunction internal photoemission detector, Tech. Digest oflEDM 6 4 1 - 6 4 4 (199()). 63. T. L. Lin, T. George, E. W. Jones. A. Ksendzov and M. L. Huberman, Elemental boron-doped p+-SiGe layers grown by molecular beam epitaxy for infrared detector applications, Appl. Phlls. Lett. 60, 3 8 0 - 3 8 2 ( 1992 ).

Silicon infraredfocal plane arra!lS 391

64. J. S. Park, T. L. Lin, E. W. Jones, H. M. del. Castillo, T. George and D. Gunapala, Stacked long-wavelength heterojunction internal photoemission infrared detector using multiple Si~_xGex/Si layers, Proc. SPIE 2 0 2 0 , 12-21 (1993). 65. H. Wada, M. Nagashima, K. Hayashi, J. Nakanishi, M. Kimata, N. Kumada and S. Ito, 512 • 512 element GeSi/Si heterojunction infrared focal plane array, Opto-Electronics Review 7.3() 5- 311 ( 1999 ). 66. S. M. Sze, p-n junction diodes in Ph!lsics of Selniconductor Devices, Chap. 3, pp. 7 7 - 1 4 9 , John Wiley & Soms Inc., New York, (1969). 67. T. Ishikawa, M. Ueno, K. Endo, Y. Nakaki, H. Hata, T. Sone. M. Kimata and T. Ozeki, Low-cost 32()• uncooled IRFPA using conventional silicon IC process, Proc. SPIE 3 6 9 8 , 5 5 6 - 5 6 4 ( 1999 ). 68. T. Ishikawa, M. Ueno, Y. Nakaki, K. Endo, Y. Ohta, J. Nakanishi, Y. Kosasayama, H. Yagi, T. Sone and M. Kimata, Performance of 32()x24() uncooled IRFPA with SOI diode detectors. Proc. SPIE 4 1 3 0 , 1 5 2 - 1 5 9 (2()()()). 69. M. Kimata, H. Yagi, M. [Jeno. J. Nakanishi. T. Ishikawa, Y. Nakaki, M. Kawai, K. Endo, Y. Kosasayama, Y. ()hota, T. Shugino and T. Sone, Silicon infrared focal plane arrays, Proc. SPIE 4 2 8 8 , 2 8 6 - 2 9 7 (2()() 1 ). 70. R. A. Wood, C. J. Han and P. W. Kruse, Integrated uncooled infrared detector imaging arrays, Tech. Digest Solid-State Sensor and Actuator Workshop, pp. 1 3 2 - 1 3 5 (1992). 71. M. H. Unewisse, B. I. Craig, R. J. Watson, O. Reinholed and K. C. Liddiard, The growth and properties of semiconductor bolometers for infrared detection, Proc. SPIE 2~ 54, 4 3 - 5 4 ( 19951. 72. J.-L. Tissott, J.-L. Martin, E. Mottin. M. Vilain, J.-J. Yon and J.-p. Chatard, 3 2 0 x 2 4 0 microbolometer uncooled IRFPA development, Proc. SPIE 4 1 1 0 , 4 7 3 - 4 7 9 (2000 ). 73. J.-L. Tissott, F. Rothan, C. Vedel, M. Vilain and J.-J. Yon, LETI/LIR's uncooled microbolometer development, Proc. SPIE 3 4 3 6 , 6() 5-61 () ( 1998 ). 74. E. Mottin, J.-L. Martin, J.-L. ()uvrier-Buffet, M. Vilain, A. Bain, J.-J. Yon, J.L. Tissott and J-P. Chatard, Enhanced amorphous silicon technology of 3 2 0 x 2 4 ( ) microbolometer array with a pitch of 35 Bm, Proc. SPIE 4 3 6 9 . 2 5 0 526 (2001). 75. J. Brady S. Scimert, D. Ratcliff R. Gooch, B. Ritchey P. McCardel, K. Rachels, S. Ropson, M. Wand, M. Weinstein and J. Wynn, Advances in amorphous silicon uncooled IR systems, Proc. SPIE 3 6 9 8 , 1 6 1 - 1 6 7 ( 1999 ). 76. G. L. Francisco, Amorphous silicon bolometer for fire/rescue, Proc, SPIE 4 3 6 0 , 1 3 8 - 1 4 8 (2001). 77. D. S. Tezcan, S. Eminoglu, (). S. Akae and T. Akin, An uncooled microbolometer infrared focal plane array in standard CMOS, Proc. SPIE 4:288, 1 1 2 - 1 2 1 (2001). 78. S. Eminoglu, D. S. Tezcan and T. Akin, A CMOS n-well microbolometer FPA with temperature coefficient enhancement circuitry, Proc. SPIE 4 3 6 9 , 2 4 0 - 2 4 9 (20() 1 ). 79. T. Kanno, M. Suga, S. Matsumoto, M. Uchida, N. Tsukamoto, A. Tanaka, S. Itoh, A. Nakazato, T. Endoh. S. Tohyama, Y. Yamamoto, S. Murashima, N.

392 Handbookof Infrared Detection Technoloqies

Fujimoto and N. Teranishi, Uncooled infrared focal plane array having 128 • 128 thermopile detector elements, Proc. SPIE 1 1 6 9 . 4 5 0 - 4 5 9 (1994). 80. O. Paul, N. Schneeberger, U. Mfinch, M. W~ilti, A. Schufelbfihl, H. Baltes, C. Menolfi, Q. Huang, E. Doering, K. Mfiller and M. Loepfe, Thermoelectric infared imaging microsystem by commercial CMOS technology, Proc. 28th European Soild-State Device Research Conference, pp. 52-55 (1998). 81. A. Schaufelbuehl, U. M/inch, C. Menolfi, O. Brand, O. Paul, O. Huang and H. Baltes, 256-pixel CMOS-integrated thermoelectric infrared sensor array, Tech. Digest IEEE MEMS, pp. 2 0 0 - 2 0 3 (2001 ). 82. A. D. Oliver and K. D. Wise, A l()24-element bulk-micromachined thermopile infrared imaging array, Sensors and Actuators 7 3 , 2 2 2 - 2 3 1 (1999).

Chapter 11

Infrared silicon/germanium detectors Hartmut Presting

11.1 Introduction Generally, the infrared (IR) spectrum can be divided into the short wave or near IR (NIR, 1-3 gm), medium wave IR (MWIR, 3-5 [am), long wave IR (LWIR, 8 - 2 0 lam), and very long w a v e IR (VLWIR, > 20 gm) regime. We report here about silicon/silicon-germanium (Si/SiGe) detectors grown on Si substrate for the SWIR (~1.3/am), the MWIR and the LWIR regime which are envisaged for different application fields such as fibre optical c o m m u n i c a t i o n as well as inter (and intra) chip-to-chip optical c o m m u n i c a t i o n on Si IC chips, and thermal imaging detectors for civil and military applications. Sil_xGex alloy grown on Si for that purpose is an ideal material because, due to its 100% complete miscibility (O~<x~< 1) the SiGe bandgap can be continuously tuned from the Si (1.1 eV) down to the Ge bandgap (0.66eV). Both Si and Ge crystallize in the diamond lattice having an fcc (face-centered cubic) crystallographic unit cell w i t h a cubic point group (Oh, 4 / m 3 2/m). Since Ge has a 4.17% larger lattice constant, the lattice mismatch between Si and Ge leads to strained hetero-epitaxy which means that for a given Ge-content x, a Sil_xGex alloy can only be deposited on Si up to a critical thickness t~ without the formation of dislocations. 1 The strain also leads to a substantial change in the SiGe b a n d s t r u c t u r e (especially a shift of the bandgap). It lifts the six-fold degeneracy of the c o n d u c t i o n band m i n i m u m in the (100) direction (A-direction) of the Brillouin zone into a 2-fold degenerate set of electronic states extending in the (001) growth direction and a set of 4-fold degenerate states which extend in the x-y plane perpendicular to growth direction. The bandgap curve of a (strained and unstrained) Sil_xGex alloy as a function of x is illustrated in Figure 11.1. One can see the big influence of the strain to the bandgap when one compares the unstrained curve (dashed) with the strained one (solid lines). The sharp drop of the bandgap for x > 0.85 for the unstrained alloy, indicates the transition point to the Ge-like bandstructure for

394

Handbook of hlfrared Detection Technoloflies

12

l

,

,

,

,

,

,

,

, /

.~, %

1.1

, .,".L,

1.1

.." : .,

"~ %, u n s t r a i n e d

"".....

1.0

SiGe..~]log.." ~ .

"

"'

1.2

~

: 1.3

,.,

:'l

O.9

S i G e on Si, incl. siz~ uantization effect~

-

1.4

~'. ,, -

1.5

=08

."

~-

'%o .... .,'"

1.6

"~ _

0.7 _

strained

SiGe

on Si

2

0.6 l 0.0

I

0.2

i

I

0.4

J

1.8

I

1

0.6

0.8

2.2 1.0

Ge content

Figure 11.1 Fundamental bandgap of SiGe allo!ls versus (;e-content. The dashed line corresponds to unstrained SiGe bulk samples, the solid line to strained Si(;e with the lateral lattice constant of Si. For the dotted lines the effect of critical la!ter thickness and size quantization is included.

this high Ge-content where the conduction band minimum changes from the Apoint to the L-Point (111 ) of the Brillouin zone. In addition, the height of the valence band offset at the Si/SiGe interfacerelevant for intraband transitions from the SiGe well to the Si boundaries in the MWIR detectors- can be continuously adjusted by changing the Ge-content of the alloy. Thus SiGe detectors offer sensitivity in the whole IR (SWIR, MWIR and LWIR) wavelength regime 2's'6 as schematically depicted in Figure 11.2. In this figure the sensitivity range of different IR detectors, such as silicide detectors (IrSi, PtSi) as well as III-V and II-VI semiconductor detectors (InGaAs, HgCdTe) is also shown. On the bottom of this figure the spectrum of the transmission of the atmosphere is shown, which defines the technological important wavelength regime of the atmospheric transmission window between 1.5 and 2 pm, 3 pm and 5 Jam and 8 l.tm and 12 pm. Furthermore, one of the key advantages of the SiGe material system is that the detector devices such as large area focal plane arrays (FPA) can be fabricated using the cost effective and mature Si technology and can then be monolithically integrated with commercially available Si readout circuits. The ubiquitous silicon technology which governs todays and tomorrows microelectronic market (99% of world semiconductor market in year 2 0 0 1 , total volume 320 bill. US $ t) leads to the effort of fabricating Si based devices for Future H o r i z o n s (2001).

Infrared silicon~germanium detectors

IR detector technologies NOVEL Si IR TECHNO LOGY

395

Si/SiGe heterostructures for IR detection 9 1.3to~2pm SiGe QW (Quantum Wells) 300 -77 K

IR DETECTOR TECHNO LOGIES

9 2 t o ~ 2 0 pm SiGe HIP ~ e t e r o i n t e r n a l 12hotoemission) 40 - ~ 80 K

ATMOSPHERIC TRANSMISSION

9 .~ 15 to 30 pm .SiGe (QW) SLS (superlattice 9 structures) 9 4-77K WAVELENGTH (pm)

Figure 11.2 Comparison of different IR detector technolo[lies and their sensitivit!t spectrum. Si(;e is the on~!~ material from which one can fabricate detectors for the whole infrared regime (near. mid and far IR : 1 l~m30 l~m ). On the bottom the atmospheric transmission is depicted showing the technologically important three transmission windows between 1.5 I~m and 212m. 3 l~m and ~ lain and 8 lira and 12 ~m.

fibre-optical communication. In essence the hope is that the material inherent disadvantage of Si as optoelectronic material - its indirect bandgap - can in some respect be overcome by the introduction of the Si/Ge heterosystem which allows application specific tailoring of the electronic and optical material properties up to a certain extent. The realisation of integrated optical circuits on a Si wafer requires Si based emitter and receiver device functions which can be monolithically integrated on a Si IC chip with their corresponding electronic driver circuits and digital signal processing units as schematically depicted in Figure 11.3(a). A scheme of possible realisation of interchip and intrachip coupling via Si/Ge optical devices can be seen in Figure 11.3(b). A great impact to this field has been given by the realization of 1.3 ~m SiGe waveguide photodetectors with external efficiencies of r1~12% and response times of 4 0 0 ns from strained Si/SiGe q u a n t u m well layers. 6 In addition, about the same time short-period SimGen strained layer superlattices (SLS) 7 have been grown with strong optical transitions at 1.3 ~m up to room temperature. Other passive optical device functions such as modulators and interferometers with SiGe waveguides on Si substrate became available thanks to the progress of silicon-on-insulator (SOI) technology and the excellent work of the group at the Technical University of Berlin 6'~ as well as work from a group at the University of California, Los Angeles. 9 This opened the way of a monolithic integration of Sibased optical devices integrated with Si electronic driver and signal processing circuits as schematically depicted in Figure 11.3(b) also gives an estimation of the total loss (propagation and coupling losses) for an optical free space interchip coupling between two Si IC chips. Besides the big effort to enter the market of

396 Handbook of Infrared Detection Technologies

Figure 11.3 (a) Concept of Si optoelectronics with SiGe photodetectors and waveguides and the electronic driver circuit on board. The external clock rate is provided b!t an external optical signal. (b) Concept of optical transmission between two Si IC chips with active SiGe optoelectronic components such as laser, LED, photodetector and modulator together with the electronic driver and signal processing unit. The optical signal losses due to coupling and propagation are estimated.

fibre optical communication devices with SiGe, there have not yet been any published efforts to fabricate Si/SiGe imaging detectors in the NIR regime for the so-called third transmission window, i.e. in the 1.5 ~m-2 ~m regime. In the MWlR regime, however, imaging systems of infrared detectors with FPA technology have important applications in both commercial and military areas. Important commercial applications are medical diagnosis, fire and combustion

Infr~tred silicon/,qermanium detectors

397

control, surveillance and driver's vision e n h a n c e m e n t . The military applications include night vision, rifle sight, military surveillance, guidance and tracking for missiles as well as for interceptors. For exo-atmospheric, space-based surveillance sensors, where cool targets with low background irradiance levels are often present, LWIR and VLWIR are the appropriate wavelength bands. For tactical military applications as well as for the major part of commercial applications, the important wavelength bands are determined by the atmospheric transmission windows in the NIR. MWIR and LWIR regime. Therefore high resolution, large area IR FPAs in the two atmospheric windows (3-5 IJm and 8 - 1 2 ILtm) with high sensitivity, high uniformity and stability are demanded. For the 3-5 IJm window large area FPAs are fabricated in the well developed and cost effective PtSi technology. However, for the 8 - 1 2 IJm regime there is no completely satisfactory detector material available for fabricating large area staring FPAs. HgCdTe offers, for example, excellent detector performance, but large substrates are not available and costs are high. Due to the lack of large substrates one often has to adopt the line-scanning technique, which is more expensive and less reliable because of fast moving mechanical parts in the system. In addition, bonding of the III-V detector array to the Si readout circuit is highly challenging, especially for larger arrays, lc~'ll For nA1GaAs/GaAs q u a n t u m well infrared detectors (QWIP's) coupler gratings have to be added due to their very low sensitivity to normally incident radiation. ~2, ~3 Consequently, there has been a recent interest in fabricating MWIR detector arrays using the Sil _xGe• alloys grown on Si. 14 These SiGe detectors operate on the principle of hetero-internal photoemission (HIP) within the valence band (intraband transition) from a highly p-doped Sil_• well into an undoped Si layer where the cutoff wavelength is determined by the valence band offset at the Si/SiGe interface (i.e. by the distance of the Fermi energy in the SiGe well to the top of the Si valence band). This type of internal photoemission detector which we refer to as a HIP detector consists of a degenerately p-doped semiconductor electrode in place of the metallic silicide contact in the well established silicide detectors, such as PtSi and IrSi. This variant to the Schottky barrier detector was first proposed by Shepherd et al. in 19 71.1 s The concept was then for the first time implemented in 1990 by a group at the Jet Propulsion Laboratory. ~ They fabricated Si based heterojunction detectors which consisted of a heavily p+ doped GexSil_x epitaxial layer (with a Ge-content of x~().2-(). 3) on a p-doped Si substrate. These pioneering HIP devices were sensitive up to 1 ()IJm but suffered from a large dark current due to an imperfect Si/SiGe heterojunction which made it necessary to operate them at very low temperatures. Apparently, it is now more than 10 years since the first single SiGe/Si HIP detectors were realized and in 1994 the first large area FPA of SiGe HIP detectors with up to 4()()• 4()() pixels was demonstrated by B.Y. Tsaur. 1, We report here about state-of-the-art near, mid and long wavelength infrared SiGe detectors from the material growth, fabrication and performance point of view. We will concentrate on near and mid-infrared devices for imaging systems since SiGe has a clear advantage in the use of the cost effective and m a t u r e Si technology which has the main advantage over competing devices.

398

Handbook of Infrared Detection Technologies

11.2 Near infrared detectors 11.2.1

General operation principle

Most of the near IR (NIR) Si/SiGe detectors are designed for the fibre optical communication wavelength which means that they should have their m a x i m u m sensitivity at the m i n i m u m dispersion wavelength (1.3 lJm: ().95eV) and the m i n i m u m absorption wavelength ( 1.5 5 ~m: ().8 eV) of fibre optical propagation. The SiGe approach for integrating NIR detectors on Si has been thoroughly discussed in the pioneering work of Lury et al. 1~ The key challenge is the lattice mismatch between Ge (or SiGe) and Si. When Ge or SiGe alloys are deposited directly on Si, misfit dislocations which form due to lattice mismatch relaxation occur and cause residual threading dislocations in the SiGe epilayers which are known to affect negatively the detector performance. Many different methods have been proposed to overcome this problem. Among them has been the growth of pseudomorphic q u a n t u m wells (OW) 2 and strained layer superlattices. 7'2{J'21 The usage of a thick ( ~> ] ~m), compositionally graded, fully relaxed buffer layer to avoid threading dislocations in the active layers and for strain adjustment of the subsequent strained layer superlattice, has been also demonstrated with s u c c e s s . 22 Although the up to several microns thick, fully relaxed S i l _ y b G e y b buffer layer is able to solve the lattice mismatch problem between the Si-substrate and the active SiGe/Ge layers, its growth requires deposition times m u c h longer than for the active absorption layer, which makes it rather troublesome and expensive. The direct deposition of Ge photodetectors on Si substrate, both as polycrystalline Ge and as thick, relaxed Ge grown on a Ge-buffer layer, has also been demonstrated.-' ~.24 However, the huge n u m b e r of misfit dislocations at the Si/Ge interface leads to device unacceptable high dark currents (jd > ,--,1 m A / c m 2 at RT, 1 V). A reasonable compromise between performance and costs might be the use of polycrystalline Ge which can be deposited at rather low temperatures (~ 3 5()~ even on a readily processed IC chip for the fabrication ofa Si optoelectronic integrated circuit (OEIC). 2s The two types of NIR SiGe detector structures which we want to focus more closely on are the GeiSi2(~Gei (j= 2 . . . 4 ) double OW and the short-period SimGe~ SLS structure. Both structures work as interband detectors, i.e. they are based on the bandgap transition (interband) from the valence to the conduction band. In the latter case the transition is spatially localized at the Si/SiGe interface. Due to a change in the Ge-composition of the structure (Ge-ML j resp. n/(n+m)) the bandgap can be adjusted roughly from (). 7 eV to 1 eV. The GeiSieoGe j double OW (DQW) structure just described is, for practical applications, a very promising structure for 1.3 13m light detection, since it is pseudomorphically grown on Si substrate and the electronic transition originates from quantized electronic valence band states (hh, lh) in the Ge well to the Si-conduction band. Figure 1 1.4 shows a typical band alignment of a DOW GeiSieoGe i structure deposited on Si. The corresponding optical transitions which occur around 0.9 eV to 1 eV (~ 1.3 btm) are schematically depicted. The Ge

lnfi'ared.,;ilicon/germanium detectors

399

Figure 11.4 Band alignment of GejSi2~)(;e i double QI~' structure ( j = 2 - 4 ) used.for near-infrared detection. Tire hole wavefunction in tire valence band is located near the (;e wells, whereas the electron wavelilnction is located in the 20 ML wide Si well which results in a strong excitonic intertnmd transition around 1. J ltm (0.95 eV). The SiGeo. 1~ allo!t cladding la!lers improve the localization of the electrons and hoh's which allo~vs fl~nctioning of this device up to room temperature.

double wells which are j atomic monolayers (ML. j = 2 - 4 ) wide, are separated by 20 ML of Si, the OW layers are embedded by two Sil_•215 alloy cladding layers. It is well known that for fully strained Ge on Si (1 ()()) substrates a large conduction band offset occurs only for the conduction band states with k-space minima in the growth direction. These 2-fold degenerate states lie m u c h higher in the strained Ge wells than in the unstrained bulk Si, giving rise to confinement of the electronic states in the Si well in the growth direction. 26 The two thin Ge wells together with the Si layer in between and the band bending effect ensure, in an ideal material (no defects, sharp interfaces etc.), that the electron state confined in the central Si well constitutes the lowest electron level in the system. In addition the discrete states of the Ge wells are the topmost valence band states. The sharp Si/Ge interface provides the symmetry breaking potential which leads to the spatially localized strong excitonic transition as schematically indicated in Figure 11.4. The Sil_• cladding layers provide the strong localization of the electron and hole wavefunction which makes the structure suitable for a room temperature device. Photoluminescence {PL)27 and electroluminescence (EL)2 have been measured in these structures up to room temperature which is shown in Figure 11.5(a) and (b). Figure 1 1.5{a) shows the temperature dependent PL spectra of 2 GejSi2oGej OW structures from ref. 2 7. At lower temperatures the no phonon (NP) and the transverse optical (TOt phonon replica lines are well separated in the PL spectrum. At room temperature, there is a broad peak present

400

Handbook of Infrared Detection Technologies

Figztre 11.5 (a) Photolmninescence ( PL ) spectra.l)'om the dozihh' (.)~" (;eiSi2o(;e i strm'tzlres for d(flerent well width (left plot) and measzired at d(fferent temperatz~res (right plot) Zll~to room temperatz~re (taken from Ref. 27). The total PL intensitH is onl!l slightl!t decreasing at room temperature, tlze sz~bstrate related PL lines (at 1.06 and 1.08 eV) disappear above 50 K. (b) EL spectra at room temperature from d(ff. erent Si(;e I)C)W mesa diodes taken from ref. 2. A red shift o.fabozlt A/, ,~0.21~m with increasing (;e well widthJ?om j=2 to j=4 can be seen. This occurs because of tire correspolrdiprg shift of the valence band groz~nd state (*denotes samples with unsHmmetrical SiGex cladding la!ler). (c) EL spectra from a Si6(;e4 SLS mesa diode grown on a thick. compositionall!l graded and fldl!! relax'ed Si ~ -,t,(;e,i, buff. er at various temperatures up to room temperature. 7

In h'ared,,;ilicon/germanimn detectors

401

Figure 1 1. ~ (continued).

at 0.95 (1.35 IJm) eV with an integral intensity which is only a factor of 2 less compared to lower temperatures. Figure 11.5(b) shows the EL spectra at room temperature from different double QW structures with the parameter j (Ge-well width) changing. A clear red shift in the transition energy with increasing well width can be seen in Figure 11.5(b) due to lowering of the electronic ground state in the Ge-well with increasing size (AE--,(). 16 eV (().2 13m) from j=2 to j=4 1. Please note that, because of band filling effects due to a higher excitation density. there is a shift in the transition energy between the EL and the PL experiment. 2 Short-period SimGen SLS consists of an alternating sequence of m monolayers (ML) of Si and n ML of Ge stacked on top of each other and repeated N times (N ~> 100). By introducing a new super-period of Psl.s (ps~.s-n+m) in the Brillouin zone, a folding of the bandstructure in k-space occurs, which leads according to the calculations of Gnutzmann and Klausecker 2s for appropriate values of PsLs to a direct bandgap semiconductor in one dimension parallel to the growth direction (see Figure 11.6). The lattice mismatch between Si and Ge can be accommodated either by strain or by formation of misfit dislocations. According to the equilibrium theory of V. der Merwe 2'~ Ge deposited on Si is fully compressively strained up to the critical thickness t,. above which misfit dislocations form, which partly relax the strain. However, the experimentally found critical thickness is higher than predicted by equilibrium theory, esp. at low growth temperatures ~r which defines a metastable regime of pseudomorphic growth. To achieve a device meaningful thickness of the SLS ( > 200 nm) and to provide sufficient stability of the superlattice layer as a whole, one has to adopt

402

Handbook of Infrared Detection Technologies

Figure 11.6 Brillouin zone folding of a Si-like bandstrlwture introduced b!l a superlattice with a period L= 5ao (ao lattice constant of Si ). The bandstructure is.folded into the first minizone (bold) which has the extension of 2Kst, =2(rc/5 ao). By this folding process the original bandstructure with the indirect bandgap transforms into a direct bandgap structure.

the strain symmetrization of the SLS which requires that the average strain within one period adds up to zero. Under the approximation of equal elastic constants of Si and Ge 31 this leads to ESLS _SLS __ 0 Si -~ ~Ge

with asi _SLS --rice~L,

ESIeS - -ftsi/L.

(1)

and f=2(ace-asi)/(asi+a(,,,) being the lattice mismatch between Si and Ge t(;,,=na• and L=tsi+t(,,,, being the individual layer (f - 0.0417), tsi=m a• and total thickness of one period of the SLS (aj is the perpendicular lattice constant in growth direction (z), all is the in-plane lattice constant in the interface plane (x-y)). 8si SLS ,8(; eSI,S are the magnitude of the strain of the Si and Ge layers of the SLS. For the most important case of equally thick layers (tsi-tce=L/2) equation ( 1 ) reduces to _SLS ESiS L- - S--~Ge

-f/2

(2)

The symmetrical strain is of special importance because it marks the lowest energy state of the SLS. It is only in this strain situation that the superlattice has its lowest energy content (the SLS produces zero average strain) and can be

Infrared silicon~germanium detectors

4()3

stable grown up to infinite thickness. However. to fulfil this condition for arbitrary superlattices one needs to grow an intermediate relaxed Sil_ybGeyb alloy buffer layer between substrate and superlattice ! which provides the average in-plane lattice constant of the subsequent superlattice allt3~'a s~.s Figure 11.7 depicts schematically the strain relation (compressive or tensile) of the single layers in a SimGen SLS grown on a strain adjusting buffer layer. If the sum of the Si and Ge ML in an SLS is a multiple of 10 {(m+n)=s• 10, s = 1 , 2 . . . } strong interband transitions have been predicted by theory ~2 and experimentally found. ~~ Further on. electroluminescence up to room temperature has been observed from fabricated mesa d i o d e s - schematically shown in Figure 1 1 . 9 ( c ) - from a strain symmetrized Sir SLS for different temperatures from 159 K up to room temperature. 7 The EL has been measured with a constant injection current density (3 A/mm 2) up to room temperature. One can see a strong peak indicating an interband transition at 0.78 eV (1.5 l~m) from the SLS, which roughly agrees with the calculation. ~2 11.2.2 Detector

growth and fabrication

All structures reported here are grown on a 1()() mm diameter (4 inch), (100) oriented silicon substrate by molecular beam epitaxy (MBE). This UHV epitaxy technique which operates at a base pressure of 1() -l{~ mbar far from thermodynamic equilibrium, with slow deposition rates, is the method of choice for the growth of the QW and SLS structures with dimensions of atomic monolayers. 35

Figure 11.7 Strain symmetrization b!! a virtual substrate ( Si+Si~ _~1(;e,t bz(ffer) in a Si,,(;e,, SLS. Dots depict the natural lattice spacing, arro~vs the lattice spacing distortion b!t strain. Ill the strain s!immetrized state the average strain of the SLS over one period is zero.

404 Handbookof Infrared Detection Technologies

GemSi2oGem double quantum well detectors The QW samples with Ge wells of several atomic MLs are grown as pn-doped structures on a p+-doped (1()()) Si substrate by molecular-beam epitaxy (MBE). Figure 11.9(a) schematically depict the grown OW structures. Details of the growth process (MBE machine, growth parameters) and the substrate preparation which can be found in refs. 34 and 35. After growth of a 10()nm Si start layer, one ML of Sb has been deposited for surfactant mediated growth 3~,. Subsequently, the active Ge/Si/Ge QW layer stack, together with the surrounding SiGeo.ls cladding layers, is deposited with a p-doping level of 5 x 101 s cm-3. After growth of a 50 nm Si barrier layer, the whole layer sequence is repeated ] 0 times to enhance the absorption of the active layers and provide device relevant thickness. On top of the p-doped OW layer stack, a 400 nm thick, n-doped SiGeo.o25 alloy layer has been grown as a waveguiding layer to improve the coupling of an optical beam from the side facet of the detector and to enable its waveguide propagation in the detector. The structure is terminated by a 10 nm n+-doped Si contact layer deposited on top of the waveguide layer. Figure 11.9(a) shows this structure schematically. As indicated, a pre-deposited ML of Sb is used as surfactant during growth to enhance the two-dimensional layer growth and the Si/Ge interface sharpness by hampering the segregation of Ge on the Si surface during growth. ~' The process temperature, growth rate and boron flux must be carefully chosen in this structure to overcompensate the n-type background level from spontaneous incorporation of the pre-deposited antimony. Figure l 1.8(a) shows a high resolution cross-sectional transmission electron micrograph (TEM) of the OWs in the (110) direction. Very sharp interfaces and fiat layers indicating excellent growth quality can be seen. SimGen short-period superlattice and SiGe quantum well detectors The short-period SimGen SLS samples are grown by MBE on highly n-doped substrates as pn diodes (n + substrate for SimGen SLS). After deposition of the 100 nm thick Si start layer and one ML of Sb acting as surfactant, a partially or fully relaxed Sil_ybGeyb buffer layer has to be grown for strain adjustment of the highly strained SLS. 37 To provide the strain adjustment of the subsequent SLS a rather thick (0.65 ~m), fully relaxed step-graded Sil_y~,~tGeyiz~ buffer layer with a slow stepwise increase of the Ge-content from y=() at the bottom to yf-O. 39 at the top buffer interface in 13 steps has been deposited (rate Ay/Az=().03/ 50 nm). The slow increase of the Ge-content provides a slow build-up of the strain in the buffer which drives the misfit dislocation segments - nucleating at the Si/SiGe interface since the buffer thickness exceeds the critical thickness to the edge of the sample which leaves large parts in the centre of the sample with a low density of threading dislocations. As schematically depicted in Figure 11.9(b) the graded Sil_ybGeyblz~buffer which is almost fully relaxed, is followed by a 5()0 nm thick, constant composition Sil_yfGeyf buffer layer which stabilizes the mean in-plane lattice constant required for the strain symmetrization of the subsequent SLS (yf,~(n/(n+m)). On top of the buffer the 14 5 periods of the Si6Ge4

Infrared silicon/germanium detectors

4()5

Figure 1 1.8 High-resolution cross-sectional transmission elect ton micro~lraph of: ( a ) Si~,(;e4 SLS gro~vn on a compositionally graded and constant composition, full!l relaxed Si l ~(;e,~ allo!! bz((fer la!ler. Ver!l ~,lood Si/(;e interface sharpness and layer qualit!t can be seen lor the ,\IBE gro~vn SL$. (b) (;e4Si2~(;e4 double quantztm ~vell ( DOW) detector structure surrounded b!l a one-sided ~,'i(;ec~.2claddin~t la!ler.

406

Handbook of Infrared Detection Techllologies

GejSi~Gej double QW structure on Si (B2817, B2818, B2975, B3185 )

2. p=l 10 ~" B/cm .3 , j=2, 20ML Si (B2975) 3. p=l 10 ~7 Blcm -3 , J=4, 30ML Si (B3185) (a) Strain symmetrized Si~Ge 4 SLS on step-graded buffer with waveguide layer (B2805)

m•6

p-n junction

! 0 " Nh/cm ~ (=1MI.) Nb--surfactant

! OOn

<100>n + Si I Si T, =

600C

substrate

65Ohm

~=o.03Si~-~iG~'%~n~s~ ~,=0.39a 600C-520C

step-graded buffer

(at N=130 periods)

50Ohm

Si~176r~)OC

constant buffer

200 n m

2_~0 n m

Sio.(Geo.,

145 x (SisGe , }

n-

P 500(."

SLS

I0 n m

P

Si

P*

500("

waveguide

cap

y(z)

(b) Figure 11.9 Layer sequence of MBE grown (;(,jSi2t~(;('i l)(3it ~(a) with unS!lmnletrical (a 1 ) and symmetrical (a2) SiGex alloy cladding layers and Si(,(;e4 SLS structure (b) tjrown on aridly relaxed. Ce-~lraded Si l~jt:~Ge,l,(:~ buffer layer. In (c) a processed mesa detector is schematically depicted, which is also used for EL experiments.

Infrared silicon/germanium detectors

407

Figure 1 1.9 (continm'd).

SLS follow, which result in a total SLS thickness of 2()() nm. Buffer and SLS layers are n-doped due to incorporation from the residual Sb adlayer which has been used as surfactant and has been deposited prior to growth and co-evaporated during growth to e n h a n c e the Si/Ge interface sharpness in the SLS. On top of the SLS, a 4 0 0 n m , p-doped Si~.6Ge~.4 alloy layer is deposited to enable 1.3 l~m waveguide propagation from a into the side facet of the crystal coupled in optical beam. Finally the SLS sample is terminated by a 1 () nm thin, p+-doped Si contact layer. Figure 11.8(b) illustrates a cross-sectional high resolution TEM micrograph of a Si6Ge4 SLS showing an excellent layer quality with good interface sharpness. The diodes are fabricated by standard semiconductor processing techniques, a detailed description of this fabrication process can be found in ref. 2 ]. Circular mesa and waveguide structures with different diameters ranging from 1 ()() ~tm to 1500 Bm and waveguide ridge lengths up to 4 mm have been fabricated. Figure 11.9(c) depicts schematically such a processed mesa diode. A very promising device structure for a Si based NIP, detector, from the integration aspect, is a ridge type waveguide/detector combination first realized by A. Splett. 6 The schematic layout of this device is shown in Figure 11.1 (). On a SIMOX (Separation by IMplantation of OXygen) substrate with a 38()nm Si02 layer u n d e r n e a t h the surface, a 11 ~m thick Si layer is grown by chemical vapour deposition (CVD). After a pre-epitaxial wafer cleaning, the growth is continued by MBE with a 1 ()()nm thick, p-doped bottom contact layer on top of which the active OW layer stack is deposited. On top of the OW an n-doped Si top contact layer is deposited and the structure is terminated by a 20 nm thick n + contact layer. The fabrication technology is done as follows. In the first mask

408 Handbook of Infrared Detection Technologies

Figure 1 1.10 Waveguide/detector combination device fabricated on a SIM()Xsubstrate. The waveguide layer has been grown by CVD, the active la!lers of the detector are ,\IBE grown NiGe OWs (shown is GeiSi2oGe i DO W structure).

layer the detector mesa is dry etched by an SF~/O2 plasma using a photoresist mask. Then the whole structure is covered by 1 l~m SiO2 deposited by plasma enhanced chemical vapour deposition (PECVD). The oxide serves as etch mask for the subsequent wet chemical waveguide etch using potassium hydroxide (KOH). In the second mask step, the Si waveguide layer is defined and subsequently etched to a depth of 5 ~m to 6 btm. The anisotropic etch of the KOH creates the (111 / oblique side walls of the ridge spanning an angle of 54 ~ to the (()()1 / face which is in favour of the propagation of an optical mode. In the last step the ridge structure is metallized by 3()()nm aluminum, bottom and the top contact pads are formed by lift off. The processed device with its layer thickness and waveguide dimensions can be seen in Figure 11.1 (). The from a single mode fibre into the Si waveguide coupled in optical mode travels in the waveguide. It will then be attracted by the higher refractive index and coupled into the active detector layers where it will be absorbed and detected. The ridge length of the detector is optimized for maximum coupling efficiency between waveguide and detector at the operating wavelength at 1.3 l.tm. 11.2.3 Results and discussion

Table 11.1 lists the up-to-now realized Si/SiGe NIR detectors giving the reference, the device structure and the figure of merits in terms of external

Table 11.1 Si/SiGe photodetectors for the NIR range Keference

Structure

Efficiency at operating wavelength

(;c p-i-n Si(;e/ Si MOW/ S1.S (PC')

qt,,l=40'%, irr 1 .45 prn qV,!=iY'%,crr I . 3 11m 11,,,~= 300'%,(11 I . 3 pni l l , . x l z ~ O (11 ' x ~I. 3 h1r11

--

-

1,ury ('I trl. 1984 Ic' 'I'emkin 1.1 ttl. I 9 8 h

''

'l'ernkin 1.1 ttl. I 98f1"" tluangct trl. 1 '1')i"' Iluangc,f (11. 199f>." Splelt 1.1 (11. 199 4" I3.Schiippcrt 1.1 nl. I 99 3 'l'ashiro 1.1 trl. 1997'' S:rmavedam t,t trl. 1998"" Shaoc,tctl. 19C)X4i C'ol:rcer~tt~l.1999'"'

'

S i ( k MOW/ S1.S ( p i n ) wavcguidc Si(;e MOW, Al'l) wavcguidc Si(;c MOW Si(;eC' S i ( k MOW pin waveguide Si(;c SI, pin I'urc ( k - pn diotle Si(;cC' pin MSM (;e

~ 1 ~ , , ~ = 4 0I I I0 1. ' %3 ~11111 1 (XI 111 I . 3 111n 11,.,,= I 'XI ( ( 1 I . 3 pr11 11,,,~=I 1 'XI (11 I. 3 prn 11,.,~=40'%, (11 1 . 3 ~1111 I ~ < . , ~5'%, = Lcrr O.C)X11n1 q c y= l I 2 .O'%,(11 I . 3 ptn

.-

Dark current density - ~ - - ~

Remarks ----

-

--

-

-

- --

i 0 niA/cmL at K'I'

dem. LOO Mh/s li)r L 5 krn

7 rnA/cm2

mi2/crn2 7 1111\/c111~ I mA/cm2

f l = 1 (;lIz. I:WIIM= 300 ps dcm. 8 0 0 Mb/s f o r 1 5 km

(10

50 ~cA/crnL 0 . l 5 mA/c&

f , = L . i (;IIz

410

Handbook of Infrared Detection Technolo~ties

q u a n t u m efficiency rlext ("~Pel/P{~) and dark current density as well as the system application of the used detector. Even though this list does not lay claim to completeness it tries to give an historical overview about some milestone works on Si/SiGe detectors. The pioneering work of Lury et al. lu from 1984 dealt with an MBE grown, ll, tm thick Ge pin diode deposited on a Si substrate where an 0.18btm intermediate Sil_xGex graded alloy layer has been deposited for strain adjustment and to getter the threading dislocations. Even though some of the dislocations have been kept in the graded alloy, there have been a substantial n u m b e r of threading dislocations r u n n i n g through the active Ge-diode to the top, thereby deteriorating the detector's performance. They lead to large dark currents and hence lower detectivity of the detector. The achieved external q u a n t u m efficiency (QE) has been 40% at )v= 1.4 5 l,tm. In addition, a substantial shift of the m a x i m u m responsivity wavelength has been found with wavelength. The work of H. Temkin et al. from 1986~s'394t~ concentrated on Sil_xGex/ Si superlattice detectors at 1 . 3 p m set up as photoconductor with a pseudomorphically deposited Sil_xGex/Si superlattice (SL) on Si-substrate. The SL consisted of 20 periods with the Ge-fraction varying from x=0.2 to x-().8 and a corresponding decrease in well thickness from 1()0 nm to 1.5 nm to meet the critical thickness limit. A m a x i m u m responsivity of 0.6 A/W at a bias of 7 V has been obtained which corresponded to a m a x i m u m butt coupled efficiency of 59% from a single mode fibre at 1.3 pm which - taking into account a coupling efficiency from the fibre to the semiconductor device of no more than 2 0 % results in an internal QE of rlint,-~300%. Several months later the same group realized an avalanche photodiode at 1.3 l,tm with the above described SL stack of 20 Geo.6Sio.4/Si layers. A rib waveguide with 12 ~tm width and 1.5 l,tm height which consisted of the active layers and a p-type layer on top, has been fabricated. Light from a single mode fibre was coupled into the side facet of the 500 lam long, cleaved devices which resulted in a m a x i m u m responsivity of 4 A/ W at 1.3 l,tm or an external QE of rlext,~,4()()%. Much larger DC gains in the order of 1 0 0 - 2 0 0 have been reported by the authors. The work of the group of K.L. W a n g at UCLA (University of California, Los Angeles, USA) concentrated on pin photodiodes from SiGe().5/Si superlattices ( l O x 1 0 / 4 0 nm) pseudomorphically deposited on a p+ Si substrate. An external efficiency of 1% at 1.3 btm and 17% at 0.85 pm at 4 V bias has been obtained by this group. The dark currents of these non-passivated mesa detectors with a lOOxlOOl~m area have been quite high (7 m A / c m 2 at 0 . 5 V bias). The latter structure showed its m a x i m u m photoresponse at 0.85 l~m, the peak wavelength could be shifted to 1.55 pm applying an external bias voltage. The first integrated waveguide/detector device on SIMOX substrate has been realized - as already mentioned - by A. Splett from the group of K. P e t e r m a n n at the Technical University of Berlin with external efficiency of rlext,-~11% at 1.3 p6. A very good review work of realized components on Si based integrated optics has been given by B. Schfippert from the same group, s Passive key components such as modulator, waveguide coupler and TE/TM splitter realized on Si substrates are reviewed, in addition a waveguide/detector combination

Infrared silicon~germanium detectors 411

with an MBE grown active SiGe OW layer has been fabricated by this group as described above. Another approach showing the feasibility of the latter idea has been the use of G e l _ y C y structures with carbon percentages as low as ()~
11.3 Mid- and long-wavelength SiGe IR detectors 11.3.1 Introduction

MWIR and LWIR detectors cover the range between 3 ~tm and 20 I~m and are mostly designed to be sensitive in the two atmospheric transmission windows 3 lam-5~m and 8 l~m-12 lam. MWIR detectors are mostly used for thermal imaging which means that they have to be cooled during operation and that they work only at cryogenic temperatures (T < ~. 15() K, usually to 77 K). Usually the detectors have to be mounted on a standard IC package and placed into a sealed housing which has to be cooled down to temperatures substantially lower than the temperature of the scene they want to image (mostly room temperature) before operation by a Stirling engine or a thermoelectric cooler (Peltier cooler). The reason for that is that the dark current increases exponentially with temperature which limits the m i n i m u m detectable photocurrent. Therefore considerable effort in v a c u u m and cooling technology has to be invested to operate these detectors. However, compared to thermal detectors (bolometers0 pyroelectric detectors, thermocouples etc.) which work at room temperature, photon detectors have a higher efficiency and a m u c h faster response time. In addition the technology for making large focal plane arrays with more than 1 million pixels (1024 • 124) and a pixel size of several microns is available since it can be transferred from microelectronic process technology. As already shown in Figure 11.2, quite a n u m b e r of different detector types and technologies are available and for (almost) every spectral range there is a different detector type or material (intrinsic semiconductor or extrinsic elemental, photoconductors etc.).

412

Handbook of Infrared Detection Technologies

As mentioned before, only detectors from the Si/SiGe material system are capable of covering the whole NIR and MWIR spectral range. In this chapter, we report the fabrication and characterization of Sit_xGex/Si HIP and Sil_xGex/Si multiple quantum well (MOW) detectors, realized as a focal plane array (FPA) and as single mesa detectors. We investigate and compare - theoretically as well as experimentally - novel structural concepts of these detectors, such as introducing a gradient in doping and composition, use of doping setback layers and multi-stack designs in terms of detector performance, i.e. q u a n t u m efficiency and dark current. The main goal therefore is to increase the efficiency and maintain low dark current density. We will see that the sensitivity spectrum of the SiGe HIP detectors can be tuned by change of their composition and well dimensions. We will also describe the detector fabrication, for the single mesa as well as for FPA detectors and its impact on the detector performance, i.e. quantum efficiency and dark current. For example, we will show the critical importance of the doping setback layers between the heavily doped active SiGe layer and the undoped Si barrier layer on the performance of the detectors in terms of dark current (detectivity) and responsivity. 4 After a brief introduction into the principle of operation of Schottky barrier type detectors in Chapter 11.3.2 we focus in Chapter 11.3.3 on the material properties of our MBE grown detector samples measured by secondary ion mass spectroscopy (SIMS), X-ray diffraction (XRD) and Rutherford backscattering (RBS) and we briefly outline the fabrication technology of the investigated detectors. We report on our own as well as on the results of other groups in Chapter 3.4 concerning photocurrent (i.e. quantum efficiency) and dark current of the various investigated MWlR SiGe detector structures and we compare the optical and thermal barrier energies for these structures. In Chapter 11.3.5 we present a bandstructure calculation using a full scale pseudopotential calculation. We also calculate the absorption, the charge density and the density of states of the investigated samples. In the last chapter 11.3.6 we outline the current status of SiGe MWIR detectors. 77.3.2 Principle of operation of HIP detectors The operating principle of the Sit_x Gex HIP detectors is illustrated in Figure 11.11. The detector concept was first proposed by Shepherd, Vickers and Jane is which is very similar to the operation of silicide-type detectors. The first HIP detector has been realized by T.L. Lin from Jet Propulsion Laboratories, 3 his devices were designed for a broad sensitivity spectrum and a long cutoff wavelength, but suffered from a high dark current. In this device only holes are excited over the valence band barrier and transported to the contact by an electric field. Infrared photons with energies less than the bandgap of Si are transmitted through the Si substrate where a fraction of the energy is absorbed in the degenerately p++ doped SiGe epilayer by free holes . These are excited to energies greater than the barrier height @t and emitted into the Si substrate and finally collected by the contacts to produce the detector photocurrent. The energy difference @t between the Fermi energy and

Infrared silicon/gernlanium detectors

41 3

Figure 11.1 1 Band alignment and principle of operation of Si(;e/Si hetero-internal photoemission (HIP) detector.

the valence band edge (which is approximately determined by the valence band offset AE,,) determines the cutoff wavelength and can be tailored by deliberately changing the Ge-composition of the Sil _xGex alloy layer. The spectral response of the Sil_xGe• HIP detector differs from that of other internal photoemission detectors and is given by the modified Fowler equation, expressed here as a function of wavelength r/= 1.24C1

(1/,;.-

1/5.,o) 2

(3)

where C1 is the Fowler emission coefficient. ,;. is the wavelength, and ,;.,,,, is the cutoff wavelength of the detector. The thermal barrier ~t which is defined as the energy difference between the Fermi energy EI: and the edge of the Si valence band (see Figure 11.11) is usually some meV lower than the optical barrier q~o ( 0 . 0 2 - 0 . 0 5 eV). 47 From equation (3) it can be seen that it is interesting to model the detector's response close to the cutoff wavelength, since there is a fast response increase when the wavelength decreases from cutoff which allows the detector to have its maximum sensitivity just before the cutoff wavelength. The detector's response increases initially with increasing wavelength and then decreases monotonically to zero at the cutoff wavelength. 11.3.3 Growth and material characterization

The Si/Sil_xGex MWIR detectors discussed here have been grown by molecular beam epitaxy (MBE) on a double sided polished, weakly p-doped (p= 50 facm). 4 inch (~=1()0 mm) Si-substrate. All samples have been deposited at medium process temperatures (4 50~176 after a pre-epitaxial wafer cleaning and in

414

Handbook of Infrared Detection Technologies al) SiGe HIPs with varying Ge-con~ent

a2) SiC,e HIPs with and without doping setlmek layer

a3) SiGe HIPs with doping- and gemmnium gradient

a4) S i G e H I P s w~th double sided doping setback l a y e r

aS) SiGe H I P s with vm-iation of growth p m-mneters

Figure 11.12 Layer sequences of the various AIBE grown Si/Si(;e HIP and M O W detector structures: (a 1a 5) Single well HIP sample with various designs as mentioned in tire text. (f trot otheru,ise spec(fied, tire p++doping level is 5 • 102~ cm-

situ vacuum desorption of Si native oxides before growth. Details of the MBE

growth process and the pre-epitaxial sample preparation can be found in refs. 34 and 35. Figures l l.12(a) and (b) depict the layer sequences of a series of MBE grown HIP detectors with various well designs such as different Ge content (Figure 11.12(al)), doping setback layers (Figure 11.12(a2) and (a4)), boron and Ge-gradient (Figure ] 1.12(a3)), change of growth conditions (Figure 11.12(a5)), and several multi-well designs (Figure l l . 1 2 ( b l - b 3 ) ) . Figure 1 1 . 1 2 ( c l - c 3 ) shows the grown layer stacks for the multiple quantum well (MOW) detectors with and without use of undoped doping setback layers between the highly doped SiGe well and the weakly doped Si-substrate. All structures were grown on a l()()nm-3()()nm thick, weakly p-doped (1 • 1016 cm -3) Si start layer to provide an epitaxial surface for the subsequent layers. For the HIP detectors this is in general followed by a 2.5 nm undoped Sil_xGex doping setback layer which provides a better Si/Sil_• interface by spatial separation of the hetero- and doping interface, thereby suppressing the escape of the carriers from the well via interface states situated on the boundary of the doped well. We have grown a sample series with changing Ge content in the well (Figure 11.12(al), sample 4325. 4535. 4536 and 43()8), a series with symmetrical, unsymmetrical and without doping setback layers ( 4 8 1 9 : 4 1 6 5 ;

Infrared ,,;ilicon/,qermanium detectors bl ) 3-well HIPs with changing Ge-content

415

b3) 3-well HIP with constant Ge-content and equal thickness

b2) 0-well HIP with Oe-gradient inside each well

Figure 11.12 (b l-b3) Multiple well HIP detectors (482 ~. 4 8 2 ~, and 4 740) with constant and (;e-gradient in each well. The Ge-content in each well is d(flerent it1 structure ( b 1 ) and ( b2 ).

cl) MOW series with different number of wells

Samples 1949

c2) MOW series with doping setback layers on one side

(N=I), 1948 (N=5), 1947 (N=IO)

Samples 1979 (p+--5 1018cmS), 1 9 8 0 (p+= 1,1019cmS)

c3) MOW series with doping setback layers on both sides

Samples 1 9 5 0 (N = 5), 1 9 7 4 (N=I O)

Figure 1 1 . 1 2 ( c l - c 3 ) Multiple quantum well (,\IQit') detectors: ( e l ) 1 9 4 9 N - 1. 1 9 4 8 N - ~. 1 9 4 7 N - 10 wells. (c2) 1 9 7 9 . 1 9 8 0 N - 10 (c ~ ) 1 ~) r ( N= ~ ). 1 9 7 4 ( N - 1 ()).

3794, Figure 11.12(a2 and a4)) and a third series where we introduced a boron (B)- and a Ge-gradient inside the well to create a built-in electric field {Figure 11.12(a3); sample 4684, 4685). We have also grown samples with an additional undoped Sil_xGex doping setback {or spacer) layer on the sample

416

Handbook of Infrared Detection Technologies

surface which should provide a more efficient carrier injection through the top metal contact (-f~-4819, Figure 11.12(a4)). 4g Finally. we have grown three multi-well HIP samples with three thin SiGe wells, separated by 6 nm of undoped Si" one with different Ge content in the three wells (sample 4823" from bottom x=0.25, x - 0 . 3 to x=0.4 in the uppermost well, Figure 11.12(b1 )), another one with constant Ge-content in the three wells (x=(). 3, 482 :3, Figure 11.12.(b3)), and a third one where the three w e l l s - in addition to having a different average Ge content in each w e l l - have a gradient of the Ge concentration inside each single well (4740, Figure 11.12(b2)). Figure 11.12(c) depicts schematically the layer structure of the MQWs. We have grown MOW detectors with a 2.5 nm thick SiGe~.4 well and a 1 nm SiGe~.4 setback layer on one side (1980, 1979, Figure 11. ] 2(c2 i) as well as on both sides (1950, 1974, Figure 11.12(c3)) of the well. The p+ doping level of the OW samples was chosen to be 5 x 1() ~s cm-3. Due to the operation from quantized states in the well rather than from a c o n t i n u u m of states, as in the HIP case, MOW detectors must not be doped too high (<,-~1() 19cm-3). For all mentioned MOW samples the wells together with the Si layers are usually repeated N= l, 5 and N= 10 times. Characterization by SIMS, RBS and XRD The boron and germanium content of all samples have been measured by SIMS. Figure 11.13(a) shows the B-content measured by secondary ion mass spectroscopy (SIMS" 3 keV 02 + primary ions, inclination angle ~=6() ~ for a ' l well' and two '3-well' HIP samples (4819. 4823 and 4825) which basically confirms - within the measurement resolution of SIMS - the intended doping levels. Figure 11.13(b) plots the boron content of the MQW series from sample ] 9 4 7 - 1 9 4 9 as measured by SIMS (primary ions 5 keV Cs+. inclination angle ~=60~ The plot in Figure ] l . ] 3 ( b ) shows a boron doping level of roughly 2x1018cm -3 in the OW's as measured by SIMS with the SIMS experiment limited background concentration of 1 x 1017 cm- 3for all three samples. The ten (resp. five) p-doped wells of the MQW samples 1947 and 1948 which are separated by 50 nm of undoped Si can be clearly seen in this plot. However. one has to realize that for these thin QW layers the SIMS experiment is rather inaccurate and the determined levels have only a significance relative to each other. Figure 11.14(a) compares an X-ray rocking spectrum of sample 4825, taken in (004) reflection geometry by a Philips 4+1 X-ray monochromator, with a numerically calculated spectrum based on the dynamical diffraction theory. From the excellent fit we have deduced values for the Ge-content and the thickness of all three wells, the model structure is shown in the inset of Figure 11.14(a). The same comparison is made for the MQW sample 1975 (N=]0, Figure 11.14(b)). Here we achieve an even better agreement of the calculated spectrum with experiment, the deduced model structure with its structural parameters is again given in the inset of the spectrum. Due to the larger period thickness and the higher number of periods, there are much more resolved satellite peaks in this spectrum, the spacing between two neighbouring satellites

Infrared silicon~germanium detectors ' ' ' '1

....

I ' ' ' ' } ' ' . ' ' l ' ' ' ' l . ' ' ' ' l ' ' ' ' . l ' ' ' ' l ' ' : ' l ' ' ' ' l

X ~ , 4 ; ~ X

1E20

,'-'.

O'3

"

:

i -:~.. .•

i

%" . . . . . . . . . . .

-~ t ~-, A ~ ~ :~: ,< . ~ - -~'-~ ~,w,,, ,~..... ! 9

~..

o

:

9 ~ .......

1F..19

El

""

:

zg'.

1E18

[]

4 819

4~123 ........... :. ...........

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

~.......... :

1El?

::

,~..-

I

I

~

I ' ' ' '

A

~x;';i.~ i ~

. . . .

417

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

0

!....

9

% '

9

:

[]

Q::~:~-.,:. .

~.,-~,,...-",:

~............

~ .....

:

: O [ ] l-l-I

10

20

:

........

z.),~

m 4:]

[

:

3keVO,+, c,r ~

o ....

:

30

:

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

x

s

, " ::,,..

i ..........

:

.oo.~o~:.d~..o..-,,,,4

:I-I

x

E]D

40

l-l:

1

:

]

.:&,

.x:

~

•

[]

. -t

[]

•

o[l:::r3

50

"-~..L.]

El:] L[:I

60

70

depth ( nm ) (a) '

~~

1El8

"~ ~

!

'

'

'

- -! i

'

I.

'

'

'

'

I

'

'

'

'

I

'

'

.'

'

I.

'

:

i

i

0,0

0,1

'

'.

+

. . . .

'

I.'

0

i

1947s

1E16

o

=

,~

1El5

1El4 0,2

0,3

0,4

O~

0,6

0,7

0,8

0,9

1,0

depth (Bm) (b) Figure 11.13 (a) Boron concentration as measured b!t SIMS ( 3 keV 02 +primar# ions. inclination angle 60 ~) for the samples 4 8 2 3 , 4 8 2 5 and 4 8 1 9 . (b) Boron content measured b!! SIMS for the MOW sample series 1 9 4 7 - 1 9 4 9 . Note that for better clarit!l the curves are split (two lower curves are divided by 10 and 1 O0 respectively).

418

Handbook of Infrared Detection Technologies

Figure 11.14 (a) X-Ray diffraction ( XRD) rocking spectrum of the I-well HIP sample 482 5 compared with a numerical simulation. From the good fit to the experimental cz4rve we deduce values for the Ge-concentration and well thickness of the model structure which are shown in the inset. ( b ) XRD rocking spectrum with a numerical simulation of the MQW structure 197 ~ with N= 10 wells. Due to the higher number of periods and larger thickness of the period, more satellites are resolved in this spectrum, the distance between two satellites gives the period thickness.

is inversely proportional to the total period width. From the good fit of the theory to the experimental spectrum we conclude an excellent structural quality and an experimental confirmation of the intended structural values for this sample. Figure 11.14(c) shows the channelling (dotted line, upper trace) and the random (straight line, lower trace)spectrum of the Rutherford backscattering (RBS) experiment from the same sample 4825 which has been measured by XRD. Figure 11.14(d) exhibits a magnified resolution of the random RBS spectrum

ln/?ared silicon/germanimn detectors

419

Figure 11.14 (c) Rutherford Backscattering ( RBS) random and channelling spectrum of 3-well HIP sample 482 5. Tire energy of the primar!l He + ions ~vas 1.4 .\le~'. To enhance tire resolution, a gracing incidence angle of 80 ~ has been chosen. (d) RBS random spectrmn plotted ~vith a numerical simulation in the Ge-recoil energz3 range of the above sample.

(full line) compared with a numerical simulation (crosses) in the recoil energy range of the Ge atom. F r o m the Si peak of the channelled spectrum in Figure 11.14(c) (channel -f~65()) we evaluated a minimum yield of ~min~4% which confirms the excellent crystal quality of this sample. The model structure deduced from this RBS analysis, basically confirms the designed structural parameters and the XRD analysis result of this sample shown in Figure 11.14(a). We have fabricated FPAs with 2 56 • 2 56 pixels as well as single mesa detectors from the grown Sil_xGex/Si HIP samples using standard Si microelectronic

420

Handbook of Infrared Detection Technoloqies

Cross section of processed 256x256 Si/SiGe FPA before flip chip mounting on a Si readout circuit using the BLM process (a)

Technology steps of single mesa detector 1.) Photolithography I (layer 1): mesa mask 2.) SF~ / 02 plasma etching of mesas 3.) Deposition of 650nm SiO2 (ECVD) (anodic oxidation with an additional sacifice

~yer)

4,) Photolithography II: exposure with top and bottom contact masks (layer 2+3)

5 .) Wet oxide etch of contact windows. 6.) Evaporation of 300nm aluminium 7.) Lift-Off-process (for tx)p contact ring and bottom contact pad) 8.) Photolithography 13.I: exposure with top mirror mask (layer 4) 9.) Evaporation of aluminium (lOOnm) 10.) Uft-Off-process (top mirror) 11.) Dice of mesa detect~r chips and mount on a DIL housing

Processed single detector mesa with top mirror

(b) Figure 11.1 5 Cross section of fabricated SiGe IR detector. (a) a processed 2 ~6 x 2 ~6 SiGe focal plane arra!l before the array is flip chip mounted to a Si readout circuit using a bump limiting metallization process (b) cross section of fabricated single mesa with additional top mirror. The fabrication process steps of the mesa detector are listed.

processing techniques. The pixel square in the 256 • array (total dimension 7.1 • 7.1 mm 2) has a length of 21 l~m, the array has a pitch of 24 ~m (separation between nearest neighbour pixels is 3 l.tm), for isolation between the pixels, a thermally grown field oxide has been used. For the pixel and the substrate contact, implanted guard-rings were employed. For the common p-substrate contact, one

hlj?ared silicon/germat~itml detectors 421 guard ring surrounds the whole array. The detector array is backside illuminated, therefore a cavity oxide with an aluminium mirror on top has been deposited on each pixel. Prior to aluminium deposition a contact hole is etched into the oxide for pixel contact. A cross-sectional view of the pixel in the completely processed array is schematically shown in Figure 11.15(a). For the single mesa detector fabrication we used various diameters from 1 mm down to 125 l.tm. The circular mesas have been plasma etched (SF~,/Oe) using a photoresist mask deposited in the first lithography step. After the deposition of 6 5 O n m passivating oxide by plasma enhanced chemical vapour deposition (PECVD, deposition temp. Ta,,~,~ 3()()~ or by anodic oxidation of the uppermost Si sacrifice layer (Taet,~room temperature) on top of the active layer a contact ring on top of the mesa and a bottom contact pad were formed in the second lithography step, both on the epitaxy side of the substrate. Finally, there is an aluminium mirror deposited within the inner illumination ring on top of the mesa by a separate lithography step {Figure l l.15(b)). The thickness of the passivating oxide has been chosen according to the quarter wavelength condition to be ~.o/(4 • n(~.~.~)~65() nm where ~.~jis the wavelength of the incident wave (~4 l.tm). Under this condition the wave reflected at the top mirror creates a wave crest at the active well of the SiGe detector (constructive interference). The processed mesa detectors as well as the arrays are mounted on a commercially available DIL 24 housing equipped with a hole on the bottom for backside illumination. Figure 11.15(b) lists the processing steps for the mesa detector fabrication according to the just described processing sequence: the cross-section of the processed pixel array as well as mesa device can be schematically seen in Figure 11.1 5(a)and (b). 11.3.4 Experimental results and discussion The photocurrent as well as the absorption of the detectors have been measured with a Fourier transform infrared spectrometer (FTIR, Bomem DA8.2) equipped with a spectrally broad emitting glowbar lamp (1 l.t ~<)V~.,nit~< ] ()0 ~m). The samples were immersed in a He bath cryostat during measurement, which was usually performed at 77 K. CaF, beam splitters, filters and cryostat windows were used in the experimental setup. The photocurrent signal of the measured detector was amplified, analog-to-digital (A/D) converted and the photocurrent versus wavelength spectrum was calculated by the processor unit of the FTIR and transferred to a computer. The determination of the absolute efficiency was done by dividing the measured spectrum by the signal from a calibrated PtSi or InSb reference detector which had been cooled down to 77 K and placed in the same position and under the same aperture as the detector under test. Due to the steep cutoff of these reference detectors for )~ > 5 l.tm the measured spectra usually cut off for wavelengths beyond 5.5 ~m, and therefore most spectra here are shown between 2 lam and 5 ~m. The quotient is then multiplied by the known efficiency of the used reference detector to obtain absolute OE numbers of the detector under test. The dark current of the detector was measured inside the cryostat with a metal cap on the DIE 24 housing as a cold shield. The

422

Handbook of Infrared Detection Technologies

Figure 11.16 (a) Absorption (c~xd;i~;,,: c~. ahsorption coeflicient: d~ic;,., thickness of SiGe well) of single well SiGe HIP samples grown b!l us at di~erent temperatures. (b) ,-tbsorptiopl lot x d~.i~;,.) of ,\IC)W samples grown with different doping levels as measured in tel 17.

Infraredsilicon~germaniumdetectors 42 3 current-voltage (I-U) and the capacitance-voltage (C-V) characteristic are measured at 77 K using a programmable current-voltage source unit and an automated L-C-R (inductivity, capacity, resistance) bridge which is capable of measuring capacitances up to 1 () MHz. An important physical quantity to know for each detector is the absorption of the detector material itself, in our case the absorption of the SiGe HIP and MQW structures in the infrared spectral wavelength range. Figure 11.16(a) shows the absorption spectra taken at room temperature of three single-well HIP samples grown by us at different temperatures with a well doping level of p++-5• 49 For wavelengths longer than 3.2itm the absorption is mainly governed by the strong free carrier absorption caused by the degenerate doping in the SiGe layer, the intraband absorption at the Si/SiGe interface dominates at smaller wavelengths as can clearly be seen from the kink in the slope of the absorption curve ()~> 3.2 Itm) in Figure 11.16(a). The absorption of the detector yields its upper limit of the photocurrent, i.e. the QE of the detector and provides a fundamental material property from which, together with the photocurrent (i.e. quantum efficiency), the recombination rate of excited carriers inside the material can be determined. Other authors have reported absorption data for SiGe/Si MOW samples shown in Figure 11.16(b). 17 The three different MOW samples in Figure 11.16(b) have 1() wells with 3()nm thickness with a Ge concentration of x-().26 and different doping levels from 4 x 10 ~ cm-~ up to 4• 1019 cm -3. As expected and clearly confirmed in Figure 11.16(b) the highest absorption is for the MOW sample with the highest doping level, i.e. for the sample with the highest free carrier concentration. Even more important is the evaluation of the dark current and the photocurrent of the detector at its designed operating temperature, namely 77 K. As can be seen from the absorption data, the high boron doping level of 5 • 102{~cm- 3 leads to a strong absorption of the incident light due to free carrier absorption. Figure 11.17(a) and (b) show the QE and the dark current of a series of HIP samples with different germanium fractions (4 3()8-4 336' from x-().2 5 to x=0.5" see Figure 11.12( a 1 ))" First of all we see that the cutoff wavelength and the position of the peak responsivity decreases with increasing Ge fraction, due to the higher threshold energy. The relatively large valence band offset between the heavily p-doped Sil_xGex layer and the Si substrate for x=().5 (43()8) results in a cutoff wavelength of approximately 3 It, whereas the low Ge fraction of sample 432 5 leads to a broad absorption of free carriers in the entire measured spectrum. QEs of up to 0.43% at 2.7Itm were measured for sample 4:335 containing a Sio.6sGeo.3s layer. As shown in Figure 11.17(b), the dark current densities of these samples vary accordingly in a wide range between 10 -9 A/cm 2 (4 308) and 10 -2 A/cm 2 (432 5) at 77 K according to the effective Richardson equation that describes the thermal emission current over the barrier.

jdark- A*T2exp ~RT

14)

424 Handbook of Infrared Detection Technoloqies

Figure 11.17 Quantum efficiency ( OE ) ( a ) and dark current (b) ~[mesa detectors from an HIP series having one SiGe ~vell with different Ge content in the well (4 ~2 ~. 4 ~ ~ ~. 4 ~ ~6. 4 ~08: Figure 11.12(a 1 ).

Here A* is the effective Richardson constant, k~ is the Boltzmann constant, ~t is the thermal barrier height and T is the temperature. The threshold energy of the HIP structures was determined from the temperature dependent measurement of the dark current where I n ( j D / T 2) is plotted versus 1 / ( k l ~ T ) (Figure 11.17(c) and from the slope of the curve in this Richardson plot, the thermal barrier energy is determined. The optical barrier ~'o is determined using the Fowler equation (see equation (3*)) written in a slightly modified form, that correlates the measured optical barrier ~ with the efficiency rl of the detector (Figure 11.17(d)):

Infrared silicon~germanium detectors

425

Figure 11.17 (c) Richardson plot and (d) Fowler plot of these HIP samples. All QE spectra have been taken at 77 K. Note that the low temperature dark current level depends on the (;e content as discussed in text.

,

/7 -- (71

(by-

r

hi~

2

(3*)

Here hv is the photon energy and C1" is the modified Fowler emission coefficient. 14 The results are listed in table 11.2 below: The inelastic scattering of photo-excited holes prior to the emission over the barrier, results in a slightly higher optical barrier ~o compared to the thermal

426 Handbookof Infrared Detection Technologies Table 11.2

Thermal and optical barriers as d e t e r m i n e d from R i c h a r d s o n and Fowler plots

Sample

x

Thermal barrier qlt {Richardson )

Optical barrier ~,, (Fowler)

4325 4535 4536 4308

0.25 0.35 0.40 O.50

N/R 191 meV 286 meV 390 meV

157 meV 216 meV 299 meV 405 meV

N/R no results due to veryhigh dark current. barrier 1//t. To maximize the detectivity of such IR detectors, high q u a n t u m efficiency and a low dark current are desirable. From this result we conclude that the optimum g e r m a n i u m fraction in the Sil_xGex layer is between O. 3 and O. 35 where the detectors achieve high responsivities between 3 and 4 lam (0.4% at 7 7 K and 5 V reverse bias)combined with low dark currents (jd,-~l() -s A/cm-2). Because detector performance of a given SiGe structure is mainly influenced by crystal quality and growth conditions of the epitaxial layers, we have also studied the effect of various changes in this process: particularly the boron doping level and the growth temperature of the Sil_xGex-layers, which are discussed in the following section. The incorporation of electrically active boron atoms into the Sil_xGex-lattice is limited; at high doping levels, more and more boron atoms are incorporated on interstitial lattice sites which causes additional strain and lattice defects. To evaluate the optimum doping level and growth temperature of conventional HIP-structures, four samples with different growth parameters were analysed. All of them contain a heavily doped 1Onto thick Sio.~,~Geo.35-layer separated by the thin ( 2 . 5 n m ) undoped Si~.6sGeo.3s setback layer on an undoped Si-buffer-layer ( 3 0 0 n m ) and a 50 facm Si substrate (Figure 11.12(a5)): 4677 (p++=5xl0X~ -3) and 4 6 7 8 ( p + + = 2 x l 0 2 ~ -3) were grown at 550~ 4679 (p++=5xl0a~ -3) and 4 6 8 0 ( p + + = 2 x l 0 2 ~ -3) at 450~ Figure 11.18(a) and (b) show the experimental results of the photoresponse and the dark current from the HIP detectors. One can see that changing the doping level and growth temperature results in a different absolute magnitude of the photoresponse in the range between 2 lam and 4 lam. At the peak response position of 2.7 lam, the q u a n t u m efficiency varies from ().21% (4678) to 0.33% (4680), whereas the difference in the long wavelength section of the spectrum is only small and results from slightly different barrier heights caused by fluctuation in the Ge-flow during the MBE growth process. The small difference in the threshold energies also induces the shift in the dark current curves of Figure 11.18(b). Generally, samples grown at higher temperatures (46 77 and 46 78) show less photocurrent t h a n samples 4 6 7 9 and 4 6 8 0 grown at lower temperatures, because reduced surface migration during the growth process leads to a better electrical activation 5o of the doping atoms which causes higher absorption and enhanced photocurrent. On the other hand, a change of doping concentration from p + + = 5 x l 0 2 ~ cm -3 to p++=2x 1() 2~ cm -~ at lower growth temperatures,

Infrared silicon~germanium detectors 427

Figure 11.18 Quantum efficienc!!(a) and dark current (b) the ofdetector samples 4677-4680.

does not actually result in less photocurrent, because defect densities decrease and crystal quality improves with lower doping concentration, which leads to a reduced trapping and recombination of photo-generated carriers and thus to a higher efficiency (see also remarks in the theory Section 3.5). The same result can be obtained by absorption measurements from these samples: According to ref. 17 the absorbance of heavily doped HIP-structures is given by

428

Handbook of Infrared Detection Technologies

Figure 1 1 . 1 8 (c) Absorption measurements at 77 K - plotted as absorbance divided b!t 22 - versus NAd (product of doping concentration and thickness of the active ~vell). (d) OE spectrum of a PtSi detector and a SiGe HIP detector in comparison.

Infraredsilicon/,qermaniumdetectors 429 ABS - 22 NAdS*

(5)

where NAd is the total quantity of dopant atoms. S* includes constant terms like effective mass, optical constants, velocity of light etc. Thus the slope of the plot ABS/,;, 2 over NAd should be constant. Results of the absorption m e a s u r e m e n t s of samples with different doping levels but the same growth temperature are presented in Figure 11.18(c)" ABS/,;. 2 increases linearly with increasing NAd in the low-doping section of this curve. At a m a x i m u m of about 3 x 1()2~ c m - ~ the absorbance cannot actually be further increased by higher doping concentrations or thicker Sil_xGex-layers as is evident from the connection line between the data points. This result has also been previously reported by other groups 17 who also obtained absorbance data for higher N~. From this result we can conclude that best detector performance is achieved with a doping concentration of about 2 - 3 • 1()2~ c m - ~ and growth temperatures of ,~4 5()~ where the activation of the incorporated boron atoms is optimized. A comparison of the QE between a platinum-silicide (PtSi) and a SiGe HIP detector with x=0.25 (p++=Sxl()2~cm -~. ds~;,,-15nm) is shown on Figure 11.18(d). Whereas the PtSi detector still has a higher QE for wavelength smaller than 3.5 lxm compared to the HIP, we can clearly see that the SiGe HIP detector has a broader photoresponse spectrum with a tunable cutoff wavelength. The ability of the Ge-composition to tailor the photoresponse curve has important advantages for the MWIR detector since PtSi has a fixed m a x i m u m responsivity at 2.7 ~m with a steep cutoff slope which means that for the range beyond 4 lain, PtSi detectors cannot be used any more Inote that the ordinate scale is logarithmic in Figure 11.18(d). As mentioned earlier, we have used an undoped Si~_xGex-doping setbacklayer between the heavily doped Si~_xGe~ layer and the undoped Si layer in contrast to conventional HIP-structures. ~4~ s This thin doping setback layer forms an additional barrier for excited carriers, resulting in higher threshold energies than determined for samples without these setback layers. This effect is caused by the reduction of the charge densities at the interface of the heavily doped Si~_xGex and the undoped Si-layer which confines the holes within the active layer. Figure l l . 1 9 ( a ) shows the comparison between the measured q u a n t u m efficiencies of these two different variants. The 15 nm thick Si~. sGeo. slayer (p++-5 • 102o c m - 3) of sample 3794 is grown directly on the undoped Sibuffer layer, whereas an additional 1.5 nm thick, undoped Sir162 setback layer (p,-~1 • 1()1 r c m - 3) is added at the Si/SiGe interface of sample 4165 (Figure 11.12(a2)). Though the g e r m a n i u m and boron doping in the active layers are equal for both samples, the threshold energy is increased from 161 meV to 254 meV due to the doping setback-layer (determined from the corresponding Fowler-plots). Therefore the photoresponse of sample 4 1 6 5 drops m u c h faster and the OE decreases from ().2% to ().()7% at 4 ~m. But, as expected, the dark current of these structures is also extremely influenced by the additional layer. At temperatures of 77 K and reverse bias of 5 V, the dark current density drops about four orders of magnitude from 2 x 1 ()- ~ A/cm 2 to 2 • 1 ()-7 A/cm 2

430

Handbook of Infrared Detection Technologies

Figure 11.19 Comparison of quantum efficienc!l (a) and dark current (b) of two HIP samples (p++-~ 102~ c m - ~. dsi~;,,=l O nm ) with ( 4 1 6 ~ ) and without ( 3 7 9 4 ) an undoped doping setback la!ler. ( c ) Comparison of QE of two samples with (4 7 ~8 ) and without (4 7 ~ 7) top-doping setback la!ler.

at 2V due to the additional doping setback-layer (Figure l l.19(b)). As a result, the detectivity D* increases over two orders of magnitude from 1.5• 10 8 c m H z l / 2 / W to 3• 1() lr 1 2 / W at 4 13m which has been calculated assuming the shot noise approximation (see equation (6) below). We measured the dark current and the OE (Figure 11.2()(a+b)) in comparison to the HIP's m u c h lower doped SiGe MOW detectors and from that we evaluated

lnflared silicon~germanium detectors 431

Figure 11.19 (continued).

the detectivity from samples 1 9 4 7 - 1 9 4 9 at 77 K and 5 V bias. The result can be seen in Figure 11.20(c). Quite surprisingly, the OW sample 1949 with only one well exhibits the highest efficiency (QE~,().()7%) which is nevertheless about an order of magnitude lower than the value for the HIP samples. The lowest photocurrent is measured for the N=I() well of sample 1947. This sample, however, also shows the lowest dark current in this series. This results in the highest detectivity D* of 8x 1() ~ cmHzl 2/W at 4.2 ~tm which is confirmed in Figure 11.20(c) where D* has been evaluated using the shot-noise approximation according to

D*- rl ~ q hu

2jd

(6)

Here q is the elementary charge, hv is again the photon energy and Ja the dark current density. To get further increase in detector performance we improved the active layer composition and developed some new layer concepts. The main problem with conventional HIP-detectors is the reduced Debye length in the heavily doped Si/Sil_xGex layers. For doping densities of about 3 • 1 ()2~ cm- 3 LD is less than ] nm, so the applied electric field can only support the transport of excited carriers in a small region next to the Sil_xGex-layer interfaces. Most of the excited carriers have undirected momentum and reach the Si-interface by multiple elastic scattering events if their mean free path is sufficiently high. 17 By introducing an internal electric field by means of a germanium or boron gradient in the Sil_xGex-layer, the momentum would have a preference direction and the escape probability of excited carriers is therefore increased. Figure 11.21(a) shows the comparison between three different types of layer concepts: sample 4680 (as described above), 4684 containing a l ( ) n m

432

Handbook of Infrared Detection Technolo~lies

Figure 11.20 Dark current ( a ). quantum e~cienc!l (b) and detectivit!l ( c ) of the ,\lOW series 194 7 - 1 9 4 9.

In.l?ared silicon~germanium detecwrs

4 33

Figure 11.21 OE (a) and dark current (b) of HIP samples with a boron ( 4 6 8 5 ) and a (;e (4684)-gradient samples ( 4 6 8 4 . 4 6 8 5 ) . Comparison of phowcurrent with and ~vithout A1 top mirror (c) (see text for mesa fabrication).

Sio.6sGeo.3~-layer with a (from bottom to top, see );=0.3 to x=0.5 in the p++-2• All

boron-gradient from 3.9• 1() 2() c m - 3 to 1 • 1() TM cm -3 Figure 1 1 . 1 2 ( a 3 ) and 4 6 8 5 with a Ge-gradient from Sil_xGe• layer but with constant boron doping of samples were grown with the above-mentioned,

434 Handbook of Infrared Detection Technologies

1,21 0,8 9 .

.

.

0,6 .

0,4

energy (eV)

.

0,35

0,35

0,30

~ ' " ~,,.,,~:...

o~" 0,25 i >, 0

c .o_

E C

0,30

with mirror and cavity-oxide

~...

0,25

0,20

0,20

0,15

0,15

0,10

0,10

0,05

0,05

r

without mirror

0,00

9 . ...... 9..........

9

. 1

~

2

.

,

0,00

.

.

~

3 4 wavelength (pm) (c)

.

,

5

. 6

Figure 1 1.21 (continued).

undoped Sio.6sGeo.3 s-setback-layer and a 3()()nm undoped (weakly p-doped) Sibottom contact layer. Under thermal equilibrium conditions, the built-in gradients create an internal electric field with the desired direction. Though SIMS-measurements confirm the intended values for doping concentration and Ge content, the optical measurements indicate different barrier heights for each sample" the graded structures seem to have lower threshold energies than sample 4 6 8 0 with no gradient at all. The values evaluated by Fowler analysis for the given samples are 215 meV for 468(), 203 meV for 4684 and 200 meV for sample 4685. These results are in good agreement with the above made prediction concerning the effect of the built-in electric field. By adding a preference direction to the momentum of excited holes before and after scattering events in the active Si~_xGex-layer, the total amount of scattering events before reaching the Si-interface is reduced compared to the ungraded structure. Thus the effective barrier height of graded samples decreases because of less energy loss of the excited carriers. Nevertheless, the absolute magnitude of the measured QE cannot be increased by introducing a boron or germanium gradient in the heavily doped Sil_xGex-layer. All samples have a peak response at 2.6 pm with values from 0.24% to 0.28% for the graded samples. 4 6 8 0 with a QE of 0.33~ That result becomes more clear if we compare the mean free path length of excited holes within the doped Si~_xGex-layer and its thickness" Strong et al. 17 have determined the values L,,,~ 1.0 nm for elastic scattering and Lie between 140 and 250 nm for inelastic scattering. Thus. at a given Sil _xGex-layer thickness of about 10 nm, even with no internal electric field at all, most excited carriers with sufficiently high energy would be emitted over the barrier by an accidental redirection of their momentum. Furthermore, the dark current density of graded

Infrared silicon/gernmnium detectors 435

samples as shown in Figure 11.21(b) is much higher than for the samples without grading. That is caused by the barrier lowering and the additional momentum of injected holes as mentioned above. To improve the detector performance, we also used an Al-mirror on top of the detector mesa combined with a cavity oxide below (see Figure 11.15(b)). The thickness of the oxide is optimized for peak wavelengths of about 4 ~tm. Due to constructive interference the photoresponse of the IR-detectors can be increased by a factor of 2-3 (wavelength dependent)S~ compared to samples without AImirror and cavity oxide. The reflected wave at the top mirror (Figure 11.21(c)) shows the comparison in OE between two different processing variants of the same sample: 4822 is a conventional HIP detector with a heavily doped (p++-3 x lO 2~ cm -3) Sio.TGeo.3 layer. This sample has been processed with and without the A1 top mirror (see Figure 11.15(b)) and the appropriately chosen cavity oxide, the result of it can be seen in Figure 11.2 l(c). One can see the increase in QE for wavelengths between 2.5 and 5 l~m by the additional top mirror. The thickness of the cavity oxide beneath the Al-mirror was chosen to be 650 nm for maximum amplification at ,-~4 l~m. At this wavelength the response increases by a factor of about three (().17% with mirror, ().O6% without). The peak wavelength is shifted from 2.3 ~tm to 3.1 ~m with values up to ().34%. Consequently, the Fabry-Perot resonator also produces a minimum in the response spectrum, where the incoming wavelength satisfies the )v/2 condition, for sample 4822, roughly at 2 ~ m as expected. For optimum detector performance it is necessary to adjust the thickness of the cavity oxide to the response of the original spectrum to make sure that the reflector minimum does not coincide with the peak wavelength of the detector system. Finally, to get higher photoresponse in the long wavelength regime and reduce the dark current of the detectors, we fabricated multi-well HIP structures. These samples consist of three rather thin (,-~5 nm), heavily doped Sil_xGexlayers separated by thin, undoped (~1 • 101" cm-3) SiGe and Si layers (Figure 11.12(b l-b3)). However, these undoped layers are only a few nm thick (~6 nm) and prevent the increase of dark current due to the additional potential barriers. Due to the heavy well doping these samples are not comparable with common MOW structures, where thick Si-buffer layers (,-~5Onm)separate moderately doped (p~,,10 is cm -3) Sil_xGex layers. Some of the samples were grown with a Ge-gradient in the active layers and an overall reduced Ge content in the active layers compared to the above mentioned 'simple' HIP structures. Figure 11.12(b2) shows the layer design of such a multi-well HIP-structure: sample 4740 contains three Sil_xGex wells, each having an individual linear Ge gradient (Ge fractions are given on the right side) and different thickness, to adjust for the energy-level difference of the ground state to the top of the Si valence band edge with changing Ge content in each well. They are doped only in the centre region (p++-3• and are separated by 6 n m thin undoped Si layers. Sample 4823 has a different Ge concentration in each of the three wells (stepwise Ge grading) and only one undoped Sil_xGex spacer layer in addition to the undoped Si layer (Figure 11.12(bl )). Finally we have grown the '3- well' sample 4825 with constant Ge-content in each well (Figure 11.12(b3)).

436 Handbookof Infrared Detection Technologies

Figure 11.22 Comparison of photocurrent spectra (a) and dark clirrent (b) of' J-well' HIP samples (4823, 4740and4825).

Experimental results of the measured quantum efficiencies and dark current densities are presented in Figure 11.22(a) and Figure 11.22(b). The threshold energies were determined to be 140 meV (48 2 3 ), 190 meV (482 5) and 20 5 meV (4740). Especially, sample 4823 shows an improved photocurrent in the long wavelength regime with a broad response within the whole spectrum. Remarkable is the high quantum efficiency (up to ().4% at 4~m) of 4823

Infrared silicon/germanium detectors

437

Figure 11.22 (c) Comparison of C)Eand the dark current densit!l (d) of the 'l-well' and '~-well' samples 4 8 1 9 and 4 8 2 3 with the same integral number o.fcarriers processed with additional top mirror on the mesa.

combined with a low dark current (conventional HIP-detectors with such low threshold energies show dark current densities up to j,r.~().O 1 A/cm2). The peak response of sample 4 7 4 0 is shifted to smaller wavelengths (().43% at 3 l.tm) combined with less photocurrent in the long wavelength regime due to the overall higher Ge content and broader Sil_xGe-~ wells. The increased dark current of sample 4 8 2 5 compared to 4823 is in contrast to the higher optical

438

Handbook of Infrared Detection Technologies

barrier which has been determined by the Fowler plot. Leakage current over the mesa surface may be responsible for this effect. Nevertheless, all multi-well samples show increased photocurrent in the long wavelength regime (4-5 Bm) combined with rather low dark current densities. So, detector performance of common 'l-well' HIP structures can be increased by the use of the multi-wells which are separated by thin, undoped Si layers. The effect of linear or stepwise Ge grading within the wells seems to be negligible. Further increase of photoresponse in the desired wavelength region could be achieved by optimizing thickness and depth of the Sil_xGex wells. Now it is interesting to compare the '3-well samples' with a common 'l-well' HIP-structure. To take into account the effects caused by the undoped SiGe spacer layers at the top and bottom of the active region, we have grown a ' 1-well' HIP sample (4819) with a l O n m heavily doped (p++=3• 1()2{~cm-3) Si{j.TGeo.3 layer surrounded by two 3.5 nm undoped Si{~.7Ge~.3 setback layers onto a Sibottom contact layer (1OOnm, p - = l • Figure 11.22(c) shows the comparison between the measured OE of 4819 with one well (constant Ge content, x=O.3) and 4823 with three small separated wells (and decreasing Ge content). The spectrum of the photocurrent between 1.5 and 3.5 ~m is nearly equal, reaching a peak at 3.2 ~m with a maximum OE of 0.45% for both samples. At higher wavelengths the 3-well sample 4823 achieves more OE due to the combination of thin SiGe wells combined with the decreasing Ge content throughout all wells. Note that the thickness of each well is different, to adjust for the energetic position of the ground state of each well in the entire structure (for the higher Ge-content wells the ground state is shifted up to the bottom of the Si valence band which can be compensated by increasing the well thickness). Because the total number of doping atoms is nearly the same for the two samples, the amount of excited and collected carriers of 482 3 indicates that all three wells must contribute to the measured photocurrent, thus scattering and recapture of photogenerated holes in a following well is of minor importance. On the other hand, we have seen no further improvement in detector performance, if we use samples grown with more than five wells. This means that, for the given level of doping and Ge content, the optimum number of separated, heavily doped wells lies between three and five. In addition, the lower dark current density of 4823 compared to the 'l-well' sample 4819 is also remarkable (Figure 11.22(d)), because the higher cutoff wavelength of sample 4823 (kco[4823]=6,71 l.tm, )V~o [4819]=5,631.tm, determined by Fowler analysis) should result in an increased dark current. 11.3.5 Calculation of optical properties of SiGe HIP detectors

Band structures of common Si/SiGe HIP structures are well known and reported in a number of papers. 15-17.52 A simple sketch of the band diagram showing the valence band profile of a heavily p-doped (2 x2() 2c~cm- 3) SiGeo.3s well has been sketched in Figure 11.11. Note that the Fermi level lies within the valence band edge of the well at 77 K, above the well there is the continuum of unoccupied states. To get additional information about the effect of the above mentioned Ge

Infrared silicon/germaniun7 detectors

439

Figure 11.23 Simulation of the band structure and electric Jieht for "gradient'samples 4 6 8 4 ( a ) and 468 ~ (b).

and B gradients within the active Sil_xGe• layer of our detector structures, it is necessary to determine carrier densities and potential profiles in thermal equilibrium. These simulations were made with a computer program, solving iteratively Schroedinger and Poisson equations for a given Sil_•215 structure,

440 Handbookof Infrared Detection Technolo#ies doping profile, and external bias voltage. For the gradient structures 4 6 8 4 and 4 6 8 5 - mentioned in Section 11.3.4 - the following results of the valence band profile and built-in electric field have been achieved, which is shown in Figure 11.23(a) and (b). Only the first 30 nm of the structure are shown in the plots the 10 nm heavily doped Si] _• layer, 2.5 nm undoped Si 1_xGe• and part of the intrinsic Si-bottom contact layer. The simulation does not include an external bias. One can clearly see the valence band discontinuity between the Si~_xGex layers and the Si-bottom contact layer at 12.5 nm for the two gradient structures of samples 4 6 8 4 and 4685. The calculated threshold energies of 232 meV (4684) and 226 meV (4685) agree quite well with the experimental results described in the previous chapter. The valence band profiles are quite similar for both samples except the band bending in the low doping section of sample 4684. Because the Sil_xGex-layer of 468 5 is heavily doped (p++=2 x 1()2~ cm- 3), the Ge gradient does not actually result in considerable band bending. The desired effect of the built-in gradients is shown in the upper part of the graphs in Figure 11.23(a) and (b) where the resulting electric field is shown. As expected, both samples have positive field strengths in the active Si~ _~Gex layer that should give photoexcited carriers an additional m o m e n t u m towards the Si interface. Simulations of conventional HIP structures without gradient do not show any field in this part at all. The undoped Sil_xGe~-doping-setback-layer and the Sibarrier cause strong negative peaks of the field at ~ 1 ()nm. In addition to these band structure simulations, we present calculations on the effect of undoped Sil_xGex-spacer-layers that separate the heavily doped Sil_xGex-layers from the Si layers, as mentioned above. We concentrate our attention on the density of states, which should give useful information for the discussion of dark current, and also present calculations of the optical lineshape produced by these systems comparing our calculations with the experimental results. Starting with pseudopotential calculations, including effects due to spin, non-parabolicity and mixing between valence minibands, we introduced a screening Coulomb potential, representing the high concentration of impurities. The alloy concentration in the well is taken to be Sio.7Ge{j. 3 throughout. For doping concentrations of 3 x 102o c m - 3 a self-consistent method was used to calculate the density of states within the well. We model highly-doped SiGe heterostructures where 'dilute models' do not apply due to the large number of impurity potentials per unit volume. We use as a starting point our full-scale pseudo-potential calculation of the valence band where spin and the non-parabolicity of the valence minibands are implicitly included. 53-56 We model the impurities using an added Coulomb potential to represent the impurities, self-consistently screened assuming the random phase approximation. The inclusion of dopants into semiconductor heterostructures can be tackled in three ways. The calculation to establish the miniband structure could be performed and the doping included only through the position of the Fermi level, chosen such that the number of holes introduced by the doping are accommodated in the top valence band levels, s7'~ Thus no account is taken of the extra potential

Infrared silicon~germanium detectors

441

introduced by the impurities in establishing the miniband structure and hence the spread of acceptor levels into a wide band as the doping increases is not included. Secondly, we have our large cluster calculation presented here, where the additional potential due to the presence of the impurities is explicitly included when calculating the miniband dispersion. This results in a full description of the spread in energy of the acceptor states in the high doping limit, allows a more realistic Fermi level to be established, and provides acceptor states into which excitation can occur within the absorption process. Further improvement would require a full ab initio calculation, s€ i.e. a many-electron calculation, which greatly complicates the situation and is outside our scope for large clusters at present. The impurities are considered to be randomly distributed t h r o u g h o u t the well layers of our cluster, which contains some 47 000 atoms. We consider doping concentrations of up to 5 • 1 ()2~ c m - ~ which results in one in every 100 atoms within the alloy well layers being replaced with an impurity atom. We also considered a range of doping concentrations from 1 x 1() ~8 cm-~ to 5 x 102o cm -3. In Figure 11.24 we show the change in the density of states of a 12.5 nm wide Sio.TsGeo.2s alloy well including an undoped 2.5 nm spacer layer, equivalent to the structure of sample 4325 shown in Figure l l . 1 2 ( a l ) , as the doping concentration is increased. We can clearly see the formation of a wide band of acceptor states as the number of impurities increases. For 5 x 1019 c m - 3 this band is some 200 meV wide increasing to 800 meV width for 5 • 102o cm-3. The full description of this acceptor band is one of the key features of our model, and, to our knowledge, has not been previously achieved. The position and nature of these impurity states is crucial to the resulting optical lineshape, as, at the low temperatures considered there, it is these impurity states into which the carriers are excited in the optical absorption process.

Figure 11.24 The change in the densit!l of states with doping concentration,for a Sio.zsGeo.2s allo# well, 12.5 nm wide, incorporating a 2.5 nm undoped spacer la!ler, equivalent to the structure of sample 432 5. The zero of energy is taken to the top of the unperturbed valence band at the zone centre. The vertical lines indicate the top and bottom of the valence well.

442 Handbookof Infrared Detection Technologies We next consider four different situations based on the structure 4 8 2 5 , namely a 6 nm Sio.7Geo. 3 alloy well: 9 9 9

9

Structure 1. A doped region of 3.5 nm with a 2.5 nm undoped spacer, as in experimental structure 4 8 2 5 as shown in Figure 1 1 . 1 2 ( b l ). Structure 2. A 3.5 nm doped region with 1.25 nm undoped spacers on either side of the doped region. Structure 3. No doping setback layers (=spacers), a 6 nm well with doping distributed t h r o u g h o u t the well set to give the same number of impurities as present in structures 1 and 2. Structure 4. No spacers, a 6 nm well with doping distributed t h r o u g h o u t the well at a concentration of 3 • 1 ()2c~c m - 3.

Structures 3 and 4 both have doping distributed t h r o u g h o u t the well, i.e. no spacer layers. However, structure 3 has the same number of impurities as structures 1 and 2, thus a lower actual doping concentration, whereas structure 4 has the same concentration of impurities, therefore, as the doping width is greater, it has a larger n u m b e r of actual impurities present. Experimental results presented in Figure 11.19 for structures with and without spacer layers show little change in the q u a n t u m efficiency spectra on introducing spacers but a dramatic change, of more than three orders of magnitude, in the dark current density. In Figure 11.25(a) we show the change in the density of states produced by the different spacer configurations. We see that there are minor changes to the distribution of the states for the different spacer configurations. In particular there is a change to the distribution of states around the bottom of the well, (zero energy on the diagram). For the structures with no spacers, structures 3 and 4, there is a peak in the density of states just below the well, (zero energy on the diagram). For the structures with spacers the density of states in this region is flatter, with no peak. The states positioned at the bottom of the well are of particular importance to the dark current characteristics of the structure, as it is these states from which carriers can be thermally excited into the continuum, thus contributing to the dark current. However, it seems unlikely that this small redistribution of the states could result in the dramatic reduction in dark current that was observed experimentally. Our calculations of the normal incidence absorption response for structures with and without spacers, presented in Figure 11.25(b), similarly show only small changes in the optical lineshape with n u m b e r and width of spacers. As structure 4 has doping spread t h r o u g h o u t the well, i.e. a larger doping width, it therefore contains more impurities than structures 1, 2 and 3, hence the greater magnitude of the response. The change from one to two spacers results in a shift of the peak absorption of 0.5 l~m. This can be accounted for by a slight difference in the spread of energies t h r o u g h o u t the acceptor band for the two different structures. The main difference in the optical lineshape between the spacer and non-spacer configurations, is the lower magnitude of the absorption above 3 l~m. There is no obvious change in the density of states that would account for this difference. However, the charge densities of the acceptor states will be confined

hffrared silicon~germanium detectors

443

Figure 11.25 (a) The change in the densit!t of states with the spacer width and position. For details of the structure configurations, see text. The vertical lines indicate the top and bottom of the valence band well. The zero of energz3 is taken to be the top of the unperturbed valence band at the zone centre. ( b ) The variation of the normal incidence absorption response with structure, calculated at T - O K. The structures are labelled as described in the text.

within the doped region, therefore for structures with doping setback layers present there will be a smaller degree of overlap between these acceptor states and the states towards the bottom of the well. which are in general centrally located.

444 Handbookof Infrared Detection Technologies Structure 1 is equivalent to the structure of sample 4825. The experimental q u a n t u m efficiency for the sample is presented in Figure 11.22(a). We can see that some of the gross features of our calculated curve are in good agreement with the experimental results, namely the turn-on of the absorption occurs at 2 l~m and peaks at around 4 I~m. The experimental curve sharply drops off at longer wavelength, whereas the calculated curve flattens off. The absorption m e c h a n i s m is excitation from occupied valence states into unoccupied acceptor states. As our calculation produces a large n u m b e r of closely spaced levels around the Fermi energy for the distribution of impurities considered, a large response at low energy (long wavelength) will always occur. A well defined peak would only arise if there was a non-uniform distribution of acceptor states, i.e. the acceptor states form discrete bands. This may result if the added impurities form clusters rather than being distributed randomly t h r o u g h o u t the well - an interesting topic for further investigation. The reasons for the dramatic reduction in the dark current with the introduction of the spacer layers is still not fully understood on a microscopic level. The presence of the spacer layers could cause further confinement of the wavefunctions of states lying within the well but within a kRT spacing from the continuum. The leakage of the wavefunctions out of the well for states near the c o n t i n u u m results in a 'ladder' of states which can be easily populated by thermally generated carriers resulting in a high dark current. If the confinement of the wavefunctions is enhanced by the presence of spacer layers, then this effect, and hence the dark current, may be reduced. However, the interaction of the excited carriers with the interface, the non-ideal nature of the real interface, i.e. roughness, islanding 6~ etc, may also be important factors. In a paper by Shaw 5s on the fundamental physics of SiGe structures, it was demonstrated that the microscopic interaction between the defects and the Si-Ge interfaces results in changes to the electronic wavefunctions that can drastically affect the optical properties of these systems, and concludes that localized interface features will play an important role in determining both the optical and transport characteristics of SiGe structures. At the extremely high doping concentrations considered here, where up to 1% of the original atoms are replaced by the dopant boron atoms, the interaction of these dopants with the interfaces may well be highly significant. Further investigation is needed to determine the ultimate effect of the interaction. This is a first attempt at modelling these highly doped structures, and although some success was achieved, there needs to be more investigation of how the impurities are included for these very high doping concentrations, the sensitivity of the optical response to the type and position of these impurities, as well as a more detailed investigation of the effect of the interaction of the defects with the interface. 77.3.6 R~sum~ and outlook for SiGe MWIR detectors

We have discussed several types of Si/SiGe HIP and MQW detectors grown by MBE and fabricated as single mesa as well as focal plane array detectors. Various structural designs such as multi-well, Ge and B gradient and the use of doping

hffrared silicon/germanium detectors 445

setback layers on one and on both sides of the highly doped SiGe well have been applied and the detector characteristics have been compared. Broad photoresponse curves with a m a x i m u m OE of ~O.5% at a peak wavelength between 3 l.tm and 4 l.tm, at 77 K have been obtained. By using doping setback layers in addition to an appropriate passivation to suppress surface leakage currents detectivities of up to 9• 1()11 cmv/Hz/W can be achieved, a value which, to our knowledge, has not been obtained by any other III-V detectors for this wavelength range. Reports of higher QEs by other authors must be carefully analyzed. The reported m a x i m u m OE of 8% from ref. 14 has been measured at a m u c h lower temperature of 30 K. In addition, these structures had unacceptably high dark currents for device fabrication purposes. Other authors report very similar values for the OE and the dark current as we have for comparable structures, a6'17 SiGe/Si MQW samples have a lower doping level compared to SiGe HIPs therefore, in general, lower dark currents and higher detectivities are possible in these structures. A detectivity of D*> ~l()l(~cmv/Hz/W has been achieved for a ] 0 well OW sample. However. with an order of magnitude lower photoresponse compared to a standard HIP structure. In comparison to PtSi SiGe, HIPs do not show a higher OE between 2.5 and 3 ~ m (Figure 11.18(d)). However, by changing the well geometry and/or composition, one is able to change the cutoff wavelength and tune the m a x i m u m of the photoresponse curve over a rather broad wavelength range (2 ~tm,-~4 ~tm). Since the barrier height strongly determines the dark current at a fixed temperature (see Richardson equation (2)) SiGe HIP's with rather short cutoff wavelength have also lower dark currents and higher detectivities. Furthermore, there is strong evidence that in multi-well HIP samples with three wells, all wells contribute to the photocurrent. This means that by proper choice of the well geometry and separation, the carrier trapping during transport of the excited carriers to the contacts can be overcome, which gives some potential for further improvements of the SiGe HIP detectors. The SiGe material system is also unique, not only because of the already mentioned ability to realize near infrared as well as mid-infrared and longinfrared detectors in the same material system but also because of the favourable selection rules of the quantized states for normal incident radiation, which do not require expensive and complicated coupler gratings as in the case of A1GaAs/ GaAs OW infrared photoconductors (QWIP). This makes the SiGe intersubband detectors for the LWIR spectral range attractive. In addition the compatibility of SiGe detectors to the Si FPA fabrication process could - provided there are further improvements in the O E - make SiGe MWlR detectors an attractive alternative to the commercially available silicide on one side and the III-V OWIP detectors on the other side.

Acknowledgements I would like to acknowledge the work of m a n y of my colleagues, for XRD m e a s u r e m e n t s we are grateful to Mr. H.-J. Herzog, fo RBS m e a s u r e m e n t s to Mr.

446

Handbookof Infrared Detection Technologies

K. Schmidt (from KFA Jiilich) and for the sample growth to Mr. Horst Kibbel. For reading the manuscript and numerous measurements we are especially indebted to Mr. Johannes Konle.

References 1. E. Kasper, H.-J. Herzog, H. Jorke and G. Abstreiter, Mat. Res. Syrup. Proc. 1 0 2 , 3 9 3 (1988). 2. H. Presting, T. Zinke, A. Splett, H. Kibbel and M. Jaros, Appl. Phys Lett. 69, 2376(1996). 3. T. Lin andJ. Maserijan, Appl. Phys. Lett. 57, 1422 (1990). 4. J. Uschmann, H. Presting, T. H. Kibbel, K. Thinke and R. Sauer, Thin Solid Films 2 9 4 , 3 4 0 - 3 4 2 (1997). 5. T. Fromherz, E. Koppensteiner, M. Helm. G. Bauer, J. F. Ntitzel and G. Abstreiter, Extended Abstracts of the International Cotfference on Solid State Devices and Materials, Makuhari (Japan), p. 41 ()-412 ( 1993 ). 6. A. Splett, T. Zinke, K. Petermann, E. Kasper, H. Kibbel, H.-J. Herzog and H. Presting, Photonics Technology Letters PTL-6, 42 5 ( ] 994). 7. J. Engvall, J. Olajos, H. Grimmeiss, H. Kibbel and E. Kasper, Appl. Phys. Lett. 63,491(1993). 8. B.Schtippert, J. Schmidtchen, A. Splett and K. Petermann, Integrated Optics in Silicon, Conference on Microsystem Technologies ed. by H. Reidel, p. 27B (1993)" B. Sch(ippert, J. Schmidtchen, A. Splett, U. Fischer, Th. Zinke, R. Moosburger and K. Petermann, Journal ofLightwave Technolog!t 14, 2 311 (1996)" see also R. A. Soref, Si based optoelectronics, Rev Article, Proc. IEEE 81, 1687 ( ] 993). 9. B. Jalali, S. Y e g n a n a r a y a n a n , T. Yoon. T. Yoshimoto, I. Rendina and F. Coppinger, Advances in Silicon-on-Insulator Optoelectronics, IEEE Jollrnal of Selected Topics in Quantum Electronics, Special Issue on Silicon-Based Optoelectronics, 4(6), 9 3 8 - 9 4 7 (December 1998 ). ] O. R. Breiter, W. Cabanski, R. Koch, W. Rode, J. Ziegler, K. Eberhardt and R. Oelmaier, SPIE Proceedings 3 3 7 9 , 423, (1998) and R. Breiter, W. Cabanski, K. H. Mauck, W. Rode, J. Ziegler, K. Eberhardt and R. Oelmaier, SPIE Proceedings 3 3 7 9 , 344 (1998). 11. S. D. Gunapala, T. N. Krabach, S. V. Bandara, J. K. Liu and M. Sundaram, SPIEProceedings 3 0 6 1 , 2 9 2 ( 1 9 9 7 ) a n d S. D. Gunapala, S. V.Bandara, J. K.Liu, W. Hong, M. Sundaram, R. Carralejo, C. A. Shott, P. D. Maker and R. E. Miller, SPIE Proceedings 3 0 6 l, 722 ( 1997). 12. H. Schneider and Eric Lerkins, Semicond. Sci. and Technol. 10 1329-1338 (1995). 13. H. Schneider, C. Mermelstein, R. Rehm, C. SchOnbein, A. Sa'ar and M. Walther, Phys. Rev. B57, 15()96 ( 1998 ). 14. T. L. Lin, T. George, E. W. Jones, A. Ksendzov and M. L. Huberman, Appl. Phys. Lett. 6 0 , 3 8 0 (1992).

hlfrared silicon/germanimndetectors 447 15. T. L. Lin, J. S. Park, S. Gunapala, E. W. Jones and H. M. Del Castillo, Opt. Eng. 3 3 , 7 1 6 (1994). 16. B.-Y. Tsaur, C. K. Chen and S. Marino, Opt. Eng. 33, 72 (1994). 17. R. Strong, R. Misra and D. W. Greve, ]. App1. Phys. 82, 5191 ( 199 7): and R. Strong andD. W. Greve, ibid 5199 (1997). 18. F. D. Shepherd, V. E. Vickers and A. C. Yang, Schottky barrier photodiode with a degenerate semiconductor active region, U.S. patent No. 3 603 84 7 filed Sept. 7,1971 19. S. Lury, A. Kastalsky and J. C. Bean, IEEE Transactions on Electron Devices 31,1135(1984). 20. E. Kasper, H. Kibbel and H. Presting, Thin Solid Fil,ls 183, 87 (1989). 21. H. Presting, H. Kibbel, M. Jaros, R. M. Turton, U. Menczigar, G. Abstreiter and H. G. Grimmeiss, Semicond Sci. atld Techllol. 7, 112 7 ( 1992 ). 22. E. A. Fitzgerald, Y.-H. Xie, M. L. Green, D. Brasin, A. R. Kortan, J. Michel, Y.-J. Mii and B. E. Weir, Appl. Phys. Lett. 59, 811 ( 1991 ). 23. G. Masini, L. Colace, G. Assanto, H.-C. Luan and L. Kimerling, IEEE Trans. on Electron Devices 48, 1092 (20() 1 ). 24. L. Colace, G. Masini, F. Galluzi, G. Assanto, G. Capellini, L. DiGaspare, E. Palange and F. Evangelisti, App1. Phys. Lett. 72, 3175 (1998). 25. G. Masini, V. Cencelli, L. Colace, F. De Notaristefani and G. Assanto, to be published App1. Phys. Lett. (2002). 26. M. Jaros, Semiconductors and Sel~lilnetals 32, 1 75 (1990). 2 7. M. Gail, G. Abstreiter, J. Olajos, J. Engvall, H. G. Grimmeiss, H. Kibbel and H. Presting, Appl. Phys. Lett. 66, 29 78 ( 1995 ). 28. U. Gnutzmann and K. Clausecker, Appl. Phys. 3, 9 ( 19 74). 29. V. d. Merwe, Surf. Sci. 31, 198 ( 19 72 ). 30. R. People and J. C. Bean, App1. Phys. Lett. 47, 322 (1985): Erratum: APL 4 9 , 2 2 9 (1986). 31. W. A. Brantley, J. Appl. Phys. 44, 534 ( 19 73 ). 32. K. B. Wong and M. Jaros, Semicond. Sci. and Technol. O, 790 ( 1991 ). 33. G. Abstreiter, K. Eberl, K. Friess, W. Wegscheider and R. Zachai, J. Cryst. Growth 9 5 , 4 3 2 (1989 ). 34. E. Kasper and K. W6rner, ]. Electrochem. Society 1 3 2 , 2 4 8 1 (1985). 35. H. Kibbel andE. Kasper, Vacll,m 4 1 , 9 2 9 (199()). 36. M. Copel, M. C. Reuter, M. Horn v. Hoegen and R. M. Tromp, Phys. Rev. B42, 1 ] 682 (1990): E. T. Croke, T. C. Mc.Gill. R. J. Hauenstein and R. H. Miles, Appl. Phys. Lett. 56, 367 ( 1989 ). 3 7. H. Presting and H. Kibbel, Thin Solid Fihns 222, 2 1 5 - 2 2 0 ( 1992 ). 38. H. Temkin, J. C. Bean, T. P. Pearsall, N. A. Olsson and D. V. Lang, AppI. Phys. Lett. 49, 155 (1986). 39. H. Temkin, T. P. Pearsall, J. C. Bean, R. A. Logan and S. Lury, App1. Phys. Lett. 48, 963 (1986). 40. H. Temkin, A. Antreasyan, N. A. ()lsson, T. P. Pearsall and J. C. Bean, Appl. Phys. Lett. 4 9 , 8 0 9 (1986 ). 41. F. Y. Huang, X. Zhu, M. O. Tanner and K. L. Wang, Appl. Phys. Lett. 67, 566(1995).

448 Handbookof Infrared Detection Technologies 42. F. Y. Huang and K. L. Wang, AppI. Phys. Lett. 69, 2 3 3 0 (1996). 43. T. Tashiro, T. Tatsumi, M. Sugiyama, T. Hashimoto and T. Morikawa, IEEE Transactions on Electron Devices 44. "545 ( 1997). 44. S. B. Samavedam, S. M. T. Currie, T. A. Langdo and E. A. Fitzgerald, Appl. Phys. Lett. 73, 2125 (1998). 45. X. Shao, S. L. Rommel, B. A. Orner, H. Feng, M. W. Dashiell, R. T. Troeger, J. Kolodzey, P. R. Berger and T. Laursen, Appl. Phys. Lett. Y2, 1860 (1998). 46. L. Colace, G. Masini and G. Assanto, IEEE I. Quantum Electronics 3 5, 1843 (1999). 4 7. F. D. Shepherd, Infrared Internal Emission Detectors. State of the Art, ed by W . H . Makky. Proc. SPIE 1 7 3 5 , 25()-261 (1992). 48. H. Presting and M. Jaros, SiGe IR detector with ballistic carrier injection, patent filed at the German patent office, Oct. 1998. 49. H. Presting, J. Uschmann and H. Kibbel, Thin Solid Films 3 2 1 , 1 8 6 - 1 9 5 (1998). 50. C. M. Parry, S. M. Newsstead, R. D. Barelow, P. Augustus, R. A. Kubiak, M. G. Dowsett, T. E. Whall and E. H. C. Parker, APL 58, 481 (1990). 51. R. T. Carline, D. J. Robbins, M. B. Stanaway, W. Y. Leong, App1. Phys. Lett. 68,544(1996). 52. R. Strong, D. W. Greve, R. Misra, M. Weeks and P. Pellegrini, Thin Solid Films 2 9 4 , 3 4 3 (1997). 53. E. Corbin, C. Williams, J. P. Hagon. M. Jaros and H. Presting, Thin Solid F i l m s 2 9 4 , 186 (1997). 54. E. Corbin, K. B. Wong and M. Jaros, Phys. Rex,. B 50, 2339 (1994). 55. M. J. Shaw and M. Jaros, Semicondllctors and Semi-metals 56, ] 69 (1999). 56. K. B. Wong, M. Jaros, I. Morrison and J. P. Hagon, Phys. Rex,. Lett. 60, 2221 (1988). 57. E. Corbin and M. Jaros, Semicond. Sci. and Technol. 1 2 , 1 6 4 1 ( 1997). 58. D. C. Herbert, Semicond. Sci. and Technol. 13, 1090 (1998). 59. M. J. Shaw, P. R. Briddon and M. Jaros, Phys. Rex,. B, 54, 16781 (1996). 60. M. Jaros and A. W. Beavis, App1. Ph!ls. Lett. 6 3 , 6 6 9 ( 1993).

Chapter 12

PolySiGe uncooled microbolometers for thermal IR detection Chris Van Hoof and Pier De Moor

12.1 Introduction This section introduces polySiGe bolometers and its relative place in the field of resistive bolometers. Even though other chapters in this book are devoted to microbolometers, the subject of microbolometers is very briefly discussed at the end of this section for reasons of portability and to ensure uniform terminology. 12.1.1 Uncooled resistive microbolometers

Uncooled thermal imagers using (mainly resistive) microbolometers have caused a boom in thermal imaging because of reduced sensor system cost. 1 The yearby-year improvement in noise-equivalent-temperature-difference (NETD) from approximately lOOmK in 19932 to 18mK in 2001 for imagers with 50 gm• 50 ~tm pixels and the simultaneous improvement in higher-resolution imagers (also using 25 l~m• ~m pixels) has increased their competitiveness in high-end application areas such as military and medical. The first decade of microbolometer development focused on vanadium oxidebased resistive microbolometers, following the Honeywell reports and patents. 2 The potential of this material system and the ensuing concentrated research was largely reponsible for the observed improvements in the technology. In spite of the superlative uncooled performance of VOx microbolometer arrays, uniform sputter deposition of low-noise VOx material on foundry CMOS wafers proved a challenge and the optimum deposition process remains difficult (e.g. due to the tight oxygen content control needed). Especially when low-cost thermal imaging for, e.g., automotive night vision became a possible

450 Handbookof Infrared Detection Technologies mass-market application, a far lower-cost solution was needed than could be achieved using Vox microbolometers. This lead to a cost versus performance trade-off which had to be made for all parts of the system and mainly concerned the sensor array, the optics and the packaging. In the latter field it has lead to the development of zero-level v a c u u m packaging R&D and wafer-scale packaging, 3 and in the former to low-cost molded optics development. To reduce the sensor die cost (in essence a yield factor), it was necessary to re-evaluate the alternative bolometer schemes as lower-cost candidates. Many materials have served as resistive elements in microbolometers, metals (e.g. Ti4), crystalline material (pn diodes in SIS), polycrystalline material, 6 amorphous material (a-Silicon 7). However, low-cost thermal sensing dies imply, not only adequate sensor material, but also surface-micromachining compatibility and either CMOS back-end compatibility or direct front-end integration. The latter category covers bulk-micromachined crystalline Si, Si pn diode, polySiGe, and the former category concerns a-Si and Ti microbolometers. At present, the quest for low-cost uncooled imaging has not been concluded and several options which all share CMOS process compatibility are being developed in parallel: on one hand Ti-metal bolometers and on the other hand amorphous or polycrystalline Si or SiGe alloys. The choice for either system is not an easy one since both have advantages and drawbacks. Whereas metal microbolometers feature low TCR (0.1-0.2% 4) this is compensated by the low noise in these metals and the very high wafers-scale uniformity. In addition, thermal isolation of metal microbolometers is generally worse t h a n that of their polycrystalline or amorphous counterparts and the lowest NETDs reported for metal bolometers are 2 ()0 mK. The higher TCR (2-5 %) in polycrystalline or amorphous bolometers is accompanied by increased noise, but the thermal isolation levels achieved in these microbolometers are superior (e.g. below 8E-8 W/K thermal conductance) leading to reported NETDs below 60 mK for a-Si. The fact that both systems are easier to fabricate than Vanadium oxide and are more reliable materials, compensates for their higher NETD compared to Vanadium oxide. PolySiGe was indicated as a promising material system because it features low thermal conductance, high TCR and moderate noise levels combined with low stress, suitable for surface micromachining. Furthermore, it is a CMOS frontend material used in BiCMOS processes and its process modules are available in the IMEC pilot line. Therefore, compared to nearly all microbolometer processes, deep-submicron lithography is used which (among other outcomes) leads to superlative uniformity in direct material parameters (TCR, noise) and indirect process-related properties (thermal isolation)over entire 8" wafers. This chapter is devoted to a discussion of the merits and shortfalls as well as the present state-of-the-art of polycrystalline SiGe microbolometers and is organized as follows: In Section 12.2, the different deposition techniques and mechanical properties of polySiGe are presented. Section 12.3 presents polySiGe bolometer pixels and their electro-optical properties (such as TCR, noise, thermal isolation,

PolySiGe uncooled microbolometers for thermal IR detection

451

uniformity .... ) Section 12.4 concerns array development, readout requirements and system development. Section 12.5 discusses zero-level v a c u u m packaging development of polySiGe microbolometer arrays. Section 12.6 concludes and provides an outlook for the future. 12.1.2 M i c r o b o l o m e t e r t e r m i n o l o g y

A bolometer is a thermal infrared detector. In order to obtain significant heating it must be thermally insulated from its heat sink (typically the substrate) by a support having a low thermal conductance G. The active element in a bolometer is typically a resistor with a large temperature coefficient of the resistance (TCR). As a consequence, a significant change in resistance occurs when the detector is heated by incident infrared radiation. A basic thermal block diagram of such bolometer is shown in Figure 12.1. The effect of varying incident thermal radiation creates a temporary thermal imbalance since the pixel absorbs the radiation and warms up faster than the heat sink. The heat capacity C of the pixel determines the response rate and the thermal balance is described by dT

ot P - C-d-i+ G ( T - To)

(1)

where P is the incident infrared power, :x the absorption coefficient and To is the heat-sink temperature. As a consequence, in steady-state, the temperature increase of the bolometer is proportional to absorbed power and inversely proportional to the thermal conductivity: high thermal isolation leads to more heating. By solving the above differential equation, the thermal time constant for reaching the new thermal equilibrium is given by r = C / G . It is clear that the signal is m a x i m u m when pixel heating is m a x i m u m and this is achieved by optimizing the absorption efficiency and decreasing the thermal conductance G.

Figure 12.1 Thermal block diagram of a microbolometer listing the key parts and the physical quantities that govern the microbolometer behavior: thermal capacitance, thermal conductance, incident power, temperature of pixel and heat sink.

452

Handbook of Infrared Detection Technologies

Given a constant temperature rise, the higher the TCR of the pixel material, the higher the signal will be. For a practical implementation, it is only the thermal isolation requirement (G) which sets microbolometers apart from other temperature sensors which need to measure the surrounding temperature rather than incident radiation. Thermal isolation can be obtained by removing most of the underlying material as is done in bulk silicon micromachining. A superior technique used in nearly all microbolometer implementations involves surface micromachining: a sacrificial layer is removed in the last process step. thereby creating a suspended structure. A typical microbolometer is shown in Figure 12.2. 12.1.3 M i c r o b o l o m e t e r

process options

Two distinct types of microbolometers exist (Figure 12.3): either self-suspended microbolometers or bolometers supported by an insulating membrane. In the self-suspended microbolometer system, the resistive sensor provides at the same time structural support and electrical contact. Although this puts a requirement of low-stress on the material, it generally allows to maximize the thermal isolation of the pixels. In the second approach, the microbolometer material still has to fulfill certain mechanical requirements, but the support is created by a membrane (e.g. silicon nitride) which is typically below the sensor, but can also be above the bolometer pixel material. Because electrical contacts to the pixel are

Figure 12.2 (a) Translation of the basic bolometer diagram of Figure 12.1 into a practical scheme indicating the different parts. (b) SEM photograph of a :,012 • :~0 12m pol!lSiGe microbolometer suspended b!! two approximately 50 12m long legs.

Figure 12.3 Two types of microbolometers: microbolometers ~vith separate membrane for suspension and self-suspended microbolometers. The bolometers supported b!l a membrane typicall!t have separate sensor and interconnect materials (left). ~vhile the electrical and mechanical fi~nctions are combined in one material in the self-suspended bolometer (right).

Pol!lSiGe uncooled microbolonwters for thermal IR detection

4 53

required, the thermal isolation of such microbolometers is generally worse, but it is the only way to make, for example, YBaCuO-based microbolometers, since this material is not suited as structural material for micromachining. Ti-metal, a-Si and polySiGe bolometers are examples of self-suspended microbolometers, whereas all other schemes use a separate electrical interconnect and m e m b r a n e suspension.

12.2 Structural, thermal and electrical properties of polySiGe This section discusses the mechanical, thermal, and electrical properties of polycrystalline silicon-germanium thin films in relation to the deposition conditions. 8-1~ SiGe is well established as a front-end material as high-speed HBTs use SiGe epitaxial base layers and polySiGe base electrodes within a modified BiCMOS front-end process. The high deposition temperature is perfectly compatible with the front-end CMOS but presents a formidable challenge for the MEMS sensor processing as this type of processing ideally takes place after the full CMOS process and therefore requires lower deposition temperatures. Low-temperature deposition is discussed further in this section. For application of polySiGe in microbolometers, the Ge content is increased with respect to that of SiGe HBT (i.e. from 10% to 30%) which decreases the thermal c o n d u c t i v i t y of the material.

12.2.1 Deposition of polySiGe Polycrystalline silicon-germanium alloys are typically grown in a chemical vapor deposition reactor. The reactor used at IMEC is an ASM Epsilon I consisting of a horizontal, lamp-heated quartz chamber with a SiC coated graphite susceptor. Layers were initially grown on 6" silicon wafers and since 2()()() on 8" silicon wafers, covered where needed by 2 ~m of sacrificial TEOS. Samples can be grown at atmospheric pressure (APt or at a reduced pressure (RP) of 40 Torr. The deposition temperature is around 650 ~ For the deposition a mixture of germane and dichlorosilane is used and the proportion of the gases is adjusted to give a thin film containing 30% of germanium. The growth rate in such conditions amounts to O. 5 nm/s. As practical bolometers are between 1 O0 and 250 nm thick, the deposition time is limited to an acceptable 3-9 minutes. Thicker bolometers have also been made (up to 1 ~m). The nucleation of polySiGe on top of the sacrificial oxide layer and using the above gases is very slow. To improve the nucleation, a thin polySi nucleation layer is used which is approximately 10 nm thick and grown at 6()() ~ Our initial work used boron implantation to achieve the desired electrical resistivity of the polySiGe layers and of the supporting legs. After ion implantation (3x 1() 1~ B/cm 2 for the pixels and 1() l~' B/cm e for the legs), a hightemperature anneal was done (at approximately 85()~ The low-dose implantation was later replaced by in situ doping which provides more constant doping concentration in the pixel and does not require a high-temperature

454

Handbook of Infrared Detection Technologies

anneal step. The low leg resistance is still achieved by high-dose implant followed by anneal at 650 ~ 12.2.2

Structural properties

As data ofpolySiGe are limited, it is instructive to compare Si and Ge. The melting point of Si is 1415 ~ and that ofGe 9 3 7 ~ The density of Si is 2.32 g/cc and that of Ge 5.33 g/cc. The modulus of Young of polySi is 173 GPa and that of polyGe is 132 GPa (each about 10% lower than their monocrystalline counterparts). The fracture strength of polySi is 2.6 GPa and that of polyGe is 2.2 GPa. Combining these data gives a rough indication that polySiGe requires lower temperature processing than poly Si and may have adequate mechanical properties. For a material to be used as structural MEMS material, the stress in the layers is of prime importance. The stress of the layers can be measured prior to processing by analyzing wafer curvature. The resulting stress in layers grown by RP or AP CVD is shown in Figure 12.4 as a function of subsequent annealing temperature. The stress in AP CVD polySiGe layers is typically compressive but becomes nearly zero for annealing temperatures around 90() ~ The stress in RP CVD polySiGe is tensile and almost independent of annealing. By varying the pressure, the stress in the polySiGe can be further tuned. Although low stress is important, its importance for microbolometer application should not be exaggerated, as the polySiGe is just one of the materials in the pixel stack (microbolometers also feature isolator and absorber material). As a consequence, the residual stress or stress gradient in the multilayer stack is critical. It is, however, possible to compensate stress and stress gradients in multilayer stacks and this is demonstrated in Section 12.3.

Figure 12.4 Stress of different materials measured as a function of annealing temperature. Low stress is obtained particularly in RPCVD pol!tSiGe. This method was used for all actual device work and deposition took place at 62 5-6 50 ~

Pol!lSi(;e uncooled microbohmleters for thermal IR detection

455

12.2.3 Thermal properties Compared to metals and monocrystalline silicon, polycrystalline material shows much lower thermal conductivity #. Whereas aluminum and monocrystalline silicon have a thermal conductivity of approximately 200 W/mK and 1 O0 W/mK, that of polycrystalline silicon is 18 W/mK (due to phonon scattering at the grain boundaries). By adding germanium, phonon scattering at point defects further reduces the thermal conductivity, and polySiGe with 30% Ge achieves a thermal conductivity of 3 W/mK (i.e. 2 5 times less than monocrystalline silicon and more than 50 times lower than aluminum). It will be shown in the next section that polySiGe shows similar or slightly superior electrical performance over polySi, so the key advantage of higher thermal isolation remains. It is instructive to calculate the thermal conductance for a typical poly SiGe pixel. Assuming a beam section of 1 square micron, and a beam length of 50 micron, the thermal leak through (typically) two beams equals 1.2 x 1 ()-7 W/K. When reducing the poly SiGe thickness, beam sections below 1 square micron are possible, resulting in a thermal conductance close to the radiation limit (the thermal leak through radiation). 12.2.4 Electrical properties

As polySiGe is to be used as a resistive element, the interrelation between resistivity, temperature coefficient of the resistance (TCR) and noise is essential. Data have been collected for ion-implanted AP CVD and for P,PCVD polySiGe. The electrical properties of implanted or in s i t u doped polySiGe are found to be identical. The 500 nm thick samples were implanted at a fixed energy of 7:3 KeV by a boron doping dose varying from 1()11 boron/cm 2 to 101 ~ boron/cm 2 and annealed at 850 ~ for one hour. Sheet resistance and temperature coefficient of the resistance have been measured by monitoring the current as a function of temperature at fixed bias. Care has been taken to avoid contact resistance effects on resistivity determination. Figure 12.5 shows the resistivity of APCVD polySiGe layers and the temperature coefficient of the resistance (TCR) versus implantation dose. It is possible to achieve 4% TCR in highly resistive layers, but high resistivity leads to high ]/f noise. The figure shows that a practical 1% TCR is achieved for approximately 1 f~cm resistivity. A similar result is obtained in RPCVD polySiGe. This general trend of resistivity and TCR versus implantation dose is similar to the one observed in poly Si: one can distinguish a low doping region where the sheet resistance is doping independent and a region where sheet resistance sharply decreases with the increase of doping. We note that the sheet resistance starts to decrease steeply when the doping exceeds 1 ()1~ boron/cm 2. which corresponds to an average doping density of about 2 x 1 ()~ 7 boron/cm 3. Since this is similar to observations in polySi, this indicates that grain size and grain defect density are similar in the two materials. The comparative values of TCR and resistivity of polySiGe and polySi are summarized in Table 12.1 where polySi was annealed for one hour at 850 ~ and

456

Handbook of Infrared Detection Technologies

Figure 12.5 Dependence of APCVD polySiGe resistivit!l ( II ) and of TCR ( 9 ) on boron implantation dose. For low-dose implant, the TCR is as high as 4% but practical (lower noise) la!ters fi'ature approximatel!l 1% TCR and 1 f2cm resistivity.

T a b l e 12.1 D e p e n d e n c e of sheet resistance and TCR in AP poly SiGe and AP poly Si o n doping doses. The electrical properties are s i m i l a r . Doping Dose

1.5 x 1019 3 x 10 ~~ 9 x 1013

Sheet resistance

TCR (%)

AP poly SiGe

AP poly Si

AP poly SiGe

AP poly Si

3 5 0 KF2/[-] 173 K~2//-/ 19.5 KF2/[-1

4.3 Mr2/[-] 58 5 K f 2 / ~ 2 4 . 7 Kf2/[--1

-2.54% -2.1% -().96%

- 3.29% -2.2% - 1.1%

polySiGe at 650~ The table also demonstrates that for the same doping dose, poly SiGe has lower noise since 1/f noise scales with resistivity. Further evidence is obtained from noise measurements on these layers. Figure 12.6(a) compares noise spectra ofpolySi and polySiGe for 1 ~tm thick samples. At low frequencies, the noise spectrum is dominated by 1/f noise as can be expected in polycrystalline (or amorphous) materials. However. the 1/f noise in polySiGe is a factor of two lower than that of polySi. Figure 12.6(b) compares layers of polySiGe with different resistivities and it is indeed observed that the 1/f noise scales with the square-root of the resistivity. As both the bolometer signal (given by TCR which depends on resistivity) and bolometer noise (dominated by 1/f noise which depends on resistivity) are resistivity-dependent, an optimum signal-to-noise ratio has be found which is a compromise between noise and TCR. However, the problem is more complex than that as noise of the readout circuitry has to be included. This figure depends on the readout architecture and this in turn is affected by the absolute resistance of the bolometers which has to be read out by the circuit. In practice, it will be necessary for polySiGe bolometers to achieve read out noise below or equal to the

Pol!lSiGe uncooled microbolometers for thermal IR detection

4 57

$ 9

0%% ee e 9

0,5

0,1

9e 4~ee ~ 1 7 6

~

1

~

0.2

. . . . .

1

S

20 Frequency (Hz) (a)

200

0.01 1 Frequency(KHz)

100

(b)

Figure 12.6 (a) Noise spectra for 50 l~m• ~0 l~m. 1 I~m thick pol!tSiGe samples: O A P poly Si. 9 AP poly SiGe. The 1/f noise of the pol!lSiGe la!ters is approximatel!l half that of the pol!!Si. (b) Noise spectrum of polySiGe layers with varying resistivit!l. B!I changing the resistivit!l a fi~ctor of 10. the 1If noise scales with the square-root of the resistivitz3.

bolometer 1/f noise. This contrasts with metal bolometers where the readout noise is dominant.

12.2.5 High-temperature versus low-temperature polySiGe The deposition conditions described in Section 12.2.1 are in line with a CMOS front-end but clearly conflict with CMOS post-processing. Several groups have demonstrated low-temperature polySiGe deposition for MEMS applications. 11 Thin films of polySiGe were deposited in an Oxford Plasma Technology reactor at 2 Torr which allows to achieve sufficiently high growth/deposition rates in spite of the reduced substrate temperature. As modern CMOS back-end processes can withstand temperatures of 450 ~ ~ for an extended period of time, 12 the deposition temperature was between 4 5 0 ~ and :360 ~ and the deposition time between 20 minutes and one hour. Practical layers were those deposited at 520 ~ which featured 0.8 f2cm resistivity and a stress below 50 MPa. These films have higher Ge content (50%) than their high-temperature counterparts. 13

12.3 PolySiGe bolometer pixels 12.3.1

Process development

The polySiGe bolometer processing is schematically shown in Figure 12.7 and is run in an 8" CMOS pilot line. Although a typical bolometer process encompasses between 50 and 100 lot turns, the process can be summarized as follows: First, a sacrificial silicon oxide (i.e. TEOS) layer is deposited and patterned. Then, polySiGe is deposited on the wafer followed by silicon nitride isolator deposition. In a third stage, the polySiGe/silicon nitride stack are each patterned

458

Handbook of Infrared Detection Technologies

Figure 12.7 Schematic process flow of pol#SiGe microbolometers. 1) TEOS deposition,patterning and etching, 2) deposition, patterning and etching of pol!tSiGe SiN. 3 ) Metal contact and absorber definition.

and the bolometer legs undergo high-dose implant. Then, metal contacts are deposited and patterned and the IR absorber is deposited on top of the pixel. Finally, the sacrificial TEOS layer is removed (either in a HF solution or in vapor HF, see further). The SEM photograph of a single-pixel microbolometer can be seen on Figure 12.2 (Section 12.1.2). While the initial process and material development was carried out on 6" wafers, 90% of the current microbolometer process is done in our (sub-) 100 nm CMOS pilot line on 8" Si wafers. Therefore, we benefit from sub-micron stepper lithography and advanced deposition/etching technology resulting in excellent uniformity of the material. 14 In a typical design, each 8" wafer contains 1300 arrays with more than 80 000 pixels in total. Roughly half of the pixels are being probed using an automatic probe station. First, the resistance variation was studied. On full 8" wafers, no single pixel having a resistance out of a +2% range around average was found. A +1% resistance uniformity requirement results still in a yield of 99.9%. The typical resistance uniformity over a 128 element linear array is shown in Figure 12.8. We measured a 0.4% peak-to-peak and +().1% standard deviation nonuniformity. However, these values include the m e a s u r e m e n t uncertainty. This excellent resistance uniformity is the result of both a very good poly SiGe deposition control and an accurate lithography and etching. This is only possible thanks to the advanced equipment of the 8" IMEC CMOS pilot line. The high level of uniformity allows longer integration time and significantly simplifies the offset and gain correction problems in a current integration based read-out scheme.

Pol!tSiGe uncooled microbolometers for thermal IR detection

459

Figure 12.8 Typical resistance measurements of a linear array of 128 detectors: (a) the discrete steps in the measured values are due to digitizing of the measured voltage. ( b ) histogram of the measured resistance.

Using m e a s u r e m e n t s of resistance at 2() and 40 ~ the TCR of the device can be calculated. Its absolute value being around 1% (depending on the doping), the uniformity over an 8 " wafer is measured to be only a few percent. This includes, however, a possible error due to the temperature stability of about O. 5 %. Last but not least, the noise uniformity over a 128 element linear array was measured. Figure 12.9 displays the 1/f noise at 1 Hz for a typical device. The noise uniformity over the array is about 6% for the lffnoise. The Johnson noise contribution is about 2 orders of magnitude lower at 1 Hz, and moreover its variation is proportional to the excellent resistance uniformity. 12.3.2 Absorber comparison and trade-offs

It is possible to rely on polySiGe absorption without the need of an additional absorber but this compromises absorption and hence signal, in the case of thin polySiGe films. As a consequence, nearly all bolometer developments include suitable absorber layers. A practical approach consists of a thin metal absorber with sufficiently high sheet resistance. ~s Although almost any highly resistive metal could be used (and NiCr resulted in the observed absorption spectrum), the CMOS front-end compatibility limits the options to materials such as TiN and TaN. Even though the thin absorber does not absorb all radiation, part of the remainder is reflected by the silicon substrate and hence an unoptimized quarterwavelength stack is created consisting of absorber, the a i r / v a c u u m gap under the pixel, and the silicon substrate. For this reason, the thickness of the sacrificial layer should indeed be a quarter wavelength of the radiation to be detected ( 8 - 1 2 l.tm): as a consequence, the TEOS absorber is between 2 ~m and 2.5 ~m thick. This results in approximately 60% absorption over the 8-121.tm wavelength range (Figure 12.10(a)). If the application permits, a superior absorber can be constructed by means of a quarter-wave stack on top of the bolometer pixel. Such quarter-wave

460 Handbook of Infrared Detection Technologies

Figure 12.9 Typical uniformity of the 1ff noise at 1 Hz for 128 detectors. The non-uniformit!l is about 6%.

Figure 12.10 (a) Schematic implementation and absorbance spectrum (b) for a thin metal fihn based absorber, and the implementation (c) of a quarter wavelength absorber and its associated absorption spectrum (d). Thanks to the vacuum gap which creates an unoptimized quarter-wavelength stack, the thin-metal absorber achieves high absorption while maintaining low thermal capacitance.

absorbers have resulted in nearly 100% peak absorption although they come with a penalty of increased thermal mass and therefore increased time constant (Figure 12.10(c.d)). 12.3.3 Pixel optimization

Optimizing the pixel response starts simply with the basic requirement that the pixel has to be operated in vacuum. This is the only way to minimize conductive heat loss through the air surrounding the device. If conductive heat loss is further mimimized by using long narrow pixel legs, the pixel temperature and hence the pixel resistance will change as much as possible during IR irradiation.

Pol#SiGe uncooh'd microbolometers for thermal IR detection

461

Figure 12.11 shows the thermal conductance G of a polySiGe pixel (60 ~ m • ~m side, 1 ~m thick, two 30 l~m long legs with 0.8 l~m2 cross-section) as a function of ambient pressure surrounding the pixel. The thermal isolation increases over a factor of 200 between atmospheric pressure and 1 0 - 3 mbar. 1~ The polySiGe bolometer of Figure 12.11 shows a thermal conductance G of 2 x 10 -7 W/K. During the course of the polySiGe development (see further), this value was further improved to below 5x1() -~ W/K. Figure 12.11 also demonstrates the use of bolometer as vacuum gauge (micropirani) as over four orders of magnitude of pressure difference can be resolved. The complexities of v a c u u m packaging of MEMS devices in general, and microbolometers in particular, are discussed in detail in Section 12.5. Optimization of microbolometers for imaging purposes results in conflicting requirements: in order to increase the sensitivity, the thermal conductance G should be minimized. However, for a given device volume, this results in a larger time constant. As a consequence, the pixel response becomes too slow for the imaging frame rate (i.e. typically 30 Hz or higher). Therefore the only practical solution to this problem is to reduce the device thickness. Changing the pixel size is not an option since this is governed by the required temperature resolution, spatial resolution and focal-plane (i.e. optics) size. Figure 12.12 shows the microbolometer pixel heating as a function of time during pulsed excitation. As the TCR is negative, the voltage signal decreases over time with an exponential time constant. For a 1 lam thick polySiGe pixel having a G of 1 • 10 -7 W/K the thermal time constant is approximately 40 ms whereas for a 0.2512m thin polySiGe pixel with similar thermal isolation the time constant is 10 ms.

Figure 12.11 Thermal conductance G of a microbolometer as a h~nction of the ambient pressure.

462 Handbook of Infrared Detection Technologies

Figure 12.12 Pixel response during pulsed biasing of a 1 I~m thick bolometer showing a time constant of 40 ms, and a 0.25 ~m thick bolometer showing a time constant of 10 ms. Because the TCR of polySiGe is negative, the resistance decreases and hence the voltage signal decreases as a flmction of time.

Reducing the thickness of the bolometers results in a large decrease in mechanical stiffness of the pixel. The pixel deflection is inversely proportional to its m o m e n t of inertia which in turn is proportional to the cube of the pixel thickness. The stiffness, however, is critical for the release process (i.e. the etching of the sacrificial layer) and for the presence of strain and strain gradients in the pixel material. As a consequence, it is virtually impossible to realize extremely thin devices. A pragmatic solution to the release process is presented in Section 12.3.4 and a fundamental solution to overcome the loss of stiffness is discussed in Section 12.3.5.

12.3.4 Vapor-HF processing Many surface micromachined devices use a silicon oxide as sacrificial layer and this is also the case in the polySiGe bolometer process. In the last process step, this sacrificial layer is removed by etching. This is undoubtedly the most critical step in a surface-micromachining process, as shown in progress in Figure 12.13. A conventional technique consists of etching in an HF solution followed by rinsing or, better, freeze-drying to avoid condensation. Condensation has proven detrimental as it leads to sticktion: the surface tension forces in liquid droplets pull down thin microbolometers and once bonded to the substrate, the long beams do not have sufficient stiffness to restore the pixel. In addition to the condensation and sticktion, wet HF attacks a l u m i n u m interconnects and bond pads. The problem of sticktion during wet release has been solved by the development of a completely dry process consisting of vapor-HF performed at elevated temperatures. 17 The process tool is a modified Gemetec PAD-Fume vapor HF wafer cleaning system with heated wafer chuck. It has proven to be

PolySiGe uncooled microbolometers for thermal IR detection

463

Figure 12.13 Subsequent steps in the release etch of the sacr(ficial oxide (black) underneath the pol!lSiGe pixel. The gradual etch of the sacrificial oxide is clearl!i seen as well as absence qf residue due to the clean vapor HF process.

very efficient in releasing extremely thin microbolometers with nearly unity yield. The chemical reaction shows that water is produced in any case: Si02 + 4 HF ---+ SiF4 + 2 H20

(2)

However, by heating the wafer to 30 ~ ~ the condensation can be avoided. Pixels with a thickness of only 160 nm could be successfully released as is shown in Figure 12.14. The curvature of the device is due to the combination of an extremely low m o m e n t of inertia and the presence of a strain gradient in the pixel. Apart from the pixel yield, the vapor HF process is very clean as can be seen from the SEM photographs. An added advantage of vapor-HF processing is that other device materials such as a l u m i n u m and TaN are hardly attacked during the release etch. 12.3.5 Stiffness e n h a n c e m e n t

techniques

Theory

A fundamental problem of extremely thin device layers, both for microbolometers and for surface-micromachined MEMS in general, is their behavior as a result of strain and strain gradients. Particularly since a device may consist of a stack of materials. The very low m o m e n t of inertia of the microbolometers suspended on long legs makes them prone to curvature. The ensuing touching of the bolometers to the substrate reduces the thermal isolation and renders the pixel useless. This is demonstrated in Figure 12.15 below where a linear array of fully-processed 0.2 5 lam polySiGe microbolometers is shown. The compressive stress gradient between absorber/isolator and pixel causes the observed buckling and touching of pixel to substrate. There are two solutions to such a problem. The first is pragmatic and consists of minimizing stress in the materials or compensating stresses in the material stack. The second, and by far superior, solution is to enhance the m o m e n t of

464 Handbook of Infrared Detection Technolo#ies

Figure 12.14 160 nm thin poI!tSiGe microbolomeler reh'ased using vapor-HF at elevated temperatures.

Figure 12.1 5 Detail of a linear arra!t of full!! processed 0.2 ~, 12m thin microbolonwters. The stress gradient in the material stack causes the devices to touch the substra~e. This stress #radient is due to the three materials used (polySiGe, SiN, TAN).

inertia of a pixel by design and therefore make the pixel less sensitive to stress gradients. Typically, a combination of both techniques will lead to the most robust MEMS devices. Let us briefly consider the pragmatic solution of minimizing stress and stress gradients. The situation encountered in Figure 12.15 stems from a stress of roughly 60 MPa in the polySiGe and of the order of 1 GPa in both the TaN absorber and the SiN isolator, although with opposite sign. By modifying the TaN deposition parameters, the stress in the TaN could almost completely compensate for that in the SiN layer and the pixels remain suspended. The second and more generally applicable solution is to overcome the bending of the pixels altogether by increasing structural stiffness by increasing the moment of inertia of legs (and also of the pixel}, is Similarly to civil engineering, this can be accomplished by using the well-known I-profile (see Table 12.2) which has the highest moment of inertia for a given beam cross-section. The

PolySiGe uncooledmicrobolometersfor thermal IR detection 465 Table 12.2 files

S c h e m a t i c b e a m s e c t i o n s a n d t h e i r m o m e n t o f i n e r t i a for d i f f e r e n t b e a m pro-

reader should briefly dwell on the complexities of realizing such an I-profile in a surface micromachining based MEMS process. As this is indeed most impractical, another type of profile should be sought which combines easy manufacturability with excellent stiffness and low thermal conductivity. This is found to be the Uprofile (see Table 12.2). The m o m e n t of inertia of a U-profile is identical to that of the optimal I-profile (for the same section). More importantly, it can be processed in a very simple way which is by incomplete removal of the sacrificial layer in a second patterning step and timed etch. This is shown in Figure 12.16. This extra etch step defines the lower part of the U-profile. Thanks to its excellent step coverage, subsequent polySiGe deposition creates a U-shape in this narrow trench. A polySiGe etch will, in a following step, define the width of the U-profiled legs. The quantitative stiffness improvement can be calculated. Comparing the Uprofile to a rectangular profile with thickness H and width B, the stiffness reduction R 1 of a rectangular profile is given by:

Ij

BH 3 - b h 3

b(h) =

B

3 (3/

The reduction of the beam section R 2 is R2 .

Si B H - bh . . . . Sr BH

1

b 1l BH

(4)

Since the goal is to have m a x i m u m thermal insulation, hence a minimal section, in combination with a m a x i m u m stiffness, the ratio M of m o m e n t of inertia I over section S should be maximized. W h e n comparing regular rectangular legs to U profiles, the figure of merit m is given by

m -

(S/

466 Handbookof Infrared Detection Technologies

Figure 12.16 Schematic process flow for U-profile fabrication: patterning and partial etch of the sacrificial layer (top), conformal polySiGe deposition (middle). patterning and etching of the pol~3SiGe(bottom).

This figure of merit is plotted in Figure 12.17. It compares the I/S for regular beam with that for a U-profile beam as a function of layer thickness t. The different curves reflect different beam width B and beam height H. For all curves, the increase in figure of merit becomes more pronounced for thinner layers, which is relevant since m a x i m u m thermal isolation is obtained for thinnest layers. For rather thick devices (0.5 lum), the figure of merit is roughly 10 whereas for l O O n m (50 nm) thin layers, the figure of merit is 200 (1000). This implies a 200-fold (respectively ]()()()-fold) reduced deflection of the beams for a given force on the beam.

Implementation

We have experimentally verified the stiffness e n h a n c e m e n t on l O O n m thin cantilever beams (Figure 12.18) where standard cantilevers and U-profile cantilevers with increasing length alternated. Whereas the regular cantilevers progressively curved up due to a stress gradient, the U-profile cantilevers remain straight. Let us now consider a practical example such as a 501amx5Olam microbolometer. Such a bolometer featuring classical rectangular beams of width 1 lam and device layer thickness 0.2 lam has a very low beam section of 0.2 l.tm2, resulting in high thermal insulation. If we construct a 50 nm thin Uprofile of the same width and a 2 nd sacrificial layer etch depth of 1 lam, we obtain an identical beam section of 0.2 l.tm2. However, due to the increased m o m e n t of inertia, the deflection for a given force is reduced by a factor of 50. Yet another advantage becomes clear. The time constant of the bolometer is reduced by a factor of 4 (from 33 ms to 8 ms) since the pixel thickness is reduced by that factor. The reduced time constant is very important for 2D imaging applications. In conclusion, the use of U-profiles not only enhances the mechanical stiffness of the devices, but also considerably reduces the thermal time constant. SEM photographs of several polySiGe pixels and details of the profiles are shown in Figures 12.19 and 12.20.

Pol!lSi(;e uncooled microbolometers fi~r thermal IR detection

467

10000

1000

B= 1 pm, d= 21J.m

.,.L.'

r

~

::3

100 B=lp.m,d=l

10

I

0.1

I

9

i

0.2 0.3 layer thickness t (pm)

i

9

0.4

m

0.5

Figure 12.17 Gain in st(fl)wss over section, plotted a.s.ligure of merit m vs. the la!ler thickness.

Figure 12.18 1 O0 nm thin cantilevers of increasing length, alternativel!l with and without U-profiles.

Figure 12.19 Detailed SEM photograph of a U-proJile beam with the following dimensions: length 7 ~ I~m. top width: 1.2 #m. depth" 1 I~m.

468

Handbook of Infrared Detection Technolo~jies

Figure 12.20 SEM photograph of a linear arra!t leaturiplg pixels with d(flerent U-lwofih' pattern. From upper left to lower right: no U-profih, s. onl!l U-prqfih's in the beams. U-profih's in both beams and pixel.

Table 12.3 C o m p a r i s o n of different p a r a m e t e r s for a r e c t a n g u l a r a n d a U-profile b e a m w i t h identical section

I

Beam shape"

~'---~"~*

Section"

G(W/K) Deflection (a.u.): Timeconst.=-

C G

~ t

II.u5.mn

ii ,~m

(). 2 t.tm 2 3xl() s

(). 2 l.lm 2 3• s

1

().()2= 1 / 5()

3) ms

8 ms=l/4

Figure 12.20 shows four neighboring 13() nm thin polySiGe bolometer pixels. The one on the left features no U-profiles and is heavily curved. The second one features U-profile legs which effectively straighten the legs but the unsupported parts of the pixel still curve. The other two feature U-profiles in both the legs and in the pixel and it is obvious that the added stiffness keeps the entire bolometer perfectly suspended. One could argue that pixels that curve upwards are not dead since they do not touch the heatsink. However. the bending changes the optical properties (projected cross-section, quarter-wave effect with silicon surface) and should therefore be avoided.

12.4 Readout and system development 12.4.1 Introduction

Due to the high-temperature processing of polySiGe, the IMEC polySiGe bolometers are not monolithic but hybrid arrays since postprocessing on backend CMOS cannot withstand the deposition temperature of polySiGe nor that of the sacrificial layer.

PolJISi(;e mlcooled microholometers Jbr thermal IR detection

469

As a c o n s e q u e n c e , their m a i n relevance is for linear arrays or small 2D arrays w h e r e the bolometer matrix is wired to b o n d p a d s w h i c h are read out by a dedicated ASIC (see Figure 12.21). This allows re-usability of an N - c h a n n e l r e a d o u t for a variety of 1D or small 2D arrays, w h i c h t h e n need a limited n u m b e r of CMOS lot t u r n s to produce. F u r t h e r m o r e . the r e a d o u t can be designed in a less expensive CMOS front-end process (e.g. (). 713m or (). 5 lJm). 12.4.2 Readout of polySiGe microbolometer arrays

The effect of the absorbed IR radiation is a resistance c h a n g e of a microbolometer. This c h a n g e must be detected from the r e a d o u t circuitry and in the case of a m i c r o b o l o m e t e r a r r a y the produced signal m u s t be multiplexed to a c o m m o n output. Various r e a d o u t schemes h a v e appeared in the literature as the i n t e g r a t i o n of the c u r r e n t flowing t h r o u g h the microbolometer, 1~ the m o d u l a t i o n of the phase of an RC-oscillator due to the resistance c h a n g e 2~ and a copper stabilization t e c h n i q u e for noise reduction. 21 Implementation o[ a direct current readout In this section a new readout scheme is presented that provides a pulsed current bias, measures and amplifies the voltage drop across each microbolometer and multiplexes this signal to a c o m m o n o u t p u t . 22'2~ With this scheme and in contrast to integrating readout, a continuous time measurement of the voltage drop across the microbolometer is performed. The advantage of this circuit is the high dynamic range that allows the accommodation of the large signal swings due to the self-heating effects of the microbolometers without adding extra circuitry. 24 Based on that scheme, a CMOS chip capable of reading linear arrays or small 2D arrays of microbolometers with pixel counts up to 144 has been fabricated and tested. The technology used is CMOS with ().7 Bm m i n i m u m feature size and 0-5 V power supply. The readout chip is connected with the sensor by means of a ceramic substrate using a Multi-Chip-Module (MCM) technology (see Figure 12.21 ). The same R()IC design can be used for monolithic sensors when integrated on the same substrate with the microbolometers. Each microbolometer has its own dedicated readout circuit referred to as pixel amplifier. The chip contains ] 4 4 pixel amplifiers, an 8 bit address decoder, an output driver and the bias stages. The genera] architecture is shown in Figure ]2.22, together with the connectivity with the sensor chip on the MCM. The pixel amplifiers are designed on a pitch of 5() l~m in order to match the pitch of the bolometers to be tested. This simplifies the connectivity and most importantly, minimizes the design effort for a future integration with the microbolometer array. Employing an address decoder instead of a shift register for the multiplexing a|]ows for dynamic windowing and, most importantly, eases the connection when a (small) 2D microbolometer array is used. The schematic of a pixel amplifier is shown in Figure 12.22. Pixe]s are selected according to the digital addresses that are applied in the address decoder. The m i c r o b o l o m e t e r is biased using a c u r r e n t mirror circuit. A pulsed c u r r e n t biasing

470

Handbook of Infrared Detection Technologie~

Figure 12.21 H!lbrid interconnection of a linear boh,neter arra!l (bottom chip) with its dedicated R()IC (top chip) using wire bonding and an MCAI substrate.

Figure 12.22 Schematic design and electrical scheme of the readout chip.

is performed that minimizes the self-heating of the microbolometer due to the power dissipated from the bias current. The output of the amplifier is: Vo :

v,.,,s -

IH(O)I (v,.,:r

-

n,o,)

(6)

where H(O) is the low frequency gain of the amplifier and Vl,,,1 is the voltage drop across the bolometer. The second opamp that is placed in a unity-gain configuration drives the output bus. The opamps are realized using differential pairs. The above equation implies that the signal difference between two pixels is the difference in the voltage drops across the two microbolometers multiplied by the low frequency gain of the amplifier. In other words, the pixel amplifier enhances the differences in the voltage drop of the microbolometers. In the current implementation, the pixel amplifiers can operate with a gain of 1 () or 1.

Pol!!Ni(h' ltm'ooh'd microl~olometer,,; for therlnal IH detection

471

Selection of gain 1 results in the direct output of the voltage over the microbolometers. The readout chip has been connected and tested with a linear array of 64 microbolometers based on poly Si-Ge. The microbolometers exhibit a typical resistance of 50 kOhm and they are biased with a pulsed current of 1 () up to 8() microAmps amplitude. Figure 12.2 3 shows the output during the selection of all consecutive pixels. The exponential drop is attributed to the self-heating of the microbolometers. In an imaging system, the actual signal corresponding to one pixel must be a sample of this waveform while this pixel is selected. In the current implementation, this sampling is performed off-chip because the whole output waveform is desired in order to study the characteristics of the microbolometers. An advantage of this readout scheme is the high dynamic range. Depending on the actual microbolometer resistance, the bias current can be adjusted in order to set the operating point within the input range of the pixel amplifiers ((). 5-4 V). The total input referred noise has been measured as 22()laV which in theory would allow to measure 3()()mK noise-equivalent temperature difference NETD (with f/1 optics) provided the bolometers are biased to 4V and show 6()% absorption, 1% TCR and 1• 7 W/K thermal isolation. This noise level exhibited by the sensor can be translated in a m i n i m u m resolvable resistance change of almost O. 9 Ohms.

Figltre 12.2 ~ Direct cttrrent readoltt oZltl~Ztt of a 64 eh'n~ent lirlear bolomewr arra!l, tnea~tlred at anlbient

pressltre atld in vacuttm. The itzset ,,;/tow,,,"the t!lpical e.rl~onential self-heating bo/ometer reSl~Onse aml the of]._ chip sampling.

472

R~,o V

Handbook of Infrared Detection Technolo#ies

1~

Rr~

/

V1/

,i II

] I

v.

/

Lv tv iv Lv

//

y

Figure 12.24 Scheme of a CTIA based microbolometer readout (left). A serial and paralh'l readout architecture for a linear arra# of microbolometers is shown ri~lht.

A very important issue in multi-element sensors is the response uniformity. For a uniformly bright scene the response of various pixels is not the same, mainly due to mismatches in bolometer resistance, amplifier gain and bias current. Therefore, a complete sensor system must utilize signal calibration. This can be performed on-chip or off-chip by software. The second approach has been taken in the present phase of this work. The non-uniformity due to the bias current has been minimized by the proper design of the current sources. According to data provided by the silicon foundry the relative error in the current is 0.3%. The gain uniformity is also high because the gain depends on capacitance ratios that are well controlled in CMOS technologies. The total ROIC non-uniformity has been measured as (). 3'Y~,.

Integration readout principle

The most commonly used readout scheme uses CTIA integration of the net IRinduced current. 19 This is accomplished by means of a reference bolometer or reference current source so only the change in IR radiation provides a signal. Even for the same readout noise, the dynamic range of this approach is superior to the direct readout scheme. For a linear array (or small 2D array), two different readout architectures are possible: serial or parallel. In the first the voltage over subsequent detectors is switched in a serial way. while the current is integrated in one CTIA. The parallel architecture provides a CTIA for each detector (see Figure ] 2.2 4).

12.5 Zero-level vacuum packaging 72.5.7 Introduction

Using the above poly SiGe based process on 8" enables low cost production of large quantities of IR imaging systems for. e.g., the automotive industry. An often overlooked cost is the packaging at low pressure (typically 1()-~ mbar).

Pol!lSi(;e uncooled microbolometers for thermal IR detection

473

Commercial hermetic packages are available, but their cost is prohibitively high for (by definition) low-cost automotive and other non-military applications. Even structural stiffness enhancement techniques cannot overcome the forces encountered during the final dicing and pick-and-placing of the dies into packages. Bolometer packaging, therefore, is a critical item. The requirement for vacuum packaging of microbolometers further complicates this issue as hermeticity of many electrical feedthroughs has to be guaranteed over an extended period of time. As a consequence, much effort is currently devoted to so-called zero-level hermetic packaging techniques where the microbolometer device is protected at an early stage of the process to improve yield and reduce cost. The additional vacuum requirement makes this approach even more challenging. When waferlevel or on-chip encapsulation takes place, this is commonly termed zero-level (or O-level) packaging. It means that a package is realized at the die level. This can be accomplished by wafer bonding, by flip-chip bonding or by surface micromachining. The electrical feedthroughs are routed on-chip. After realizing such a zero-level package, further standard (hence low cost) packaging techniques can be applied. IMEC has developed a patented zero-level packaging technique called IndentReflow Sealing (IRS) which is described in Section 12.5.2. An alternative approach using BCB instead of metals as soldering material currently under investigation is presented in Section 12.5.3. 12.5.2 Indent-Reflow Sealing using metal solder We proposed the so-called indent reflow sealing (IRS) process as a new O-level packaging approach for, e.g., MEMS.2s In this process, two chips are hermetically sealed using an electroplated ring (Figure 12.25). Because most MEMS devices including microbolometers have to stay clean, a fluxless solder joining is required. This is realized by using the eutectic bonding between tin lead (SnPb) and gold (Au). This process is such that no vacuum flip-chip bonder is needed. The IRS processing details are as follows. First, the under bump metallization (UBM) ring is formed on the bottom device wafer, i.e. the Ni/Au layer is deposited on a SiO2 isolation layer using electroless plating. The electrical connections are processed on the bottom chip under (and electrically isolated from) the plated

Figure 12.25 Scheme of the on-chip hermetic packaging assembl!l b!l indent-reflo~v sealing ( IRS).

474 Handbook of Infrared Detection Technoloc3ies rim. Figure 12.26 shows a linear array of microbolometers each with an electrical 'fan-out' structure u n d e r n e a t h the Ni/Au v a c u u m rim. On the cover substrate, the same geometry is created by electroplating PbSn. We have used Si, Ge (which is I1R t r a n s p a r a n t ) or glass (for application in the visible w a v e l e n g t h range or for d e m o n s t r a t i o n purposes) as cover material. 26 A groove is made in the ring to allow evacuation before the reflow. In the second step, the two substrates are prebonded in a flip-chip aligner. Then the cavity is evacuated in a reflow oven t h r o u g h the indent in the ring structure. As the indent is relatively large with respect to the overall volume of the cavity, both the evacuation and the filling is fast. Finally, the assembly is reflown at a typical t e m p e r a t u r e of only 2 4 0 ~ resulting in a eutectic solder bond rim sealing and hence a hermetic micropackage. This is s h o w n in Figure 12.2 7 for the case of a glass top cover. The fact that the actual sealing takes place after the flip-chip assembly, results in a marked cost reduction, since no v a c u u m flip-chip aligner is needed and,

Figure 12.26 Detail of the Ni/Au ring processed around a linear arra!! of microbolometers (a). and Pb/Sn electroplating pattern on the top (glass) substrate ( b ).

Figure 12.27 Nanoliter package formed b!! the IR S process rising glass as a top cover.

Pol!lNi(h' uru'ooh'd microbolometers lbr thermal IR detection

475

moreover, m a n y hermetic micropackages can be reflown in one operation. We realized cavities with typical surface of a few square millimeters and an inside height of around 10 micron, resulting in a hermetic volume of only several nanoliters. The total height of the system is marginally more than the thickness of the combined substrates.

12.5.3 Zero-level packaging using BCB Although metal-based reflow soldering seems the method of choice for v a c u u m packaging. BenzoCycloButene (BCB)is also being investigated as a sealing material. 27 Photosensitive BCB can be easily patterned, displays minimal outgasing and has a low reflow temperature ( < 2 5() :C). Zero-level packaging of microbolometers using BCB is a relatively simple process compared to the metal solder IRS technique. No further processing on the microbolometer wafer is required since the BCB rim can be processed on the top substrate. Moreover. since BCB is an insulator, there is no need for an insulating layer on top of the interconnects. Figure 12.28 shows a BCB-based zero-level package containing a linear array of microbolometers.

12.5.4 Hermeticity testing using microbolometers The hermeticity of BCB capped zero-level packages containing released microbolometers has been tested according to the military standard MIL-STD883D. Gross leaks are tested using tluorocarbon liquids with different boiling points. After immersion in a first liquid with low boiling point for several hours. the package is transferred to the second liquid (having a higher boiling point). When heating above the boiling temperature of the first liquid, the appearance of bubbles (due to the presence of the first liquid in the package) indicates a leak. With this method, leaks larger than 1 () 4 mbar.l/sec can be detected. The fine leak test procedure is as follows. First the package is exposed for several hours to

Figl~re 12.28 Example of a :ero-level package of a lim'ar arra!l of microl~olometers tisinfl BCB (a). Detail of the BCB rim viewed throztgh the glass top mtbstrate {17).

476

Handbook of Infrared Detection Technologies

a He environment at high pressure (typically 3 bars). Next, the samples are transferred to a He leak detector. The latter has a He leak rate sensitivity down to approximately 10 -9 mbar.1/sec. No leaks were found when applying both the gross and fine leak tests to the BCB packaged microbolometers. In situ characterization of the microbolometers, however, proved the existence of a leak. As discussed already in Section 12.3.3, microbolometer characteristics can be used to measure (local) pressure. When pumping (and venting) a test chamber containing the capped microbolometers, the device response upon a pulsed bias was monitored (Figure 12.29). The typical exponential selfheating characteristic at low pressure, and a square signal at high pressure clearly indicated that the pressure inside the package was exactly following the pressure of the test chambers. Hence the package was leaking. This result conflicts with the hermiticity tests according to the MIL-STD-88 3D standard. However, from a careful theoretical analysis of the military standard leak testing procedure we concluded that it is not applicable to the small volumes typical for wafer-level packages. Depending on the volume, there is indeed an undefined leak regime which is not covered by the gross leak, nor by the line leak procedure. 27

12.6 Conclusions and outlook This chapter gave an overview of polySiGe microbolometer development. In the introduction, polySiGe was compared to other microbolometer process schemes. The structural, thermal and electrical properties of polySiGe were presented, from which it followed that polySiGe is a very suitable structural material for MEMS applications and a good microbolometer material as well. The latter stems from a very low thermal conductivity combined with a reasonably high TCR and

Figure 12.29 Schematic test setup (a) for testing the hermiticit!l of zero-h'vel packages using the pressure dependence response oll a pulsed bias of microl~oh, m, ters ( b ).

l)ol!lSi(;e uncooh'd microl~olometers for thermal IR detection

477

the former from low stress and stress gradients in the material. Since the detector processing takes place in an 8" CMOS pilot line, one can take full advantage of the submicron lithographic capabilities as well as the superlative uniformity: on an 8" silicon wafer, the pixel-to-pixel resistance non-uniformity is better than 2%. The different techniques to optimize pixel response were discussed. Whereas these, to a large extent, involve polySiGe bolometers, two generic approaches were highlighted: ultra-clean oxide sacrificial layer removal using vapor HF processing and the use of structural stiffness e n h a n c e m e n t to make stiffer MEMS components. Since current production techniques use a high-temperature deposition/ anneal process incompatible with CMOS postprocessing, polySiGe microbolometers are, in essence, hybrid sensors which can be integrated to a readout circuit by multi-chip-module technology. A direct current readout for polySiGe microbolometers was discussed in detail and conventional integrating readout was presented. Results on 64 element linear arrays were shown. MEMS packaging in general and microbolometer packaging in particular is a challenge due to the requirement of long-term v a c u u m packaging to ensure proper operation. Only wafer-scale or zero-level packaging can result in sufficiently low-cost packaging, provided that the technical challenges are solved. The zero-level packaging approach at IMEC using the patented indentreflow-sealing method (IRS) was presented. In view of the current high-temperature process technology used for polySiGe, these sensors are hybrids and therefore presently limited to linear arrays or small 2D arrays. As there is a definite market for small imagers, this hybrid approach may be very competitive, particularly given the increased flexibility. An Nchannel readout circuit can be connected to a variety of 1D and small 2D detector geometries.

Acknowledgements This work has only been possible thanks to m a n y co-workers in the Microsystems, Components and Packaging division of IMEC. as reflected by the list of references. It is particularly necessary to acknowledge A. Verbist, S. Sedky. P. Fiorini, A. Witvrouw, B. Du Bois, S. Kavadias, G. Ruttens, Y. Creten, P. Merken, M. Gastal, A. Jourdain, H. Tilmans, K. Baert, E. Beyne, R. Van Hoof for their contributions. Furthermore, the collaboration with XenICs (C. Goessens, V. Leonov, J. Vermeire) is acknowledged.

References 1. P. W. Kruse and D. D. Skatrud, Uncooled Infrared Imaging Arrays and Systems, Semiconductors alld SelT~i171etals 47, Academic Press ( 199 7 ). 2. R. A. Wood, High-performance infrared thermal imaging with monolithic Silicon Focal Planes operating at room temperature, Intenmtiona] Electro11

478

Hamtbookof Infrared Detection Technologie,s

Devices Meetin# 1993. Technical Digest (Cat. No.93CH3361-3). IEEE, p.175 (1993). 3. H. Baltes, O. Brand and M. Waelti. Packaging of CMOS MEMS, Microelectronics Reliabilit!! 40, 1 2 5 5 - 1 2 6 2 (2()()()). 4. V. K. Jain and C. R. Jalwania, [Jncooled IR-sensor array based on MEMS technology, Proc. SPIE 3 9 0 3 , 2()6 (1999 ). 5. T. Ishikawa et al., Low-cost 32()x24() uncooled IRFPA using conventional silicon IC process, Proc. SPIE 3 6 9 8 , 556 ( 1999 ). 6. M. H. Unewisse, K. C. Liddiard el al., The growth and properties of semiconductor bolometers for infrared detection. Proc. SPIE 2 5 5 4 ( 4 3 ) , 4 3 - 5 4 (1995). 7. J.-L. Tissot et al., LETI/LIR's uncooled microbolometer development, Proc. SPIE3436, 605-610(1998). 8. S. Sedky, P. Fiorini el al.. Structural and mechanical properties of polycrystalline silicon g e r m a n i u m for micromachining applications, Jozlrttal of Microelectromechanical S!lstelns 7(4), 3 6 5 - 3 7 2 ( 1998 ). 9. S. Sedky, P. Fiorini et al., IR bolometers made of polycrystalline silicon germanium, Sensors and Actuators A A 6 6 . 1 9 3 - 1 9 9 ( 1998 ). 10. S. Sedky, P. Fiorini et al., Characterization and Optimization of Infrared Poly SeGe Bolometers. IEEE Trails. olt Eh'ctrolt Devices 4 6 ( 4 ), 67 "3-682 ( 1999 ). 11. A. Franke et al., Optimisation of poly-silicon-germanium as a microstructural material, Transdt~cers ' 9 9 , 4 9 2 - 4 9 5 ( 1999 ). 12. S. Sedky, A. Witvrouw et al.. Experimental determination of the m a x i m u m post-process annealing temperature for standard CMOS wafers. IEI~E Traits. Electron Devices 48(2), 3 7 7 - 3 8 5 (2()() 1 ). 13. S. Sedky, A. Witvrouw et al.. Effect of in silu boron doping on properties of silicon g e r m a n i u m films deposited by chemical vapor deposition at 4()() degrees C, Journal of Materials Research 16(9 ). 26() 7 - 2 6 1 2 (2()() 1 ). 14. V. Leonov et al., Optimization of design and technology for uncooled polySiGe microbolometer arrays in Illhared Detectors atut Focal Platte Arra!ls VII. Proceedings of SPIE 4 7 2 1 (2002). 15. A. D. Parsons and D. J. Pedder. Thin-film infrared absorber structures for advanced thermal detectors, J. Vac. Sci. Techttol. A6( 3 i. 1686 ( 1988 ). 16. P. De Moor, C. Van Hoof et al.. Process development of fast and sensitive polySiGe microbolometer arrays in Photodetectors: A laterials atut Devices VI. SPIE, pp. 9 4 - 1 0 0 (2001 ). 17. A. Witvrouw, B. Du Bois et al.. A comparison between wet HF etching and vapor HF etching for sacrificial oxide removal in MicrotTtachittitt# atut Microfabrication Process Tech17olo#!! ~I. SPIE 41 74. pp. 1 3()-141 (2()()()). 18. P. De Moor, C. Van Hoof et al., I~inear arrays of fast uncooled poly SiGe microbolometers for IR detection in hq)'ared Detectors atut Focal Platte At'rails VI. SPIE 4 0 2 8 , pp. 2 7 - 3 4 (2()()()). 19. A. Tanaka et al., Infrared focal plane array incorporating silicon IC process compatible bolometer, IEEE Trails. Electrott I)evices 43, 1 8 4 4 - 1 8 5 ( ) (1996). 20. U. Ringh, K. Liddiard et al., CMOS RC-()scillator Technique for Digital Readout from an IR Bolometer Array. the 8 th International Conference on

PohlSi(;e lltlcooh'd nlicrol~olonlelersjbr thernlal IR detectiotl 479 Solid-State Sensors and Actuators and eurosensors IX. Stockholm. Sweden. pp. 1 3 8 - 1 4 1 (1995). 21. C. Menolfi and Q. Huang. A I, ow-Noise CMOS I n s t r u m e n t a t i o n Amplifier for Thermoelectric Infrared detectors. IEEE ]oz~rt~al of Solid State Circllits ] 2 ( 7 ) ,

968-976(1997). 22. S. Kavadias et al., CMOS circuit for readout of microbolometer arrays, Electronics Letters ] 7(8), 4 8 1 - 4 8 2 (2()() 1 ). 23. S. Kavadias, M. Gastal et al., Hybrid l o n g - w a v e l e n g t h IR sensor based on a linear array of poly Si-Ge uncooled microbolometers with a CMOS readout in Sensors, Systelns, arid Next-Getleratiotl Satellites IV. SPIE 4 1 6 9 , pp. 3 4 8 - 3 5 5

(2001 ). 24. C. Jansson, U. Ringh and K. Liddiard. Theoretical analysis of pulse bias heating of resistance bolometer infrared detectors and effectiveness of bias compensation, Proc. SPIE 2 5 5 2 . 644-6 g2 ( 199g ). 2 5. H.A.C. Tilmans et al., The indent reflow sealing (IRS) technique- A method for the fabrication of sealed cavities for MEMS devices, ]ozirnal of Microeh'ctromechanical S!lste171s 9( 2 ). 2()6-21 7 (2()()()). 2 6. De Moor, P. et al., Hermetically sealed on-chip packaging of MEMS devices in Proceedings Europeatl Space CotllpotletltS Cotlh'retlce- ESCCON 2 0 0 0 , pp. 6 7 - 6 9 (ESA SP: 439)(2()0()).

2 7. A. Jourdain et al., Investigation of the hermeticity of BCB-sealed cavities for

housing (RF-)MEMS devices in ~\II.;,~,IS 2 0 0 2 . tlle 1 5 th hit. ConiC'fence Oll Micro Electro Mechanical Sllstenls. pp. 6 7 7 - 6 8 ( ) (2()()21.

This Page Intentionally Left Blank

Chapter 13

Fundamentals of spin filtering in ferromagnetic metals with application to spin sensors Henri-Jean Drouhin

13.1 Introduction A new field of solid-state technology, spintronics, is presently emerging. Electron devices based on electron spin manipulation may exhaust our capacities in key areas. Spin-dependent transport experiments have attracted broad interest because of the new physics involved and their possibility of application. Studies on magnetoresistance effects and related phenomena are nowadays an active field of research, stimulated by industrial challenges concerning high-density recording and magnetic sensors. ~ In 1995. M. Johnson proposed the concept of the ferromagnetic bipolar spin transistor made of three metal layers: a ferromagnetic emitter: a nonmagnetic metal base: and a ferromagnetic collector. 2 Today, the progress in fabrication technology enables us to associate the properties of ferromagnetic thin films with those of metals, semiconductors or insulators. The development of MRAM magnetic memories, based on tunnel junctions, is very promising. ~ The most popular transistor concept - so far never a c h i e v e d - is the Datta and Das transistor, analogous to the electro-optical modulator, based on the Rashba spin splitting of the conduction band in some two-dimensional semiconductor structures. 4 Various other devices have also been considered, for instance, unipolar devices, which are closely related to semiconductor electronics, s Very attractive new ideas, based on magnetization reversal triggered by a spin current are emerging, (' and the idea of possible stimulated emission of spin waves (SWASER) was proposed. 7 It appears that dealing with spin-polarized electron transport in ferromagnetic multilayers presents many similarities with the techniques used in the field of semiconductors. For instance, the well known notion of ambipolar diffusion of charge carriers, electrons and holes in semiconductors, s proves to be

482

Handbookof Infrared Detection Technologies

perfectly adapted to describe the diffusion of carriers of different spins in metals. 9 Even the concept of doping (spin doping) could be extended to ferromagnetic metals. 1~ In this context, devices associating optical properties and magnetism have to be seriously considered. Atomic physics experiments have shown that optically-pumped spin filters can be achieved, ll It was also demonstrated that ferromagnetism order may be induced in magnetic semiconductor structures by near infrared excitation.l-' Photogenerated spin-polarized carriers in semiconductor q u a n t u m wells, under intense far-infrared excitation, can be directly converted into electrical currents, l~ An important step toward spin injection in semiconductors was made by Fiederling et al. 14 who demonstrated extremely efficient spin injection from a dilute magnetic semiconductor into a LED structure. Nevertheless, the understanding of spin injection from a ferromagnetic metal into a semiconductor remains under discussion and controversial. 1 S In this paper, I will describe ultra-thin ferromagnetic layers as spin-filters, analogous to optical polarizers. However, dealing with electrons offers more possibilities because the electron polarization possibly has both longitudinal and transverse components. The inelastic electron mean free path (IMFP) in metals, which characterizes the escape depth, plays a crucial role in these experiments, as in all electron spectroscopies. Its analysis in the low-energy regime (below a few tens ofeV) remained a puzzling question for decades. If it was thought that its energy dependence was following a 'universal curve', 1~' more recent experiments involving metals with d electrons, evidenced a clear material dependence. 17'1s Note that direct IMFP m e a s u r e m e n t s are extremely difficult because they imply the extraction of vanishingly-small ballistic-electron contribution from a large background of secondary and stray electrons. However, the situation is quite different in magnetic materials, because the spindependent part of the IMFP can be determined with accuracy. Indeed, reversing the sample magnetization, or eventually the primary beam polarization, eliminates parasitic effects. Today, electron dynamics in metals, and particularly magnetic metals, is actively studied using time-resolved two-photon photoemission experiments which give direct access to the electron lifetime for electrons up to a few eV above the Fermi level. TM All these results stimulated several analyses, ~s-29 most of them based on numerical calculations. In refs. 2 7 - 2 9 , I showed that an accurate analytical description of the IMFP is obtained in a model based on density-of-state effects, in the range 5-5()eV above the Fermi level. The result does not depend on the detail of the d bands, but only on the numbers of s and d electrons, and the various electron relaxation channels are disentangled. In the first part of this paper. I will analytically derive the relation between the IMFP and the electron numbers. In the second part, I will focus on the ferromagnetic bilayer structure, which associates a 'polarizer' and an 'analyzer'. This spin-valve structure plays a central role in the study of new p h e n o m e n a and presents some u n u s u a l properties, which are closely related to Stern and Gerlach experiments. In particular I will precisely define the parameters characterizing

Fundamentals of spin filtering in ferromr

metals ~vith ~pl~lic~ltion to spin sensors

483

the spin selectivity, k n o w n as the Sherman functions in spin polarimetry. 3(J Finally, I will discuss solid-state spin-detector structures, based on the exchange interaction, whereas conventional electron polarimeters make use of the spinorbit coupling. ~~ At the end, I will make a link between experiments involving hot electrons and transport experiments at the Fermi level.

13.2 Theoretical IMFP variation 13.2.7 The simplest m o d e l - mathematical bases of the calculation

In transition metals, the IMFP variation is mostly determined by density of state effects, whereas the transition matrix elements introduce weaker corrections. The crudest model was given by Sch6nhense and Siegmann, 1~ who e m p i r i c a l l y related the 'scattering cross-section' ~+(~-) for majority-spin (minority-spin) electrons to the numbers of holes per atom in the d bands. This m o d e l - which was an important contribution to the understanding of spin-dependent transport but turned out to be unphysical and led to strong deviations from the experimental d a t a - is based on the picture of a primary electron 'falling' into empty states in the d bands, with spin conservation. Thus, the majority(minority-) spin electron scattering cross-section should appear to be only related to the number of d holes in the corresponding spin subband. On the contrary, we must keep in mind that majority-spin electrons propagating in the metal can always lose a small a m o u n t of energy and excite minority-spin electrons inside the minority-spin band so that any correct expression for the majority-spin cross section should contain the number of holes in the minorityspin band. First, let us remember that. generally speaking, an electron scattering crosssection (3e is connected to the electron mean free path ;.~, by the relation ~.cc~e N = ] , where N is the density of scattering centers. In ref. 18, to make more evident that the d holes act as scattering centers, the scattering cross-section is somewhat improperly defined as the inverse of the IMFP. Here, we will keep this terminology to define notations for ease of comparison between different papers in the literature. The simplest realistic model, besides numerical calculations, is the density-ofstates model proposed in ref. 28. which only takes into account electronelectron scattering in the framework of the random-k approximation. 31.32 The band structure is described by a constant density of states n,.t, for the sp band, which is taken as unpolarized, on which is superimposed a positive continuous compact-support function nd~(u) describing the considered d band 9 n~(u)=na~(u)+nsp. The energy origin is set at the metal Fermi level, so that - EF specifies the energy location of the bottom of the sp-band. The d bands are bounded by energies E1 ~ (lower bound) and E_,~(upper bound), their common width is denoted as Wd (see Figure 1 3.1 ). The numbers Ne ~ and Nh ~ of electrons and holes in the d-bands are

484

Handbook of Infrared Detection Technologies

Ep E

8'

......

E~

.~

-/~ -.-

o

_

;.:-".5;'..-: ":-:;'.:.:.:; i

"':2;":)-?;':;??" Wd

g'- E

~I ET

mu

Density of states Figure 13.1 Principle of the calculation. For each spin, the band structure is described bt3 a constant densitt3 of states in the sp band on which is superimposed a positive continuous compact-support function describing the d band. The d bands are bounded by energies E1 • and E2 +-. their common width isdenoted as Wa. The dotted area represents the occupied states, located below the Fermi level ( energ!t origin). In the figure, a primar!t electron at energy Epwith a minority spin. loses an amount of energ!t e, which is used to excite a secondar!t electron from a negative energy E'-e to a positive energy e'. with spin conservation. This occurs either in the minorit!t-spin band (direct processes) or in the majority-spin band (exchange-like processes ).

Ne -

j()-~c

n d~(u)du; N h~-

n d~(u)du

l+0 :x:

(1)

with NeC%Nh~ The d-band centers Ed~ and the centers of the emerged part of the d bands, Eh~, are defined as

j

+3c

5E~ -

un~(u)du;

N~Eg -

j

+~c

()

un~(u)du

(2)

This implies that the densities of states are normalized to individual atoms. The total d-hole number is .A/'h:Nh-+Nh +. whereas the hole asymmetry is AN'h=Nh---Nh +. Other relevant quantities are defined in the following way: Ed=-- (Ed-+Ed + )/2, .AfhEh = (Nh-Eh- +Nh+Eh + ), A2V'hEh'= ( Nh-Eh--- Nh+Eh +), and 7 N'h= [(Nh -)2 +(Nh +)2].

Fundamentals of spin filtering in ferromagnetic metals with application to spin sensors

485

We consider a primary electron at energy Ep, with a spin ~=+. This electron loses an a m o u n t of energy e, which is used to excite a secondary electron from a negative energy e'-e to a positive energy d, with spin conservation. Under these assumptions, the transition rate from Ep to (Ep-e) for the primary electron, r~'(g), is expressed as r~(g) = n~(Ep - g) [co(~cm(e) + cord~-c~-c~(g)]

(3)

E

~'~" (~) -

I

n ~'(e')n ~'' (e' - g)de'

(4)

0

In the expression of r~(e), m and mr are taken as constant transition matrix elements characterizing direct and exchange processes. From physical arguments, it will finally be reasonable to consider them as equal. The n~(u) functions are the relevant densities of states. Although this is not necessary at this step, we more generally define the ~)~'(e) functions for cs'-_+c~. We also define d~d~'(e), which is ~ ' ( e ) calculated for the only d-bands, i.e. with nsp-O. The overall spin-dependent scattering rate at energy Ep, R~(Ep) is R~ (Ep) -

i

Ep

r~ (~)d~

(5)

O

It can be s h o w n that the qb~"(e)-functions have the following properties: (i) For small e, qb~'(g),-m~(()) n~'(())e. This results from a straightforward mathematical limit and holds in the vicinity of e=0. Physically, significant deviations can be expected at growing ~, in particular if the d-band density of states is zero at the Fermi level, because the occupied d bands lie at a lower energy, which is the case for gold, copper or silver, for which the top of the d bands lie respectively at 1.8, 1.7, and 3 eV below the Fermi level. 33 A sharp increase of the electron scattering cross-section is expected w h e n the energy transfer is large e n o u g h to allow excitation of the d electrons. In such a case, E2 ~ and E2 ~; are negative and, if ~ only slightly exceeds (-E2~S'), whe have ~ ' ( ~ ) ~ n s p 2 ~+H(~+E2 ~') nsp nd~'(E2 ~') (~+E2~'), where H is the Heaviside function. (ii) For large values of ~, and more precisely when ~ ~>W ~ ' = S u p (]EI~[, ]E2C~], IE,~"I, IE2~"], ]E2~'-E~" l, IE2~"-E,~'})]. we find qb~ ' (1r - nsplr 4- nsp A consequence is that the value of qb~'~'(g) does not depend on the band considered: also note that, ifn~p=0, ~'~'(e)=0d~'~"(g)=()when g>~Sup ({Ee~'-EI~"I, IE2~S'-EI~I). This results from a straightforward integration, after observing that, w h e n g is large enough, either g' or g ' - g is out of the support of the d-band density of states. (iii) W h e n u >~Sup(IEe~'-E, ~'l, IE_,~"-E, ~'l),

486

Handbook of Infrared Detection Technoh}gies

I

ll ~ 0 0 '

0

"t-'d

(e)de - NhN o ~o' \

(7)

]

To derive this relation, let us define the G{u} function as follows

G(u)-

i

ll {}

(a)de

'+'d

(8)

When u~>Sup{lE2~'-El'~' l, IE2'~'-E~I}, we have the relation dG{u)/ du={J)d~ Thus, G{u) keeps a constant value that we evaluate for u=+cx:. G(+ec) -

j+3C dg i,g dv n d{3"(v)n d~ (v ()

{}

=

de ()

e) (9)

dv H ( e - v ) n ~ ( v ) n f ( v - e) 0

Using Fubini's theorem and with the change w = v - a , we obtain

j0

C(+oc) - .+ ::x;dvn~(v) i+~ daH(~- v)nf ( v - 8) -

{}

dvnj(v)

()

(lo)

( - d w ) H ( - W ) n d (w)

Finally, we find

G(+cxD) --

(i

()

dv n dc~(v)

)(3"

--x.

dw n d

)

__ NhNe

(11)

For very small primary energies. R~'{Ep) is easily evaluated (using (i)} and we obtain

I(R-(Ep)2 + R+(EP))

~Ep-(n-({}) + n-((})) co n-((})-+n*(())+({or - co)n-({})n+ ({})]

(12)

If co=mr, or if n + ( 0 ) - 0 (perfect ferromagnet), or if n+{()}-n-(()){non-magnetic metal), the spin asymmetry of the IMFP is the asymmetry in the density of states at the Fermi level. In the case of a metal with all-occupied and identical d-bands (E2+=E2-=E2; nd+=nd-=nd), the overall scattering rate is R ( E p ) - n~p [conspE~ + H(Ep + E2)co nd(E2)(Ep + E2) 2]

(13)

It is possible to identify the origin of the scattering process by substituting co with the relevant matrix element of the scattering channel, in this case (2rc//i) [M~s[2

Fundamentals of spin filtering in ferromagnetic metals with application to spin sensors

487

when only sp states are involved or (2re~h)IMsdssl 2. when a d state is involved. This approach will be systematically developed in the next section. R(Ep)

27r , [ , , 2 ,] ----fi-- nsp nsp M~I-Ep + H(e + E2)nd(E2)lMsdlss (Ep + E,)-

(14)

The onset of the d-state contribution was analyzed by Echenique et al., who treated the d bands as 'boxes' (square bands). 21 Nevertheless, note that equation (14) differs by a factor of( 1/4) from the expression originally derived by Berglund and Spicer, where the spin is not taken into account, ~1 and also given in ref. 21. This factor arises because, in our model, the scattering events occur with spin conservation, so that only half of the total sp density of states ( 2 nsp) is available as a final electron state. For a large primary energy [Ep>~Sup (~,7++,~1- )+Sup((), E2-, E2+)], R~ is calculated after cutting the integration domain at energy W=Sup (W++.W - - ) W'(Ep) -

r~

-

()

r~(8)d~ + ()

r~

(15)

W

When ()~<8~<W. ( E p - 8 ) lies outside of the d-bands so that nCS(Ep-8)-nsp. Considering only qbd~176 in the expression of ~oo(8). the first integral yields

(usin g (iii)) (16)

nsp (mN~N~ + c0rN~-~ ~

The other terms are easily calculated after an integration by parts. Concerning the domain W <~8 ~<Ep, we use the large-energy expression of qb~~ (see (ii)) so that the corresponding integral reads nsp(m + mr)

i

Ep

n~

-- 8)[nsp8 4- 5]de

(17)

W

Finally, the expression of R~

reduces to

R ~ (Ep) - nsp (mN~N~ + c0fNcT~N,~~

(]g)

+ 5n~p(mE~ + tarEd ~) -- 2n~p (mE,~N,~ + mfE,~'Nh ~) + nsp(O + of) ('5 - n~pEt~) N~ + 5n~pEp 1 +

+ ? n;Ep

In ref. 27, it was demonstrated that. unless o~mf. the spin polarization of the secondary electrons has a strong memory of the primary beam polarization. This was never observed and, in the following, we assume that re=Or. In Section 13.2, it will be shown that it is relevant to distinguish the different matrix elements, depending on the symmetry (s or d) of the states they involve. Then,

488

Handbook of Infrared Detection Technologies

the low-energy scattering rates for Ep/>W+Sup ((), E2-, E2+), where W=Sup (IE~+I, IE2+I. IEll, IE2-1, Wd), is R~(Ep) -- 10 nspO~ ~-A/'h

1 -- -~ -- -~nspEh

+ nsp(Ep - Ed) (19)

-4--~nspg

p -t-- N h

1 +

g n~p(Ep- E~)

In a square-band model, assuming that the Fermi level crosses both d bands, equation (19) holds for Ep>~Wd ( l + N h - / 5 ) . In particular, it is relevant for 5eV~<Ep~<10eV. Using the relations A/'h=2 NhCS+~A.A/'h and Nh~EhCS=(1/2) (.A/'hEh--OA.A/'hEh'), we identify terms proportional to Nh r but also terms proportional to AA/'h=Nh---Nh +. It is possible to introduce proper c~{},C~dand ~'d coefficients, which are not constant (in particular C~d and ~'d depend on the 7 factor; the detailed expressions of these coefficients will be given, in a more general case, in the next section) and to write R•

17 -- - {~+ - - {7{} --k OdN~: q-- o d'

AA/'h 2

(2{})

v is the electron velocity, taken as proportional to [Ep+EF] 2/3 to ensure a constant density of states in the sp band. These relations are physically reasonable as both scattering cross-sections, for minority- and majority-spin electrons, increase with the hole number. As already mentioned, this arises because, if a majorityspin electron cannot 'fall' into empty majority-spin d states, it may undergo small energy losses, exciting a minority-spin electron inside the minority-spin d band. Note that these small energy losses, which are equally possible for up- and down-spin electrons, do not contribute to A{~, but only to the spin-averaged IMFP. Then, we define (o). the spin-averaged scattering cross-section, Ac~, the spin-dependent component of the scattering cross-section and we calculate the scattering rate asymmetry, which is also the usual IMFP asymmetry A/'h

(o-) -- {7{, + {3"d 2

(21)

(22)

Ao" . . . . A 2(~s)

1

A./M"h

1

1 + - nsp(Ep - E'h) ~ 1 4 ( o 0 / ~ d ) + "/V' 2 h 1 --]-{) ~/- -}- ~ nsp(Ep -- 3 Eh)

(23)

13.2.2 A more c o m p l e t e treatment

Whereas the preceding model gives a good insight into the physics, a precise comparison with experimental data needs to take into account the relative

Fundamentals of spin filtering in ferromagnetic tnetals with application to spin sensors

489

weight of the different transitions, an approach related to the careful analysis of ref. 34. For that, we perform the R ~ calculation step by step, following the procedure described above, but just performing the integration on the a m o u n t of energy loss e after splitting the domain into three parts: from 0 to W (A domain). from W to EM (B domain), and from EM up to Ep (C domain): EM is a somewhat arbitrary cut-off energy which verifies Ep-Ex1~>Sup ((), E2 +. E2-). In such a model, the electrons are treated as distinguishable particles, which appears to be a reasonable approximation in the energy range we consider (a justification is given in ref. 34: in Appendix A, eq. (A1), the cross terms in the first matrix element are neglected, which is equivalent to ignoring the wave-function antisymmetrization: this approximation is further discussed in Section III of ref. 34 and is based on the numerical results of ref. 35). The matrix elements are assumed to depend only on the (s or d) nature of the states involved, in each given energy domain. Both spin channels are treated on an equal footing. To track the origin of the different terms, we refer to the weight of a transition between k and k' states (where k - s or p) by introducing the factor Wkk,=Wk,k into the calculation. At the end, to make the link with the notations of ref. 34, we perform the identification to~.~ to~,~=(2~/fi)IMSSss]2, toss tOsd=(2~/]l) IMSdss]2, tOsd (.Osd=(2~/h) [MSddsJ2, toss r (the superscript indicates the nature of the initial states, the subscript the nature of the final states). These matrix elements are proportional to the Coulomb matrix elements (they should include a factor accounting for the ion density: an estimate of this low-energy matrix element is extremely difficult because of the importance of screening). A straightforward calculation yields the following expression, which is a natural extension of eq. (19)

R~

1N'h [(1

10 nsp(2gfli) = 2

~/)

sd(a)]2

- -5 IMsd

2 nspEh IaSd(A) ]2]

--5

ss

+ n s p ( W - Ed)lM~d(A)l-+-f-~n~pW2 ]M~(A)I -

sd

~

1

~

~ __ W 2) ]Mss(B)] ss 2

+ nsp(EM -- W) Mss (B)]-+-i-~ nsp(E~a

(24)

-4- nsp(Ep -- EM) M~d(c)l'-+-f~n; E~ - E~ IMSs~(C)l2

{,

,1

+ NG M~d~(c)]-+Nns~(Ep -- E~)]Ms~ (C)]

21

(A), (B) and (C) refer to the energy domain where the matrix element has to be evaluated. An important conclusion is that the energy dependence of A(~ directly reflects the IMSdds] exchange-matrix-element energy dependence. In ref. 34, it is empirically concluded that the cascade polarization value, or equivalently, the IMFP asymmetry (see ref. 28), is not sensitive to the ]MSdsdl2/IMSdssl2 ratio, which is not true in general, and the variation of the squared exchange matrix element is approximated by the law [1+(~/B)2] -1, where a is the energy transfer. For simplicity, we only retain the energy dependence of the exchange matrix element, which is expected to fall offmuch faster with increasing energy. 34 Then, the whole

490

Handbook of Infrared Detection Technoloqies

energy dependence contained in the matrix elements is restricted to the IMSdds(C)l term, which will be evaluated at Ep and referred to as IMSdd~(Ep)l. In the terms proportional to the hole numbers, the domain indication will be kept in the IM~ds~lmatrix elements only, to recall the origin of these contributions. This also allows us to keep the full energy dependence of A cs. In the absence of domain indications, the matrix elements are evaluated in the A region. Consequently R ~,

10 n~p (2x//i) - aCh

[ 2 ( - S T- )

+ nsp(Ep - Ed)

IMsd , dl 2--~lnspEhlM:~tl2""

M~I-+ Tr

,1

+ N h MSd(Ep)l-§

]M~ -

(25)

, MSdss(Ep)l :1 - Eh)

As in eq. (20), we can calculate c~j, (3"d and &d, or equivalently csc~, (5"d and Ars. Assuming that the Coulomb matrix elements do not depend on the material considered, a simplifying assumption which m a y be rather intuitive in a rigid band model, we see in the above expressions that the Ed, Eh, and E'h terms (the two latter arising from the Nh~Eh c~product, see the derivation of eq. (2())), generally smaller than Wd, still contain some material dependence. In usual cases, they can alternatively be viewed as related to the hole numbers. In a square d-band model and assuming that the Fermi level crosses both bands (obviously, the results will hold if the Fermi level is not far from the d bands, the energy scale being their and E'h=./V'hWd/lO, energy width Wd) it is readily shown that Eh=u Ed=Ed~ O, where Ed {) is the spin-averaged center of the d bands, if both d bands are completely full. Concerning Ed, we also observe that the expression holds w h a t e v e r the d-band shape, in the extreme case where each d band is either completely full or empty. The validity of this linear approximation in real cases is supported by the calculation given in ref. 33 (Figure 13.2, p. 22). In the following (except when N't,=O) we will retain these expressions and we observe that 2 nspEF=./V'esp, where A/'eSp is the electron n u m b e r in the sp band, which appears to be of the order of unity, due to s-d transfer. ~3 This means that the ratio of the densities of states in the sp to the d bands is about n~p Wd/5=( 1/1 ()) (Wd/EF)./V'e sp, i.e. of the order or smaller, than (). 1. We write

cro -

VF

o'~o'o

(26)

O-d----VFO-~)O"d V

(27)

V

where VFis the electron velocity at the Fermi level and

O.~])- (nsp 27/" sd 2) \ VF --ff-I[adsl

and we obtain

(28)

Fundamentals of spin filtering in ferronlagnetic metals with application to spin sensors

IE-

~_o-SsV~p PEr

El?IM~d2 M~d2

.....:}

1

491

(29)

lVids

JV~p Wd M~d[~] O'd -- .~ e 1

JV.eJV.h

M~tdi,

nul0 I'Mds(Ep)[ [ sd 2 q- .,~,P ( Ep [M~dsI2

~

Ar~--[10 M~ds(Ep)I2--~-.~;P(gp

Madl2

fv

1() Et: ~ d

~Vids]

(3o)

y Wd'~ [M:~(Ep)-~.] 1()-~t."J I ~V,ds ~,,,~d [-'

sd

.A/'h Wd ) Mss . . . (gps.)l . 1() E,:

M~d-

-']

__VF0.~)A.Afh ,'

(31)

where JV'eml0-./V'h is the number of electrons in the d bands. In the _Qd expression, the following relations have been used N'h 7 --~+

1

AN"h 2.Af---~

~/-- 1 J~e[1 5 - 1--6

(32)

'/k'A'/"h1

AfeAfhJ

(33)

AJV'h2/jV'e JV'h is zero for a non-ferromagnetic material and is much smaller than unity for materials with small or large hole numbers and can sometimes be neglected. When N'h-0, because the top of the d bands may be located significantly below EF, Ed has to be kept in the ~r expression instead of Edr The importance of the terms involving A/'~~p in these expressions depends on the matrix element ratios, which are expected to be larger than unity as the denominator is an exchange-type matrix element. 34 These results have been derived assuming a constant density of states nsp in the sp band. This is reasonable for Ep 4 Ev. For larger Ep, it is more realistic to cut this density of states below the energy - EF, through multiplication by a Heaviside function. Here again the calculation is straightforward, assuming EF >~W, in the domain Ep>/Ev+2 Sup (0, E2+, E2-). In a square band model, this condition implies that Ep ~>Ev+2Nh-Wd/5. In the calculation of R ~', we perform the integration on the amount of energy loss c after cutting the domain at energies W, EF, EF+SUp (0, E2e~) for direct (involving the +~ channel) and exchange-like (involving the - ~ channel) processes, EM, and Ep [with Ep-EM ~>Sup (0, E2 +, E2-)]. We obtain the important relations

492

Handbookof Infrared Detection Technologies

R

~

10 nsp (2~fli)

1 (Ep + Eh _ EF)IM~dj2] =.A/'h [2 ( 1 - - ~ ) l M : d ] 2 ---f~nsp + n~p(gp - Ed)IM~I-+~n;EF

+gh

Hh)

Up --?EF

,1

O'd--.,~ e

1

(34)

,

1 - 10 I~d(Ep)]-+-n~pEF]M~d(Ep)]*"~ds 5 ss

o.o_5N.:p[(Ep-E~)]M~dl2 1 .~p(Ep 1)M~:] EF

Ia:Ssl-

IM~d] 3 f- ~ "A/e ~ . . . .

M,~ds

(35)

A.,/~h ~ Msd 2 sd 2 sd ~ -Jr-.Ale Mds(gP)[ .,~eJV.hj M~tds I~/[sd2*''ds] (36)

+ A/.~p 2 EF + (1 -- O.17)Wd -- Ep MSdss2

EF Ao" -- I.,~/" ]Mds sd(E_P) 2 e M ds,

M~dl2

M~d2

VFO~IA.~ h v

(3 7)

In eq. (34), the energy domain where IMsd~]2 has to be evaluated is only indicated in the term proportional to Nh c~. This leads to a simpler expression, whereas the full matrix-element energy dependence is kept in eq. (37). The comparison of the regimes Ep ~<EF and Ep>~Ej: shows that, in materials with a large hole number, the spin-averaged cross-section is strongly reduced when increasing the energy, because the excitation of electrons from the sp band into the empty d states is no longer possible (IM~dds(Ep)]2 in eq. (2 5) is multiplied by (./V'e/10) in eq. (34). This unusual effect was in fact observed in Gd (see the discussion in ref. 28). 13.2.3 An intuitive derivation

In this section, we do not perform any detailed calculation but we simply suppose that the scattering rate is proportional to the product of the numbers of states involved in the process, an intuitive idea. We introduce a hierarchy of the scattering events as a function of the number of d states they involve (they have the larger density of states) and classify them as a function of the amount of energy transfer. We consider an energy domain where Ep exceeds the d-band width, so that the relevant events are of the following types (see Figure 13.2): (i) Events involving 2 d states and large energy transfers. The primary electron, initially in an sp state, falls into the empty d states of the same spin, undergoing a large energy loss (close to Ep), and excites an electron with either

Fundamentals of spin filtering in ferromagnetic metals with application to spin sensors

Ep

Ep

~I

I

49 3

lk

E I

J

~xO ~0

-'l

i....'..

;..;.-;

iii;.ili'".-... j'..'..

_EF

_

iili

Density of states Figure 13.2 Simple picture. Left: (ii) process. The primar!! electron, with either spin. injected at energy Ep into the sp band loses a small amount of ener#!! e which is used to excite a secondar!l electron at energ!t e' from the occupied d states into the empt!l d states, with spin conservation. Right: ( i ) process. The primar!l electron (here with a minorit!l spin)falls into the erupt!! minoritJI-spin d states. Secondar!l eh'ctrons can be excited from the occupied d states into the empt!l sp states or. (f tlp <~Er. from the occupied sp states into the empt!t d states.

spin from the occupied d states into the empty sp states, or from the occupied sp states into the empty d states. When Ep ~<Ev, they contribute to the scattering rate proportionally to IMSdds(Ep)l 2 Nh+(Nh++Ne++Nh-+Ne - }= 10 IMSdds(Ep)12Nh+ . The Ep dependence of the matrix element is the dependence on the amount of energy transfer. When Ep/> Ee, electron excitation from the occupied sp states into the empty d states becomes impossible, due to energy conservation, so that this contribution reduces to IMSdds(Ep}r2Nh+(Ne++Ne-)=lMSdds(Ep)lZNh+./V'e 9Most of the spin asymmetry of the scattering rate arises from this term. (ii) Events involving 2 d states and small energy transfers. The primary electron loses a small energy amount and falls into an sp state. It excites an electron with either spin from the occupied d states into the empty d states. The corresponding contribution to the scattering rate is proportional to IMSdsdl2 [Nh+Ne++Nh-Ne-]. This term does not contribute to the spin asymmetry of the scattering cross-section as the scattering process is equally probable for majority- and minority-spin electrons.

494 Handbook of Infrared Detection Technoloqies (iii) Events involving one d state and large energy transfers. The primary electron falls into the empty d states of the same spin, undergoing a large energy loss (close to Ep), and excites an electron with either spin from the occupied sp states into the empty sp states. These events introduce additional spin dependence. When Ep/> EF, they contribute to the scattering rate proportionally to IMSdss(Ep)]2 Nh+A/'esp, where the number of sp electrons is ./V'eSP=2 nsp Ev. When Ep~Ev, as the number of accessible occupied sp states is (Ep/EF)A/'~sp. their contribution reduces to IMsd,~,~(Ep)]2 Nh+(Ep/Ev)A/'r ~p. (iv) Other scattering events, involving one d state, where the primary electron falls into the empty sp states and excites an electron with either spin from. or into, the d states. When Ep~<Ev. their contribution is proportional to (Nh++Ne++Nh-+Ne-)nspEp ]Msd~]2=l() nsp Ep ]MSdssl2. because all d states are involved. When Ep >/EF, the events with an energy transfer smaller than Ev yield the total contribution 10 nsp El-. IMsdssl2. Concerning the events involving an energy transfer larger than Ev. the contribution of the empty d states is forbidden so that they yield the additional term (Ne++Ne-)nsp (Ep-Ev) [MSdss]2 =(10-A/'h)nsp(Ep-Ev) tM~dssl2, with A/'h=(Nh++Nh - ). Thus, the overall contribution is proportional to 1() n~p El: ]M~d,~,~12+(l()--A/'h) nsp (Ep-Ev) ]M~dss]2=10nspEplM~d~s]2-A/'hnsp {Ep-Ev)IM~'~.~I2.These events do not contribute to the spin asymmetry. (v) Events involving only sp states. Obviously~ these do not contribute to the spin asymmetry. When Ep ~<El-., they yield the familiar quadratic contribution: (1/2) (nspEp) (2nspEp)tMSSss]2: the (1/2) factor arises from integration on the amount of energy transfer. When Ep >/El-, the contribution of the events with an energy transfer smaller than El-, is then (n~pEv) (2n~pEv) IM~~,~,~I2. The additional contribution of the events involving an energy transfer larger than Ev is expressed as: [n~p(Ep-Ev)] A/'~~p ]M~~,~]2 because the number of occupied sp states involved in the transitions is .A/'esp . Finally, we obtain the overall contribution (nspEl:) (nspEv)]MSSss]2+[nsp (Ep-Ev)] J~'esp ]MSSssl2=nsp [Ep-( 1/2 )Ev].A/'esp IMSSssl2. Consequently, when Ep ~<Et:, the scattering rate for majority- and minorityspin electrons is proportional to

1 Af.sp2(Ep)4-'e ~

)

ss 2+Sj~ ,p~gp Mss sd]2-]- (1 "~'h'/~'e AJ~.h) ]msd ]Mssl sd2] -

-

__b_N~ [101 Mds(Ep) sd I2-qL.A/'~P~ Up ]Mss sd(Ep)I 2]

(38)

When Ep >/EF, the scattering rate is proportional to ~'-e

~--

+X1 (.Af.hA/ee _

Mss I- + 5"/V'~,PLF Mss - ~-/v hJ~ c

Ev

(39)

~/lsd(Ep) 12+./~p ]M:~t A./V'~)Im:~l2+Nil~_[j~eel,,,ds ,, (Ep) ]2]

These relations coincide with eqs. (2 5) and (34). except for some small terms proportional to Ed, Eh+, and Eh. which are missing here.

Fundatnet~tals of spitl filterin# itl ferrotna~,ltu'tic tm'tals with applicatiotl to spit1 setlsors

49 5

13.2.4 Comparison with the Schdnhense and Siegmann model In ref. 18, S c h 6 n h e n s e a n d S i e g m a n n empirically propose the relation (40)

(5• -- cy()+ OdN~:

w h e r e c~() a n d (3d respectively refer to sp and d c o n t r i b u t i o n s a n d are a s s u m e d to be m a t e r i a l i n d e p e n d e n t . C o n s e q u e n t l y , the s p i n - a v e r a g e d cross-section is

(~}

-

~- + ~+ 2

ZCh

(41)

= cs(i + ~ d - ~

and A ~ = ~aAN'h

(42)

The IMFP a s y m m e t r y A is expressed as A -

c~- - ~+ c~- + ~s+

=

)v+ - )vX+ + Z,-

=

An (cs()/~d) + 5 -- n

1 = -

~

A.A/'h ./~'h

(43)

-(~,,/~d) +7-

These expressions h a v e to be c o m p a r e d to eqs. { 2 1 - 2 3 ) . For a m a t e r i a l with a small hole n u m b e r a n d at m o d e r a t e p r i m a r y e n e r g y , we observe t h a t the S c h 6 n h e n s e a n d S i e g m a n n expressions o v e r e s t i m a t e Acy a n d A by a factor of two. For a s t r o n g f e r r o m a g n e t , eq. (43) predicts a m a x i m u m spin a s y m m e t r y A=IO()(~) (Jk/'h=/XJV'h=Nh-. O'()=()). w h e r e a s eq. (23) yields A ~ 5 ( ) % . This c a n be simply u n d e r s t o o d by the simple picture given in the p r e c e d i n g section. Indeed, m o s t of the spin a s y m m e t r y o r i g i n a t e s from the (i) events. We refer to their c o n t r i b u t i o n as OdINh+. A large c o n t r i b u t i o n to the s p i n - a v e r a g e d s c a t t e r i n g cross-section o r i g i n a t e s from the (ii) events. We refer to this c o n t r i b u t i o n as r A/'h/2. Thus, the overall c o n t r i b u t i o n of the (i) a n d (ii) e v e n t s to the s c a t t e r i n g AA/'h/2. ~d' a n d ~"d h a v e n o w a rate is GdfNh-++O"'dA/'h/X=(Gd'+G"d) Nh-+• simple physical m e a n i n g as they express the c o n t r i b u t i o n to the s c a t t e r i n g crosssection of the t r a n s i t i o n s with a large a m o u n t a n d a small a m o u n t of e n e r g y transfer. Obviously, we h a v e AC~=CSd'AA'rh a n d Gd=r 9If all the t r a n s i t i o n m a t r i x e l e m e n t s are equal, Gd/Gd '= 2( 1 --7/1 ()), w h i c h explains the m a i n t e r m s in eq. (23). On the c o n t r a r y , the S c h 6 n h e n s e a n d S i e g m a n n f o r m u l a overlooks the (ii) e v e n t s (i.e. c o r r e s p o n d s to c~"d=(), or e q u i v a l e n t l y to M'~dsd=()) SO t h a t it c a n n o t fit b o t h Ac~ a n d (cs} data, even a p p r o x i m a t e l y , w i t h a single set of p a r a m e t e r s , 3~, a n d has to be rejected.

13.3 Experimental study of

A c~

Cobalt a n d iron t h i n films h a v e been extensively studied a n d h e r e a f t e r we focus on these materials. In the following, e x p e r i m e n t a l data from several g r o u p s are

496 Handbookof Infrared Detection Technoloqies analyzed in the framework of the above described model. The spin-dependent IMFP component is of particular interest as it mostly originates from electronelectron interaction and can be measured with accuracy. The dark squares and crosses are Co data from ref. 37 (ll, 2.5 nm-thick layer, X, 4 nm-thick layer). The spin-dependent IMFP in Co was measured at several energies in refs. 38 and 39 and these data correspond to the circles ( 9 and to the cross (+) respectively. Similar measurements were performed in Fe, and the Acy(Fe) values have been scaled by multiplication by A.A/'h(Co)/A.Afh(Fe)=0.77, a good approximation in the relevant energy domain (from eq. (37), assuming that IMSdds(Ep)] has strongly decreased; see below). The corresponding points are indicated by the triangles (A) (ref. 17), (V) and ( A ) ( r e f . 39). Low-energy measurements performed on Co and giving high Acs values (Acy~,().8 nm -1 at 1.7 eV), which have been obtained through a very reliable technique, are indicated by 7-1, after ref. 40. Recent results obtained by the BEEM technique in the range 1-2 eV, taken from ref. 41, are indicated by the A symbol. The very-low-energy determination of Filipe et al. 42 in Fe is marked by the symbol ~ (without any scaling, see the discussion below). To determine the Cyd~ value in Co, we use eq. (3 ] ) and the data from ref. 37. For a first estimate, we consider the measurements of Acy at E p : 6 e V (almost the lowest energy where the model strictly applies; 6.7:3 eV). Because the different samples give somewhat different results, and as stray effects will reduce the spin asymmetry, we use the largest values and take Acy(Co)=0.50 nm -1. Neglecting the term proportional to ,Ale sp in eq. ( 31 ) and taking [MSdd~(6 eV)12/lMSddsl2 : 1 , we -1, i.e. CydC~().()43nm -1. At Ep=lSeV, Acy is obtain (vv/v) cyd~ strongly reduced, being of the order of (). 15 n m - l : assuming that the exchange matrix element at this energy is small, we deduce from eq. (37), [M~d~s(1:5eV)12/ IMSddsl2 ~<4.2. The cy+ values measured in Fe in ref. 17 keep a sizable asymmetry at Ep=41 eV, where A cy~().l nm -1. Using the CYd~ value determined for Co and assuming that [MSdds(41eV)[2=(), with eq. (37), we find JMSd~s(41eV)[2/ [MSdds]2=3.5. Because at such large energies, a decrease of [M~dss(Ep)[2 with energy should probably be considered, we conclude that 3.5~
Fundamentals of spin filtering in ferromagnetic metals with application to spin sensors

49 7

1.0

0.8

..-,, '~

0.6

s . o.4

0.2 ,'I 0.0

0

.

. . . .

J

i

L

i

10

20

30

40

50

Ep (eV) Figure 13.3 Act in Co. The dark squares and crosses are data from ref. 37 (11, 2.5 nm-thick layer, X. 4 nmthick layer), the circles ( 0 ) f r o m ref. 38, the squares ([-])from ref. 40. the diamond from ref. 42. The triangles (A) have been measured in ref. 17 for Fe and the values have been multiplied by AN'h (Co)/AA/'h (Fe)=0.77. The ( V ) and (+) points, taken from ref. 59. respectiveh3 refer to Fe (multiplied by 0,77) and Co, on a W substrate; the ( V ) point, from the same reference, refers to Fe (multiplied by O. 77) on a Cu substrate. The ( A ) points are taken from ref. 41. In the calculations, the values EF=8 eV, Wa=5 eV, N'e sp= 1 are used. The dotted parabola is the low-energy variation (eq. (44)) with ]M~aadl2/]M~d,t.~]2= 1.5. The full curve is a fit through eq. (38), valid close to 7 eV. The lowest dotted curve is a fit through eq. (37). strictly valid above 11 eV. Both fits use the a [ 1 +(Ep/4) 2 ] law to describe the exchange-matrix-element variation. The values ~d~ nm -~ and ]Ma~12/lM~aa~]2= 3. 5 have been chosen. The upper dotted curve is calculated using a constant exchange matrix element, with the 5 e V value of the preceding calculation.

eq. (31) in the 6-8 eV range, the application domain, does not significantly differ, although it is slightly better. It is thus remarkable that the simple law expressed by eq. (37) provides an excellent fit of the experimental data in the 6-5OeV range, with the s a m e ~d ~ value for Co and Fe. An exception are two values in Fe and Co at 14 eV determined in ref. 39 which, although consistent as the two points are almost superimposed, are significantly higher than other determinations. These points refer to bcc Fe and hcp Co on a W substrate. Another determination, from the same reference, for fcc Fe on a Cu substrate gives a significantly lower value (V). One could wonder whether this discrepancy reveals some dependence on the metal structure, in particular on the electron density. However, it must be remarked that the data of ref. 38 concern fcc Co on a Cu substrate, the data of ref. 37 and 40, hcp Co on an Au substrate, and the data of ref. 17, fcc Fe on a Cu substrate. The perfect consistency of these various measurements indicates that the dependence of Ar on the metal structure should not be very important. This conclusion contrasts with the statement made in ref. 38, 'The complex morphology of the Co films on Cu (111) is a major problem in the determination of the absolute values of the

498 Handbookof Infrared Detection Technologies IMFP' and supports the idea that Ac~ is a much more intrinsic parameter than {cs). Let me stress that the spin-dependent component of the electron scattering cross-section, A~, is almost entirely determined by the electron-electron interaction whereas the spin-averaged cross-section, (o). is partly determined by other interactions, so that the analysis of A may be intricate. This is perhaps the origin of the notably smaller asymmetry deduced from time-resolved two-photon photoemission experiments, which probe the ratio of the minority- to the majority-spin electron lifetime. ~~ To analyze the lower-energy domain, we first assume that the exchange-matrix element is constant at energies below 5 eV, retaining its 5 eV value which corresponds to a 1.27 ratio. The corresponding curve is drawn in Figure 13.3 and agrees well with the data from ref. 4(). For very low energies, the result may be truly material dependent as it involves the densities of states at the Fermi level. To get a rough idea, we follow the procedure yielding eqs. ( 1 2 - 1 3 ). we neglect the density of states in the majorityspin band at the Fermi level, we take nd-(())= 5/Wd, and we assume that nsp is exactly ten times lower. Thus. we obtain A~ ~ 125c~ J vf, ,'

Ep

"~ddl -tMc~C~2

"'"

(44)

The bracket mainly involves transitions within the d-bands, with the weight O)sd

O)dd--(X/z//~) ]MSdddl2, but also a sum of small contributions originating from the

FMSddsl2, IMSdsdl2, IM~d~12. and [M~S~12 matrix elements. The corresponding parabola is plotted in Figure 13.3. with an overall bracket value equal to 1.5. The result is consistent with the data ofFilipe eta].. 4 2 who estimate A~ ,-,,0.4 n m - ~ at about I eV above the Fermi level in Fe. Thus, m a n y experimental data are now available which give a quantitative picture of the energy variation of the exchange matrix element, t h r o u g h o u t a wide energy range. From the detailed analysis performed in ref. 28. the following matrix-element ratios, normalized to the exchange matrix-element value close to the v a c u u m level, are selected: IM~Sssl2/lM~dds21~6, bM~ds~[2/IM~dd~]2--3. 5, IM~d~dl2/IM~dd~12--1.5, and IM~dddl2/ ]MSdd~]2,'~l.5. These values are consistent with those used in refs. 21 and 34 and support the intuitive idea that the larger the n u m b e r of s states involved, the larger are the matrix elements, a noticeable exception being the relatively large ]M~ddd]2 value. These results, and in particular the precise knowledge of the exchange-matrix-element energy variation, should certainly stimulate theoretical investigations. 13.4 Spin precession and spin filters Spin-dependent electron transport of low-energy electrons in a ferromagnetic metal arises because majority- and minority-spin electrons have different relaxation channels due to different final densities of states in the d spin

Fmldamentals Qf spin filterin# in ferroma[Inetic metals with al)plication to spin sensors

499

subbands. Now, consider the one-dimensional experiment where an electron beam with a longitudinal polarization (propagating along the z axis) crosses a ferromagnetic layer magnetized along z (the vectors are indicated by boldface characters). Referring to the initial polarization as P~, the polarization P of the transmitted beam is P=(P~+S)/(I+SPII). In this expression, S=(t++-t+_)/ (t+++t+_) is the asymmetry of the transmission coefficients. The electron transmission only depends on the relative orientation of the incident spin and the majority-spin direction in the ferromagnetic laver and t++ ( t + ) is the transmission coefficient of an u p - ( d o w n - ) s p i n electron when the majority spins in the ferromagnetic layer are parallel to the direction of the z quantization axis (indicated by the first+index). Following the terminology which is widely used in spin polarimetry, we refer to the S parameter as the Sherman function. ~c~In the case where the incident beam polarization (along the z axis, the propagation direction) and the layer magnetization axis (in the (x,y) plane, say along the x axis) are orthogonal, the electron spin both undergoes spin filtering in the x direction, which tends to align it in the x direction, and spin precession around the x axis, in the exchange field of the ferromagnet, which results in the emergence of a y component. This effect was indeed observed by Oberli and coworkers. 37 In their experiments, a longitudinally spin-polarized electron beam was injected trough a free-standing ultra-thin ferromagnetic film with an inplane magnetization. Because the primary beam is injected in the sp bands (at a few eV above the metal Fermi level), spin precession originates from the sp-band splitting, whereas spin filtering is a consequence of the spin-dependent inelastic scattering, i.e. of the d-band splittitlq. An analysis of these effects can be found in refs. 43 and 44. I will focus on the bilayer structure, which plays a key role in spin valve-devices as well as in polarimeters, and may be important in the study of current-induced magnetization reversal. r 13.4.1

Density-operator formalism

The interest of a density-matrix description of the spin filters was pointed out by H. C. Siegmann. 45 More generally, the electron beam entering a ferromagnetic layer is described by the density operator

1~+ 1 (S).6-

(45)

where IS) is the mean value of the electron spin, 6- the Pauli operator (the operators are indicated by the symbol ^ ), and I the identity. 46 The time evolution of the density operator in the spin filter is given by [)(t) - l~(t)D U(t) +

(46)

where the evolution operator U(t) is related to the hamiltonian I2I through the relation U(t)=exp-i(I2It///). 46 Consider a ferromagnetic layer where an orbital eigenstate with a given wave vector corresponds to two different energies E+ and

500 Handbookof Infrared Detection Technologies E_ depending whether its spin state is a majority (+) or a minority ( - ) spin state. An electron injected at this wave vector but with a spin state which is not a pure state will undergo spin precession. As long as only precession is concerned, H is diagonal with E+ and E_ real eigenvalues 9 the energy spin splitting is A E = E _ - E + - h f2 and, as the majority-spin states should correspond to a lower energy, f2 should be positive. Electron absorption, i.e. spin filtering, can be empirically taken into account by adding imaginary energy contributions. 47 Thus, we write ^

I2i_ (E+-ia+

0

0

E_ - i~_

)

(47)

Let us define T+(t)=exp-i (E+-ia+)t/h and T _ ( t ) : e x p - i (E_-ia_)tffL From eq. (46), we obtain [)(t)- (

d++lT+! t)12 d~__T+ (t)T_(t)

d+_T+(t)T_ ~)* ] d--lT+(t)l

(48)

Where dij (i, j - + ) are the matrix elements of D. Let us denote as I)' the density matrix of the beam leaving the ferromagnetic layer, then the number of transmitted electrons with a majority or a minority spin is Tr (I)' I+ > < +l), the transmitted intensity is Tr D' and the spin-polarization component of the emerging beam in the ~ direction (~=x, y, z), P~:Tr (IY&a)/Tr D'. Note that this statistical picture is conceptually different to the coherent q u a n t u m transmission for pure spin states proposed in ref. 3 7. The electrons cross the layer in the time "c=d/v, where d is the layer thickness and v, their group velocity. The relations IT+(~)12-exp-(2~+d/vf/)-exp-(~+d) and IT_(~)12-exp-(2~_d/vf/):exp-(~s_d) connect the imaginary energy components ~_+ to the spin-dependent scattering cross-sections cr_+ calculated in Section 13.2. From the D' expression, it is straightforward to show that the polarization P' of the emerging beam is 43 + p,__fi~-.m~u,P (

S

I+S.P

(49)

where S=S u, u being the unit vector parallel to the direction of the majority spins in the ferromagnetic layer, l ~ ( u ) is the matrix corresponding to the composition of a clockwise spin rotation of the angle f2~ around u (spin precession) with the multiplication of the spin component in the plane normal to S by the (1-$2) 1/2 homothety ratio (spin-filter effect). It is also readily checked that the transmitted intensity is

I(u) - I(1 + S.P)

(50)

where I is the intensity which would be transmitted if the layer were nonmagnetic (S=0) or equivalently if the primary beam were unpolarized (P=O).

Fundamentals of spin filtering in ferromagnetic metals with application to spin sensors

501

13.4.2 Electron transmission through ferromagnetic bilayers Consider a ferromagnetic bilayer with an arbitrary magnetization of each layer. Since in the absence of q u a n t u m interferences the transmitted current I(u, v) through the multilayer is the product of the transmitted current of each layer, I(u, v) = I(1 + $1 .Po)(1 + S2.P, )

(51)

In this expression, S1=$1 u ($2=$2 v) is the Sherman vector of the first (second) layer, specifying the spin selectivity, 3~ and P~_~ (P~) is the polarization of the beam entering (emerging from) the ith layer. I is the current which would be transmitted through the structure if the layers were not magnetized (Sl =$2-0), which is no longer equivalent to the transmitted current for an unpolarized primary beam, because the first layer acts as a polarizer. Using eq. (49), we obtain I(u, v) - I l l + S1S2u.v -}- S1P(I.u -+- S213,n~(u)Po.v]

(52)

Equation (52) can be rewritten in the form

I(u, v) -- Io(u, v)[1 + $(u, v).Po]

(53)

where Io(u,v)=I [1+$1 $2 u.v]. Analogously to eq. (50). eq. (53) defines the overall Sherman function S (u,v) of the structure and Io(u,v) is the transmitted intensity when the primary beam is unpolarized, which depends on the {u,v} configuration. This dependence is closely related to a magnetoresistance effect. It can be easily shown that the last term in eq. (52) verifies the 'commutation' relation I t ~ ( u ) P o . v - 1Rfz~(u)v.P'o

(54)

where P'o is the mirror symmetric of P~ with respect to the (u,v) plane, i.e. changing the sign of the polarization component along (uxv) when u and v are not collinear. When u and v are collinear, the P'~ component in the plane perpendicular to u is arbitrary and we can take P'~)=P~: this case is obvious. In the first case, the relation is derived after developing P~ on the (u, v, u x v ) basis (which, in general, is nonorthogonal). The following properties have been used: l ~ ( u ) u = u ; l ~ ( u ) ( u x v ) - [ ( l ~ ( u ) u ) x ( l ~ ( u ) v ) ] " l ~ ( u ) v . u - u . v . We define p(u,v) as the polarization of the transmitted beam when the primary beam is unpolarized and first impinges onto the second layer (v axis), then crosses the first layer (u axis). From eq. (49), we have p(u, v)(1 + S1S2u.v) = S,u + S21~n~(u)v

(56)

Consequently, we find that eq. (52) can be written

I(u, v) = Io(u, v)[1 + p(u, v).P',,] = I,,(u, v)[1 + p'(u, v).eo]

(57)

502 Handbook of Infrared Detection Technologies where p'(u,v) is obtained from p(u,v} by mirror symmetry, in the same way as P'{} is obtained from P{}. When u and v are collinear, p'{u,v)=p{u,v}. This demonstrates that the overall Sherman function S {u.v} is nothing but p'(u,v). This vector is easily expressed in the set {u. [v-iu.v}u], (u• which constitutes an orthogonal basis if u and v are not collinear. [1 +

Se(u.v)],S(u, ,9 - (s, + S,u.

)u

+ $2 g/1 - S~coss'2r[v- (v.u)u]

(s8)

+ S2 V/1 - S~sinf2r(u • v)

When using a bilayer as a spin polarimeter, different combinations of the transmitted intensities, which will be easily measured, allow determination of P{}. The following equations are of particular interest because they are directly related to the three polarization components. I(u, v) + I(u, ~) - 2I[1 + S1 (P{,.u)]

(59)

I(u, v) + I(fi, v) -- 2 I[1 + S2(u.v)(P{,.u) +$2 V/1 - S~cos~rP{,.[v - u(u.v)]

(6o)

I(u,v) + [ ( f f , ~ ) - 2 I l l + S]S2(u.v)+ S 2 V / 1 - S~sinf2rP{,.(u x v)]

(61)

I(u, v) + I(~. v) + I(u. ~) + I(~. ~) - 4 I

(62)

where f i = - u ( f = - v ) . Note that only two layers may allow to determine the three components of the incident polarization. Other combinations, obtained from intensities differences are also useful, like the following I(u, v) - I(~, V) =

2 I [($1 § S2(u.v))(P{}.u)+ S2 V/1 - S~cosf2rPc,.[v- u(u.v)] 1

(63)

13.4.3 The bilayer with collinear magnetizations

In this case, we have u=+v. The bilayer in the collinear geometry allows measuring of the polarization component along the magnetization axis, P{}.u. This case has been analyzed in detail in ref. 28. The simplest way to perform a polarization measurement makes use of eqs. 61 and 63 which can be written as (1/2)[I(u, v) + I(~, v)] - I[1 + SlS2(u.v)] - I{,(u, v)

(64)

Fundamentals of spin.filtering in ferromagnetic metals with application to spin sensors

(]/2)[I(u,

y) -- 1(7, V)] -- [(e().u)(S1 § S 2 ( u . v ) ) - A I ( u . y)

50 3

(65)

and consequently

AI(u, v)

$1 + S2(u.v)

Io(-, v)

1 + s~s:(u.~)

(e().u)

-

(66)

8(u. v)(e().u)

This relation allows us to determine (P().u) once $ (u. v) is known. It is easy to verify that

I()(u, v) 2 -

(AI(., v)): = I()(u. V ) ,- - \ ~o~u

P,,.u

-i:(1-

(67)

so that one obtains

I()(u, v

AI(u. v) -~

(68)

Then, the two Sherman functions S (u, u) and $ (u, ~), for parallel and antiparallel layer magnetizations, can be determined only from intensity measurements, with no need of independent characterization of the bilayer. In that sense, the structure appears as 'self-calibrated'. 13.4.4 The bilayer with perpendicular magnetizations

This case, where u.v=(), has been discussed in ref. 4 3. There, eqs. ( 59-62 ) yield

I(u, v) § I(u, V) - 2 I{1 § S1 (P(,.u)]

(69)

I(u, v ) + I(~, v ) - 2 I[1 + $2 V/1 - S~cosf2r(pt,.v)l

(70)

I(u, v ) + I ( ~ , ~ ) - 2 II1 + S2 V / 1 - S~sinfZrP,,.(u x v)]

(7~)

I(u, v) + I(~, v) + I(u. ~) + I(~, f) - 4 I

(72)

The structure allows to determine the three components of the polarization, as if there were a third 'virtual' layer magnetized along the (uxw) axis. provided the Sherman function is known. This additional calibration can be performed either by starting with a primary beam of given polarization or by measuring the

504 Handbookof Infrared Detection Technologies polarization of the transmitted beam when an unpolarized primary beam impinges on the flipped structure (see eq. (58)). It has been shown in ref. 43 that, with a proper choice of materials and thicknesses, the spin sensitivity of such a structure, in the three directions, can compete with the best existing polarimeters.

13.5 Discussion and conclusion In this paper, I have established simple relations between the number of holes in the d spin subbands and the IMFPs. This information is essential for the conception of devices based on spin filtering. However, in ballistic-electron devices like hot-electron transistors, only electrons which have undergone almost no energy loss are collected. In that very case, elastic scattering limits the device operation. 4s Indeed, an electron ballistically injected into the metal layer and which undergoes a change in its wave-vector direction will no longer travel along the device axis, and will thus travel a longer path. Finally, it will be lost due to inelastic scattering. This effect is well known and has supported unidimensional transport equations in field-assisted photoemission. 49 Nevertheless, in this low-energy range, typically less than 3 eV above the Fermi level, the excitation of spin waves by minority-spin electrons could also play a significant role. 24.5~ An other important conclusion, which has to be kept in mind for electrontransport modeling in magnetic metals, is the surprisingly large value of the [MSddd[ matrix element. This means that the s-d transfer has to be taken into account. In particular, if most analyses deal with a two-channel transport model for up- and down-spin electrons, it may be worth considering a four-channel transport model, with up- and down-spin electrons in s and d channels. Due to their higher effective mass, the d electrons have a lower mobility so that one may wonder if p h e n o m e n a similar to those arising in transferred-electron devices will occur.S 1 However, the consequences of the s-d band hybridization have certainly to be examined, s2 The efficiency of ferromagnetic thin films as spin filters was experimentally demonstrated in ref. 5 3. In these experiments, a low-energy (a few eV) spin-polarized electron beam emitted from a GaAs photocathode under optical pumping s4 impinged on to a flee-standing Au/Co/Au multilayer under ultra-high vacuum. The transmitted current was energy analyzed. Sherman functions as high as 0.6 were measured. The case of ferromagnetic bilayers with collinear magnetizations is reported in refs. 40 and "55, in complete agreement with the above analysis. Going a step further towards application requires the deposition of the films on convenient substrates. The possibility of growing spin filters on semiconductors was demonstrated in ref. 42, in the case of Fe on n-type GaAs. There, a spinpolarized electron beam was injected into a Pd/Fe/GaAs structure and the current collected in the semiconductor was detected. The system appears analogous to a transistor where the emitter is spatially separated from the base/ collector part. Here again, a spin asymmetry in the collected current of about 0.6

Fundamentals of spin filterinq in ferroma#netic metals with application to spin sensors

505

was measured, while the 'base' and the 'collector' were maintained at the same potential. The absolute current asymmetry AI, which is the difference of the transmitted current for two opposite polarizations of the primary beam obviously do not depend on the 'base' access resistance. On the contrary, it was observed that the relative asymmetry AI/I~, where I~ is the averaged collector current, decreases when the access resistance increases. This is not surprising, because the electrons which are not transmitted through the ferromagnetic layer accumulate and polarize the structure, which results in a current flow which has no memory of the primary beam polarization, but emphasizes the importance of the boundary conditions which lead from ballistic filtering to the spindiffusion regime where the transport is governed by a chemical-potential diffusion equation. 56's7 The electrons captured by the ferromagnetic layer transfer their charge, but also their spin momentum, which may lead to magnetization switching, 6 although this may also result from the torque exerted by the ballistically-transmitted spin current on the magnetization, ss's9 The complete microscopic description of the bilayer structure with non-collinear magnetizations in the diffusion regime may provide valuable information, and these effects are presently under investigation. For the moment, the potential of ferromagnetic bilayers as three-dimensional spin polarimeters has been clearly evidenced, the standard Sherman function, which is a scalar, has been generalized to a Sherman vector and its direct link with the spin polarization acquired by an unpolarized primary beam propagating backwards has been established. The large precession angles measured in ref. 57 for a spin-polarized beam transmitted through Co an Fe films (~=20 and 30 deg. nm -1 resp.) is consistent with s-band splittings AE of 0.2 and 0.3 eV, estimated from the relation 0= A E ~ , with "r=d/v1:,~l O-is s and v v ~ l 06 cm s -1. These values are in good agreement with the calculations of ref. 3 3. These precession angles are comparable to those measured in photoemission from (110) GaAs, where the spin precesses during the subpicosecond photoemission time (assuming an escape depth of the order of 100 nm and a group velocity of about 106 cm s -1) in the effective magnetic field associated with a conduction spin splitting of the order of tens of meV. ~,c~ In contrast, the precession angle measured in Ni in ref. 59 appears surprisingly low (6~7 d e g n m -1) as the s-band spin splitting in Ni seems to be comparable to the spin splitting in Co. 33 It has to be underlined that the manipulation of spin packets in solids, experiencing coherent spin precession, also brings technological challenges to new frontiers. 61 The development of magnetic semiconductors at room temperature could be a crucial step towards spintronics. 62 The absorption or emission of spin waves and their coupling with light waves may also lead to applications involving long wavelength excitations. 63 Recently, unexpected light transmission through gold films structured with a submicron-hole array was discovered. 64 The excitation of surface plasmons is probably at the origin of the effect. One could wonder if the insertion of a ferromagnetic bilayer, with parallel or antiparallel magnetizations, acting as a switchable surface resistance, would affect this phenomenon.

506

Handbookof Infrared Detection Technologies

Acknowledgement I thank the D~16gation G~n0rale pour l'Armement for support.

References 1. E. V(~lu, C. Dupas, D. Renard. J.-P. Renard and J. Seiden, Phys. Rev. B 37, 668(1988). 2. M. Johnson, ]. Magn. Magn. Mater. 1 4 0 - 1 4 4 , 21 ( 1995). 3. J. Daughton, J. App1. Phys. 81, 3758 (1997). 4. S. Datta and B. Das, App1. Phys. Lett. 56, 665 (1990). 5. M. E. Flatt6 and G. Vignale, AEp1. Phys. Lett. 78, 12 73 (2001). 6. J. E. Wegrowe, D. Kelly, T. Truong, Ph. Guittienne and J.-Ph. Ansermet, Europhys. Lett. ~ 6, 7 4 8 (2 O01 ). 7. L. Berger, Phys. Rev. B 54, 9353 (1996). 8. E. Rashba, Phys. Rev. B 62, R 1 6 2 6 7 (2()()()). 9. R. A. Smith, Semiconductors (Cambridge Univ. Press. London, 1959). 10. K. N. Altmann, N. Gilman, J. Hayoz, R. F. Willis and F. J. Himpsel, Phys. Rev. Lett. 87, 137201 (2001). 11. H. Batelaan, A. S. Green, B. A. Hitt and T. J. Gay, Phys. Rev. Lett. g2, 4 2 1 6 (1999). 12. S. Koshihara, A. Oiwa, M. Hirasawa, S. Katsumoto, Y. Iyes, C. Urano, H. Takagi and H. Munukata, Phys. Rev. Lett. 78, 4 6 1 7 ( 1997). 13. S. D. Ganichev, E. L. Ivchenko, S. N. Danilov, J. Eroms, W. Wegscheider, D. Weiss and W. Prettl, Phys. Rev. Lett. 86, 4358 (2001). 14. R. Fiederling, G. Reuscher, W. Ossau, G. Schmidt, A. Waag and L. W. Molenkamp, Nature 4 0 2 , 787 (1999 ). 15. A. Fert and H. Jaffr6s, Phys. Rev. B 64, 1 8 4 4 2 0 (2001) and references therein. 16. M. P. Seah and W. A. Dench, Surf. hlterface Anal. 1,2 ( 1979). 17. D. P. Pappas, K.-P. Ktimper, B. P. Miller, H. Hopster, D. E. Fowler, C. R. Brundle, A. C. Luntz and Z.-X. Shen, Phys. Rev. Lett. 6 6 , 5 0 4 ( 1991). 18. G. Sch6nhense and H. C. Siegmann, Ann. Phys. (Leipzig) 2, 465 (1993). 19. R. Knorren, K. H. Bennemann, R. Burgermeister and M. Aeschlimann, Phys. Rev. B 6 1 , 9 4 2 7 (2000). 20. F. Passek, M. Donath, K. Ertl and W. Dose, Phys. Rev. Lett. 75, 2746 (]995). 21. E. Zarate, P. Apel] and P. M. Echenique, Phys. Rev. B 60, 2326 (1999). 22. I. Campillo, V. M. Silkin, J. M. Pitarke, E. V. Chu]kov, A. Rubio and P. M. Echenique, Phys. Rev. B 61, 13484 (2000). 23. P. M. Echenique, J. M. Pitarke, E. V. Chulkov and A. Rubio, Chem. Phys. 2 S l , 1 (2000). 24. J. Hong and D. L. Mills, Phys. Rev. B 62, 5589 (20()0). 25. A. Campil]o, A. Rubio, J. M. Pitarke, A. Goldmann and P. M. Echenique, Phys. Rev. Lett. 8 5, 324 ] (2000).

Fundamentals of spinfiltering in ferromagnetic metals with applicationto spin sensors 507 26. J. S. Dolado, V. M. Silkin, M. A. Cazalilla. A. Rubio. and P. M. Echenique Phys. Rev. B 64,, 1 9 5 1 2 8 (20() 1 ). 27. H.-J. Drouhin, Phys. Rev. B 56, 14 8 8 6 ( 1 9 9 7 ) . 28. H.-J. Drouhin, Phys. Rev. B 62, 556 (2()()()). 29. H.-J. Drouhin, ]. App1. Ph!ts. 89, 68()5 (2()()1 ). 30. J. Kessler, Polarized electrons (Springer, Berlin 1985 ). 31. C. N. Berglund and W. E. Spicer, Ph!ls. Rev. 136, Al()3() (1964). 32. E.O. Kane, Phys. Rev. 1 5 9 , 6 2 4 ( 1 9 6 7 ) . 33. D. A. Papaconstantopoulos, Handbook of the band structure of elelnental solids, (Plenum press, New York, 1986). 34. D. R. Penn, S. P. Apell and S. M. Girvin, Ph!ts. Rev. B 3 2, 7753 (1985). 35. R. W. Rendell and D. R. Penn, Ph!ls. Rev. Lett. 4 5 , 2 0 5 7 (1980). 36. H. C. Siegmann, Surf. Sci. 3 0 7 - : 1 0 9 , 1 ()76 (1994). 37. D. Oberli, R. Burgermeister, S. Riesen, W. Weber and H. C. Siegmann, Phys. Rev. Lett. 8 1 , 4 2 2 8 (1998). 38. E. Vescovo, C. Carbone, U. Alkemper, O. Rader, T. Kachel, W. Gudat and W. Eberhardt, Phys. Rev. B 52, 1 3 497 (1995). 39. M. Getzlaff, J. Bansmann and G. Sch~nhense, Solid State Commun. 87, 467 (1993). 40. H.-J. Drouhin, C. Cacho, G. Lampel, Y. Lassailly, J. Peretti and A. J. van der Sluijs, Proceedings of Low energy polarized electron workshop, edited by Y. A. Mamaev, S. A. Starovoitov, T. V. Vorobyeva, and A. N. Ambrazhei, PESLab-Publishing (St. Petersburg ] 998), p. 79 : H.-]. Drouhin, C. Cacho, G. Lampel, Y. Lassailly and J. Peretti, unpublished. 41. W. H. Rippard andR. A. Buhrman, Phlts. Rev. Lett. 8 4 , 9 7 1 (2()()()). 42. A. Filipe, H.-J. Drouhin, G. Lampel, Y. Lassailly, J. Nagle, J. Peretti, V. I. Safarov and A. Schuhl, Phz3s. Rev. Lett. 8 0 , 2 4 2 5 (1998). 43. N. Rougemaille, H.-J. Drouhin, G. Lampel, Y. Lassailly, J. Peretti, and A. Schuhl, 46th Annual Conf. on Magnetism and Magnetic Materials, Seattle (2001); to be published in ]. App1. Ph!ls. 44. H.-J. Drouhin and N. Rougemaille, to be published. 45. H. C. Siegmann, proceedings of the 5th European Conference on Atomic and Molecular Physics, Edinburgh, 1995: in Selected Topics on Electronic Physics, edited by D. M. Campbell and H. Kleinpoppen (Plenum press, New York, 1996, p. 221). 46. R. Balian, Du microscopique aH tnacroscopique (Ellipses, Paris, 1982): p. 85 andp. 69. 47. C. Cohen-Tannoudji, B. Diu and F. LaloiS. Mdcanique Ouantique, Hermann (Paris, 1977): p. 939. 48. R. Vlutter, O. M. J. van't Erve, R. Jansen, S. D. Kim, J. C. Lodder, A. Vedyayev and B. Dieny, Phys. Rev. B 65, 0 2 4 4 1 6 (2001 ). 49. H.-J. Drouhin, D. Paget and J. Peretti, Surf. Sci. 2 1 1 / 2 1 2 , 593 ( 1989). 50. R. Vlutters, O. M. J. van't Erve, S. D. Kim, R. Jansen and J. C. Lodder, Phys. Rev. Lett. 88, 02 7202 (2002). 51. S. M. Sze, Physics of Semiconductor Devices (John Wiley, New York, 1969). 52. ]. N. Chazalviel and Y. Yafet, Phz3s. Rev. B 15, 1062 ( 1977).

508 Handbookof Infrared Detection Technolo~lies 53. H.-J. Drouhin, A. J. van der Sluijs, Y. Lassailly and G. Lampel, J. App1. Phys. 79, 4 7 3 4 (1996). 54. H.-J. Drouhin, C. Hermann and G. Lampel, Phys. Rev. B 31, 3872 (1985). 55. C. Cacho, Y. Lassailly, H.-J. Drouhin, G. Lampel and J. Peretti, Phys. Rev. Lett. 88, 0 6 6 6 0 1 (2002). 56. T. Valet and A. Fert, Phys. Rev. B 48, 7099 ( 1993). 57. J. E. Wegrowe, Phys. Rev. B 62, 1067 (2()()()). 58. L. Berger, J. Appl. Phys. 89, 5521 (2001). 59. W. Weber, S. Riesen and H. C. Siegmann, Science 2 9 1 , 1 0 1 5 (2001). 60. H. Riechert, H.-J. Drouhin and C. Hermann. Phys. Rev. B 3 8, 4 1 3 6 (1988). 6 I. J. M. Kikkawa and D. D. Awshalom, Natllre 3 9 7 , 139 (1999). 62. T. Dietl, H. Ohno, F. Matsukura. J. Cibert and D. Ferrand, Science 2 8 7 , 1019 (2000). 63. S. Venugopalan, A. Petrou, R. R. Galazka. A. K. Ramdas and S. Rodriguez, Phys. Rev. B 2 5, 2 6 81 ( 19 8 2 ). 64. T. W. Ebbesen, H. J. Lezec, H. F. Ghaemi, T. Thio and P. A. Wolff, Nature 391,667(1998).

Index

A

absorption coefficients 11, 1 2 , 1 5 2 , 1 9 5 - 1 9 7 , 2()9,226. 2 8 2 , 3 0 9 . 451 polySiGe microbolometers 459-46() SiGe detectors 4 2 1 - 4 2 3. 4 2 6 - 4 2 7 . 442 absorption-edge spectroscopy (ABES I 312. 31 3315,325 active imaging 299. 300-301 active IR countermeasure systems 2()1 AEG 32 AFM (atomic force microscopy l 148.216 AIM 53,283. 302 A1, and vapor-HF processing 463 Al-free materials 124-12 5. 142 M-mirror 4 3 4 - 4 3 5 A1GaAs barriers 55, 99. 102.1 1(). 114 A1GaAs/GaAs 153.2 75. see also QWIps ALE (atomic layer epitaxy) 246 annealing 147-148. 153. 319. 342. 357. 453. 454 anodic oxidation 420 APDs see avalanche photodiodes As diffusion/implantation 34 As doping 240, 2 5 5 , 2 5 6 . 291. 302. 311.32 33 2 4 , 3 2 5 - 3 2 6 . 326 astronomy applications 2 9 . 3 3 . 2 33.2 51.2 71. 300 AT&T Bell Laboratories 85 atomic force microscopy (AFM) 148, 216 atomic layer epitaxy (ALE) 246 Au doping 283 Au films 5 () 5 Au pixel separators 385 Au/AuZn electrodes 142 Au/Co/Au multilayer 504 Au/Ge and Au ohmic contact 5 5 . 9 5 . 1 0 4 . 1 () 7 Au/n-Si SB diode 356 Au/SnPb eutectic bonding 473 Auger exclusion 285 Auger recombination HgCdTe (MCT) 1 7 . 7 0 - 7 1 . 163. 187. 2 4 2 245,280,303

InAs/iGaln) Sb superlattices 163, 1 76,187 lead salt detectors 275 type-II superlattices 191.2()1, 2()3, 2()4, 228 automotive applications 58, 70.3()5. 397. 4 4 9 45(). 472 avalanche gain 3()1.333 avalanche photodiodes IAPI)s) 2 33. 31(). 331337.41() B

BAESvstems 3 1 , 2 7 2 , 2 7 4 , 28 3 ball limiting metallurgy (BLM) 1 39 band alignment 191-192 modeling, GaSb/InAs s u p e r l a t t i c e s 2()()-2()1 bandgap, HgCdTe {MCT) 2 33-2 34, 2 36-238 bandgap engineering 69, 194,202, 3() 5 barium strontium t i t a n a t e (BST) 61,62 BCB t BenzoCycloButene) 4 7 3 , 4 7 5 , 4 7 6 Be doping 84, 144, 1 7(). 212, 215 BEEM technique 496 Bi-Sb-Te thermoelectric materials 65 Bi-Te thermoelectric materials 65 bias-selectable d e t e c t o r 3()2 BIB {blocked impurity band)devices 31, 37-38, 68 biCM()S processing 58, 45(), 453 BID I Blocked Intersubband Detector) 114-115 BLIP condition 16-17, 25 HgCdTe 2D arrays 2 7 2 , 2 7 3 , 2 7 4 , 2 9 6 , 3 0 3 InSb photodiodes 3()-33 QWIPs 57, 90, 91,94, 1()(), 1()7 BLM {bal/limiting metallurgy) 139 blocked impurity band {BIB)devices 37-38, 68 blocked intersubband d e t e c t o r {BID) 114-1 ] 5 blooming 274,296, 361-362 blue shift 129-1 3() Boeing see Rockwell/Boing bolometers 5, 9, 2 3, 26, 31-32, 62, 68,122, 353, 379-38 3. 411,451,seealso microbolometers boron IB)37(), 3 7 3 , 4 0 4 , 4 1 3 , 4 1 6 , 4 2 3 , 4 2 9 , 431,444,455

510

Handbook of hlfrared Detection Technoh)qies

bound-to-bound (B-B)intersubband absorption 128-129 bound-to-continuum (B-C) intersubband absorption 85. 100. 111. 128-129 bound-to-minibound (B-M)intersubband absorption 1 2 8 - 1 3 0 bound-to-quasibound (BQB)intersubband absorption 8 6 - 8 7 . 9 3 , 102.1 () 3.11 ()- 111. 115. 128-129. 146 Bridgman crystal growth method 2 4 7 . 2 8 4 . 2 8 5 broad-band OWIP imaging camera 102-11 () buffered direct injection 2 78 bulk crystal growth 3(). 245. 2 4 7 - 2 4 8 . 274. 284 Burstein-Moss effect 33 C C-QWIP (corrugated QWIP) 152 capacitative transimpedance amplifier (CTIA) 278.472 carbon 4 1 0 - 4 1 1 carrier concentration. HgCdTe (MCTI 2 38-2 39. 258-260 carrier generation rate. OWIPs 72.1 31 carrier lifetime HgCdTe (MCT) 241-245.3()(). 3()9. 33(). 338 type-II superlattices 2 0 4 . 2 2 5 carrier mobility. HgCdTe (MCT) 24()-241.2 58260 cavity oxide 4 34-435 CCDs (charge coupled devices) 2 9 . 3 6 . 3 7 . 2 76. 360-363. 365. 367,373. 384 CdTe, in MCT 2 33.234.2 35. 246.252.281. 285. 316. 317. 344 CdTe buffer layers 249,251,286.3()(). 341342,344 CdTe substrates 2 8 . 3 3 . 3 5 , 49,249.25(). 252 CdTe/CdZnTe buffer layers 2 5 5 , 2 8 6 CdTeSe 285 CdZnTe layers 2 55 CdZnTe substrates 2 8 . 3 3 . 3 5 . 4 9 . 249.25(). 254. 255. 283. 284-285. 286. 291. 296. 302. 318. 322. 325. 341-348 ceramic thin films 6 2 - 6 4 characterization, of HgCdTe (MCT) 2 56-26(). 334,345 charge injection efficiency 95-96. 107-1 ()8 charge sweep device (CSD) 362-363. 366 chemical vapor deposition see CVD chip carriers 1 38 CI Systems Inc. 313 CMOS 29-3(). 3 7 , 4 8 . 5 7 . 5 8 . I()7, I()8. 112. 141. 2 7 6 - 2 7 9 microbolometers 449-45(). 4 5 3 . 4 5 7 . 4 6 7 468.477 Si infrared FPAs 360. 374. 377. 38 3. 385 Co thin films 4 9 5 - 4 9 8 . 504 Co,St detectors 367-368 commercial/industrial applications 3 8 . 5 8 . 6 5 66.88. 304. 312. 396-397 communications applications 2()1 _

composite detector technology 64 condensation, polySiGe microbolmeters 4 6 2 463 conduction band HgCdTe 28(), 281 lnAs/IGaln )Sb superlattices 16(). 1 7() ()\VlPs in-type ()WIPs) 12 3. 128-1 3(). 14(.)147 Si (;e detectors 39 3. 398 type-II superlattices 19(-), 2()()-2() 1 cooled Si FPAs 374- 386 cooled type-II photoconductors 2()8-211 cooled type-II photodiodes 215-22 7 coolers 9 4 . 9 7 . 9 8 . 9 9 . 1 1 2 , 1 2 4 , 2 7 3 - 2 7 4 . 3() 3.367. see also cryogenic cooling corrugated ()WIP (C-QWIP) 1 52 corrugation 14(-), 151-152 crosstalk 274. 288. 291.29(.). 297. 33{). 3(.)3 cryogenic cooling 5 . 7 . 1 1 . 1 6 . 2 1 . 2 3.58. (.)8. 272. 337.411 cost 49.58.2()9.27() see also cooled systems: coolers cryopumps 31() crystal growth see growth technologies crystal mining operation 24 7 crystal structure. HgCdTe (MCT) 2 35-2 3(.) CSI) (charge sweep device) 362-3(.) 3.3(.)(.) CTIA (capacitative transimpedance amplifier) 278.472 Cu at I,PE grov,'th stage 289 Cu doping 28 3 Cu substrates 4 9 7 - 4 9 8 CV1) (chemical vapor deposition) 245.2 51. 2 52-2 54. 3 7 1 . 4 5 3 - 4 5 5 . seealso M()CV1) Czochralski crystal growth method 247 D dark current 39-41 ttgCdTedetectors 39-41.44, 51. 332. 33533(.). 337 InAs/( Gain )Sb superlattices 173 QWIPs 4()-41. 47. 51. 85-87, 94, 95, 1()41()8. l l l . 1 1 5 . 1 3 1 - 1 3 2 . 1 3 5 , 1 4 2 , 1 4 5 , 148.276 Schottkv-barrier FPAs 356. 37() SiGedetectors 41(). 411. 4 1 2 . 4 2 3-438.44(), 442. 4 4 4 - 4 4 5 type-II superlattices 2()1-2()2.22 7 I)AS (digital acquisition system)98 densities of states 48 3-49 5 I)ERA 117KI64 detectivitv (I)') extrinsic silicon detectors 224 ttgCdTe 292. 2 9 5 - 2 9 6 InAs ((;aln)Sb superlattices 184-185 photon detectors 1 ( } - 1 1 . 2 1 . 2 5 . 4 1 - 4 3 . (.)7. 71, 12 3-124 QWIps 4 1 - 4 3. 7 3. 89-9(). 1() 5-1()6.115. 125-126. 13(). 133.135.15()-151 Si(;e detectors 412. 4 2 4 - 4 3(). 4 4 4 - 4 4 5 thermal detectors 21

hlde.v

type-II superlattices 2 0 6 - 2 ( ) 7 . 2 1 3.214. 224.227 dielectric bolometer 62 diethyl telluride {DETe) 2 "32 Difference ofGaussian {DoGI filter 3()4 digital acquisition system {DASt 98 dimethyl cadmium {DMCd) 2 ,52 dipping. LPE crystal growth 2 ,5(). 2 ,51. 284.29() direct injection 2 77-2 78 direct Schottky injection (DSIt FPA 36 3 dislocation densities, HgCdTe MBE technology 3 2 0 - 3 2 1 . 341. 342 dislocations in SiGe epilayers 398 doping 2 34.282 BIB devices 3 "5, 37 HgCdTe (MCT) 2 39-24(). 2 .5.5.28()-281. 28.3,286. 2 8 7 , 3 0 9 . 311. 322-326. 338 InAs/(GaIn)Sb superlattices 1 6 7 - 1 6 9 . 17{)171,181-183 polySiGe microbolometers 4,53-4 .54.4 ,5,5 QWIPs 8 4 - 8 6 . 9 3 . 9 9 - 1 ( ) { ) . 1()4.11()-111. 1 1 4 , 1 2 8 . 1 3 3 - 1 3 4 . 1 34. 142.14,5. 146 Schottky-barrier FPAs 37() SiGe detectors 4 1 2 . 4 1 3 - 4 1 6 . 4 2 .5-4 38. 44()-44,5 spin doping and ferromagnetic metals 482 type-II superlattices 1 9 2 . 2 ( ) 9 . 2 1 2 . 2 1 .5.22 7 double planar heterostructure {DLPH I 291 DOW {GeSiGe double QW)structure 398-411 DRS Infrared Technologies "31. ,53.28 3.297. 302, 30 3 d u a l b a n d FPAs .52-.58.69.99-1()2.11.5. 126. 301-3()3 dual ion beam sputtering 6 3 - 6 4 E

e-beam technologies 34. 3,37, 368 early warning systems applications 88.2{) 1. 233 Eastman-Kodak 32 ECR-RIE 1 4 0 . 1 4 , 5 , 1 4 6 electroluminescence (ELl 399-4()1 electron probe microanalysis 2 ,57 electron transmission through ferromagnetic bilayers ,5()1-.3(),3 electroplating 139 ellipsometry see spectroscopic ellipsometrv envelope-function approximation i EFA I 161 162.16,5. 197-2()() envelope function quantum number 19 ,5 epitaxial technologies HgCdTe{MCT) 33-3 .5.68.69. 246. 2 4 8 2 . 5 6 , 2 8 4 - 2 8 7 , 2 8 8 . 3()9-348 HgCdTe on Si 4 7 . 4 9 . 2 ,5.3.286 multispectral HgCdTe ,53. ,5,5 inQWIPs 1 2 7 . 1 3 4 . 1 37 see also MBE epoxy underfill 138 ethyl iodide 2 "56.28:5 extrinsic detectors 7, 13.18, 21, 3 6 - 3 8 . 7 2 - 7 3 . 122,226

,51 1

extrinsic silicon detectors 2{)8. 2 2 4 - 2 2 7.2 3423,5 F

fast fl)urier-transform {FFTt spectrum analyser 1 3,5.2()4.21

3

Fermi level 48 3 ferroelectric ITFFEt detectors 6 2 - 6 4 . 6 6 . 7 ( ) , 122.2()1. 3,53 ferromagnetic metals, spin filtering 481-,5() "3 fibre optical applications 393. 398 fire and combustion control applications .58. 396-397 tlip-chip bonding 2 6 . 2 8 . 1 37-1 38. 147. 473 Fourier-transform {FT} interferometer 69 Fourier-transform IR {FTIRI spectrometer 93. 13 ,5. 146. 2{)2. 2()9. 2 1 3 . 2 1 9 . 22(). 2 ,56.

321 FPAs Ifocal plane arraysl .5.7-9.2 3-26.41-,5(). 6 8 . 8 3 - 1 1 7 . 1 2 2 . 363-367. 3 7 4 - 3 8 6 for photon detectors 2 6 - . 5 8 . 1 2 4 for thermal detectors 6..58-67 three- and four-band 69.11 ()-11 3.11 ,5 see also dual band FPAs G

GaAs interfaces 9 9 . 1 1 ( ) - 1 1 1 . 1 6 5 - 1 6 6 GaAs lasers 124 GaAs substrates HgCdTe {MCI)) 2 4 9 . 2 8 3. 296 QWII's .355.9 3.1{)2.1()4.1()7.1 1{)-1 1 1. 1 14.12,5.1 34. 142. 144 in spin tiltering ,5{)4.5{),5 type-II superlattices 2{)2.2{)7.2{)9 (;aAs ~AIGaAs-based detectors 1 2 4 - 1 2 .5.126. 14 3. 147. see also QWIPs (;aAs GalnAsI ) {p-typelQWIP 14 3 (;aAs/(ialnl) (p-typel QWIP 1 4 2 - 1 4 3 (;alnAs,Int) (n-type} QWIPs 1 4 4 - 1 4 6 . 1 ,51 (;alnAs/InP {p-type)QWIP 144 (;alnAs/InP QWIP-on-InP 148-1 ,51 GalnAs/InP QWIP-on-Si 148-1.31 (;alnAsP. for API)s 332. 336 (;alnAsP barriers 14 3 (;alnAs{ P1 material system ()WIPs 12 ,5 (;alnAsP/(;aAs 12 ,5 ( ; a I n A s P (;alnAsP Ip-tyel ()WIt)s 144 (;aInAsP'InP 12,5.1 34 (;alnl' barriers 142 (;aSb buffer contact layers 2()2, 212.21 ,5 (;aSbsubstrates169.181. 187.192.212.21,5 GaSb I n A s superlattices 191-2 3{). see also typeII superlattices gated pillars 2 2 8 - 2 2 9 gated-imaging applications 2{) 1.3(){) (,d {gadolinium1492 (,e doping 41 3.416. 4 2 9 - 4 31,43,5 (;elenses .59.98.1{)8.112 (]e photoconductors 2 3 . 3 6 - 3 8 Ge substrates 249 GeC structures 41 ()-411

512

Handbook of Infrared Detection Technologies

GeSi HIP LWlR FPA 374, 386 GeSiGe double OW (DOW) detectors 398-411 GEC Marconi 32.51 General Motors 62 GEN III detectors 2 6 9 , 3 0 1 - 3 0 4 glass 4 74 graphite 25 O, 453 growth technologies HgCdTe (MCT) 2 4 5 - 2 5 6 . 2 7 1 . 2 8 4 - 2 8 7 QWIPs-on-Si 14 7 see also epitaxial technologies GSMBE 1 3 4 , 1 4 4 , 1 4 7 H

Hall effect 2 5 8 - 2 6 0 , 3 2 2 - 3 2 3 HAWAII multiplexers 291 HDVIAP process 2 8 9 - 2 9 0 . 2 9 6 . 2 9 7 helium refrigerator 2 1 6 - 2 1 7 hermeticity 4 72-476 hermocouple 383 heterojunction internal photoemission (HIP1 detectors 3 5 3 , 3 7 1 - 3 7 4 , 3 8 6 , 3 9 7 . 4 1 1 - 4 4 5 heterojunctions 16, 33, 69, 70,200, 2 34.2 _52. 282,290-292,309 DLHJ structure 3 3 , 2 9 0 - 2 9 2 , 2 9 6 , 302 type-I 191 type-II 191-192 HgCdTe detectors 6 . 1 7 - 1 8 , 21.2 3, 2 6 . 3 1 - 3 6 . 37, 6 7 - 7 1 , 1 2 3 - 1 2 4 , 2 0 1 , 2 6 9 - 3 0 5 . 394. 397 compared with InAs/(GaIn )Sb superlattices 26,159,163,186,187 compared with OWlPs 3 9 - 5 2 . 6 9 , 12 3-124 compared with type-II superlattices 201. 207. 208-209, 213. 214, 2 1 7 , 2 2 3 - 2 2 4 . 2 2 7 dual-band 53-:54 HgCdTe on Si 47, 49. 255,286. 305. 310. 341-348 multi-junction 70 HgCdTe semiconducter technology see MCT HgTe, in MCT 233, 234, 2 3 5 . 2 4 6 . 2 5 2 , 285 high temperatures 50, 225,300. 341. 347348, see also operating temperatures high-performance MWIR detectors 3 3 7 - 341 high-resolution PtSi FPAs 363-367 higher operating temperature (HOT) IR detectors 233, 303 hillock formation 254. 2 8 6 . 3 1 8 HIPs see heterojunction internal photoemission (HIP) detectors hole impact ionization resonance 333 hole intersubband absorption 12 3 . 1 4 1 - 1 4 4 hole mobility, HgCdTe films 323 hole states, InAs/(GaIn)Sb superlattices 160163 homojunctions 2 8 1 , 2 8 2 , 2 8 7 , 29(), 297. 299 Honeywell 58, 62,449 hot electron transfers 191 HOT (higher operating temperature) detectors 233,303 HRL Laboratories, LLC 342

human eye 2, 3()4 hybrid FPAs 26-28, 5(). 68,363 hybrid polySiGe microbolmeters 4 6 7 , 4 7 7 hybrid pyroelectric/ferroelectric bolometer 62, 66 I

I-profile 4 6 4 - 4 6 5 IL-CCD (interline transfer CCD) 362. 365 IMEC 450. 45 3. 4 6 7 . 4 7 3 , 4 7 7 IMFP {inelastic electron mean free path) 482. 483-495 IMP (interdiffused multilayer process) 2 5 2 . 2 8 5 impact ionisation 282 in situ process sensors. HgCdTe MBE technology 3()9. 31(). 312-317 indium bumps 26, 34.50.51, 53, 57, 95. 107, 137140.216. 255,288. 302,322. 327 doping, in HgCdTe r 240, 2 55,286. 311. 322.32 3 - 3 2 4 , 3 2 6 levels, in InAs/( Galn)Sb superlattices 160, 163 InAs:Si superlattices 2 1 2 . 2 1 5 InAs:Si/GaSb superlattices 212, 215 InAs/IGaln)Sb superlattices 159-187 InAs/GalnSb superlattices 209. 2 2 2 - 2 2 7 InAs/GaSb:Be superlattices 212, 215 InAsSb detectors 33.7(). 201.21 3 InGa/InGaP quantum dot detector 1 53 InGaAs 32. 33.3()(). 394 InGaAs/AIGaAs and GaAs/AIGaAs MOWs 126 InGaAs/AIGaAs strained layer system 57 InGaAS/GaAs (n-type) OWIP 146-147 InGaAsP/InP in-type)QWIPs 146 InC,aP/GaAs 125 InGaSb 33 InP barriers 144. 145. 146. 147 InP substrates 1 2 5 . 1 3 4 , 1 4 4 . 147 InSb 7-9 InSb detectors 3()-33, 37.68, 7 1 , 2 7 4 - 2 7 5 , 3()5. 344. 346-348 InSb interfaces 165-166. 212 InSb substrates 249 Indent-Reflow Sealing (IRS) 4 73-475 Indigo Systems 9 4 . 9 8 . 9 9 . 1 0 7 industrial applications see commercial/ industrial applications inelastic electron mean free path (IMFP) 482. 483-495 integrated detector technology 64 integrated detector-cooler assembly (IDCA) 272-273 interband tunnelling 281 interdiffused multilayer process (IMP) 2 52.285 interferometers 6 9 . 3 9 5 intersubband absorption 8 3 . 1 1 4 . 1 2 3, 1261 3(). 141-144. 276. see also b o u n d - t o continuum: bound-to-minibound: bound-toquasibound intrinsic detectors 7 . 8 . 1 7 . 122. 225,361

Index

iodine doping 240, 255, 285 ion beam milling 2 8 7 , 2 8 9 ion implantation 2 8 7 , 2 8 8 , 3 0 0 , 370, 455 ion pumps 310 IR light 1 2 1 - 1 2 2 IR transmission 256, 3 1 6 . 3 3 4 iron, in thin films, 4 9 5 - 4 9 8 , 5 0 4 . seealso ferroelectric detectors; ferromagnetic metals IrSi detectors 3 6 7 , 3 6 8 - 3 6 9 , 3 9 4 ISOVPE (isothermal VPE) 245

I

Jet Propulsion Laboratory 3 2 . 4 5 . 1 2 6 . 3 9 7 . 412 Johnson noise 20, 4 1 , 1 3 2 , 1 8 5 . 2 0 5 . 2 1 3 , 337-338,459 Joule-Thompson cooler 2 73 junction forming 2 8 7 , 2 8 8 , 2 9 6 . 2 9 8 - 2 9 9 , see also heterojunctions: homojunctions K

Kentron 275 Kopin Inc. 148 Kronig-Penney model 128 L lasers 6 4 , 1 2 4 , 1 5 9 , 1 9 1 . 1 9 2 , 2 0 1 . 206. 271. 299,300,331-332 lead salt detectors 2 74.2 75. see also Pb leakage currents 281, 2 8 2 . 2 8 5 . 2 9 1 . 2 9 7 - 2 9 8 . 436 LED structures 124, 153 lenses 50, 59, 9 8 , 1 0 8 , 1 1 2 LETI 5 3 , 3 8 1 - 3 8 3 LIDARs 201 light coupling 38, 57, 84, 87-88, 9 5 , 1 0 1 . 1 0 5 , 1 0 9 - 1 1 0 , 1 1 1 . 1 3 3 , 1 3 4 . 146. 152, 201 linear arrays 6 4 - 6 5 , 6 6 , 2 7 0 , 2 7 2 , 2 9 6 , 4 6 7 . 471,477

lithography 172, 450, see also photolithography Lockheed Martin 45, 53, 56, 58, 6 0 , 2 2 4 long wavelength IR (LWIR) 5 - 6 . 8 , 2 1 - 2 6 . 3 1 32,121,283 d u a l b a n d 53, 54, 9 9 - 1 0 2 , 1 1 5 , 1 2 6 HgCdTe (MCT) 9, 17, 2 1 - 2 6 , 3 3 . 4 9 . 6 7 . 2 55, 2 7 0 , 2 7 2 , 2 7 4 , 2 7 7 - 2 7 8 , 2 8 2 . 283. 284. 296-299,303. 305,321,342,344 MEMS 70 QWIPs9, 45, 57, 69, 88, 9 2 - 1 0 2 , 1 1 4 - 1 1 6 . 126,144, 147,274,276 Si and Si/Ge detectors 3 5 3 , 3 6 9 . 3 7 3 - 3 7 4 . 393-397, 411-445 thermal detectors 2 1 - 2 6 Type-II superlattices 1 9 2 , 2 0 2 , 211 loophole interconnection 26, 5 1 , 2 8 9 - 2 9 0 . 296,299 low temperatures 5, 38, 45.5(). 1 1 3 - 1 1 5 . 201. 2 2 5 , see also operating temperatures LP-MOCVD 144, 145, 146 LPE (liquid phase epitaxy) 34, 53. 2 4 0 . 2 4 5 ,

513

2 4 9 - 2 5 1 . 283. 2 8 4 - 2 8 5 , 2 8 7 , 2 8 8 . 290, 296. 299. 3 0 0 , 3 0 9 . 346 L W I R / L W I R dual band 5 3 . 5 4 . 9 9 - 1 0 2 . 1 1 5 LWIR/VLWIR dual band 9 9 - 1 0 2 , 1 2 6 LWIR see also long wavelength IR M

Marconi 32 MBE (molecular beam epitaxy ) dual-band HgCdTe 53.55 (;alnAstPt based QWIPs 1 3 4 . 1 3 7 . 144. 147 GaSa/AIGaAs based QWIPS 8 4 . 9 3. 1 0 4 . 1 1 1 . 114 GaSb/InAs superlattices 202, 2 1 5 . 2 2 7 HgCdTe (MCT) 33.34, 35.69, 24(), 245. 247. 2 5 4 - 2 5 6 , 2 5 7 . 2 8 3 , 2 8 6 . 291. 296. 3 0 9 348 HgCdTeonSi49. 255. 286 HIP detectors 372. 373 InAs/(Galn)Sb superlattices 1 6 3 . 1 6 5 . 1 6 9 Schottky-barrier FPAs 3 6 9 - 3 7 1 SiGe detectors 4 0 3 - 4 0 8 . 410, 413. 426 MCT (Mercury Cadmium Telluride) semiconducting material 5 . 7 . 1 2 3-124, 2 33-260. see also HgCdTe detectors medical applications 8 8 - 8 9 . 305. 396. 449 megapixel arrays 301. 310. 366 MEMS technology 70. 453. 4 5 4 , 4 5 7 . 461. 4 6 3 - 4 6 4 . 476. 477 Mercury-Cadmium-Telluride {material} see MCT mesa 2 9 0 - 2 9 1 metal absorbers 4 5 9 - 4 6 0 metal-insulator-semiconductor (MIS) structure 26 metallurgical defects 35 MFPAs see multispectral focal plane arrays microbolometers 5 8 - 6 1 . 6 6 . 7 0 . 2 0 1 , 2 0 7 2 0 8 , 2 1 4 . 353. 3 8 3 . 4 4 9 - 4 7 7 microelectromechanical structures see MEMS technology microlenses 50.22 7 middle wavelength IR (MWIR) 6 . 9 . 2 4 - 2 6 . 3 1 32. 121. 2 7 4 . 2 8 3 dual band 5 3 . 5 4 . 5 7 . 302 HgCdTe (MCT) 7 . 3 3 . 3 5 . 2 5 5 . 2 7 0 , 2 7 1 . 2 7 4 , 2 7 7 . 2 8 2 . 284. 293. 296. 303. 3 0 5 , 3 1 0 , 318. 321. 322. 3 2 8 - 3 2 9 , 3 3 7 - 3 4 8 InSb detectors 33, 2 7 4 , 3 4 6 - 3 4 8 QWIPs 57. 1 2 6 , 1 4 2 - 1 4 4 , 1 4 7 , 1 5 2 - 1 5 3 Si, PtSi and Si/Ge detectors 3 5 3 . 3 5 7 . 3 9 3 397.411-445 type-II superlattices 192, 211 military applications 33.38, 5 0 . 5 8 . 6 1 , 65, 66, 2 2 3 . 2 3 3 , 3 0 4 . 3 1 2 , 3 9 3 . 397. 449 mine detection 58 MIRIADS programme (miniature IR imaging applications development system} 303 missile applications 201, 233, 2 73, 397 Mitsubishi 31.59 MOCVD {metalorganic CVD) HgCdTe(MCT) 33, 34.53, 69. 2 4 0 , 2 4 5

514

Handbook of Infrared Detection Technologies

OWIPs 134, 137, 14(), 142 see also LP-MOCVD: MOVPE model-solid theory 200 modeling IMFP variation 4 8 3 - 4 9 5 SiGe s t r u c t u r e s 44() type-II superlattices 197-2() 1 modulators 1 9 1 , 1 9 2 , 3 9 5 , 41() MOMBE (metallorganic MBE1246, 256, 311 MOS 37,360, 374, 377, see also CMOS MOSFETs 1 0 8 , 2 7 6 , 2 7 7 , 2 7 8 , 3()2-3{) 3 MOVPE (metal-organic vapour phase epitaxy 2 8 3 , 2 8 5 , 2 9 6 , 3()(), 302, see also M()CVl) MOW s t r u c t u r e s 84, 92-93, 1()(), 1()2, 11(), 126, 1 3 4 , 1 4 7 , 2 7 5 - 2 7 6 , 4 1 1 - 4 4 5 Multi-Chip-Module (MCM) 469 multicolor operation 5, 9 HgCdTe 5 0 . 5 3 - 5 4 , 68, 69 OWIPs 54-58, 69, 11()-11 3. 126.144. 147. 152-153 see also multispectral focal plane arrays (MFPAs) multijunction HgCdTe photodiodes 7() multiplexers/readouts 26, 34.5(). 5 1 . 6 9 . 9 4 . 122,269. 2 7 0 , 2 7 6 - 2 7 9 . 288. 297.3()2. 467-472 noise 4 7 . 5 7 , 1 0 1 - 1 0 2 , 1 0 7 see also CCDs: CMOS: ROWs multiquantum well structures see MQW structures multispectral focal plane arrays (MFPAsl 2 33, 234.26(), 3()1-3()3,305. 324-331. s'eealso multicolor operation MWIR/LWIR dual band 5 3 . 5 4 . 5 7 . 302 MWIR/MWIR dual band 53.54. 302 MWIR see also middle wavelength IR N nanoscale structures 153, 228 near-IR (NIR) 54, 153.2 33, 31 (). 39 3. 396. 398-411. see also avalanche photodiodes: short wavelength IR Ni, in spin filtering 4 9 6 , 5 0 5 Ni/Au layers 4 73-4 78 NiCr 459 NiSi detectors 367-368 night vision applications 58, 88. 201. 397. 449-45O nightglow 2 9 9 NIR see near-IR noise 10-11, 2 0 - 2 1 , 2 3 - 2 6 HgCdTe detectors 299.3()(). 332.336. 346 high-performance MWlR detectors 337- 341 InAs/(Galn)Sb superlattices 183-186 polySiGe microbolometers 4 5 6 - 4 5 7 . 4 5 9 QWlPs 9 7 , 1 0 1 - 1 0 2 . 107, 1 1 2 . 1 3 2 - 1 33. 276 Schottky-barrier detectors 358- 3 5 9 . 3 6 3 365,374,379 thermal detectors 2 0 . 5 8 . 6 0 type-II superlattices 2 0 4 - 2 0 6 . 2 1 3

noise equivalent temperature difference 2 3-26, 67,69 dual-band FPAs 56.57, 99-1()2 HgCdTe 21) arrays 2 7 7 , 2 9 2 - 2 9 3 lnSb photodiodes 73-75 metal bolometers 4 5() photon detectors 36, 67.69, 73-74 poly,Si(;e microbolometers 449.45(). 471 Q\VIPs 6 9 . 9 1 - 9 2 . 9 4 . 9 7 . 9 9 . 1 ( ) ( ) - 1 ()1. 1()6. 1()8.115. 133 Si bolometers 58.59.6()-61. 38(). ~8 3 thermal detectors 6 7 . 7 4 - 7 5 . 2 ( ) 8 non-equilibrium device 7() non-unitbrnaitv 2()8. 2 2 8 , 358. 367. 379 correction tNI~C~ 273 in QWIPs 97,1()7.1()9. 112. 115 see also unifl)rmitv O ()EIC ioptoelectronic integrated circuit ) 398 ()FXl ioptical absorption tlux monitoringt 312, 316-318 ()NIVPE Iorganometallic vapour phase epitaxy) 2 8 6 - 2 8 7 . 31)9 operating temperatures 4(). 45.5(). 11 3-11 5. 2 ( ) 1 - 2 ( ) 2 . 2 2 5 . 2 33. 3()(). 3()5. 341. 356 BI1)detectors 11 3-11 5. 11 7 ttgCdTe 35.68-69. 341. 347-348 InSb photodiodes 3()-31 pyroelectric arrays 61-62 O\VIPs 38.39.56. 124 see also cooled s y s t e m s : coolers: uncooled SvsIelllS

optical absorption tlux monitoring {()FMI 312, 316-318 optical communication applications 393 optical concentrators 9 optical conversion efficiency 39, 51 optical immersion 7() optical modulators 61 optoelectronic integrated circuit IOEICI 398 organometallic vapour phase epitaxy {()MVPE} 286-287. 3()9 oxygen 358.4()7.42(). 449 P p-type QWIPs 142-144 PA-XI()CVI) (photo-assisted MOCVDt 2 53-2 54 PACE process 2 8 6 - 2 8 7 packaging 45(). 4 6 1 , 4 7 2 - 4 7 6 . 477 passivation (;aInAs(P) based QWIPs 1 37.1 39.151 HgCdTe 33-34. 281,291. 298 InAs/(Galn)Sb superlattices 1 59, 1 7(). 1 72. 178. 187 type-II superlattices 22 7.2 3() Pb-based "perovskite" oxides 6 1 - 6 2 . 6 4 PbSn 474 PbSnTe 1 7 Pd,Si 367 t'ECVI) Iplasma enhanced chemical vapor

Index

depositiont 1 3 7 . 3 8 0 . 4 ( ) 8 . 4 2 ( ) PEM (photoelectromagnetic) detectors 7 "pendell6sung-fringes" 1 6 4 - 1 6 5 phase diagrams, MCT 246 photo-assisted MOCVD (PA-MOCVDt 2 53-2 54 photoconductive detectors 7.2 3 . 3 6 - 38.69. 72-73, 2 7(), 284. see also QWIPs: type-II photoconductors photoconductive gain 9 . 6 9 . 9 7 . 1 1 4 - 1 1 5.1 35. 152,204 photocurrent injection techniques 2 77-2 78 photodiodes 7 - 7 5 , 2 6 9 . 279, 41(). seealso avalanche photodiodes: type-II photodiodes photoelectric gain 9 - 1 0 photoelectromagnetic (PEM) detectors 7 photoemissive detectors 7, 36. see also Schottkvbarrier detectors photolithography 139. 140. 1 4 2 , 1 4 5 . 2 1 6 . see also lithography: MEMS photoluminescence (PL) 16,3, 1 6 8 - 1 6 9 . 399401 photon detectors 5 . 6 - 1 8 , 2 3 . 2 6 - 5 8 . 6 7 . 1 1 2 . 121 compared with thermal detectors 9 . 2 1 - 2 3 . 25-26,121 limitations 201 photovoltaic detectors 7 . 2 8 . 3 6 . 2 1 5 Physical Electronics Laboratory of ETH Zurich 384 physical vapor deposition (PVD) 2 4 5 . 2 5 1 - 2 5 2 PiN diodes 301 pm photodiodes 41 () pixel amplifier 469 pixel fabrication 1 3 4 . 1 3 7 pixel optimization 4 6 0 - 4 6 2 pixel size reduction 59 pixelless large-area applications 153 planar device structure 2 7 4 . 2 8 8 - 2 8 9 . 2 9 6 plasma enhanced chemical vapor deposition see PECVD plasma enhanced milling 2 8 7 . 2 8 9 PMN (lead magnesium niobate) 6 1 . 6 2 polarization selection rule 195 pollution monitoring 208 polyGe 454 polyimide 382 polySi 65, 454 polySiGe uncooled microbolometers 4 4 9 - 4 77 pre-amplifiers 213, 214 process sensors in HgCdTe MBE technology 309. 310. 3 1 2 - 3 1 7 proton implantation 153 proximity fuzes 201 PST (lead scandium tantalate) 3 2 . 6 1 . 6 2 . 6 4 PteSi 3 5 7 - 3 5 8 Ptlr silicides 369 PtSi 7 - 9 . 3 1 - 3 2 . 3 6 . 4 9 . 6 8 . 9 2 . 394. seealso Schottky-barrier detectors PT (lead titanate) 6 1 . 6 2 PVD (physical vapor deposition) 2 4 5 . 2 51-2 52 pyroelectric detectors 5 . 9 . 3 1 - 3 2 . 5 8 . 6 1 - 6 4 .

515

66. 122.411 pyrolytic XI()CVI) of ~ICT 2 52 PZT Ilead zirconate titanatel 61

O

quantunl cascade lasers 1 9 2 . 2 ( ) 6 . 2 1 4 quantum confinement 228 quantum dot detectors 1 53. 228 quantum efficiency 24, 2 34. 2 8 2 . 2 9 1 extrinsic silicon detectors 2 2 4 . 3 5 4 . 356, 36{). 365. 3 7 2 - 3 7 3 photon detectors 9. 33. 36. 3 9 . 6 8 . 1 2 3-124 Q\VIPs 3 9 . 4 7 . 8 9 - 9 ( ) . 9 3 . 1 ( ) 5 . 115.1 31. Sit]e detectors 41 (). 4 1 2 . 4 2 1 - 4 38. 442. 444-445 tvpe-II superlattices 2() 3-2()4.22 5 quantum grid IR detector 153 quantunl mechanical tunneling 85 quantum mechanics 8 3 - 8 9 . 1 2 3 quantum pillars 2 2 8 - 2 2 9 quantum well detectors 7.8 quantum wire detector 153 quaternary p-type OWIP 144 quench and anneal ~ICT crystal growth 247 QWIP-I,EI) 1 53 Q~.'IPs (quantum well IR photoconductors t 6.7, 9. 12-1 3 . 1 8 . 2 1 . 3 2 . 3 8 - 3 9 . 8 3-11 7 . 1 2 6 1 ~5.2()1. 397. 445 applications 8 8 - 8 9 compared with HgCdTe detectors 3 8 . 3 9 - 5 2 . 69. 1 2 3 - 1 2 4 detinitions 122.12 3-126 dual-band 5 3 . 5 4 - 5 8 . 9 9 - 1 ( ) 2 . 1 1 5 fabrication procedures 1 3 4 - 1 4 1 . 1 5 1 four-band 11()-11 3.11 5 n-type Iconduction bandt 12 3. 124. 1 2 8 1 3(). 1 4 6 - 1 4 7 new approaches 151-1 53 p-typelvalenceband) 123. 124. 128.1 3(), 142-146 parameters 1 3()-1 33 types 5 3 . 5 4 - 5 8 . 9 2 - 1 1 7 . 1 2 3 - 1 2 4 . 1 2 8 1 3() R

R~jA product HgCdl"e 29(). 328. 337-34(). 346 InAs/IGaIn)Sb superlattices 1 7 5 - 1 8 1 . 1 8 5 186 photon detectors 3 5 . 3 9 - 4 0 type-II superlattices 22 3-224 radiation hardness 367 radio frequency magnetron sputtering 63 radiometers 66, 2 33 Ravtheon 3 1 . 5 3 . 5 8 . 6 0 . 6 2 , 64, 2()8.28 3. 312. 325. 342. 345,381. 383 RBS {Rutherford back-scattering) 4 1 2 . 4 1 6 - 4 1 8 reactive ion etching see RIE readouts .see multiplexers/readouts Real Time Imaging Electronics {RTIE) 98

516

Handbook of Infrared Detection Technologies

recombination processes, in MCT 24 ] - 2 4 5 . see also Auger recombination red shift 401 reflection high energy electron diffraction see RHEED reflectors for 0WIPs 8 7 - 8 8 . 9 4 - 9 5 . 106-1()7 remote sensing applications 88. 3 6 6 - 3 6 7 resistive bolometers 4 4 9 , 4 5 0 . 4 5 1 . 4 5 5 - 4 5 6 resonant tunneling diodes (RTD) 1 9 1 . 1 9 2 responsivity HgCdTe (MCT) 2 3 4 . 3 2 9 - 3 3 0 InAs/(gaIn)Sb superlattices 1 7 2 - 1 7 3 OWIPs 1 0 . 1 0 4 . 1 0 5 . 114. 1 3 1 . 1 3 3 . 142. 144. 1 4 5 - 1 4 6 . 1 4 9 - 1 5 0 SiGe detectors 412. 4 2 9 , 4 3 4 - 4 3 8 type-II superlattices 2 0 2 - 2 0 3 . 206. 2 0 9 211.213,219 see also uniformity retina level processing 2 . 3 0 4 reverse bias 2 1 7 . 2 9 7 - 2 9 8 . 298-299. 303. 327. 356-357,369 RHEED (reflection high energy electron diffraction) 2 0 2 , 2 1 2 , 3 1 8 RIE (reactive ion etching) 2 2 9 - 2 3 0 . see also ECR-RIE Rockwell/Boeing 31, 53, 58, 283. 3 0 0 , 3 2 2 ROCs 9 5 - 9 6 . 9 7 , 9 9 , 1 0 2 , 1 0 7 , 1 0 9 , 1 1 2 - 1 1 3 . see also multiplexers ROICs (readout integrated c i r c u i t s ) 2 9 . 3 9 . 4 7 . 48, 6 4 , 1 2 2 , 1 3 7 , 1 4 0 - 1 4 1 , 2 4 9 , 2 5 5 . 3 { ) 1 . 336,341,468-472 fabrication 1 4 1 , 1 4 7 , 1 5 1 . 271. 302. 3 6 0 363 see also CCDs; CSD; multiplexers roiling readout device 2 79 RTD (resonant tunneling diodes) 1 9 1 . 1 9 2 Rutherford back-scattering (RBS) 4 1 2 . 4 1 6 418 S

SAM-APD architecture 3 3 2 - 3 3 3 sapphire substrates 2 8 , 2 4 9 , 2 5 0 , 251, 283. 300 Sarnoff 32 Sb doping 38 Sb surfactant 404 SBRC 46 scanning see multiplexers/readouts scattering effects 3 5 4 - 3 5 6 scattering, in spin filtering 4 8 3 - 4 9 5 , 5 0 4 - 5 0 5 Schottky-barrier detectors 5, 18, 26, 36.68, 69. 2 7 4 , 2 7 5 , 3 5 3 - 3 7 1 , 4 1 2 , 4 2 9 . 445 scientific applications 2 3 3 , 3 0 4 SE (spectroscopic ellipsometry) 2 5 7 - 2 5 8 . 3 1 2 . 315-316,318,333-335 search and rescue applications 58 security applications 305 SEIR image processing station 112 semiconductors 5, 6 , 1 2 1 , 1 9 9 - 2 0 0 , 2 3 3 , 3 5 4 and spin filtering 4 8 1 - 4 8 2 , 5 0 4 - 5 0 5 temperature sensors 3 7 4 - 3 7 9

Sensors Unlimited 32 sequential tunneling 85 sequential-mode detector arrays 3 2 6 - 3 3 0 Sherman function 483. 499. 505 short wavelength IR (SWIR) 3 1 - 3 2 . 3 3 . 121. 122.283 HgCdTe detectors 2 9 . 4 9 . 251. 255.27(). 2 9 9 - 3 0 0 . 301. 318 PtSiSBdetectors 353. 357. 366 Si/Ge detectors 3 9 3 - 3 9 7 see also near-IR silicon 7 - 9 . 1 3 . 2 6 . 353. 357. 3 7 4 - 3 7 9 . 394. 395 amorphous (a-Si) 32. 60-61. 3 7 9 - 3 8 3 . 450, 453 bolometers 5 . 2 6 . 5 8 - 6 1 . 353. 3 7 9 - 3 8 3 buffer layers 4 2 9 . 4 3 1 . 4 3 5 doping 1{)4. 145. 146. 170 for dual-band devices 302 HIP detectors 3 5 3 . 3 7 1 - 3 7 4 . 4 1 1 - 4 4 5 nitride 2 30. 452. 457 oxide 4 5 7 . 4 6 2 photoconductors 2 3 . 3 6 - 3 8 readouts 2 9 - 3 0 . 9 5 . 108. 302. 394. 395. see also multiplexers: ROICs Si On Insulator (SOl) technologies 353. 3 7 4 379. 395 silicon substrates 2 8 . 3 5 . 4 7 . 4 9 , 50.69. 410. 413.453 in HgCdTe detectors 2 8 3 . 2 8 6 , 2 9 6 . 3 1 9 , 341-348 in QWIPs 125. 1 4 7 - 1 5 1 silicon, see also extrinsic Si detectors: Schottkybarrier detectors SiGe 453 SiGe buffer layers 4 0 2 - 4 0 3 . 404 SiGe detectors 3 9 3 - 4 4 5 SiGe films on Si substrates 3 7 0 - 3 7 1 . 3 7 2 . 3 7 3 SiGe/Si HIP detectors 353. 3 7 1 - 3 7 4 . 4 1 1 - 4 4 5 SiGe/Si MQW detectors 4 1 1 - 4 4 5 SiGe/Si superlattices 41 () SiN absorbers 464 SiN-based passivation 1 7 0 . 1 7 2 SIMOX (separation by implantation of oxygen) substrate 407 SIMS (secondary-ion mass spectrometry) 2 56, 287. 316. 323. 3 2 5 - 3 2 6 . 4 1 2 . 4 1 6 . 4 3 1 simultaneous unipolar multispectral integrated technology (SUMIT) 302 simultaneous-mode detector arrays 326. 329 slider growth apparatus 2 4 9 - 2 51. 2 8 4 . 2 8 5 . 290 slush method. MCT crystal growth 2 4 7 - 2 4 8 SnPb473 Sofradir 3 1 . 4 6 . 4 8 . 283 SOl (Si On Insulator) technologies 3 5 3 . 3 7 4 379. 395 sol-gel processing 64 solid state recrystallization (SSR) 247 space applications 5 . 6 . 3 3 . 3 7 , 1 2 1 , 2 0 8 , 2 2 5 . 233. 3 3 7 , 3 6 6 - 3 6 7 , 397

Index

spectrometers 9, 122.233, see also Fouriertransform IR (FTIR) spectrometer spectroscopic ellipsometry see SE spectroscopy, absorption-edge (ABES) 312. 313-315,325 spectroscopy applications 2 9 9 , 3 0 0 spin doping 482 spin filtering 4 8 1 - 5 0 5 spin-on metal-organic decomposition 63 spin-orbit split-off 301, 332 spintronics 4 8 1 - 5 0 5 sputtering 34, 63-64. 357. 368. 380 sticktion 4 6 2 - 4 6 3 stiffness 462, 4 6 3 - 4 6 7 , 4 73 stimulated emission of spin waves (SWASER t 481 strained layer systems 57, 1 3 0 . 1 6 0 . 1 6 9 - 1 7 ( ) . 2 3 3 , 3 7 0 - 3 7 1 . 395. 398-411 submonolayer OWIP structure 152 substrates 2 8 , 1 2 5 , 1 3 4 . 407 debiasing 297 for epitaxial growth 2 4 8 - 2 4 9 . 2 5 0 . 2 52. 285-286 removal 51.138 in spin filtering 4 9 7 - 4 9 8 . 504-5()5 thinning 1 3 . 1 8 1 , 2 8 8 see also CdTe: CdZnTe: GaAs: GaSb: InP: InSb: Si substrates SUMIT (simultaneous unipolar multispectral integrated technology) 302 superlattices AnAsSb/GaSb (type II) 70 GaAS/A1GaAs-based (BID) 114-11 5 . 1 1 6 117 InAs/GaInSb 26 InAs/(Galn)Sb 1 5 9 - 1 8 7 n-type OWlPs 144-14 7 p-type OWlPs 1 4 2 - 1 4 4 SiGe 398-411 type-I 193 type-II 70, 193-230 surface roughness, type-II superlattices 216 surveillance applications 58, 88, 2 33.337. 39 7 SWASER (stimulated emission of spin waves) 481 T TaN 459, 4 6 3 , 4 6 4 TCR (temperature coefficient ofresistancel 58. 6 0 , 4 5 0 , 4 5 1 . 455-456. 459. 476 Te alkyls 2 5 2 - 2 5 3 temperature coefficient of resistance see TCR temperature-sensing diodes 374- 379 temperatures, operating see operating temperatures TEOS 453. 457.458. 459 tetramethyl ammonium hydroxide {TMAHI 38 TFFE see ferroelectric (TFFE) detectors thallium TI) doping 370 thermal conductivity, polySiGe 455.461. 476 thermal contrast 24

517

thermal detectors 5 . 6 . 7 . 8 . 1 8 - 2 1 . 2 3 . 5 8 - 6 7 . 7O. 121 compared with photon detectors 9 . 2 1 - 2 3. 25-26. 121 compared with type-II superlattices 2()1. 2O7-2O8 see also bolometers: pyroelectric detectors: thermopile detectors thermal diffusion 2 8 0 - 2 8 1 . 2 8 7 thermal fluctuation noise 20 thermal imaging 9 . 2 3 . 5 8 - 6 7 , 2 3 3 . 2 6 9 - 2 7 0 . 271-274. 299. 300. 324. 393. 397. 411. 449 thermally-assisted tunneling 40.85 thermionic emission 4 0 . 4 7 . 8 5 . 8 6 . 107. 142. 356 thermistor bolometers see Si bolometers thermo-electric cooling 2 73-274.3{)() thermoelectric arrays 64-65 thermoelectric cooling 58. 305.411 thermopile detectors 6 4 - 6 5 . 3 5 3 . 3 8 3 - 3 8 4 . 411 thin-film ferroelectric {TFFEt detectors 6 2 - 6 4 threat-warning applications 52 3I) imaging 301 Ti metal bolometers 453 Ti microbolometers 450 Tishortcut layer 1 39. 140 Ti sublimation pumps 31 () Ti/Au/Pt/Au metallization 170 Ti/Pt/Au metallization 170 TIN459 tipping. I.PE crystal growth 2 5(). 2 51.284 TI {thalliuml doping 370 transistors 4 8 1 - 4 8 2 . 504 transportation applications 58.7(). 88.2 33. 3()5, 397 traveling heater method {THMt. MCT crystal growth 247 TV-compatible formats 29. 274. 364. 386 twin formation 318. 342 2-D arrays 69. 269-305. 4 6 7 , 4 7 7 two colour arrays 301-303, 3 2 3 . 3 2 5 - 3 2 6 . see also dual band FPAs type-II band alignment 3 . 8 . 2 6 . 1 6 ( ) - 1 6 3 . 191194 type-II photoconductors 201-211 type-II photodiodes 2 1 1 - 2 2 7 type-II superlattices 159-2 31 compared with extrinsic silicon detectors 208. 224-227 compared with HgCdTe detectors 2{)1.2{)7. 2()8-2{)9. 214.21 7 compared with thermal detectors 2{)1. 207 modeling 197-201 U [!-profile 4 6 5 - 4 6 7 UBM tunder-bump-metallurgyl layers 139 uncooled systems 6 . 9 . 2 5 - 2 6 . 6 5 - 6 6 . 7 0 . 2 0 1 214. 354-374

518

Handbook of hlfrared Detection Technologies

uniformity 2 34.2 76 extrinsic silicon detectors 224.358. 366 HgCdTe(MCT) 256,296. 329. 336. 342 polySiGe microbolometers 45(). 4 5 8 - 4 5 9 . 472 type-II superlattices 2()8.211 see also non-uniformity: responsivity V

vacuum packaging 4 5 0 , 4 6 1 , 4 7 2 - 4 7 6 valence band HgCdTe 281. 332 InAs/(GaIn)Sb superlattices 160.1 7() OWIPs (p-typeOWIPs) 123.124, 128.1 3(1. 146-147 SiGe detectors 394. 398.438-44() SiGefilm 373,412. 413 type-II superlattices 196.2()()-2() 1 vanadium dioxide 58 vanadium oxide 31.58, 59. 449-45() vapor phase epitaxy (VPE) 245. 249, 251-252. 2 8 4 , 2 9 9 , see also MOCVD vapor-HF processing 4 6 2 - 4 6 3 . 477 very large arrays 300 very long wavelength IR (VLWIR) 5-6. 31 - 32.

\'lAP processes and devices 287.289-29(), 296.297 video processing units 98.112 visible imaging 269.2 7() void defects. HgCdTe MBE technology 321-322. 342. 348 Vl~Eivaporphaseepitaxyt 2 4 5 , 2 4 9 . 2 5 1 - 2 5 2 , 284.299. see also M()CVI) ~T

W substrate, in spin filtering 49 7 wafer-bonding 4 73 Wannier-Stark oscillation 21 7-218 waveguide photodetectors 395.4()7-4()8.41 () Wright Patterson Air Force Base ID,'PAFBI 216, 22(). 222-22 3 X

X-ray difl'raction pattern lnAs tGalniSb superlattices 16 3-165 QWII~-on-Inl~ 148 QWIP-on-Si 148 SiGe detectors 412. 416 type-II superlattices 216 Xe arc lamp 258

121.393

HgCdTe 9 . 4 5 - 4 6 . 6 7 . 187.233 InAs/(Galn)Sb superlattices 162.187 OWIPs 6, 9 . 4 5 - 4 6 . 5 1 . 5 7 . 6 9 . 1 1 ( ) - 1 1 1 . 126-127,146-147 Si and Si/Ge detectors 208. 2 2 4 - 2 2 7 . 39 3 thermal detectors 2 3 type-II superlattices 2 1 5 - 2 2 7 see also LWIR/VLWIR dual band via-hole devices 289-29(). 299.3()2. see also loophole interconnection

Y

YBaCu()-based microbolometers 45 3 Z

Zenertunnelling 175. 192, 2 1 7 - 2 1 8 zero-level vacuum packaging 45(). 4 7 2 - 4 76 zincblende crystal structure, in MCT 2 35-2 36 Zn doping 142 ZnTelaverslilms 255. 319. 342. 344 zone melting crystal growth 247