Radiation Test Challenges for the New Millennium

2000 IEEE NSREC IEEE Nuclear and Space Radiation Effects Conference Short Course

Radiation Test Challenges for the New Millennium

July 24, 2000 Sponsored by: IEEE/NPSS Radiation Effects Committee Supported by: Defense Threat Reduction Agency Sandia National Laboratories Air Force Research Laboratory Jet Propulsion Laboratory NASA – Goddard Space Flight Center Approved for public release; distribution is unlimited.

2000 IEEE Nuclear and Space Radiation Effects Conference

Short Course

Radiation Test Challenges for the New Millennium

July 24, 2000 Reno, Nevada

Copyright© 2000 by the Institute of Electrical and Electronics Engineers, Inc. All rights reserved. Instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. For all other copying, reprint, or replication permission, write to Copyrights and Permissions Department, IEEE Publishing Services, 445 Hoes Lane, Piscataway, NJ 08555-1331.

Table of Contents SECTION I …………………………………………………………... I 1-8 Introduction Lewis M. Cohn Defense Threat Reduction Agency

SECTION II ……………………………………………………….. II 1-82 Performance Characterization of Digital Optical Data Transfer Systems for Use in the Space Radiation Environment Robert Reed NASA, Goddard Space Flight Center Ray Ladbury Orbital Science Corp.

SECTION III …………………………………….……………… III 1-73 Optoelectronic Devices with Complex Failure Modes Allan Johnston Jet Propulsion Laboratory

SECTION IV ………………………………..…………………… IV 1-47 Radiation Testing of Mixed Signal Microelectronics Jake Tausch and David Alexander Mission Research Corporation

SECTION V ………………………………………………………... V 1-65 Radiation Testing and Characterization of Programmable Logic Devices Lee Hoffmann and R.C. DiBari Honeywell International, Space Division Rich Katz NASA Goddard Space Flight Center Lew Cohn Defense Threat Reduction Agency

2000 IEEE NSREC SHORT COURSE

SECTION I

INTRODUCTION

Lewis M. Cohn Defense Threat Reduction Agency

Approved for public release; distribution is unlimited.

INTRODUCTION This document contains the text of the 2000 Nuclear and Space Radiation Effects Conference (NSREC) Short Course held on 24 July 2000 at Reno, Nevada. This is the twenty-first time that a course has been offered in conjunction with the NSREC and, as in the past, it is our hope that the material presented will be interesting, informative and serve as a archival reference for members of the radiation effects community. The text for the course will be available on CD-ROM at the 2003 NSREC. The theme of this year’s Short Course is “Radiation Test Challenges for the New Millennium” or stated another way – “What is involved in the development and implementation of technically and cost-effective radiation test and characterization methods for complex microelectronic and photonic circuits and subsystems?” This theme reflects the growing use of near or state-of-the-art microelectronics and photonics technologies in space and the problems associated with the characterization of their radiation response. The rapid and continuing evolution of microelectronics and photonics technology has made possible near-revolutionary advancements in both satellite and missile system performance capabilities. We are now entering an era where we can envision swarms of micro-satellites circling the earth or totally autonomous long-lived missions probing the galaxies all made possible through the use of modern electronics. The use of these advanced technologies will provide unprecedented capability to monitor the earth’s resources, predict weather patterns, perform peacekeeping missions, support instant communications between any areas of the planet and support other yet to be conceived missions to serve mankind. However, the benefits to be derived through the use of these advanced microelectronics and photonics comes with a number of challenges. One of the foremost of these challenges is the development of technically and cost effective radiation characterization and test methods to support the use of these technologies for long lived space missions and other applications where easy replacement is not an option. The impact of our ability to unambiguously determine the long term performance of a complex circuit or subsystem in a radiation environment will have a pervasive effect on system cost and performance, since this information will influence part selection,

I-1

determine the design margins to be invoked, the level and the complexity of the testing to be accomplished and set boundaries on overall system performance. Thus, it is imperative that we be able to identify and understand the issues and problems involved with the radiation testing of modern electronic and photonic circuits and subsystems. The topics covered in this course will address many of the issues related to the use of existing test methods and the development and implementation of test and characterization strategies for both complex electronic and photonic circuits and subsystems. The course is divided into five sections as follows: Section I “Introduction.” In this section a brief introduction concerning the motivation for the development of this short course is provided. Section II “Performance Characterization of Digital Optical Data Transfer Systems for Use in the Space Radiation Environment.” In this section Dr. Robert Reed will present a discussion of the issues involved concerning the use of optical link systems, the impact of radiation on the system components, the response of the subsystem to various radiation environments and the challenges associated with the testing and characterization of the radiation response of these subsystems. As we shall see one must take a holistic approach when dealing with such technology to ensure that a valid understanding of the radiation degraded performance is ascertained. Section III “Optoelectronic Devices with Complex Failure Modes.” In this section of the Short Course Mr. Allan Johnston will provide a discussion of the issues involved with the radiation test and characterization of a variety of photonic devices that are now used (or being contemplated for use) in a number of satellite systems. The device and technology types to be addressed will include optocouplers, light emitting diodes, optical silicon detectors, optical fiber systems and other novel device structures and technologies. An overview of the radiation environments, as pertaining to these technologies, optoelectronics fundamentals and testing issues will be covered.

I-2

Section IV “Radiation Effects Testing of Mixed Signal Microelectronics.” Mr. Jake Tausch, in this section, will investigate the issues and problems associated with the radiation testing of mixed-signal microelectronics with specific emphasize on analog-to-digital converter circuits. The discussion will include an overview of radiation effects as they pertain to mixed-signal technologies, general testing issues and testing techniques appropriate to this ubiquitous circuit type. Section V “Radiation Testing and Characterization of Programmable Logic Devices.” Mr. Lee Hoffmann will provide a discussion of the issues and problems associated with the radiation response characterization and testing of the various types of programmable logic devices including programmable arrays, simple and complex logic devices and Field Programmable Gate Arrays (both re- and non- reconfigurable). The discussion will address the pertinent radiation environments, failure modes, test strategies and future issues. I want to personally thank the four Short Course presenters, Dr. Robert Reed, Mr. Allan Johnston, Mr. Jake Tausch, and Mr. Lee Hoffman their efforts in preparing and presenting the course material. It is through the hard work, expertise and diligence of these individuals that the tradition of excellence of the NSREC Short Course has been maintained.

Lew Cohn Alexandria, Virginia

I-3

Biographies Lewis M. Cohn Short Course Organizer Defense Threat Reduction Agency Lewis Cohn received his B.S. and M.S. Degrees in Electrical Engineering in 1965 and 1970 from the Milwaukee School of Engineering and Syracuse University respectively. Since 1985, he has been a Program Manager at the Defense Threat Reduction Agency (DTRA) and has been involved in a wide variety of efforts to develop and demonstrate technology to radiation harden microelectronics and photonics and semiconductor materials. Prior to his assignment at DTRA, he served as a Nuclear Engineer in the Naval Sea Systems Command Naval Reactors Group. He has also worked as an electronics design engineer at both General Electric Heavy Military Electronics Division and Rockwell International. Mr. Cohn is a member of the IEEE, is a member of the HEART Steering Committee and has served in a variety of positions for the IEEE NSREC and HEART Conferences. Mr. Cohn has published and presented over ten papers concerning the effects of radiation on microelectronics and system survivability. Robert Reed NASA Goddard Space Flight Center Robert A. Reed received his B.S in Physics from East Tennessee State University in 1990 and his M.S. and Ph.D. in Physics from Clemson University in 1993 and 1994, respectively. After completion of his Ph.D., he worked as a post-doctoral fellow at the Naval Research Laboratory and later worked for Hughes Space and Communication. He is currently a research physicist at NASA Goddard Space Flight Center, where he supports NASA space flight and research programs. His radiation effects research activities include topics such as single event effect (SEE) basic mechanisms, SEE and displacement damage effects in photonic components and systems, and space system on-orbit performance analysis and prediction techniques. He has published over 30 papers on radiation effects on microelectronic and photonic devices and systems.

I-4

Allan Johnston Jet Propulsion Laboratory Allan H. Johnston received B.S. and M.S. degrees in physics in 1963 and 1983 from the University of Washington in Seattle. He joined Boeing Aerospace Company in 1965, where he worked on radiation effects in microelectronics and optoelectronics for military and space systems. In 1992 he joined the Jet Propulsion Laboratory as radiation effects Group Leader, where he directs research and testing activities related to NASA space projects. His research interests include total dose effects in bipolar and MOS devices, the effects of device scaling on single-event effects, radiation effects on optoelectronic devices, and radiation-induced latchup. Allan is the author or co-author of more than 60 refereed publications. He has served as Assistant Guest Editor, Local Arrangements Chair, Awards Chair, Short Course Instructor, Short Course Organizer, and Technical Program Chair for the IEEE NSREC. He has also served as member-atlarge, and is currently the secretary of the Radiation Effects Steering Group. Allan received the Distinguished Poster Paper award for the 1987 NSREC, and the Outstanding Paper Award for the 1999 NSREC.

Ray Ladbury Orbital Science Corporation Ray Ladbury received his B.S. in Physics from the Colorado State University in 1982 and his Ph.D. in experimental particle physics from the University of Colorado at Boulder in 1988. He joined the Radiation Effects Group at NASA Goddard Space Flight Center in January of this year. He comes to Goddard after working on radiation hardness assurance for several commercial and government satellite programs at Hughes Space and Communications Corp. He has also worked in science education and international development. As a science journalist, he has written on subjects ranging from planetary physics to metallic hydrogen. He is a member of IEEE.

I-5

H. Jake Tausch, Jr. Mission Research Corporation Mr. H. Jake Tausch, Jr. received his B.S. in Electrical Engineering from the United States Air Force Academy in 1969 and his M.S. in electrical Engineering from the University of New Mexico in 1973. From 1971 through 1974, Mr. Tausch worked in the transient radiation effects (TREE) branch of the Air Force Weapons Laboratory, performing experimental research on electrical components. He joined Tektronix Inc. in 1974 and from 1974 to 1977 designed electronic test equipment for characterization of integrated circuits. From 1977 through 1983, he worked for BDM where, among other things, he developed test systems for characterizing radiation effects on components and systems. From 1983 through 1991, he founded and was president of Design Engineering Inc., a small business that designed and manufactured specialized systems for radiation effects testing of electronic components. He joined Mission Research Corporation in 1991 and is currently the Design and Analysis Group Leader. His responsibilities include designing and fabricating specialized electronic test equipment for ionizing radiation effects testing. Mr. Tausch has been a member of IEEE since 1967 and has authored or co-authored 12 papers relating to radiation effects and associated testing. David Alexander Mission Research Corporation David R. Alexander received his B. S. in Electrical Engineering from the U. S. Air Force Academy in 1968 and his M. S. in Electrical Engineering from the University of New Mexico in 1973. From 1968 to 1973 he was an Air Force officer assigned to the Air Force Weapons Laboratory in Albuquerque, New Mexico. In 1973, he joined the BDM Corporation and was the principal investigator for several programs in radiation response modeling of microcircuits. In 1980, he became a member of the technical staff at Sandia National Laboratories. He is currently with Mission Research Corporation and has been Manager of their Microelectronics Division since 1993. At MRC, he has been responsible for applying computer-aided design and modeling practices to microcircuits. Mr. Alexander has been active in the NSREC for several years and has served in several positions. He has numerous technical publications and was a recipient of the Distinguished Poster Paper Award in 1988.

I-6

Lee F. Hoffmann Honeywell International Space Division Lee F. Hoffmann received his B.S. in Electrical Engineering from Louisiana State University in 1983. Since joining Honeywell, Inc., he has worked in the design and development of radiation-hardened ring laser gyroscopes and accelerometers, electromagnetic pulse test equipment, and more recently on radiation-hardened computers and interferometric fiber optic gyroscopes. Mr. Hoffmann has significant experience in both weapon and natural space radiation effects and design and test techniques. He led the development of the first, qualified, true no-upset computer for the Minute Man III program. More recently, Mr. Hoffmann has been the Systems Survivability Engineer on several satellite and missile development projects. He has authored or co-authored several papers at the NSREC and the biennial SEE Symposium. R.C. DiBari Honeywell International Space Division Becky DiBari graduated from Florida State University in 1995 with a BS in physics and received her MS in physics from Rensselaer Polytechnic Institute in 1997. She joined Honeywell in 1997 where her work has concentrated on analysis and testing for space and weapon caused radiation effects. Richard Katz NASA Goddard Space Flight Center Rich Katz received his B.S. degree in both Computer Science and Applied Mathematics from the State University of New York at Stony Brook in 1982. Additionally, he received the Master of Science degree, also from Stony Brook, in 1983. After graduation, he served as a Member of Technical Staff at the Jet Propulsion Laboratory as a design engineer on the attitude control and radar electronics for a variety of deep space missions. Additionally, Rich has designed electronics for a number of spacecraft and instrument systems, both Earth orbiters as well as planetary probes. He is currently an Electronics Engineer at the Goddard Space Flight Center as well as an independent consultant. He has published numerous technical publications, writes a quarterly column on programmable logic, and is Chairman of the annual MAPLD International Conference on programmable logic. His current interests are: design of high-reliability circuits; use of programmable elements and devices in space-flight applications; high-performance, compact microelectronics for spaceborne processing applications; and missile and space computers of the late '50s and early '60s. I-7

2000 IEEE NSREC SHORT COURSE

Section II

Performance Characterization of Digital Optical Data Transfer Systems for Use in the Space Radiation Environment Robert Reed/NASA-GSFC Ray Ladbury/Orbital Science Corp.

Approved for public release; distribution is unlimited

Performance Characterization of Digital Optical Data Transfer Systems for Use in the Space Radiation Environment

1.0 Introduction 2.0 Summary of the Space Radiation Environment and Basic Radiation Effects in Devices 2.1 Space Radiation Environment 2.1.1 Trapped Protons and Electrons 2.1.2 Trapped Heavier Ions 2.1.3 Transient Environment 2.2 Basic Radiation Effects in Devices 2.2.1 Single-Event Effects 2.2.2 Total Ionizing Dose Effects 2.2.3 Displacement Damage Dose Effects 2.3 Case Study: Environment Predictions for a Near-Earth Spacecraft 2.3.1 Single-Event Effects Environment 2.3.2 Total Ionizing Dose Environment 2.3.3 Displacement Damage Dose Environment 3.0 Intra-Satellite Digital Optical Data Links 3.1 Overview of the Digital Optical Data Link 3.1.1 Network Architectures 3.1.2 Point-to-Point Digital Optical Data Links 3.2 Brief Description of Digital Optical Data Link Components 3.2.1 Optical Digital Transmitters 3.2.1.1 Semiconductor Light Emission and the Resulting Wavelength 3.2.2 Passive Optical Components 3.2.3 Receivers 3.2.3.1 Stages of a Receiver 3.2.3.2 Digital Data Coding 3.2.3.3 Receiver Noise and Decision Circuitry 3.2.3.4 Photodiodes 3.2.4 Support Electronics 3.3 Digital Optical Data Link Performance Metrics 3.3.1 Eye Diagram Measurements and Analysis 3.3.2 Bit-Error Ratio 3.3.3 Optical Power Budgets 4.0 Radiation Effects in Digital Optical Data Link Components 4.1 Permanent Degradation of Sources 4.2 Total Ionizing Dose Degradation of Transmission Media 4.3 Permanent Damage in Optical Detectors 4.4 Single-Event Transients in Photodetectors 4.5 Radiation Effects in Support Circuitry

II-1

5.0 Optical System Response to Space Radiation Environments 5.1 Optical Data Link Single-Event Effects Ground Testing 5.1.1 Particle-Induced Bit Errors 5.1.2 Particle-Induced Bit-Error Ratio Testing 5.1.2.1 Proton-Induced SETs in Photodiodes that Cause Bit Errors 5.1.2.2 General Discussion on Error Cross-Section Dependence on Data Rate 5.1.3 Optical Data Link System Level Radiation Effects Testing 5.1.3.1 Ground testing of a Digital Optical Data Bus 5.1.3.2 Other Ground Test Results 5.2 Assessing Radiation Effects on Optical Digital Data Link Performance Metrics 5.2.1 Total Ionizing Dose and Displacement Damage Impacts on Power Budget 5.2.2 Impacts of Radiation-Induced Bit Errors on On-Orbit Bit-Error Ratio 5.2.2.1 Proton-Induced Transients in Photodiodes and On-Orbit Bit-Error Ratio 5.2.2.2 Optical Link On- Orbit Bit-Error Ratio and Message Error Considerations 5.2.3 Bit Error Mitigation Approaches 5.3 Ground Testing Results and On-Orbit Use of Selected Digital Optical Data Links 6.0 Radiation Effects in Some Emerging Optical and Optoelectronic Technology 7.0 Summary and Conclusions 8.0 Acknowledgments 9.0 References 10.0 List of Acronyms

II-2

1.0 Introduction Radiation effects in photonic and microelectronic components can adversely affect the performance of high-speed digital optical data links in a variety of ways. This segment of the Short Course focuses on radiation effects in digital optical data links operating in the MHz to GHz regime. (Some of the information is applicable to frequencies above and below this regime.) The three basic component-level effects that should be considered are effects due to total ionizing dose and displacement damage dose, and single-event effects. In some cases the system performance degradation can be quantified by means of component-level tests, while in others a more holistic characterization approach must be taken. In Section 2 of this segment of the Short Course we will give a brief overview of the space radiation environment, followed by a summary of the above listed space radiation effects important for microelectronics and photonics. The last part of this section will discuss an example of radiation environment requirements for a typical mission. Section 3 gives an overview of intra-satellite digital optical data link systems, including various link topologies and their associated components. Also, we discuss some of the important system performance metrics that are impacted by radiation-induced degradation of optical and optoelectronic component performance. Section 4 discusses radiation effects in optical and optoelectronic components, focusing on degradation of passive optical components and single-event effects in photodetectors. (The other mechanisms are covered in detail in segment II of this Short Course, entitled “Photonic Devices with Complex and Multiple Failure Modes”.) Section 5 will focus on optical-data-link system response to the space radiation environment. Optical link system level single-event effects ground testing will be discussed. To close, we give a discussion of optical system level assessment of data link performance for optical systems operating in the space radiation environment. In Section 6, we end this segment with a brief discussion of the prospects for using emerging technologies in future high-speed, space-based optical data links.

II-3

2.0 Summary of the Space Radiation Environment and Basic Radiation Effects in Devices Earth-orbiting and interplanetary spacecraft face a variety of radiation-related threats. Determining the survival probability of a spacecraft during its mission requires not just accurate ground-based test data for the device and a validated model for predicting the device performance in space from the ground-based data. It also requires accurate prediction of the space radiation environment. Although a complete description of the environment is beyond the scope of this course, its importance demands that we give at least a cursory treatment here. An excellent description of the space radiation environment as well as the use, validity, and limitations of the relevant models thereof, can be found in the Nuclear and Space Radiation Effects Conference 97 Short Course [Bart-97]. The radiation environment encountered by a spacecraft depends on several factors, including the path of the spacecraft relative to the planets, the level of solar activity, and the mission duration, determine the radiation levels incident on the spacecraft. For some radiation effects, the spacecraft’s ability to shield sensitive components from radiation can be crucial in determining whether radiation effects will degrade the performance of those components. Finally, the threat of man-made radiation environments (not addressed in this course) can be an important consideration. Typically, these variables are used as inputs to computer codes that predict the space radiation environment encountered by a spacecraft and how this environment affects the spacecraft’s mission. There are two major components of the natural space radiation environment: the transient environment and that trapped by the magnetic fields of most planets. As might be expected, Earth’s trapped radiation environment is better characterized than that of other planets. Our brief discussion of the radiation environment will focus on the naturally occurring radiation environment as it affects the performance of microelectronic and photonic devices in Earthorbiting spacecraft. Deep-space missions passing near other planets, radiation dosimetry [Dyer98] and man-made radiation environments are described elsewhere [Teag-72]. Figure 2.1 is an artist’s conception of Earth’s radiation environment. The near-Earth trapped particle environment will be discussed first, followed by a discussion of the solar and galactic radiation environments. Next, we will give a brief discussion of the basic concepts of radiation effects in photonic and microelectronic devices. The last part of this section will discuss the typical environmental data needed to evaluate and predict the performance of photonic and microelectronic devices exposed to the radiation environment. 2.1 Space Radiation Environment The objective of this section is to give a brief summary of the radiation environment encountered by near-Earth spacecraft. [Bart-97] gives a more detailed overview of the basic physics and theories that describe solar processes, Earth’s magnetic field, charged particle interactions with the magnetic field and many of the other details not covered by this brief summary. Our discussion is divided into two parts: trapped and transient radiation environments. 2.1.1 Trapped Protons and Electrons Particles with the proper charges, masses, energies and trajectories can be captured by Earth’s magnetic field. See [Bart-97, Dyer-98] for details of the species, origin, confinement processes, subsequent motion, and measurement of these particles. Of the particles confined by II-4

Galactic Cosmic Rays

Solar Protons & Heavier Ions Trapped Particles Protons, Electrons, Heavy Ions

Figure 2.1: Cartoon showing the components of the space radiation environment important for microelectronic and photonic performance degradation evaluations. [Endo] Earth’s magnetic field, electrons and protons have the greatest effect on spaceflight hardware. Figure 2.2 shows cross sections through dipole plots of these two particle populations as predicted with the AP-8 and AE-8 trapped particle models. The proton environment is given on the left side of the cartoon, and the electron environment is shown on the right. In simplified approximations of these environments, they form a toroid with Earth at the center and Earth’s magnetic pole defining the toroid’s central axis. (The geographic pole is roughly 11 degrees off center from the magnetic pole.) L is the dipole shell number of the Earth’s magnetic field. At the magnetic equator L is the distance in Earth radii from the center of the earth. As the angle of inclination moves away from the magnetic equator, the value of L is corrected for the magnetic field strength. The values of L in Figure 2.2 refer to the magnetic equator. As shown in the figure, the proton flux is confined to a single toriod, and electrons form two high intensity toriods. The region between the two electron zones is known as the slot region. Earth’s atmosphere and magnetic field and their interaction with the solar wind and the solar magnetic field define the details of the flux of each particle toriod. The particles roughly follow Earth’s magnetic field lines. In the inner region, L < 3.5, electron and proton toriods overlap, while for 3.5>L>8.5, the trapped particles are mostly electrons. No significant particle trapping occurs for values of L>11. In the trapped regions the flux is considered to be approximately omnidirectional. Although Figure 2.2 presents a static view of the shape, regional flux, and orientation of the toriods, in reality, these belts are very dynamic, growing and shrinking over time. Occasionally, new torroidal regions form and disappear, especially in the slot region. The dynamic nature of the trapped radiation belts is not very well understood and is poorly modeled. Research has shown that fluxes can change dramatically with solar activity, but quantitative models of this variability do not yet exist for short term averages. One temporal variation that has been quantified is the variation of the flux levels in the toriods with the 11 year solar cycle. (During solar maximum the integral fluences for protons are lower for low Earth orbit than during solar minimum, while for electrons the reverse is true.) It is the short duration temporal variations that are the most difficult to quantify. Progress in dynamic modeling techniques has been made by [Bosc-99] with the salambo code. II-5

AP-8 Model Ep > 10 MeV 10 2 10

4

AE-8 Model Ee > 1 MeV

10 6

103 10 2 6

10

0 3x1

# / cm2 / s

6

6

10

5

10

4

10

# / cm2 / s

4 3 2 1 1 Proton Fluxes (cm-2.s-1) E>10 MeV

2

3 4 5 6 7 8 9 Electron Fluxes (cm-2.s-1) E>1 MeV

10

L-Shell

Figure 2.2: Dipole projection of the Earth's Van Allen radiation belts as predicted by the AP-8 and AE-8 trapped particle models. A cross section of the protons are depicted on the left and the electrons are shown on the right. Another modification to the simple toriodal model results from the fact that Earth’s magnetic field is multipolar in nature, causing the magnetic field strength contours to sink towards the Earth in some regions. This multipolar field causes the South Atlantic Anomaly (SAA)—a dip towards the Earth in proton and inner electron flux contours over the South Atlantic. For equal altitudes, the particle flux will be higher for locations in the SAA than for those outside of it. Further discussion of models for the trapped particle environments are given in [Bart-97, Hous-98]. For example, the NASA’s AP8 and AE8, Huston-Pfitzer, CRRESPRO and CRRESELE models are used to model the trapped proton and electron environments [see Bart97 and references therein]. 2.1.2 Trapped Heavier Ions Ions with Z > 1 can also be trapped by Earth’s magnetic field, although the intensities for these ions are lower than those for protons and electrons. The trapped heavy ions have energies on the order of 10s of MeV/amu, so most of them will not penetrate even the thinnest spacecraft shield ing [Bart-97 and references therein]. Effects of these particles on microelectronic and photonic systems are second order in most cases. 2.1.3 Transient Environment Although many types of radiation make up the transient environment, the two most important components for radiation effects in spacecraft are the Galactic Cosmic Rays (GCRs) and particles emitted during solar events. The sources of GCRs are sufficiently far from our solar system that the fluxes of these particles are essentially isotropic in free space regions. Interactions in the vicinity of Earth between the solar wind and our planet’s magnetic field change individual particle trajectories and energies. However, the net GCR flux is still essentially omnidirectional.

II-6

Figure 2.3: Correlation of the occurrence of several large solar proton events over a thirty year period (vertical light lines) with the Zurich sun spot number (single solid dash line). [Bart-97] The GCR particle composition is roughly 83% protons, 13% alpha particles, 3% electrons, and 1% heavier ions (Z > 2). The ion energies range from 10s of MeV/amu to 100s of GeV/amu and beyond, and most ions are fully ionized. Some of these ions have sufficient energy to penetrate most shielding provided by a spacecraft structure. In [Bart-97], there is a more complete description of the GCR environment, including relative abundances, energy spectra and variation over solar cycle. Not all of radiation impinging on Earth originates from such distant sources. The Sun can be thought of as a boiling pot of plasma that emits charged particles most of the time. When the Sun is “quiescent,” most of the particles it emits do not have sufficient energy to penetrate through even a small amount of spacecraft shielding. However, during times of high activity, conditions occur that can accelerate a spectrum of charged particles with a large range of energies for varying durations. The duration of such events is usually between a few hours and several days. The average frequency of these solar events varies roughly sinusoidally with the eleven-year sunspot cycle. Figure 2.3 illustrates this variation over the last three solar cycles— showing solar proton integral fluences (the spikes) for large solar proton events over a thirty-year period superimposed over the sunspot numbers (smooth curve). Two important classes of events that occur during this high-activity period are Coronal Mass Ejections (CMEs) and solar flares. CMEs have been correlated with events that have a high probability of producing protons that reach the Earth. On the other hand, solar flares seem to be correlated to heavy ion rich solar particle events [Ream-95]. Solar flares and CMEs can occur at the same time, giving rise to events with very high-intensity events which contain all naturally occurring elements (Z=1-92) [Tylk-96]. The total integral fluences during these rare events can exceed average GCR fluxes by three orders of magnitude and more. Again, the reference of choice that describes the solar event environment in detail is [Bart-97 and references therein]. Newer models for predicting the solar proton event environment can be found in [Xaps-98, Xaps-00].

II-7

2.2 Basic Radiation Effects in Devices Microelectronic and photonic components are manufactured in very controlled environments. During certain phases of the manufacturing process, even the slightest change in conditions like temperature or impurity concentration can induce changes on the overall molecular structure that cause the component to fail functional or parametric performance metrics. The point here is that devices that rely on such carefully grown, well defined microscopic structures can have very low tolerance for slight changes in their characteristics. When a component is exposed to radiation, the radiation transfers some of its energy to the component materials, changing the localized material properties. This can have significant effects on component functionality and/or parametrics with the end result depending on the type of radiation, where the energy deposition occurred, and the type of component. Three important effects that occur when a component is exposed to radiation are: SingleEvent Effects (SEEs) and effects due to Total Ionizing Dose (TID) and Displacement Damage Dose (DDD). This section will briefly define these effects and describe the basic mechanisms that cause them. It does not attempt to completely describe the impact that these basic interactions have on component performance. The reader will find detailed descriptions and discussions of the effects in other portions of this Short Course, past Short Courses, and in the many years of work published in the IEEE Transactions on Nuclear Science (TNS) and Proceedings from the Radiation Effects in Components and Systems Conference (RADECS). Because much of this segment of the Short Course is dedicated to single-event transient effects, these will be treated and described in the most detail. 2.2.1 Single-Event Effects An ionizing particle generates electron-hole pairs along its path as it passes through a material, resulting initially in a line charge distribution with equal numbers of holes and electrons. Because the function of many active electronic and optoelectronic devices is governed by the controlled injection of charge into the depletion layers of p-n junctions, the uncontrolled charge injection resulting from ionization can compromise device function. Exactly what effects result from a single-particle event (or single-event), depend on the device type and on how much charge is collected at sensitive junctions in the device [Dodd-99 and references therein]. Once the charge track is generated, it evolves rapidly (on a timescale of picoseconds) in response to the same fields and mechanisms that govern the behavior of any other charges in the medium. On very short timescales, the density of both positive holes and negative electrons along the particle track is quite high, and the dominant process at work is the recombination of these charges. Recombination decreases the charge collected within the device. The initially high charge densities also lead to diffusion of charges away from the ion track. Diffusion increases the average distance between charges, decreasing the recombination rate (and so tending to increase charge collection) over time. Charges will also be swept up by the intrinsic and applied fields in the device—a process called drift, which also tends to separate electrons and holes and increase charge collection. Drift in the vicinity of p-n junctions is especially important because of the strong intrinsic and applied fields in these regions. Although the processes governing the evolution of the charge track are the same as those governing the behavior of other charges in the medium, the charge densities along the charge track can be high enough to alter the properties of the medium in the track’s vicinity. One important manifestation of this is funneling [Hsie-81], which occurs when the charge density along the track is sufficiently high to collapse the usual p-n junction field. (See Figure 2.4.) II-8

Ion Trajectory

Depletion Region

Funnel Drift

Diffusion

Figure 2.4: Ion strike crossing a p-n junction showing the drift (in the depletion region and funnel) and diffusion components of the charge collected. When this occurs, charge can be collected from depths along the charge track significantly deeper than would otherwise be possible. Funneling can be an important contributor to charge collection for some devices. If sufficient charge is collected at a sensitive junction, a variety of single-event effects can occur, with consequences ranging from trivial to catastrophic. Single-event effects that can result in potentially catastrophic failures include Single-Event Latchup (SEL) in complimentary metal-oxide-semiconductor (CMOS), Single-Event Gate Rupture (SEGR) in power MOSFETs, Single-Event Burnout (SEB) in FETs and bipolar transistors, single-particle-induced failures of linear bipolar devices, and so on. If the collected charge results in a stable logic change in the device (that is, a bit flip), the event is called a Single-Event Upset (SEU). Alternatively, if the device changes its output state temporarily, the event is called a Single-Event Transient (SET). Generally, the variable of choice used to characterize SEE is Linear Energy Transfer (LET), which can be looked upon as the energy loss due to direct ionization of orbital electrons as a particle passes through a medium (or an energy deposited per unit path length—dE/dx) normalized to the medium’s density(ρ)—1/ρ*dE/dx (expressed, for example, in MeV*cm2 /mg) [Zieg-84]. Equivalently, LET can also be expressed as the charge generated per unit length of track (for example, pC/µm). In general, the higher a particle’s LET—or, equivalently, the denser its charge track—the higher the probability that it will cause a given SEE in a susceptible device. As in nuclear physics, the probability of a given effect is expressed as a cross-section, with dimensions appropriate to the device (µm2 or cm2 ). In general, for ions with sufficiently high LET, a device’s cross-section for a particular effect is related to the physical area on the device that is susceptible to the effect. However, the relationship is usually not a simple one. In general, if a particle’s LET is sufficiently low, it will not generate enough charge to cause SEEs and can be ignored in SEE prediction calculations. However, direct ionization—the creation of electron-hole pairs by ionizing radiation—is not the only means by which radiation can produce charge. It is also possible—although orders of magnitude less likely—that a particle can first interact with a nucleus in the semiconductor lattice. The charged products of this

II-9

BEFORE +V

DURING IRRADIATION +V

DURING AND AFTER IRRADIATION +V

IONIZATION AND RECOMBINATION

DRIFT, DIFFUSION AND TRAPPING

gate SiO2 Si

Figure 2.5: [Lera-99]

Simple model of generation and trapping of charge in an MOS device oxide.

interaction can then generate electron-hole pairs in their own right. Because nuclear cross sections are so low, such indirect ionization is insignificant for most particles. Nevertheless, because protons are highly penetrating and several orders of magnitude more numerous than other ions, it is often necessary to take indirect ionization due to protons into account in SEE calculations. Proton LETs are generally so low that proton direct ionization is negligible in SEE in microelectronics. For this reason, and because nuclear cross sections depend on proton energy rather than LET, the formalism for treating proton-induced SEEs has developed along different lines than has that for heavy-ion induced SEEs. This historical artifact has important consequences for the treatment of SEEs in photodetectors, where proton LET is not negligible. (Note: Neutrons are not treated here. However, they can also be important contributors to SEEs, particularly in aircraft avionics or for man-made radiation environments.) Because optical-based subsystems may contain any traditional electronics technology, it should be remembered that they could be susceptible to any of the above mentioned single-event effects. In addition, the optoelectronic components in these devices—particularly the detectors—may be susceptible to similar effects, especially SETs, which we will consider later. 2.2.2 Total Ionizing Dose Effects As with single-event effects, damage from TID is caused by the electron-hole pairs generated by ionizing radiation passing through a material. These charges can gradually change the performance of an electronic component, with the level of change depending on the total ionizing energy absorbed—that is, on the TID [Dres-98, Lera-99, and references given in these]. Generally, TID changes the characteristics of the materials that make up a component, resulting in gradual parametric degradation and changes in functionality. In most cases, the basic cause of TID degradation is the trapping of charge in the medium. Figure 2.5 illustrates a simple model of generation and trapping of charge in the dielectric of a MOS device with a positive bias applied to the gate. If enough charge is trapped in the oxide, the device’s performance will change. Charge may also be trapped in the field oxide of a linear bipolar device and MOS devices. A third example would be charge trapping in an optical fiber which can result in darkening. This effect is much different than charge trapping in microelectronic device oxides where there are electric fields that separate charge—this topic will be discussed in more detail later. II-10

2.2.3 Displacement Damage Dose Effects The proper functioning of many devices depends critically on the semiconductor having a pristine crystalline lattice. However, this lattice can be damaged when an energetic particle, such as a neutron, electron, proton or heavy ion displaces one or more nuclei within the crystalline lattice, creating electrically active defects. As this damage to the crystalline lattice—called the displacement damage—increases, the device can degrade parametrically, and eventually stop functioning all together. In microelectronics, DDD effects result from damage to the bulk semiconductor material, while TID effects are due to the charge trapping in the dielectric materials used in components. For the most part, because protons are penetrating, strongly interacting and abundant, the proton environment is predominant in considerations of DDD effects in shielded applications. However, for lightly shielded applications such as solar cells, low-energy electrons may also need to be considered [Dale-93a]. Moreover, for heavily shielded applications, secondary products can also be an important component of the environment. The 1999 NSREC Short Course [Mars-99] and segment II of this Short Course explores DDD effects. 2.3 Case Study: Environment Predictions for a Near-Earth Spacecraft The primary science objective of the Guided Laser Altimeter System (GLAS) instrument is to obtain day and night, long-term ice-sheet topography measurements with sufficient spatial and temporal resolution to detect regional elevation changes. The GLAS instrument will be placed in a polar orbit over the Earth at an altitude of 600 kilometers. The altimetry subsystem of the instrument uses a high-powered laser and sophisticated optical receiver to detect the outgoing laser pulse and correlate it to return signal reflected from the Earth. The radiation environment encountered by the microelectronics and photonics on GLAS is a combination of the primary space environment and a secondary environment. The secondary environment is produced when the primary environment interacts with spacecraft materials. For the most part, the primary environment dominates. Although this is true for most spaceflight missions, the use of heavy or composite shielding requires careful consideration of the secondary component of environment inside a spacecraft. An accurate description of the space environment for a specific mission can be crucial when developing an appropriate survivability test plan for microelectronic and photonic components. This section of the Short Course describes the radiation environment for a spacecraft in a polar orbit at 600 km above Earth’s surface. The description focuses on the environments that are important for most microelectronic and photonic devices. The minimum shielding considered is 50 mils of aluminum. By combining these environment predictions with accurate ground testing results, one can estimate the effects of space radiation on microelectronic and photonic components. This requires that validated models exist that allow one to correlate the measurements made using ground-based radiation sources to effects that will be observed in the complex space-radiation environment. 2.3.1 SEE Environment Single-event effects can occur via either direct or indirect ionization. Direct ionization is produced when a primary ionizing particle interacts directly with orbital electrons to free charges. The electrons and holes separate in the presence of electric fields and are collected at

II-11

various semiconductor structures. The direct-ionization environment is most often characterized by the integral LET flux distributions. Indirect ionization is a two-step process. First, a primary particle interacts with a nucleus in the semiconductor lattice, producing one or more recoiling ions. These recoiling ions then produce ionization in the medium. Because protons are by far the most numerous particles capable of producing nuclear interactions, the indirect ionization is described by the integral flux curves for protons as a function of energy. The integral LET flux curves reflect the combined direct ionization effects for all particles (Z = 1 to 92) from all sources. The major sources are the GCR background environment and the environment resulting from solar particle events. Figure 2.6 depicts eight curves showing the heavy ion (Z=1-92) LET spectra for these two sources (as computed by CREME96 [Tylk-96]), assuming different aluminum shielding thicknesses. The bottom four curves are the GCR background environment. Notice that shielding has little effect after the first few mils of equivalent aluminum. The four top curves are for the worst case 5 minutes as modeled by the extremely severe October 1989 solar particle event. Notice that these particle fluxes are strongly dependent on shielding thickness. Because such large solar events are quite rare, these curves should only be used to determine the peak SEE rates. Figure 2.7 shows the LET spectra for the trapped proton environment for four shielding thicknesses. Direct ionization effects from protons are rarely of concern for SEEs. However, as we will discuss later, for certain devices, direct ionization from protons can contribute significantly to the probability of SETs. Figures 2.8 and 2.9 give the integral flux as a function of energy for the trapped protons and protons from solar particle events for four shielding thicknesses. 2.3.2 TID Environment The total ionizing dose environment is a combination of several components of the naturally occurring space environment. Figure 2.10 shows each component and the total for silicon. The total ionizing dose is simply a summation of all the components, with trapped electrons, bremstrahlung, trapped protons, and solar particle event protons being the main contributors. 2.3.3 DDD Environment Segment II of this Short Course gives a description of the important issues to be considered when determining the DDD environment. We mention this effect here only for completeness. The major contributors to DDD when considering most applications of microelectronic and photonic components within satellite enclosures are the trapped and solar protons. Figures 2.11 and 2.12 give the differential flux as a function of energy for the trapped protons and protons from solar events for four shielding thicknesses. The details of how to compute the DDD from these data are given in segment II of this Short Course and a Short Course given at the 99 IEEE NSREC [Mars-99].

II-12

Integral Flux (particles/m2 -s-sr)

10+10 10+08 10+06 10+04 10+02 10+00 10 -02 10 -04 10 -06 10 -08 10 -10 10 -12 -03 10

WC 5 min 0 mils WC 5 min 1 mils WC 5 min 50 mils WC 5 min 500 mils GCR 0 mils GCR 1 mil GCR 50 mils GCR 500 mils

10-02

10-01 10+00 10+01 2 LET (MeV-cm /mg)

10+02

10 +03

Figure 2.6: Integral LET heavy-ion flux spectra for Z=1 to 92 at for a 600 km 99 degrees orbit for various shielding thicknesess were computed using the CREME96 codes. Values are given for galactic cosmic ray background levels and for levels modeled from the October 1989 solar particle event (5-min peaks). [Tylk-96]

Figure 2.7: Integral LET flux for trapped protons in a 600 km 99 degrees orbit for various Al shielding thicknesses. [Bart-98]

II-13

10 +12 Surface Incident

Proton Fluence (#/cm2/3 years)

50 mils 100 mils 200 mils

10 +11

500 mils

10 +10

10 +09

10 +08

10 +07 0.01

10 1 Energy (>MeV)

0.1

1000

100

Figure 2.8: Orbit-integrated integral trapped proton fluence for three years in solar maximum for an orbit at altitude 600 km and an inclination of 99 degrees orbit for various Al shielding thicknesses. (AP-8 model with NOAA-PRO correction). [Bart-98] Solar Proton Fluence (#/cm 2/3 yrs)

1011 Surface Incident 50 mils 100 mils

1010

200 mils 500 mils

1009

1008

1007

1006 0.1

1

10

100

1000

Energy (>MeV)

Figure 2.9: Orbit-integrated integral solar proton fluence for three years (95% confidence level) in solar maximum for an orbit at altitude 600 km and an inclination of 99 degrees orbit for various Al shielding thicknesses. [Bart-98]

II-14

10+07 Total - 3 Years Electron Trapped Proton - NOAAPRO Bremsstrahlung

Dose (rads-si/3 yrs)

10+06 10+05 10+04 10+03 10+02 10+01 Values Do Not Include Design Margins

10+00 1

10

100

1000

Shield Thickness (mils)

Figure 2.10: Total ionizing dose and contributions for important components thereof are shown as a function of equivalent solid-sphere Al shielding thickness. Orbit altitude of 600 km at an inclination of 99 degrees. [Bart-98]

Trapped Proton Fluence (#/cm2 /3yrs)

4.0x10+07 50 mils 100 mils 200 mils 500 mils

3.5x10+07 3.0x10+07 2.5x10+07 2.0x10+07 1.5x10+07 1.0x10+07 5.1x10+06 1.0x10+05 0.01

0.1

1 10 Energy (>MeV)

100

1000

Figure 2.11: Orbit-integrated differential trapped proton fluence as a function of energy for three years in solar maximum for a 600 km 99 degrees orbit for various Al shielding thicknesses. [Bart-98]

II-15

Solar Proton Fluence (#/cm2/MeV/3 yrs)

3.51x10+08 50 mils 100 mils 200 mils 500 mils

3.01x10+08 2.51x10+08 2.01x10+08 1.51x10+08 1.01x10+08 5.10x10+07 1.00x10+06 0.1

1

10

100

1000

Energy (MeV)

Figure 2.12: Orbit-integrated differential solar proton fluence for three years in solar maximum for a 600 km 99 degrees orbit for various Al shielding thicknesses. [Bart-98]

II-16

3.0 Intra-Satellite Digital Optical Data Links In the search for high-bandwidth digital data transfer systems, spacecraft designers are considering high-speed digital optical links as an alternative to conventional systems. Designers must determine the best link for the mission, based on performance metrics of each data link. This Short Course is intended to provide the designer with a clearer understanding of the radiation effects common in optical data links and how those effects affect link performance. This section introduces digital optical links by giving a general description of optical data transfer systems, with a brief overview of each component of the optical portion of a fiber-optic data link. We also discuss the optical system performance metrics that are important when characterizing performance degradation due to radiation exposure. Most of the material in this section represents a summary of the information gathered from these books [Agra-97, Pala-98, Lach-98]. 3.1 Overview of the Digital Optical Data Link Data links transmit large amounts of information at high data rates between spacecraft subsystems. Classically, this is done via an electrical connection using standardized data link hardware, i.e., RS232 and MIL-STD 1553. Data transfer at rates in the MHz to GHz range can be achieved using optical data links, where optical fiber and its associated components take the place of wire and electrical components. Some of the advantages and disadvantages of optical links are listed Table 3.1 [Bris-93, LaBe-98d]. A detailed comparison of optical versus electrical data links is beyond the scope of this Short Course. However, it is obvious from the list of advantages that, for many applications, optical links should be considered seriously. Table 3.1: List of some of the advantages and disadvantages of digital fiber link systems. Advantages Wide bandwidth Light weight Reduced complexity during spacecraft integration Immunity to electromagnetic interference Elimination of cross talk

Disadvantages Potential cracking of fiber Limited fiber bending radius Additional engineer expertise required to implement optical links Fiber connectors are large

Figure 3.1 shows a typical optical data link. Digital data flows from subsystem to subsystem via optical fiber. The support electronics condition the signal. Optoelectronics convert the signal between optical and electrical formats. In the next few sections, different network architectures will be discussed, followed by a discussion of the optoelectronics portion of the network (a detailed discussion of each component will be given in section 3.2). 3.1.1 Network Architectures The network architecture is the baseline for how each subsystem is connected to all the other subsystems on the network. Four popular topologies are illustrated in Figure 3.2, namely, 1) point-to-point, 2) linear-bus, 3) star-hub, and 4) ring. It is the bus protocol that determines which subsystem has control of the bus. The robustness of the bus to subsystem failure is defined by this protocol.

II-17

Spacecraft Subsystem

Support Electronics

Optoelectronic Interface

Figure 3.1: Block diagram of fiber-optic data transfer interface between spacecraft subsystems. Optical fiber is the physical medium used to connect each subsystem. Support electronics and an optoelectronic interface handle signal processing and translation of the signal between optical and electronic formats In the point-to-point network (Figure 3.2.A), two subsystems are linked via a single optical fiber. Each subsystem can communicate only with its partner and by means of a single optical fiber. To avoid the fiber being a single point failure, the system must use redundant data paths [DeRu-93]. In the linear bus network (Figure 3.2.B), the bus forms a backbone structure, with subsystems transmitting and receiving data as required. This network is very robust, in that the functionality of the network does not depend on each subsystem remaining functional. Again, unless the fiber backbone is made to be redundant it could be a potential single point failure for the entire network. The star hub network’s (Figure 3.2.C) transmission lines are brought together on the common hub. The receive lines are distributed out from the common hub. One disadvantage of the star hub is that the hub forms a single point failure. However this can be avoided, for example with a redundant second hub in the data path. Like the linear bus, the network functionality is independent of its subsystems’ performance. For discussions on star-hub and its bus protocol for spaceflight application see for example [Frit-94, Bone-96]. Finally, the ring network (Figure 3.2.D) subsystems form a continuous loop. When a message is transmitted from a subsystem it must be passed by all subsystems in the ring between the sending subsystem and the receiving subsystem. This simple implementation of the ring network is not as robust as the star hub or the linear bus, in that if one of its subsystems fails, the entire network will stop functioning. However, [DeRu-93, Bris-93] discuss the use of crossstrapping and bypassing approaches in a spaceflight hardware application that allows it to remain functional even if a subsystem fails. The optical power that is required to support the data link is one of the major factors in determining which network topology to use in a spacecraft design. For example comparing the

II-18

Linear

Point to Point

… (A) (B) Star Hub Ring

… … (C)

(D)

Figure 3.2: Four common network topologies. last three networks in Figure 3.2, the ring and the star hub have inherently lower optical power loss, because the signal passes from subsystem to subsystem [Agra-97]. In the case of the star hub, the power at the receiver decreases linearly with the number of subsystems on the hub. In contrast, the power reaching a specific subsystem in a linear bus configuration decreases exponentially with the number of subsystem between the transmitting subsystem and the receiving subsystem. As we will see later, there is a relationship between radiation exposure degradation of network optical power and data link performance. 3.1.2 Point-to-Point Digital Optical Data Link Figure 3.3 is a cartoon of two subsystems connected via an optical data link. The digital data is passed from the subsystem A to the transmitter, through an optical medium, then to the receiver, and finally to the electronics of subsystem B. The transmitter’s function is to convert the electrical signal into an optical format and send the optical signal onto the fiber. Figure 3.4 gives a block diagram of an optical transmitter. The light-source’s output power is an important parameter. Usually its magnitude is expressed in units of dBm—where the unit optical output power is 1 mW. In general, to convert from power in mW to power in dBm, use: Power (dBm) = 10 log10

( Power ), 1 mW

(Eq 3.1)

So a 0 dBm source is a 1 mW source. Power loss is defined as the difference between two values, for example power loss in a point to point link is the difference between the sent and received power in dBm. The units for power loss are dB. The role of optical fiber is to transfer the optical signal from the transmitter to the receiver. The goal is to select fiber that minimizes signal distortion. Signal distortion via loss and dispersion of the optical signal due to optical fiber are important considerations when selecting the optical fiber. II-19

Digital signal from subsystem A

Digital signal to subsystem B Transmitter

Receiver

Support Electronics

Support Electronics

Source

Detector Optical Fiber Optical Coupler

Figure 3.3: Block diagram of simple a point-to-point architecture. Data flow is from left to right.

Electrical Input Driver

Optical Source

Optical Output

Figure 3.4: Block diagram of an optical transmitter

Optical Input Photodetector

Decision Circuit

Electrical Output

Figure 3.5: Block diagram showing the functions of an optical receiver

II-20

The optical receiver’s purpose is to convert the optical signal at the output end of the optical fiber back into an electrical signal. Figure 3.5 shows a block diagram of an optical receiver. The optical signal is detected by the photodetector, and the detector's output current is related to the intensity of the optical signal. The decision circuitry identifies the value of the data bit by comparing the photodetector’s output level to a reference. In some data link designs, a large number of microelectronic devices support the transmitter and receiver sides of the data link. These support electronics may include voltage level shifters, MUX, DeMUX, encoders, decoders, shift registers, phase lock loops, clock recovery circuitry, protocol chips, and a host of other devices. Section 3.2 describes in more detail some of the components that form the optical link. We will focus on the key components that are specific to optical data links. (Section 4 will discuss the radiation effects concerns for these components.) 3.2 Brief Description of Digital Optical Data Link Components The previous section gave an overview of the optical data link. In this section, we will focus on the components that make up such a link. The discussion is confined to components that are unique to optical data links—that is, to light sources, optical fiber and other passive components, and receivers (emphasis on photodetector portion). Support electronics will not be discussed in detail. 3.2.1 Optical Digital Transmitters In this section we review transmitters used in fiber-optic data links. Transmitters contain a drive circuit and a source, as shown in Figure 3.4. Typically, Light-Emitting Diodes (LEDs) and laser diodes are used as light sources. The driver converts the input electrical signal to a current. Source modulation is achieved by varying the injection current. (Although it is also possible to modulate the light output directly, this is not as common as injection current modulation.) Sometimes, a lens (not shown) is used to couple the output light to the optical fiber. Optical sources will be described in more detail below. Segment II of the Short Course will give details of source operation, along with a detailed discussion of the radiation effects mechanisms for these devices. Most texts on fiber-optic communications [for example Agra-97, Pala-98, Lach-98] give a detailed discussion of optical sources. Semiconductor light sources, either LEDs or laser diodes, are used as sources for fiber-optic data links. These devices are quite suitable for use in fiber-based links in terms of their size, range of wavelengths, and power consumption. Sources are classified as long-wavelength or short-wavelength sources. Short-wavelength sources operate in the range of 500-1000 nm, and are typically fabricated with a ternary blend of semiconductors, e.g., GaAlAs. Long-wavelength sources produce light from 1200-1600 nm and are typically fabricated with quaternary semiconductors such as InGaAsP. The light emitting region is a well confined forward biased p-n junction or heterojunction in the semiconductor. Vertical Cavity Surface Emitting Lasers (VCSELs) can be fabricated close together, providing the ability to fabricate a high throughput parallel link (see Section 6). 3.2.1.1 Semiconductor Light Emission and the Resulting Wavelength Forward biasing a p-n junction causes holes to be injected into the n-doped region and electrons to be injected into the p-doped region. These injected minority carriers recombine with the majority carriers, releasing energy, either as a photon electromagnetic radiation (radiative II-21

Figure 3.6: Cross section of a light-emitting device. Charge carrier and light confinement occur in the active layer. recombination) or as heat (nonradiative recombination). The internal quantum efficiency, ηint, is defined as the fraction of recombinations that are radiative. LEDs emit light by a process called spontaneous emission—that is, each photon is emitted spontaneously with a random direction and polarization. Such light is also called incoherent. In contrast, laser diodes emit coherent light via a process called stimulated emission, in which a photon stimulates the recombining charge pair to emit a second photon with the same direction and polarization as the first. As current carrier concentration increases, so does the radiative recombination efficiency. Therefore, it is desirable to confine current flow to the active region in a light-emitting device (see Figure 3.6). Double heterojunction structure devices are fabricated so that electrons injected into the active region from the n-type region will see the p-type region as an energy barrier. Likewise, holes injected into the active region face an energy barrier from the n-type region. Another feature of the double heterojunction diode is that the different indices of refraction confine the light to the active region, thereby reducing absorption of light in the semiconductor and increasing the output power of the diode. During radiative recombination, the photon energy is equal to the band-gap energy (Eg ) of the semiconductor. It will have a wavelength (λ ) defined by : hc

λ= E , g

(Eq 3.2)

h is Planck’s constant (6.63x10-34 joules sec) and c is the speed of light. 3.2.2 Passive Optical Components Lenses, gratings, couplers, fibers and other passive optical devices are utilized in building fiber-optic data links. Here we focus our discussion on fiber-optic waveguides. To understand how fiber-optic waveguides confine light to a long glass strand, it is useful to review some basic optics. Scattering of light propagating in a medium other than a vacuum results in the light travelling less than the vacuum speed of light. This allows us to define a medium’s refractive index as: n=

II-22

c , v

(Eq. 3.3)

Figure 3.7: (A) Typical optical fiber housing. (B) Core and cladding configuration of a typical optical fiber. [Ott-97] where c is the velocity of light in a vacuum and v is the velocity of light in the medium. When light passes from one medium to another, its direction of propagation changes according to Snell’s law. Snell’s law also defines the maximum angle a light ray can make with the normal if it is to pass from a medium with a high refractive index, n1 , to a medium with a lower refractive index, n2 . (See segment II of this Short Course.) This critical angle is given by: n2 Sin(θc) = n , 1

(Eq. 3.4)

Light incident at an angle greater than the critical angle will be totally internally reflected back into the first medium. Optical fibers consist of an inner core with a high refractive index and an outer cladding with a lower refractive index. (See Figure 3.7) The optical signal is confined to the core of the fiber by total internal reflection. Optical fiber can be fabricated with a variety of refractive-index profiles. Step-index fibers have an index of refraction that changes abruptly at the core-cladding interface (Figure 3.8.A). Other fibers may have a triangular profile (Figure 3.8.B), or a graded-index profile, like that shown in Figure 3.8.C. Yet other fibers may have profiles with several discreet steps in refractive index. Fibers are classified into two electromagnetic categories: single-mode and multimode. The mode refers to the number of electromagnetic modes that exist in the fiber under certain conditions. Each electromagnetic mode in a fiber has its own unique electric and magnetic fields that are setup in the fiber. The mode also has a unique propagation constant. Optical fiber can attenuate and distort the optical signal. The attenuation per unit length for a typical high-quality fiber is shown in Figure 3.9. Notice that there are two minima—located at near 1.3 and at 1.55µm, respectively. Optical fiber can also distort the optical signal through dispersion. One example of such distortion is pulse broadening—a widening of the pulse in the time domain as it travels down the fiber. For a more detailed discussion on optical fibers see almost any text on fiber-optic communications. [For example Agra-97, Pala-98, Lach-98.]

II-23

n1

n1

n1

n2

n2

(A)

n2

(B)

(C)

Figure 3.8: Three different profiles of optical-fiber index of refraction. The subscripts (1 and 2) refer to the core and cladding, respectively. [Adapted from Agra-97]

Attenuation (dB/km)

10 5 3 2 1 0.5 0.3 0.2 0.1 0.6

0.8

1.0

1.2

1.4

1.6

1.8

Wavelength (µm) Figure 3.9: Fiber-induced attenuation for various wavelengths. Two minima occur, near 1.2 and 1.5 µm, respectively. [Adapted from Miya-79]

Figure 3.10: Fused biconical tapered star coupler. Optical power sent from one end of the star will be distributed equally over all outputs at the other end of the star. [Adapted from Agra-97].

II-24

Front End

Linear Channel

Data Recovery

OPTICAL DATA

ELECTRICAL DATA PHOTODETECTOR

PREAMP

AMP

FILTER

DECISION CIRCUIT

CLOCK RECOVERY

Figure 3.11: Block diagram of an optical receiver. Vertical dashed lines separate the receiver into three groups (see text). [Adapted from Agra-97]

Often fiber is used to form a coupler between data link subsystems. A star coupler can be formed by using a fused biconical-tapering method, see Figure 3.10. The role of the star coupler is to combine the optical signals entering from its multiple inputs and distribute the signals equally among its outputs. The input power from all input channels is divided equally among the all output channels. Typical operation in an optical network utilizes a protocol controller that allows only one transmitting node at any given time. For example, an N x M star coupler has N inputs and M outputs. 3.2.3 Receivers An optical receiver is a combination of an optical detector, amplifiers, and electronic processing elements used to recover the signal sent from the source. Figure 3.11 gives a detailed block diagram of a typical optical receiver [Agra-97]. The receiver is made up of three stages: 1) front end, 2) linear channel, and 3) data recovery. We will first give a discussion of the each section of the receiver and then briefly describe photodetectors. Further information may be found in many optical communication texts books. (For example [Agra-97, Pala-98, Lach-98].) 3.2.3.1 Stages of a Receiver The front end of the receiver contains the photodetector and the preamplifier. The photodetector converts the optical signal to an electrical signal. This signal is very weak and therefore it must be amplified by a preamplifier. A trade off exists between bandwidth and sensitivity of the photodetector-preamplifier pair. Although a high resistive loading across the photodetector increases its sensitivity, this loading also slows the response of the circuit. An equalizer is sometimes added to the circuit to attenuate the low-frequency components of the signal more than the high-frequency components—in effect increasing the bandwidth of the circuit. The most common method used to keep the bandwidth high while maintaining the desired sensitivity level is to use a transimpedance amplifier. Basically, in a transimpedance amplifier application, the load resistor is placed as a feedback resistor around the preamplifier rather than as a load across the photodetector. This results in more amplifier gain with little

II-25

0 0 1 1 0 1 0 1 1 1 1

(A) (B) (C) Figure 3.12: Examples of NRZ (A), RZ (B) and Manchester encoded (C) data sequences. impact on circuit bandwidth. However, stability can be difficult to achieve when using transimpedance amplifiers. The front end is followed by the linear channel, which consists of a high-gain amplifier and a low-pass filter. The amplifier boosts the electrical signal even more. Sometimes such amplifiers use an automatic gain control mechanism that self-adjusts the gain based on the input signal. The goal here is to limit the average output voltage level regardless of average incident optical power. The low-pass filter shapes the voltage pulse and removes unwanted frequency components generated by signal processing. The last stage of the receiver is the data recovery portion. In this stage of the receiver, the clock and data are recovered. Transmitters can be designed to encode the clock signal into the data. The clock signal is regenerated using the clock recovery circuitry. Another method of clock recovery is to simply pass the clock signal to the receiver. The decision circuitry is used to recover the data. It compares the output of the linear channel to a threshold level. If the signal level is above the threshold, it is considered a “1”. If the signal is less than the threshold, then it is a “0”. The data sampling frequency and sampling duration are determined by the clock frequency. As will be seen later in this tutorial, the radiation sensitivity of the receiver depends strongly on the clock frequency, the sample window width, the delay between data edge and the decision window location within the bit period, and the time relative to the decision window when the radiation event occurred. 3.2.3.2 Digital Data Coding Some fiber-optic data links utilize encoding schemes to encode the clock in the data sequence, others encode the data for error checking and correcting. Here we describe four types of common encoding schemes. Three are depicted in Figure 3.12. The NonReturn-to-Zero (NRZ) code does not necessarily return to “0” during each data bit, see Figure 3.12.A. A sequence on “1s” will hold the data stream high until a “0” is transmitted. This code does not support clock encoding. This code requires less bandwidth than the other two codes in this figure. The Return-to-Zero (RZ) code does necessarily return to “0” during each data bit, see Figure 3.12.B. One example of an RZ code is when the first half of the bit period is coded with the data

II-26

bit information, while the last half is coded with a “0”. This code does not support clock encoding. The Manchester code is a RZ code (Figure 3.12.C). The signal makes a transition at the center of the bit period, upward for a “0” and downward for a “1.” In Manchester coding, the clock is encoded by the center bit transition. The final data encoding scheme is called 8B10B [Widm-87, Cart-97]. Here 8 bits of data are sequence is encoded into 10 bit long data sequence. In essence this encoding scheme allows for clock and subsequence recover. The clock is encoded via the minimum number of bit transitions in the 10 bit encoded sequence, the data is encoded via a check technique called disparity. Disparity is the cumulative difference in the number of one’s and zero’s, disparity for a 10 bit encoded sequence never exceeds ± 1. 3.2.3.3 Receiver Noise and Decision Circuitry Although the concept of distinguishing a “1” from a “0” seems straightforward at first glance, the presence of noise in the receiver circuit causes some data bits to be interpreted incorrectly. At the frequencies with which we are concerned, there are two inherent sources of noise: shot noise and thermal noise. Shot noise results from the random nature of charge generation. Although, in general, a particular incident optical power corresponds to a particular average number of photogenerated electron-hole pairs, the actual charge distribution generated in a particular event will vary in time, space, and number. These variations are responsible for introducing shot current into the receiver. When considering shot noise one must include the effects from all components of the receiver. Thermally generated noise is an intrinsic property of any conductor. Random thermal motion of electrons in some of the front end components of the optical receiver result in the largest current fluctuations. These current fluctuations are called thermal noise. The total current is the sum of all current sources—photogeneration, shot-noise current, and thermal-noise current, as well as any other current sources in the receiver. The sum of these currents is interpreted as the data signal. If a significant amount of noise exists, the data stream will have bit errors—incorrectly interpreted data. This will be revisited in Section 3.3. Figure 3.13 is a cartoon of the sampling of a certain data bit (A) in a bit sequence and the distribution of the signal level (B) measured at the sample point over the entire bit sequence. Sampling of the low data bit over the bit sequence will result in a fluctuation, or distribution, of the sampled currents that represent a “0”. This distribution is shown as the lower Gaussian distribution in Figure 3.13.B, where (i0 ) indicates the current value that is most likely to occur. Likewise for the high “1” data bits, where the most frequently occurring value is (i1 ) peak. The decision circuitry compares each sampled data bit to the threshold value (ith ). If the measured value is less than the threshold then it is considered to be a “0”, if it is greater than it h then it considered to be a “1”. Notice that there is a finite probability (hatched area) that a data bit will be incorrectly interpreted. 3.2.3.4 Photodiodes This section presents a brief description of photodiodes. For a more detailed discussion on photodetectors see most any semiconductor device physics textbook.

II-27

iI Decision Level

iTh i0

(A)

(B) Probability

Decision Window

Figure 3.13: (A) Decision circuitry sampling a fluctuating signal from the detector. (B) Gaussian probability distributions of sampled signals. [Adapted from Mars-94a]

Light in p

Depletion Region

i n

Vout RL

Figure 3.14: Simplified schematic of a p- i-n photodiode used to detect a light signal and convert it into a digital electrical data stream. RL is the photodiode loading. The output level (Vout ) depends on the incident optical power. [Adapted from Agra-97]

II-28

Typically, if not always, photodiodes are chosen as the photodetector in high-speed fiberoptical data links. Three types of photodiodes used in fiber-optics are p-n photodiode, p-i- n photodiodes (most commonly used), and avalanche photodiodes (APD) (there are others). The first and last are less likely to be used than the p-i-n diode. The conversion of optical power into an electrical current can be understood by reviewing Figure 3.14. The p- i-n photodiode is operated in a reverse bias circuit. The incident light must penetrate to the depletion layer (a region in the diode that has a high electric field due to both the applied voltage and the semiconductor doping concentrations). If the energy of the incident photons exceeds the bandgap energy of the semiconductor (Eg ), an electron-hole pair is created. The electric field in the depletion region drives the electrons towards the n-region and the holes towards the p-region. This current is called the photocurrent, and is proportional to the incident optical power. Whether the logic output of the detector is interpreted as a “1” or a “0” is determined by the incident optical power. The proportionality constant between the photocurrent and the incident optical power is called the responsivity. The responsivity depends on the photodiode material and design, as well as the wavelength of the incident light. A maximum cutoff wavelength (λmax) can be determined based on the fact that the band-gap energy (Eg) is the minimum energy needed to create electron-hole pairs. It is given by: hc , (Eq. 3.5) Eg where h is again Planck’s constant and c is the speed of light. If the wavelength exceeds this value, the photon is not absorbed and no photocurrent is generated. Table 3.2 lists some representative photodiode materials and the maximum wavelengths they can detect. Often shortwavelength photodiodes are fabricated using Si, while long-wavelength applications use InGaAs alloys.

λmax =

Table 3.2 Approximate cutoff wavelengths for various photodiode materials. Material GaAs

Wavelength (µ m) 0.9

Ge InGaAs*

1.8 1.0 to 1.7

InP Si

1.0 1.2

*typical values depending on specific alloy. A p-i-n diode is fabricated with a large intrinsic semiconductor layer between the p and nregions—in effect increasing the width of the depletion layer and giving a large region for photon absorption. Because of the high resistance of this intrinsic layer most of the voltage drop occurs across it, and most of the charge that contributes to the signal is generated inside the intrinsic layer. The width of the intrinsic layer determines how efficient the diode is at converting optical power to current. That is, the diode’s responsivity depends on the intrinsiclayer thickness. However, the response time of the photodiode increases with increasing intrinsic-layer thickness.

II-29

Data Stream Transmitter Pseudo Random Sequence Clock Generator

Receiver

Oscilloscope

(A)

Trigger

Vertical

A

“1” level Threshold

C

(B)

Bit period Tb

B

D

“0” level

Figure 3.15: (A) Experimental setup used to measure an eye diagram (see text). Representation of an eye diagram. [Adapted from Agra-97]

(B)

Indirect bandgap semiconductors, such as Si and Ge require relatively thick intrinsic layers— typically on the order of 20-50 µm. The response times of these devices are > 200ps. In contrast, using direct bandgap semiconductors like InGaAs allows much thinner intrinsic layers (3-5 µm), so these devices will have much faster response times (<50 ps). A final note: The creation of electron-hole pairs in the photodiode by the high- energy particles found in the space radiation environment can dominate the performance of a photodiode, and in turn the performance of the data link. This topic will be explored in more detail in the sections that follow. 3.2.4 Support Electronics Most optical data links require support electronics to format the data signal for optical communication. Some examples are buffering, encoding, decoding, voltage level shifting, digital data handing, temperature compensation circuitry, clock recovery circuitry (e.g., phase locked loops), and others. To implement these functions, links may contain MUX, DeMUX, ASICs, RAMs, PROMs, protocol chips, oscillators, and many combinations of the host of linear, digital and passive microelectronics and components. A detailed discussion of the radiation effects of these components is beyond the scope. However, the reader should be warned that radiation effects in the support electronics can dominate the response of the data link. We will give some of the general concerns in Section 4 and some examples of link performance limitations due to radiation degradation of support electronics in Section 5. 3.3 Digital Optical Data Link Performance Metrics Optical data link operation can be degraded in a high radiation environment. Certain performance metrics can be used to measure the extent of this degradation. These metrics include the optical power budget and the bit-error ratio. In the sections that follow we will

II-30

discuss these metrics. However, first we will look at a measurement technique called eye diagram—a simple technique that results in an easily visualized metric for overall link performance. 3.3.1 Eye Diagram Measurements and Analysis High-data-rate optical link performance can be measured using the eye-diagram technique [for example see Agra-97]. Figure 3.15.A shows a block diagram of the setup to make a timedomain measurement of the bit stream. The pseudorandom sequence generator outputs a repeatable and predictable, but random data-bit stream. The receiver output flows into the vertical trace of the oscilloscope, while the data clock activates the oscilloscope trigger. The result is an eye-shaped trace that represents superposition of several bit sequences, as represented in Figure 3.15.B. The eye diagram contains easily observable information about the overall link performance. Some of its characteristics are: • The opening of the “eye” gives the optimum sampling time interval for the received signal. The optimum sample time is where the eye is a maximum. In Figure 3.15.B this location is labeled “D.” • The voltage spacing labeled “A” is a measure of the noise when a logic “1” is transmitted. Likewise “B” is a measure of the noise when a logic “0” is transmitted. • The time spacing labeled “C” gives a measure of timing accuracy of the data stream and is called the timing jitter. • Increasing the frequency to the maximum operating frequency forces the eye to close. 3.3.2 Bit-Error Ratio Data transfer by optical digital links (or any data link for that matter) can be corrupted by noise. Figure 3.16 depicts a cartoon bit sequence as it propagates through a data link. The top line shows the “ideal” input signal. The second line shows the degraded signal after it passes through the electronics and optoelectronics. The degradation of the signal is the result of several sources of noise in the data path. The last line represents the signal as received and interpreted. Notice that the fourth bit in the data sequence has been interpreted incorrectly, resulting in a “bit error.” Depending on the operating frequency, several millions to billions of bits travel through a system every second. The probability of incorrectly interpreting a bit is a useful metric commonly known as the Bit Error Ratio (BER): number of incorrectly transmitted bits BER =

,

total number of bits transmitted

(Eq. 3.6)

Sometimes this metric is also called the bit-error rate. (However, for the radiation effects community bit-error rate usually implies an on-orbit upset rate, we will use the more commonly accepted meaning for BER.) It can be measured using commercially available Bit Error Ratio Test (BERT) equipment. A general BERT setup will be described in Section 5.1.2. The relationship between signal level, noise level and BER is fairly simple: the more margin that exists in a digital link, the lower the probability that a bit will be interpreted incorrectly. For II-31

Time 1 0

1

0

0

1

Transmitted Bit Stream

Threshold

Signal Degrades

Interpreted Bit Stream

Threshold 1 0

1

1

0

1

Error

Figure 3.16. Cartoon of a specific bit sequence (101001) as it travels through a data link. The fourth bit in the sequence is interpreted incorrectly (101101). [Adapted from Powe-93]

a fixed receiver design, increasing the optical power launched on the photodiode lowers the BER until the optical power reaches the design limit of the receiver. 3.3.3 Optical Power Budget When designing an optical data link, the designer must have a clear understanding of the power budget. There must be sufficient link margin to accommodate power losses as the signal passes through the link. The starting point for this analysis is the source’s output optical power. The end point is the optical power that the photodiode requires to achieve the desired BER. The power budget must also consider the other losses in the digital link, including: • source-to-fiber coupling loss at the transmitter, • connector insertion loss, • insertion losses in modulators, couplers, lenses, etc. • fiber-to-receiver loss, • aging effects, and • fiber losses. Figure 3.17, represents a graphical analysis of the power budget for an optical data link. In this example, the link will transmit data with an acceptable BER as long as the fiber length is less than approximately 2.5 km. To ensure proper operation of fiber-optic data links in space, power budget analyses must include radiation effects resulting from the space radiation environment. This completes our summary of the optical digital data link. In the next section we will consider radiation effects at the component level, and in Section 5 we will discuss optical data link performance degradation in a radiation environment.

II-32

Power (dBm)

Source Power 0

Insertion loss (e.g. coupling) Power in fiber

-10

Fiber length loss

-20

Required power at detector

-30 Reflection losses -40 Detector power for BER requirement 0

1

2 Distance (km)

3

4

3.17. Graphical analysis of system power budget shows link margin as a function of fiber length. [Adapted from Powe-93] This type of analysis will be revisited when we consider degradation of optical link margin resulting from radiation effects.

II-33

4.0 Radiation Effects in Digital Optical Data Link Components The last half of the previous section gave an overview of the basic components of a Digital Optical Data Link, as well as discussing some of the metrics used to measure the performance of these systems. In this section, we will briefly consider the radiation effects to which these components may be susceptible and how these effects may degrade component performance. We will first briefly consider DDD effects in light sources—LEDs and laser diodes. This subject will be treated in more detail in segment II of the Short Course. Next we will consider the effects of TID on the passive transmission media used in fiber-optic data links. We will then examine the effects of TID, DDD and single-event transients in photodetectors. Lastly, we will briefly discuss radiation effects in the support electronics. 4.1 Permanent Degradation of Sources As mentioned in the previous section, the role of light sources in digital optical link systems—usually LEDs and laser diodes—is to convert current signals into light. Because the drive signals for these devices are usually on the order of milliamperes, the small currents generated by ionizing radiation (usually picoamperes) are insufficient to cause SEEs. As such, for these devices, we are generally more concerned with permanent degradation due to DDD TID is often a second order effect also [Barn-84, Barn-86, Mars-92]. Both LEDs and laser diodes produce light by means of radiative recombination of injected minority charge carriers with majority carriers in the depletion region. As such, any damage that decreases the efficiency of radiative recombination or introduces competing processes will degrade device performance. Irradiation by heavy particles such as protons and neutrons can displace nuclei from the crystalline lattice, causing DDD. These defects can serve as sites for nonradiative recombination, decreasing the efficiency of the light source [Barn-84]. DDD effects will be covered in detail in segment II of this Short Course. Here we merely note that displacement damage decreases the output power of these light sources, as well as increasing the threshold current for laser diodes (see Figure 4.1). We also note that with the exception of amphoterically doped LEDs, these devices can remain functional even after exposure to relatively high proton fluences (>1012 protons/cm2 ), provided the drive currents are sufficiently high to compensate for the effects of DDD. Characterization of light sources continues to be an active area of research, with substantial effort directed toward the study of damage mechanisms and annealing in LEDs and multi-quantum well laser diodes. (See, for example [Hinr-98, Zhao98, John-99a, Lee-99].) It is interesting to note that even when one early study observed degradation in the radiation response in optoelectronic systems and found loss of efficiency at relatively modest doses, the light source was vindicated upon further investigation [Mars-92]. It turned out that the degradation actually took place in a graded-index lens coupling the laser to the fiber, rather than in the laser diode itself. This anecdote demonstrates not only that most light sources are remarkably hard to radiation, but that in contrast to the situation in microelectronics, in optoelectronics one cannot ignore radiation effects in the so-called passive media. 4.2 TID Degradation of Transmission Media Although charged particles traversing an optical medium can generate photons, these processes are too weak and the path lengths traversed too short for these events to interfere with signals. For this reason, with respect to transmission media, we are generally more concerned

II-34

Optical Power (mW)

20 Before Irrad. 4.6x10 11cm -2 2.1x10 12cm-2 3.9x10 12cm-2

15 10 5 0

0

10 20 30 40 50 60 70 Current (mA)

Figure 4.1 Displacement damage from 5.5 MeV proton irradiation increases laser diode’s threshold currents and decreases the output power. [Adapted from Evan-93] with permanent, cumulative degradation resulting from TID. We note in advance that for almost all space applications signal path lengths are relatively short (<75 m), so losses due to TID degradation can be accommodated by selection of rad-hard fibers and other relatively painless measures. Photonic systems require nearly flawless optical fibers, lenses, and other transmission media to transmit light over long distances with minimal losses. Our discussion will concentrate on optical fibers, because they account for most of the pathlength traversed by the signal. It should be remembered that our treatment applies equally to lenses and other transmission media. Ionizing radiation causes charge to become trapped by defects in the optical fiber, creating color centers that absorb light and decrease transmission efficiency. TID degradation in passive components should not be overlooked when analyzing the system power budget. For this reason, we will treat it in some detail. Readers who require additional detail are referred to the 1991 Short Course given by E.J. Friebele [Frie-91], the SPIE Photonics for Space Environments journals, as well as several excellent papers in Proceedings of RADECS and TNS [for example see Frie-85, Barn-90, Mars-94b, Gris-94, Hens-97]. As in microelectronics, the phenomenon responsible for TID induced performance degradation of optical fibers is charge trapping by defects in the material. These defects may be radiation-induced themselves, or they may be the result of impurities or fabrication conditions. Regardless of the defect origin, once the color centers form, light transmission efficiency is degraded as the color centers absorb signal photons. The degradation, however, need not be permanent. As is the case for trapped charge in microelectronics, fiber optic color centers can heal by annealing. The processes of formation and annealing of color centers take place in competition, and fiber-optic degradation depends on the relative rates of these mechanisms. Even after two decades of study, not all of the processes involved in radiation-induced degradation and annealing are completely understood. However, the factors that affect the radiation response of optical fibers are sufficiently understood that the designer now has a good

II-35

104 Induced Loss (dB/km)

P-Ge doped Ge doped

103 102

Pure Silica

101 100 10-1

100

101

102

103 104 105 Dose

106

107

Figure 4.2 Phosphorus-geranium-doped, germanium-doped, and pure-silica optical fibers can show quite different susceptibilities to proton-induced attenuation loss. Dose rate is 4000 rads(Si)/minute. Measurement was in situ during the irradiation. The dose is given in rads(Si). [Adapted from Frie-91] basis for both fiber selection and for formulating a hardness-assurance test program. Among the most important factors that influence both radiation response and annealing are the properties of the fiber itself, the temperature and optical conditions encountered in the application, and the radiation environment to which the fiber is exposed. A quick glance at Figure 4.2 is sufficient to indicate that radiation response of various optical fibers can vary widely. Attenuations after the relatively modest dose of 1 krad(Si) range from undetectable to around 100 dB/km. The annealing behavior of different fibers is every bit as variable. Much of the variation can be accounted for by the properties of the fibers themselves. Among the more important fiber properties that affect radiation response is fiber composition. As seen in Figure 4.2, pure silica fibers are less susceptible to radiation damage than are those with more complex compositions. However, optical fibers are usually fabricated with dopants to tailor refractive indices of the fiber core and clad to the desired application. Both germanium and phosphorus are commonly added to the core to raise its refractive index, while fluorine is used to lower the clad refractive index. Dopant choice can have important consequences for radiation hardness. In general, phosphorus seems to facilitate the formation of color centers that attenuate light signals, while germanium actually seems to improve some aspects of radiation performance. As might be expected, manufacturing conditions can also influence radiation and annealing response. The manner in which the fiber preform is deposited

II-36

and under which the fiber is drawn are among the conditions found to correlate with radiation response. The conditions encountered during the application can also influence requirements for the radiation performance of passive optical components. Annealing in optical media—as in microelectronics—is a thermally activated process. In one study, [Mars-94b] fits to activation energies for various color-center traps have yielded activation energies on the order of 1 eV, or so, and trap lifetimes ranging from about a thousand to a few tens of thousands of seconds at room temperature. Moreover, studies have shown that the annealing rate of phosphorus-doped fibers does not increase with temperature. As such the temperature range encountered in the application can influence fiber choice as well as testing procedures. Radiation damage can also be annealed by the light signal itself. This process, called photobleaching [Frie-84a], occurs when the light supplies enough energy to facilitate annealing in the fiber-optic. Even very modest signals can significantly alter the equilibrium between damage and annealing. In space, radiation damage and annealing take place as competing processes, with temperature and signal intensity playing crucial roles in determining the equilibrium point. One of the most important application conditions in determining the effect of radiation on system performance is the signal wavelength. Most color centers have their peak absorption in the ultraviolet. The further away the source wavelength is from the absorption peak, the less will the signal be absorbed. Thus, systems that use InGaAsP laser diodes operating at 1300 nm or 1550 nm show slightly less signal loss due to radiation darkening than do those operating at shorter wavelengths. The radiation environment must also be considered when evaluating potential degradation of transmission media. For applications where the transmission length is short (most spacecraft applications are less than 100 m), and where the light source can be driven hard enough to provide sufficient end-of-life margin, radiation-induced darkening of transmission media may be negligible for radiation levels less than 105 rad(Si). The dose rate and the type of radiation also affect the extent of radiation damage. The annealing and photobleaching that compete with radiation damage give rise to an apparent doserate effect. To date, however, there is little evidence for true dose rate dependence in optical media, and most studies show good agreement between low-dose rate data and higher-rate tests followed by an anneal. In particular, work by Freibele et al. indicates that a shorter, high rate test coupled with annealing data may suffice to predict radiation behavior in the low-dose-rate, onorbit environment. The space-radiation environment may include a broad range of radiation types. Because it is not practical to provide such a radiation mixture in the lab, most radiation test sources provide only one type of radiation (for example, 1.25 MeV γ rays from Co60 ). Several studies have sought to show that, at the very least, such a simplification will not significantly underestimate radiation damage. To date, most studies have shown that equivalent absorbed doses cause roughly equal levels of damage, independent of radiation type [Frie-91]. As such, one can confidently assume that to first order, a rad is a rad, and data obtained with gamma rays can be used to bound the on-orbit behavior of the fiber. It should be noted, however, that different types of radiation might have different damage mechanisms. For example, protons, in addition to generating color centers, may also be captured within the medium forming SiOH, which absorbs strongly in the infrared [Lema-91]. Although these differences have not yet been seen to have any practical effect, care should be taken when using data taken under different conditions. II-37

The preceding discussion is intended primarily to give a reader an indication of the broad variety of factors that affect the radiation performance of optical fibers. For a deeper discussion of these factors, the reader is referred to reference [Frie-91, Mars-94b] and the references contained therein. The reader can also review [Tayl-90, Tayl-91, Tayl-92, Hens-92,] for a discussion on radiation effects in different optical couplers (albeit for dose rates higher than those encountered in the natural space environment.) Some graded index (GRIN) lenses have been shown to be especially sensitive to darkening via creation of color centers, although this is not an issue in most lenses [Frie-84, Weis-90, Mars-92]. In addition to using spacecraft structures as shielding to minimize radiation exposure, there are a number of techniques for minimizing the effects of radiation damage in optical media. Selecting an optical fiber with known and appropriate radiation response for the mission environment is a first step. It should also be remembered that radiation damage takes place in competition with thermally induced annealing and photobleaching. In most applications, the equilibrium level of radiation damage will be lower if the optical fibers are kept at a higher temperature. (Note that phosphorus-doped fiber optics are an exception [Frie-84, Frie-85].) Moreover, overdriving the light source, if feasible, not only facilitates photobleaching in some fibers, but also can compensate for any attenuation that takes place over time. Properly understanding the processes that contribute to darkening in optical media will ensure that these media are not the weak links in the optoelectronic system. Generally, darkening in well-chosen passive optical components can easily be accommodated in the link margin and does not limit the radiation of on-orbit optical data links. 4.3 Permanent Damage in Optical Detectors The purpose of the detector in an optical system is to detect a weak optical signal and convert it into an electrical current that can then be amplified sufficiently to allow the resolution of individual data bits. Each of these steps—detection, conversion and amplification—is potentially vulnerable to radiation effects. Here we consider the effects of TID and DDD on the most common optical detectors [Barn-84, Wicz-86, Mars-92], while in the next section we will examine the potentially significant role of single-event transients in these devices. It will be seen that although radiation characteristics play an important role in determining which detectors are suitable for space applications, with proper choice of detector modules, permanent damage due to TID and DDD need not be a significant concern for most space applications. Optical detectors in optoelectronic devices function by detecting the photocurrent generated by a photon with energy greater than the semiconductor band gap. For this reason, any damage to the semiconductor that changes its global properties, or that degrades the semiconductor’s ability to carry a current will potentially compromise the detector’s efficiency. Because DDD generates charge traps that significantly decrease minority carrier lifetime, minority-carrier photodetectors—such as phototransistors—are generally unsuitable in high-radiation environments. This is a prime reason why photodiodes predominate in high-radiation applications. As TID accumulates, photodiodes, like all diodes, degrade, with leakage currents increasing up to several orders of magnitude for doses up to 1 Mrad(Si). Both TID and DDD increase background noise in photodiodes, and DDD tends to decrease receiver current outputs. However, many photodetectors show remarkable robustness to space radiation environments. Figure 4.3 shows the number of years a typical InGaAs p- i-n photodiode would take to suffer a 3 dB responsivity loss for various circular orbits and behind various shielding thicknesses [DaleII-38

Years to 3dB Loss in Responsivity

100

300 mils Al 150 mils Al

10

40 mils Al

InGaAs PIN Photodiodes 1 2.5

5.0

7.5

10.0

Circular Orbit Altitude (X 103 km)

Figure 4.3 Equivalent Al shielding thickness can have a large effect in determining how long a photodetector can operate before the environment inflicts a particular damage level. [Adapted from Dale-92] 92]. Often the key to success is incorporating adequate margin in source strength to account for EOL degradation in detectors. Segment II of the Short Course discusses DDD in photodetectors in more detail. A detailed discussion of predicting on-orbit performance degradation from ground based data is given in [Mars-94b, Mars-99]. 4.4 Single-Event Transients in Photodetectors As outlined previously, photodetectors detect light signals by means of the photocurrents they generate in the semiconductor. However, these photocurrents can be indistinguishable from the single-event transients (SETs) generated by ionizing radiation. For this reason, SETs can combine with other sources of noise (particularly near EOL) to significantly increase the bit error ratio in high-radiation environments. Although the importance of these effects can vary widely from system to system, good receiver design and a robust protocol layer can significantly reduce BER [Mars-94b, Labe-96b, Jack-96, Dale-96]. Fortunately, the SEE susceptibilities of many photodetectors have already been characterized, and this work can be used to guide detector selection for high-radiation environments. Some devices are very susceptible to SETs, while others seem to be virtually immune to them. Indeed, recent work demonstrates that some detectors are even susceptible to proton-induced SETs caused by direct ionization [LaBe-93a, Mars-93a, Mars-93b, Mars-94a, Mars-96, LaBe-97, Reed-98, John-99b]. As discussed in section 2, protons can generate ionization by either direct means—in which the proton, itself generates the charge—or by indirect means—in which the products of a protonnucleus collision generate the charge. If sufficient charge is generated and collected at a sensitive node in an electronic device, an SET or an SEU may result, see Figure 4.4. For most microelectronics, proton direct ionization generates far too little charge to be an SEE concern. The high sensitivity of some photodiode circuits, however, means that even proton direct ionization cannot be ignored by the SEU analyst. II-39

Direct Ionization Across Long Pathlengths - -+++-+-+ ---- ++ +- ++ -+--+-+++ + ++ -+ + --+ -- + ----+ - -+- +-+ +- + -- + -+ - -+- -+- -+ + - -- - Output

Proton

100’s µm + ---++ ++ -++ +++-+ -+ +-+++ -++ -+- -+++++++--+ -- +- + -

Proton

Time

Nuclear Reaction Recoils Indirect Ionization

SET Cross-Section (cm2)

Figure 4.4 Protons can generate charge in a device by either direct or indirect ionization. If sufficient ionization is generated, a single-event transient (SET) may result. [Adapted from LaBe-97]

4x10-7

peak value

3x10-7

2x10-7 normal incidence 1x10-7

0 0

20

40

60

80

100

120

Angle of Incidence (degrees) Figure 4.5 The SET cross section for protons incident on a photodiode increases dramatically as the protons approach grazing incidence—an indication that direct ionization is becoming important [LaBe-97].

II-40

Normilized Cross-Section

10000 Reed-98

1000

Johnston-99

100

10

1 0.000

0.005

0.010

0.015

0.020

0.025

0.030

LET (MeV cm 2/mg)

Figure 4.6 More proof of direct-proton-ionization SETs comes from proton experiments at different energies. Protons with higher energies (68-225 MeV [Reed-98]) and therefore low LETs, demonstrating a threshold LET, while for low-energy (15, 30 and 50 MeV [John-99]) protons, which have higher LETs, the cross section increases dramatically. Figure 4.5 demonstrates this rather surprising result for the photodiode in an optocoupler. (While optocouplers use Si photodiodes the data that follow show clearly that proton-induced direct ionization is important when consider radiation-induced effects on photodetectors). As can be seen, the SET cross section for protons at near-grazing incidence is more than an order of magnitude higher than the cross section for normally incident protons—this despite the fact that the physical device cross section is actually lower for grazing-incidence protons. This result has been interpreted as an indication that for sufficiently long path lengths within the detector, even protons can directly generate enough ionization to cause SETs. Further support for this interpretation is given by comparing results for low-energy protons with those for protons of higher energy, see Figure 4.6. The ordinate was computed by normalizing the maximum grazing angle cross-section to that at normal incidence. Because LET decreases with proton energy (over the energy range of interest), direct-ionization would be expected to be less important for higher-energy particles. Indeed, no enhancement is seen at glancing incidence for 225 MeV protons (LET = 3.4x10-3 MeV-cm2 /mg), while substantial enhancement is seen at neargrazing incidence for 68, 50 and 30 MeV protons (7.8x10-3, 9.9x10-3, and 1.5x10-2 MeV-cm2 /mg, respectively) [Reed-98, John-99b]. A consequence of the increase in grazing-angle cross section is that in proton-rich environments, proton-induced SETs can dominate those caused by heavier ions and by indirect ionization [Labe-97]. The susceptibility of photodetectors to direct-ionizing proton SETs demonstrates the importance of minimizing the volume in which the detectors are susceptible to ion-generated photocurrents. This criterion can be extremely important in selecting components that may perform acceptably in space radiation environments. As was discussed in Section 3 of this Short Course, direct-bandgap semiconductors, such as GaAs and InGaAs, require much thinner intrinsic semiconductor layers than do indirect-bandgap semiconductors, such as Si and Ge. As a result, not only are direct-bandgap devices more suitable to high-data-rate applications, they are

II-41

also less likely to generate large, ion-induced photocurrents, and so are less susceptible to SETs [Mars-94b]. However, knowing the device SET rate is only one step in determining how these events will affect system performance metrics such as the bit error ratio (BER). Not all SETs cause a bit error. As is the case with most radiation-induced bit errors, system-level characteristics are crucial for understanding whether these bit-errors will affect system performance. System-level aspects of bit-error assessment for fiber-optic data links will be discussed in more detail in Section 5. Here we merely note that it is system-level characteristics that determine whether a SET produced by a radiation-event in a photodiode with a given pulse height and duration will cause a bit error. Because of the periodic nature of the signal, whether an SET causes a bit error depends critically on both its magnitude and on when it occurs with respect to when the signal is read. As such, high-rate applications will be more susceptible to SETs than will slower applications. Moreover, because receivers typically set the threshold for discrimination between ‘0’ and ‘1’ midway between the lowest and highest optical powers, increasing the source intensity significantly lowers the number of SETs that ultimately contribute to the BER. Note that with respect to detector SETs, as with respect to TID and DDD induced degradation in sources, transmission media and detectors, extra margin in light source drive can significantly improve system performance. 4.5 Radiation Effects in Support Circuitry As indicated in Section 3, the receiver must not only detect a weak light signal and convert it into a current, but also amplify it and then process it. The photodiode can only accomplish the first two of these processes. Amplification and processing require additional support electronics. The devices that perform these functions are inherently vulnerable to the same radiation effects that are found in all electronics. One may devote considerable care to selection of appropriate light sources, transmission media and detectors only to find the most vulnerable component in the system is, for example, an op amp or an A-to-D converter. For this reason, we will briefly discuss how SEE, TID and DDD may affect various support electronics technologies. Because device technology determines device radiation-effects susceptibility, it is critical to understand the technologies of support devices. Most device technologies can be categorized as either CMOS, bipolar, as a mixture of the two technologies (BiCMOS) or as belonging to various specialty or advanced technologies (GaAs, InP, InGaAs, and so on). In what follows, we merely touch on the effects that are potentially of concern for these technologies. General references include: for general background on radiation effects in semiconductors [Mess-86], for TID concern in CMOS [Dres-89], and for SEE [Mess-97]. Other good starting points for the interested reader include the 21 years of NSREC short courses now available. The primary concerns for CMOS devices are single-event effects (particularly destructive SEL) and degradation due to TID (particularly leakage current increases). Because CMOS devices are majority carrier devices, displacement damage effects are generally of secondary importance. Single-event latchup (SEL) is a high-current, positive-feedback state that can be either destructive or nondestructive. Even nondestructive SEL can disrupt operations, because recovery requires cycling power on the device. SEL occurs in p-n-p-n structures in the CMOS (see Figure 4.7). These structures can be thought of as representing a pair of connected parasitic bipolar junction transistors that can form a positive feedback loop, possibly resulting in thermal failure of the device. A large amount of information has been compiled on SEL and on II-42

V

p+

n+

r bv

p+

r sv

n+ n-well

rsi

r bi

rs

p- substrate

Figure 4.7 Single-event latchup is a potentially destructive effect that occurs in parasitic bipolar junction transistors in p-n-p-n structures—usually in CMOS. [Gall-96]

mitigation techniques for the phenomenon [Gall-96]. Here we merely end with the caveat that unless a CMOS device is specifically designed for SEL immunity, it is not possible to predict with certainty whether it will be susceptible to SEL—or if susceptible, whether one will be able to mitigate the effect—without heavy-ion testing. In addition, CMOS devices may be susceptible to single-event upsets and transients, and at rates ranging from 0 to many times per day. The basic mechanisms of TID damage were discussed briefly in Section 2. In CMOS devices, the effects of TID manifest as increased leakage currents, shifts in voltage levels (including logic signals) and other parameters and eventually functional failure. Perhaps the most remarkable aspect of TID in CMOS devices is the range of failure levels observed. Devices have been seen to fail at levels as low as a few hundred rads(Si) and to remain functional at levels well above a few Mrads(Si). Again, although similarity of technology may be looked upon as a rough guide to predict device performance, actual device performance cannot be predicted without test data. Moreover, especially for commercial parts, lot-to-lot variations can be significant. If caution is appropriate when dealing with SEE and TID effects in CMOS, it is doubly so when it comes to radiation effects in bipolar devices. Destructive SEE have been observed in bipolar devices ranging from amplifiers [O’Bry-99, McDo-00] to ADCs [Koga-94]. SEUs and SETs are also possible over a broad range of rates in these devices. It is possible for an op amp or comparator to drive the service outage rate for an entire satellite. TID degradation in bipolar devices has undergone a bit of a renaissance of late—at the expense of designers’ peace of mind. Many of these devices are vulnerable to Enhanced Low Dose Rate Susceptibility (ELDRS)—that is, the devices may suffer more damage when irradiated slowly (as in space) than when irradiated rapidly (as in the lab) [John-94, McCl-94, Emil-96]. We merely note here that our understanding of ELDRS is still evolving, and until the field reaches a more mature state, bipolar devices should probably be assumed to exhibit an ELDRS sensitivity unless testing has shown otherwise. More generally, as in CMOS devices, II-43

TID in bipolar devices results in increased leakage currents, decreased gains, along with other parametric shifts and eventual functional failure. However, TID is not the only mechanism of gradual degradation one must consider for these devices. Bipolar devices, like other minoritycarrier devices, are vulnerable to degradation resulting from DDD. Some bipolar devices show significantly more degradation when irradiated with protons than they do when irradiated with gamma rays to an equivalent TID [Rax-99 and references therein]. Parameters that may suffer from DDD include leakage currents, reference voltages and device gains. The high data rates of the optical components in fiber-optic data links demand support circuitry with commensurate capabilities. Although the development and radiation characterization of such fast technologies—many of them using novel semiconductor materials such as GaAs, InP and so on—are ongoing, a few observations are possible. Because of the technology of these devices, SEL is not an issue. However, the low capacitances of these devices (needed for fast operation) make them inherently vulnerable to SEUs and SETs. Many devices have threshold LETs for upset less than 1 MeVcm2 /mg, leading to high upset rates [Mars-95c]. TID performance for many of these technologies has been found to be generally acceptable and in many cases excellent [Hash-94]. Moreover, displacement-damage tests to date have revealed considerable robustness in these devices [Hash-94, Mars-95c]. Overall, the components in fiber-optic data links can perform well for many missions. However, ensuring such performance is a matter of following good hardness assurance practices for both optical and electronic portions of the system from the component level to the system level.

II-44

5.0 Optical System Response to Space Radiation Environments This section discusses radiation effects as they affect system-level performance of optical digital data links in space radiation environments. The goal is to help the reader better understand how component-level radiation effects (Section 4) affect the link performance metrics (Section 3). We discuss system-level ground testing for single-event effects, as well as performance metric analysis for optical digital data links and how radiation effects on selected components affect such metrics. Finally, we present some on-orbit results for a few fiber links. An optical data link can contain a host of microelectronic and optoelectronic devices as well as passive optical components. Any of these devices can suffer performance degradation when exposed to space radiation environments. It is beyond the scope of this course to cover every detail of optical-link performance degradation due to all possible devices. Using good engineering practices in selecting devices and in assessing their radiation performance within the system will ensure that any adverse impact on system performance will be minimized. Component-level TID, SEE, and displacement damage screening can be sufficient to assess the performance of some microelectronic and photonic devices in the space radiation environment. However, in-situ testing is especially important for high bandwidth electronic and optoelectronic systems that employ system-level mitigation approaches. The emphasis of this section of the Short Course is the system impacts to optical data links when the performance of their optical and optoelectronic devices degrades as a result of exposure to radiation. Vulnerable devices include optical sources, optical fibers, and photodetectors. Some of the testing techniques and analysis are general, and we have attempted to indicate when a technique or analysis approach can be used to assess other component effects. Three excellent references on radiation-induced effects in fiber-optic data links are [Mars-96, Dale-96, Labe98d]. These papers provide the basis for the discussion that follows. The successful use of fiber-optic data links in spaceflight applications is discussed in Section 5.3. These successes demonstrate that not only is it possible to develop robust fiber-optic data buses that will survive in the space radiation environment, but it is also feasible to use these systems as a primary means for intra-satellite communications. 5.1 Optical Data Link Single-Event Effects Ground Testing This section summarizes the impact of single-ionizing-particle events on optical links operating in the MHz to GHz regime. First we will describe radiation-induced bit errors, focusing on radiation events in photodiodes. Then we will describe the implementation of a generic BER test on a photodiode exposed to protons and helium ions—along with the results thereof. An example of subsystem-level testing of a full functional optical data link is then given, followed by a summary of results from in-situ experiments on several other full-functional fiber-optic data buses. Before we begin we should define the difference between a bit error and a message error. A bit error is a simple data bit that has been changed from its expected value, for example radiation changing a “0” to a “1”. This bit error may or may not induce a system level error. Some systems employ error detection, correction, or mitigation schemes that are robust enough to “repair” a bit error so that it does not impact system performance. So when we refer to bit error we are implying that it is a simple change in a data bit’s state, the system impacts have not been determined. Message errors are manifestation of bit errors at the system level.

II-45

Radiation Induced Transient

iI iTh i0 Decision Window

(A)

(B)

(C)

(D)

Figure 5.1. Particle-induced SETs occurring at different locations within a bit period and with different intensities [Adapted from Mars-96]. The event in (a) will cause a bit error, those depicted in (b and d) will not induce a bit error, and the event depicted in (c) may or may not cause and bit error. [Adapted from Mars-94a] 5.1.1 Particle Induced Bit Errors As was shown in Section 3.3, noise can cause bit errors in the data stream. Likewise, singleparticle radiation events can induce bit errors in the data sequence [Mars-96, Dale-96, Labe-98d, and references therein]. These single-particle events can occur in any of the digital or linear devices in the data path. Photodiodes are particularly sensitive to ionizing radiation, as described in Section 4.4. The photodiodes used in optical links typically have large junctions, fast time constants, and are incorporated in receiver circuits designed to be highly sensitive [Agra-97, Pala-98, Lach-98]. A key advantage of fiber-optic data links is that the optical signal for a Gbps link may be only microwatts, and conversion to electronic signal during a bit period results in only picocoulombs of charge (a few thousand electrons). Photodiodes used in fiber-optic links make perfect ionizing particle detectors. The next few paragraphs will examine the particular case of how a radiation-induced current transient in a photodiode can become a bit error when interpreted by the decision circuitry of an optical data link. (The impact of this bit error is defined by the system protocol layer and other mitigation approaches. If it is interpreted as an system level error we call this a message error). We focus on photodiodes because of their prevalence in optical data links and their susceptibility to radiation-induced transient events. However, the reader should keep in mind that any device in the data path can cause bit errors, which in turn may lead to system level message errors. Studies on several high-speed digital optical links have shown that photodiodes can be the component most sensitive to single-particle events [Mars-93a, LaBe-93c, Mars-94a, Frit-95, Dale-96, LaBe-96b, Mars-96]. These studies have concluded that direct ionization from lightly ionizing particles (protons) is the primary concern for bit errors. As such, bit errors can also be caused by heavy-ion cosmic rays traversing the photodiode and energetic products from nuclear interactions between incident protons and semiconductor nuclei in the device. Figure 5.1 depicts the effects of four different particle events on a data bit. Figure 5.1.A depicts a sketch of a “0” level data bit that is part of a NRZ data stream (dark line). Superimposed on this sketch is the current pulse from an ionizing particle interacting with the photodiode (light line). In this case, the radiation event occurs during the decision time-window and with sufficient current to induce a bit error. The current transient depicted in Figure 5.1.B does not cause a bit error. The error has sufficient height, but does not occur within the decision time-window. The transient II-46

depicted in Figure 5.1.C may or may not cause a bit error. It occurs at a time that is within the decision window, but the induced current level falls in a region that will be ambiguously interpreted by the decision circuitry. The current transient depicted in Figure 5.1.D does not cause a bit error. Adding more current when interpreting a level “1” does not affect the decision made by the decision circuitry unless the charge arrives slow relative to the bit period or persists longer than the bit period and the following zero bit is corrupted. We see that the interpretation process introduces a finite probability of a “0” level bit transitioning to a “1” level when the photodiode is exposed to the space radiation environment. BER assessment must include bit errors caused by radiation events in the photodetector. More generally, any device that is in the data path and is susceptible to radiation-induced bit-errors should be included in the assessment. A bit error may or may not induce a system-level message error—system-level errors will depend on the system protocol. Bit errors that occur in devices in the data path are somewhat analogous to logic errors in memory devices, commonly known as single-event upsets. The main difference is that although the cross-section for memory bit errors has a physical geometry associated with it, the crosssection for a bit error that occurs in a clocked bit sequence is associated with not only a physical geometry, but also with a sensitive time window. The event must occur at a certain physical location and also within the decision window [Mars-94a, Reed-96]. 5.1.2 Particle-Induced Bit-Error-Ratio Testing It is desirable to collect bit error information in real time—that is, to detect, count and display the BER of an optical link as data is transmitted and detected by the link. Bit-error-ratio testing is one method of detecting bit errors in a clocked data sequence. At a minimum, a bit-error-ratio test determines the BER of an optical data link and/or the number of errors in a given time interval. (A specific test set may have other formats for displaying the data collected and more sophisticated tests may examine bus traffic at the protocol level.) For radiation characterization, we are interested in the number of errors that occur when we expose the system to a particle beam of known energy and flux. Such measurements are used to determine the error cross section while exposing each specific component of an optical data link to the test beam. 5.1.2.1 Proton-Induced SETs in Photodiodes that Cause Bit Errors In this section we describe the use of a bit-error-ratio tester to investigate the mechanisms and effects of proton-induced single-event transients in a photodiode (PD) used in an optical data link. A diagram of the experimental setup used in this investigation [Mars-94a] is shown in Figure 5.2. The instrument control and data logging were done using customized software over GPIB interface. The PseudoRandom Number (PRN) sequence generator portion of the BERT generated the data stream. The bit-stream length was (27 –1), or 127 bits. A waveform generator (not shown) operating at 200, 400 and 1000 Mbps supplied the clock signal, which could be varied continuously for broadband data-link testing. The optical attenuator regulated the power of the optical signal incident on photodiode. An optical fiber could either route the light to the photodiode or be inserted into the lightwave meter so that the incident optical power could be measured. A 3D micro-manipulator stage was used to align the optical fiber with the photodiode. An oscilloscope was used to monitor photodiode output and maximize coupling efficiency between the fiber and the photodiode. The photodiode used in these tests was an Epitaxx ETX 75 p-i-n fabricated with In0.53Ga0.47 As. All circuitry other than the photodiode, including the transimpedance amplifier, II-47

Proton BER Measurement PN Data Sequence Generator, 27-1

Sequence Recovery

0.2, 0.4, or 1 Gbps

BER Calculation Optical Data Modulator 1300 nm Laser

Proton Beam

Oscilloscope

Shield Optical Attenuator Lightwave Power Meter

Photodiode Under Test

Amplifier Clock Recovery

TIA Data Regenerator

Figure 5.2. Block diagram of BER Test set-up. [Mars-94a] was shielded from proton irradiation. The BERT performed clock and data recovery, calculated and reported the BER, and reported the number of errors that occurred during an exposure and the proportion of error free intervals. The proportion of error free intervals helps to identify single-bit errors induced by a single proton. Prior to any radiation exposure, the BER was measured to be better than 10-9. This BER is sufficient to ensure that the link will be free from errors caused by noise during the time that it takes to make an irradiation. Exposures to 63 MeV protons were carried out at Crocker Nuclear Laboratory, University of California [Mars-94, Dale-96, Mars-96]. A single exposure, with an accumulation of at least 100 errors, took a few minutes. After a proton exposure, the BER, number of errors, percent of errorfree intervals, and proton fluence were recorded. (Recall that the error cross-section is found from the ratio of the number of errors to the proton fluence.) First, let’s consider the bit-error cross-section’s dependence on optical power. (See Figure 5.3.) [Mars-94, Dale-96, Mars-96]. Here we plot the cross-section for various optical powers and a 400 Mbps operating frequency. The proton angle of incidence was 80 degrees. Notice that there are several orders of magnitude variation in cross section, depending on the optical power. The photodiode under exposure is incorporated into a broadband receiver that includes automatic gain control circuitry, and therefore an increase in the optical power sent to the detector increases the decision level of the receiver. As a result, the photodiode is less sensitive to proton-induced events that generate low current levels—so fewer protons cause errors. From this we see that one strategy for error reduction by system design would be to use brighter optical sources on the transmitter side. Another important consideration is the effect of network topology on the optical power incident on the photodiode. For example, star-coupler-based architectures often have inherently lower power loss than other topologies. The bit-error cross-section must be characterized across the entire range of optical powers that are of relevance to the system. Next, let’s consider the cross-section dependence on both proton angle of incidence and the optical power. Figure 5.4 gives the results of measurements made on the proton-induced biterror cross-section for various optical powers at four angles of incidence [Mars-94a, Mars-96].

II-48

Optical Power (µW) Error Cross-Section (cm2)

2

4

8

16

32

64 128 256 512

63 MeV Protons @ ~80º ETX75 InGaAs p-i-n 400 Mbps

10-5

10-7

10-9 σ = # Errors / Φ

10-11 -20

-10

0

Optical Power (dBm) Figure 5.3. Variation of photodiode error cross-section over optical power. [Mars-94a, Dale- 96] Optical power considerations are critical when assessing impacts of radiation-induced bit errors on BER.

Optical Power (µW)

Error Cross-Section (cm2)

10-4

2

3

5

8

12

20

31

50

80

ETX75 InGaAs p-i-n 400 Mbps

10-6

σ = # Errors / Φ 63 MeV Protons

10-8

10-10

90 deg 80 deg 70 deg 60 deg

-25

-20

-15

-10

Optical Power (dBm)

Figure 5.4. Proton-induced bit error cross-section for various optical powers and angles of incidence. [Mars-94a] These data demonstrate direct ionization mechanism for bit-errors in a photodiode.

II-49

10-3 Error Cross-Section (cm2 )

ETX75

18 MeV He Ions at 70º

10-5

10-7

10-9

10-11 -30

1000 MHz 400 MHz 200 MHz

-25

63 MeV p+ at 50º

-20

-15

-10

Optical Power (dBm)

Figure 5.5 Frequency dependence of error cross section for various ions. [Mars-94a] System operating frequencies should be considered when characterizing bit-errors. The angles are with reference to the normal to the photodiode surface. These data indicate that some of the proton-induced events are due to direct ionization, (as was also the case for the photodetectors in optocouplers, see section 4.4). Observing a few facts about the data illuminates the dynamics of the proton-induced direct-ionization effects. Notice that there is a sharp increase in cross-section with angle of incidence at high optical power. This is inconsistent with the hypothesis that indirect ionization is solely responsible for these bit errors. Probably the cross-section is some combination of direct- and indirect-ionization effects, and most likely dominated by direct-ionization effects. Also, notice that at low optical power there is very little change in the cross-section with angle of incidence. Correcting these cross-sections for angle of incidence shows that for low optical power, the cross-sections are near the geometric area of the photodiode. At this point the optical power is so low, and the receiver circuitry so sensitive that very low levels of ionization will induce a bit error. The dependence on optical power in this receiver with automatic gain control indicates an adjustment to the critical charge required for single-event transients to become bit errors. We will treat this topic in more detail in Section 5.2.2, where we discuss on-orbit BER calculations. Finally, consider Figure 5.5, which shows the photodiode error cross-section dependence on operating frequency [Mars-94a, Mars-96]. Notice that for all optical powers for both protons and He ions the error cross section is linearly dependent on data rate. In [Reed-96] error-crosssection dependence on frequency was studied in detail, the next section summarizes this work. 5.1.2.2 General Discussion of Error Cross-Section Dependence on Data Rate At this point we will digress from our discussion of photonics for a moment to discuss frequency effects in general. Frequency effects on error cross-section have been observed and studied by several authors for several types of microelectronic and photonic devices [Turf-90, Schn-92, Buch-93, Turf-94, Mars-94a, Shog-94, Mars-95c, Buch-95, Reed-96]. Some of these papers focused on complete circuits while others studied device-level effects. Some found a linear relationship with frequency, while others found this relationship to be nonlinear. Testing

II-50

2.0x10-3

Br: ~37 MeV cm2 /mg

Cross-Section (cm2)

100331 : BNL

1.6x10-3

100307 : BNL 100324 : BNL 100324 : TASCC

1.2x10-3 8x10-4 4x10-4 0

0

200

400

600

800

1000

Frequency (MHz) Figure 5.6. Cross-section data collected on several devices over various operating frequencies [Reed-96]. A linear relationship is observed for most cases except at the maximum operating frequency for the 100324. The departure from linear behavior can also be observed when testing at the maximum operating frequency of the test set-up (see Figure 5.7). to date on fiber-links has shown a linear dependence on frequency. [Reed-96] presented a discussion of the origins of frequency effects in simple microelectronic devices. These results are particularly important when evaluating microelectronic support circuitry in fiber data links. The data presented in [Reed-96] show that for a National Semiconductor voltage-level translator (100324), XOR (100307) and D flip-flop (100331) the relationship was linear until near the maximum operating frequency, Figure 5.6. The same study showed that for a proprietary microcircuit design the frequency effects were non-linear, Figure 5.7. The linear relationship with data rate is easily understood by reviewing Figure 5.1.A. Each bit is sampled for a specified duration. There is a certain time window in which the radiation event must occur in order to be counted as an error. This window is called the vulnerable time window. The duration of the decision window and the decay time of the transient event (defined by the circuit bandwidth) define the duration of vulnerable window. Increasing the frequency increases the number of vulnerable windows. The data in Figures 5.6 and 5.7 show that the cross-section increased dramatically as the data rate approached the maximum operating frequency of the test set or the device. For example, in Figure 5.6 we see that at 440MHz the measured cross-section of the level translator increases rapidly away from the linear trend. This frequency was the maximum operational frequency for the device. Another example is shown in Figure 5.7: at frequencies greater than 1 Gsps the measured cross-section increases dramatically. These devices function at frequencies greater than 1 Gsps. The maximum operating frequency of the BERT used in this test is 1.2 Gsps. These data show that a dramatic increase in error cross-section occurs when operating near the maximum operating frequency of the test set or the device being tested.

II-51

8x10 -6

Cross-Section (cm2 )

8x10 -7 6x10 -7

6x10 -6 4x10 -7 2x10 -7

4x10 -6

0

0

200

400

600

800 1000

2x10 -6

0 0

200

400 600 800 Frequency (MHz)

1000

1200

Figure 5.7. Cross-section data showing the nonlinearity of the relationship between cross-section and data rate; inset shows a blow-up of data at less than 1 GHz. [Reed-96] This result was repeatable.

1.2x10-4 National 100324 Br:37 MeVcm2/mg

Cross-Section (cm2)

1x10-4 8x10-5 6x10-5 4x10-5 2x10-5 0 0

1.0

2.0

3.0 4.0 Delay (ns)

5.0

6.0

Figure 5.8. Cross-Section as a function of delay between the sample window and the data edge. This shows greater sensitivity when sampling near data transition. [Reed-96]

II-52

Figure 5.8 shows how the error cross-section increases as the decision window, and along with that the vulnerable window, is moved closer to the data transition. The delay is the time between a data transition and the leading edge of the decision window. This suggests that the increase in sensitivity is due to a loss of noise margin close to a data transition. 5.1.3 Optical Data Link Subsystem-Level Radiation Effects Testing In order to fully qualify an optical link for use in the space radiation environment, several issues must be addressed. Some of these can be addressed at the component level, while others must be quantified at the subsystem level. Section 2.2 gives an overview of the issues that must be addressed. The list below gives an overview of how some of these effects affect device performance in general. • Total ionizing dose (TID): long term degradation of device performance, eventually leading to device failure. These effects can occur in passive optical components as well as optoelectronic and microelectronic devices. • Examples of single-event effects (SEEs): − SEL, SEGR, SEB: single cosmic ray particle event that can cause the device to stop functioning. In some cases (especially SEL) this is recoverable. These effects can occur in microelectronic devices. − SEU, SET: Single-particle induced change in the logic-state of a digital device (SEU) or a temporary change in the output state of a analog device (SET). These effects can occur in optoelectronic and microelectronic devices. • Displacement damage: long term degradation of device performance, eventually leading to device failure. These effects can occur in optoelectronic and microelectronic devices. This section deals with the topic of in situ testing of fully functional optical data links that are representative of building blocks of an optical data network with fidelity to the hardware network layer and possibly also the protocol layer. First, as an example, proton testing on a specific optical system will described. Then, we will give results from several other system-level tests of optical data links. (In Section 5.3 we compare on-orbit results of some of these digital optical data links to predictions made using ground data.) 5.1.3.1 Ground Testing of a Digital Optical Data Bus This section will describe system-level study of bit errors resulting from radiation effects. [Dale-96] The study was carried out on the STAR Fiber Optics Data Bus (FODB), which was designed by Boeing [Chap-93, Frit-94, Frit-95]. The STAR FODB is a 200 Mbps 32 node starcoupled fiber-based data bus. The star coupler is dual redundant; the data stream is Manchester encoded, and the system also allows for error checking. It uses 1300 nm multimode fiber and optoelectronics. Figure 5.9 is a block diagram of a generic application of the STAR FODB. The Fiber Bus Interface Unit (FBIU) contains a transceiver (the transmitter and receiver) and support circuitry necessary to implement the protocol and form the host interface. Each host FBIU is connected to all others via the 32x32 optical star coupler. A block diagram of the FBIU is given in Figure 5.10.

II-53

HOST SUBSYSTEM

TELEMETRY & CONTROL BUS

MIL-STD-1773 REMOTE TERMINAL

FBIU

FBIU

Transceiver

Transceiver Other FBIUs Supporting Sensors, Processors, Downlinks ...

32x32 Optical Coupler

Figure 5.9. Schematic of a STAR FODB network. [Dale-96]

TRANSCEIVER TRANSMITTER LASER DIODE ASSEMBLY

PHYSICAL LAYER

LASER DRIVER

TEMP COMP & SWITCHING

HIGH SPEED PROTOCOL CHIP

HIGH SPEED DIGITAL FORMAT

DUAL PORT MEM

DIGITAL ASIC TANK

DELAY

FILTER

ANALOG ASIC

PIN DIODE ASSEMBLY

RECEIVER

HOST INTERFACE (16 Bits)

STATION MANAGER 3051

OSC

PROM

SM SERIAL PORT

Figure 5.10. Schematic of a fiber optic bus interface (FBIU). [Dale-96]

II-54

Header

FBIU #2

FBIU #1

Transceiver

Transceiver MicroController

32x32 Optical Star Coupler

Tokens: 3-1 1-2 2-3 3-1...

72 Mbps Data Source

Header

Tokens: 3-1...

... 251 Word Message

Gaps

251 Word Message

FRAME (55.78 microsec.)

20.60 microsec

(B) 200 Mbps STAR FODB Test Configuration

Transceiver

i1

PROTONS

ith

FBIU #3

(A)

i0

Time

Bit error can cause a message error

(C) Figure 5.11. (A) Configuration of three FBIUs during proton system testing. (B) Message packet used during test. (C) SET induced bit error that can lead to a message error. [Adapted from Dale-96] A detailed description of the FBIU can be found in [Frit-94]. Some of the key components are listed here. The photodiode is a 1300 nm laser diode, and the photodetector is the Epitaxx EXT75 p-i-n diode. The data formatter is a high speed GaAs ASIC that formats and deformats, encodes and decodes, and performs error checking. The protocol chip is a CMOS ASIC. The station manager is a CMOS micro-controller. The dual port memory is 3.3 V CMOS RAM. Clearly, there are several different issues that should be addressed when qualifying this hardware for use in the space radiation environment. Each device should be reviewed for its performance degradation and SEU sensitivities. Beyond that it is important to perform subsystem-level testing and analyses to understand how bit errors are generated, propagated and realized at the subsystem level to form message errors. Here we will describe some of the results from a proton-induced-bit-error-characterization study intended to assess proton-environment impacts at various test levels. A depiction of a subsystem-level proton test on the STAR FODB is shown in Figure 5.11.A. The test-network configuration included three FBIUs, one of which was selected for irradiation. Complete testing of the system protocol (including recovery of the bus in response to the dropout of an FBIU) required that at least three FBIUs be used. Errors were defined by the bus protocol [Chap-93, Frit-94]. Error logging during the irradiations was automatic. Figure 5.11.B depicts the message transfers used for these tests. Each frame had a 251 word long message, with each word having 16 bits. When messages were not being sent, a sequence of tokens (packets of information used to maintain network traffic according to the network protocol) was passed around the network. For these tests, a message error was defined as any loss of data— that is, any event that changed a bit in the message packet Figure 5.11.C, so all bit errors resulted in message errors. Errors in tokens were not counted as message errors. A sequence of exposures was carried out for each component in the FBIU [Dale-96]. The measured error cross-sections for the entire transceiver are given in Figure 5.12 as the triangles. Comparing this to results obtained on the photodiode module (upside down triangles) shows that the vast majority of the events are due to events in the photodiode. The squares in this figure

II-55

Error Cross-Section (cm 2)

Optical Power (µW) 10-5

2

4

8

16

32

64 128 256 512 63 MeV Protons @ 80º ETX75 InGaAs p-i-n 400 Mbps

10-7 10-9 σ = # Errors / Φ

10-11

Entire Transceiver (200 Mbps Manchester) PD Module Only (200 Mbps Manchester) Present ETX75 PD Work (400 Mbps PRN) Previous ETX75 PD Work (400 Mbps PRN)

10-13 -20

-10

0

Optical Power (dBm) Figure 5.12. Comparison of message-error cross-section to error cross-section of photodiode. [Dale-96] Note that a majority of the errors are due to transients in the photodiode. give the error cross-section measured on the photodiode from the generic BER tests, Figure 5.3. In this case the message errors measured at the subsystem level are well defined by the generic bit error (or BER) test results on the photodiode. This is not surprising since the receiver circuit used in the “generic” tests used some of the same components as the subsystem design, and more importantly, the components were used in similar ways in terms of detection concept, receiver sensitivity and bandwidth. These same components, if used in a significantly different receiver design, would exhibit different error-cross-section characteristics. Therefore, it is important to assure good component performance, but also to validate that performance with high-confidence models or particle radiation testing at the subsystem level. Results from proton exposures of other components are detailed in [Dale-96]. Here we present a summary of the results. A small number of errors occurred when the protocol chip and the data formatter were exposed to more than 20krad(Si) of protons. No errors occurred during exposures of the station manager, which was a hardened micro-controller responsible for bus initialization, fault detection, and isolation. For this system configuration, TID-induced device degradation of the commercial dual port memory was a limiting factor for certain spaceflight missions. This was expected, and the testing was useful for determining the suitability of the present hardware for various missions as well as indicating a path for hardening the subsystem for more demanding missions. 5.1.3.1 Other Ground Test Results Several subsystem-level tests of other fiber-optic data links have also been performed [LaBe91a, LaBe-92a, LaBe-92b, LaBe-93a, LaBe-93c, LaBe-94a, LaBe-94b, Mars-95b, Frit-95, LaBe96b, Mars-96, Cart-97, LaBe-98c]. The reader is encouraged to review these publications which give detailed discussions of the test setup and results. We give examples some of the these: Ø Transients in photodiodes dominate system BER in most cases [Mars-96, Dale-96, Labe-98d, and references therein]. II-56

Ø In [LaBe-94a, LaBe-94b] the authors describe system-level proton SEE test results on a data bus that uses the AT&T ODL200 transmitter and receiver and the Hot Rod protocol chip set from Gazelle Microcircuits. The test results showed that the microwave complementary bipolar integrated circuit support devices dominated the message-error sensitivity of the ODL200. Test results also showed that SEUs in the Hot Rod protocol chip set would dominated the BER. These results are somewhat unusual, given that for most systems tested, the photodiode is the major cause of biterrors that lead to message errors. This example shows how SEU in support circuitry can dominate BER. Ø Some bit-error cross-sections can depend on particle flux and beam structure—for example, in systems that utilize message retries [Dale-92, LaBe-93a, LaBe-93c]. Ø Proton-Induced bit-error testing on a Fiber Channel (FC) transmitter and receiver is described in [Cart-97]. FC is an ANSI-standard for a high-performance serial data link. It includes a mix of point-to-point, network, and active intelligent interconnection topologies. Devices from AT&T and Force, Inc. were evaluated. The optical components in the AT&T devices where reasonably tolerant to bit-error effects, but SEUs in the link’s logic portion interrupted operation and required a link re-initialization. Results on the Force devices showed them to also be tolerant to biterror effects. However, during irradiation the transmitter experienced a repeatable unexplained high current condition. The results of these fiber-optic-data-link radiation tests show the need to carefully characterize each component of the fiber-optic data link for appropriate SEE, TID and displacement damage effects (see Section 3). Some of these effects should be characterized at the subsystem level, while others can be characterized at the component level. Any device that can impact performance by causing system message errors should be well characterized under a system-like environment. The on-orbit message error rate will depend on the system protocol, message packet size, message encoding, error checking and correction, and other system and sub-system level characteristics. Similar considerations apply for BER for data link traffic below the protocol layer. We should stress here that each in-situ test poses specific challenges. The test team should carefully weigh each segment of the test plan based on the requirements and functionality of the system or subsystem under test. Some important considerations are listed below: • Determine limitations when performing subsystem-level heavy ion and proton testing. For example, packaging may limit heavy-ion bit-error testing and TID may limit proton testing. • The system being tested should simulate actual spacecraft intra-satellite optical data link system operation, so that the test will determine when bit errors propagate through the system and cause message errors. • Consider how mitigation approaches may impact testing and results. • Individually expose each component of the optical data link during system level SEE testing. • Consider whether component-level TID testing is needed to quantify the risks of parametric and functional failure of support ASICs, RAM or other support devices.

II-57

Source power Insertion Loss (e.g., coupler) Power into fiber Fiber length loss

0 -5 Power (dBm)

Power Launched on detector -10 -15

Pre-flight Link Margin before considering radiation effects

-20 Required detector power Aging allowance Receiver coupling loss Required detector power for BER

-25 -30 -35

0

1 2 Time (Arbitrary Units)

3

4

Figure 5.13. Plot of pre-launch system power budget (after Figure 3.14). The abscissa is now mission duration as opposed to fiber length. • Consider whether component-level SEE testing for SEL, heavy-ion-induced bit errors or other effects is needed. • Characterize the network over the full range of optical powers relevant to the network. (Optical power incident on photodiodes critically affects BER.) • Consider the effects of the orientation of the photodiode in the receiver package. (This can dramatically affect SEE test results and should be explored to gain meaningful information for error rate predictions. • Consider the effects of radiation-induced darkening in passive optical components during the test. (Such degradation can decrease the optical power incident on the photodiode and must be distinguished from loss of responsivity in the photodiode itself.) 5.2 Assessing Radiation Effects on Optical Digital Data Link Performance Metrics Assessing performance of an optical data link for the space radiation environment requires consideration of the effects of TID, displacement damage, and SEE on the optical digital data link performance metrics described in Section 3.3. General considerations for system-level assessment of radiation effects on spaceflight hardware has been described in [LaBe-96a, Gate96, LaBe-98b, Kinn-98, Heid-99]. In this section we discuss the issues that are specific to fiberoptic data links. First we consider how TID and displacement damage impact the power budget. Next, we discuss the techniques for predicting on-orbit BER, and the impact of power budget on BER. Lastly, a brief discussion of approaches for bit-error mitigation is then given. 5.2.1 TID and Displacement Damage Impacts on Power Budget A power budget analysis for an optical-fiber link was discussed in Section 3.3. Such an analysis of the pre-launch power-budget margin for the optical link can also be conducted for radiation effects (see Figure 5.13). The abscissa is relabeled in arbitrary units of time (instead of II-58

0

Power (dBm)

-5

Idealized TID and DDD degradation of transmitter, fiber, lens, etc... Effective TID and DDD degradation of receiver

-10 -15 -20

Actual EOL Link Margin

-25 -30 -35

Effective power requirement to maintain detector output current

Mission duration 0

1

2

3

4

Time (Arbitrary Units)

Figure 5.14 Idealized radiation induced degradation of power -budget margin. When the power incident on the detector falls below the bottom line the system BER will be above the specification limits. length as in Figure 3.17). As a fiber-based system is exposed to the space radiation environment, the power budget margin will decrease as a result of several effects. Figure 5.14 shows graphically how these effects impact link margin (note that this figure is a cartoon and is for demonstration purposes only). The slope of the top line represents idealized TID and displacement damage effects as the mission time increases. This line takes into account effects such as degradation of source optical power and the signal as it passes through the fiber and optics components such as lenses. The slope of bottom line represents idealized TID and displacement damage impacts on the optical power needed to maintain the required system BER. The upward slope reflects, for example, displacement damage degradation of the photodiode responsivity. For the hypothetical mission life marked on the time axis, the power incident on the detector at the end of the mission life will be –12 dBm. The power requirement for maintaining system BER is –23dBm. So, the net EOL link margin is 11 dB. A power-budgetmargin analysis must be preformed that considers radiation-induced degradation effects to ensure sufficient EOL margin to maintain a functioning optical link. 5.2.2 Impacts of Radiation-Induced Bit Errors on On-Orbit BER Traditional proton-induced on-orbit error rate calculations [Pete-97] assume indirect ionization to be the dominant mechanism for upset, and so they proceed as follows. The error cross-section is measured at various proton energies. These data are then fit using Bendel curves. The error rate is computed by direct integration of the on-orbit flux-energy curves with the Bendel curve-fit to the error cross-section data. Traditional heavy-ion-induced error-rate calculations [Pete-97] assume that direct ionization is the dominant mechanism for upset. For these calculations, the error cross-section is measured at various effective LETs. The error-rate calculation assumes that the sensitive volume is a right rectangular parallelepiped. One then fits the graph of cross-section vs. LET to a Weibull distribution and calculates the error rate by convoluting this function with the LET spectra appropriate for the mission environment.

II-59

Error Cross-Section (cm 2)

10 -3 Protons

Alphas

10 -5

10 -7

Optical Power -25 dBm -15 dBm

10 -9 10

100 LET in InGaAs (MeV * cm 2/g)

1000

Figure 5.15. Plot of the 400 Mbps data in Figures 5.4 and 5.5 using conventional techniques to correct LET and cross-section for beam angle. [Adapted from Mars-94a] All traditional approaches for computing on-orbit error rates separate the contributions due to direct and indirect ionization effects. Although such techniques are adequate for predicting error rates for most microelectronic devices used in support circuitry for fiber optic data links, separate consideration of direct and indirect mechanisms is not the best approach for predicting protoninduced error rates for optical data links where bit errors occur due to direct ionization from protons in photodiodes. 5.2.2.1 Proton-Induced Transients in Photodetectors and On-Orbit BER Direct ionization by protons striking photodiodes can be the dominant source of single-event induced bit errors in digital optical data links (see Section 5.1.2.1). As discussed above, traditional proton SEE rate prediction approaches that only consider indirect ionization effects are inadequate for links that are susceptible to errors caused by proton direct-ionization. References [Mars-94a, Mars-96, Mars-99] show that for fiber-link applications that use photodiodes that are very susceptible to SETs due to proton direct ionization, bit error data can be analyzed in terms of particle LET. The authors use standard methods from heavy-ion upset calculations to account for geometric effects of changes in incident particle angle, including: a) using “effective LET” to account for the increased charge deposited in the sensitive volume particles at grazing incidence, and b) the decreased cross-sectional area projected by the sensitive volume as the particle angle of incidence increases. The resulting plot presents convincing evidence that it is possible to understand the data using the conventional methods of fitting a Weibull distribution to a plot of error cross section vs. effective LET (see Figure 5.15). The plot shows the cross section for a bit error while operating the link at 400 Mbps for various values of effective LET in InGaAs (note units are MeV cm2 / g). The data for LETs < 100 MeV * cm2 / g are from proton exposures, while data for LET > 100 MeV * cm2 / g are from He exposures. Predicting the proton-induced on-orbit bit-error rates using data like that in Figure 5.15 was demonstrated in [Mars-94a, Mars-96]. The approach was to determine the best Weibull fit to the II-60

BER (errors/bit)

10-8 10-9

10-10

10-11

10-12 -20

-10

0

Optical Power (dBm)

Figure 5.16. Predicted on-orbit BER during solar particle event at various optical powers [Adapted from Mars-94a] data and use the standard tools of rate calculation for heavy-ion induced SEE to calculate the proton-induced error rate in the photodiode. Figure 5.16 shows the predictions of such calculations for various optical powers for a 400 Mbps application during a solar particle event. (Recall that the error cross-section data vary with data rate, so any error rate calculation on any clocked device must consider the application operating frequency [Mars-94a, Reed-96].) This type of calculation assumes that bit errors are dominated by direct ionization effects at all angles. 5.2.2.1.1 Methods for Predicting On-Orbit SETs for Mixed Direct and Indirect Effects In the paragraphs above we address the specific case where proton-induced direct ionization is the dominant mechanism for SETs in photodetectors (this has been the case for most high bandwidth fiber-optic data links.) SETs in other optical data link configurations can be a mixture of direct and indirect effects. One specific case is high-speed optocouplers, which can be thought of as simplified optical data links. Accurate predictions of SETs in a photodetectors like those used in optocouplers, must include a combination of direct and indirect effects [LaBe-97, Reed-98, John-99b]. Characterization of SET cross section of any data link over angle of incidence will reveal which mechanism (direct ionization or indirect ionization) is dominant. In [LaBe-97], the authors suggest using a combination of the existing traditional methods for predicting SETs in for optocouplers. Ground data is used to determine the angle where direct ionization effects begin to dominate the cross-section. One then applies traditional approaches for direct ionization effects for angles greater than this cutoff and uses indirect-ionization methods indirect effects for angles less than this cutoff. In [John-99b] an empirical approach was suggested for photodetectors used in optocouplers. This approach requires data to be collected at several angles of incidence and at several proton energies. One then integrates these data over the proper on-orbit proton spectra. The first step is to integrate the cross-section over all angles at each energy. Angular data like that in Figure 4.5 must be collected at all energies and then integrated over all angles to arrive at an effective crosssection curve like that in Figure 5.17. The next step is to integrate this effective cross-section with the space proton environment. II-61

Cross Section (cm2 )

10-5

Revised Cross Section taking angle dependence into account

10-6

10-7 Cross Section without angle dependence

10-8 0

20

40

60

80

100

Proton Energy (MeV)

Figure 5.17. Effective cross-section—including the angular distribution like that in Figure 4.5— for various proton energies on the HP6N134. [John-99] Note : A detailed formulation of a model for computing proton-induced transients in photodetectors or photodiodes that incorporates direct and indirect ionization effects, i.e. directly incorporates angular effects, has not been published to date. The reader should keep in mind that not all device-level effects cause system level errors. 5.2.2.2 Optical Link On-Orbit BER and Message Error Considerations The previous sections demonstrate that the BER of an optical data link decreases as the optical power incident on the photodiode increases in links that employ automatic gain adjustment. Figure 5.18 shows a hypothetical example of this effect. In this plot, we have assumed a bit-error rate like that in Figure 5.16 and 2) that the optical power incident on an ideal photodiode degrades as in the top curve in Figure 5.14. (We note that the data in Figure 5.14 is idealized and does not represent a quantitative analysis of the relative contribution of each link component to performance degradation.) The radiation environment was assumed to be typical of an anomalously large solar proton event. To accentuate the role of incident optical power, we have also assumed constant receiver response (no degradation) from launch to mission EOL. Assuming an initial optical power of 0 dBm yields the curve traced by the triangles, the squares represent the results for the same system assuming an additional initial source optical power of just 5 dBm. Operating the source output 5 dB higher decreases the radiation-induced BER. Another consideration is the complex problem of computing the on-orbit rate for message errors from ground based data. These calculations are dependent on the network protocol and how message errors are defined. This calculation may be as simple as collecting in situ message error data on each sensitive component, computing the message error rates for each and then summing them up to get and aggregate rate. In most cases it is not as simple as that, Section 5.3 suggests references to find information on this type of calculation for a optical link network that resends failed messages. II-62

BER (errors/bit)

10-09

10-10

10-11 Initial Source Power = 0 dBm Initial Source Power = 5 dBm

10-12

0

1

2

3

4

5

6

Time (Arbitrary Units) Figure 5.18. SEE-induced bit errors can be reduced by increasing link margin. This figure illustrates such a reduction for the case where the source power has been increased. These data also demonstrate how BER might change over mission life. These data are based on the idealized curve in Figure 5.14. 5.2.3 Bit Error Mitigation Approaches There are many ways to mitigate bit errors—at the component level as well as the subsystem and system levels. Such mitigation techniques almost always involve tradeoffs between performance and error mitigation. We direct the reader to the several references to fiber data link systems given above for a detailed description of mitigation schemes. As with any system, computing the net on-orbit error rate can be complicated by the use of error mitigation techniques. Some examples are described here: • Reduce detector depletion layer thickness. This decreases the target size and limits charge collection. Indirect band gap (Si) diodes have thicker depletion regions than do direct band gap (III-V) diodes [Mars-96]. • Reduce the photodiode area. Again this reduces the target size, but it also reduces the received optical power. As such, the amount of area reduction possible may be limited. • Over-sample and compare single data bits. Over-sampling is not possible at high data rates where the duration of the transient event is on the same order as a bit period [Thel-94]. • Use encoding and parity checking to detect errors and enable retry commands for error correction. The impact is effectively a 50% loss of bandwidth. This technique also requires buffer memory and a handshaking architecture that can be difficult to implement in high data rate systems [LaBe-93c].

II-63

• Increase the optical power incident on the photodiode. This decreases the photodiode’s sensitivity to particle-induced ionization, but must be weighed against other system considerations such as power budget [Dale-96]. • Use multi-level filtering and receiver design approaches such as those used in Boeing’s implementation of the AS1773 [Bone-96]. (See the next section for a more complete description of this mitigation approach.) • Use good system engineering practices for handling SEEs at the system level. This will minimize the effects of device errors on the BER, and perhaps even more important, applying these techniques will minimize the risk that TID, DDD, or SEE will cause degradation in overall spacecraft performance. Several papers and short courses address this topic for general applications [LaBe-96a, Gate-96, LaBe-98b, Kinn-98, Heid-99]. 5.3 Ground Testing Results and On-Orbit Use of Selected Digital Optical Data Links Several optical components and optical links have been used or tested in spaceflight hardware. The Long Duration Exposure Facility (LDEF), launched in 1978, [Tayl-91b, John-92] was the first experiment to examine fiber-optics in space. LDEF orbited for 5 ¾ years and then was retrieved. The dose levels varied from 200 to 25,000 rads. No significant attenuation was observed for any of four data links, each operating at 830nm. In 1993 Boeing developed the Photonics Space Experiment (PSE) [Cros-93, Frit-93]. The PSE was a five-segment set of experiments designed to measure the performance of selected fiber optic data links and certain components over a 5 year mission. It contained experiments that measured the long term TID degradation of optical fibers and passive components. The fibers included step and graded index multimode fibers designed to operate at 850 and 1300nm. It also studied the on-orbit stability of a laser diode and selected LEDs. Finally, the first on-orbit BER test of a full function MIL-STD 1773 was performed. Preliminary on-orbit results were presented in [Frit-96]. The results favored using these components and data links in space systems. The NASA Small Explorer Program’s implementation of a fiber-optic data bus, named Small Explorer Data System (SEDS), was used to transfer telemetry and commands between subsystems. The standards community identifies the SEDS fiber-optic system structure as MILSTD-1773. This 1 MHz star bus is a slave/master configuration that has two sides, side A and the redundant side B. The messages are Manchester encoded and can be up to 32 words long. In 1995 the standard was revised to include data transmission at either 1 Mbps or 20 Mbps. The new standard was renamed AS-1773 [A-1773]. The functional requirements of AS-1773 remained the same as MIL-STD-1773. A detailed discussion of the bus protocol can be found in [A-1773]. One notable error detection and correction feature for this standard is that after every received message the receiving node sends a short return message to the master identifying the status of the received message. The status message tells the master to retransmit the message if the transfer was unsuccessful. Retransmission of a message pack can be suppressed. Two generations of the MIL-STD-1773 SEDS data bus have been developed. Major changes to the link from one generation to the other were mechanical and electrical, but the same optical components were used in both generations. The first generation SEDS (SEDSI) data bus was used on NASA’s Solar Anomalous Magnetospheric Particle Explorer (SAMPEX). SAMPEX was launched in July 1992. The SEDSI radiation test effort is described in [LaBe-93a, LaBe-

II-64

20

16 14 12 10 8 6

Points are average per day of 6 month intervals Predicted retry rate 18 retries/day

4 2

1996/2

1996/1

1995/2

1995/1

1994/2

1994/1

1993/2

1993/1

0 1992/2

Average Number of Retries per Day

18

Year/1st or 2nd half

Figure 5.19. Comparison of predicted retry rate to the measured value for SEDSI on SAMPEX. [LaBe-97c] 93c]. The second generation of SEDS (SEDII) was used on the Hubble Space Telescope (HST) Solid State Recorder (SSR). The SSR was installed on HST in February 1997. A comparison of the radiation performance of the first and second generation SEDS data bus was given in [LaBe98c]. The data of [LaBel93c, LaBel98c] show that the photodiode is the sensitive component to radiation-induced message errors. This work compared on-obit message error data to predictions made from ground data for proton- induced errors. The ground results were identical for the two generations. (This is not unexpected because the same optical components were used in both generations.) The operation of SEDSI on SAMPEX has been successful. No bus outages or failed messages have occurred. However, there have been several corrupted data messages that required a retransmission of the message for correction. Typical numbers of message retransmissions are shown in Figure 5.19. This figure also shows the predicted number of retransmissions. Also as expected, most of the retransmissions have occurred in the SAA. The number of bus SSR SEDSII retransmissions per day for the first two months of operation is plotted in Figure 5.20. SEDSII on HST is fully operational. No bus outages have occurred. Errors requiring retransmission have occurred with a frequency of about 1 every 2 days. These errors are believed to result from software or hardware error rather than single-event effects. An error cross-section calculation and on-orbit prediction approach for multiple retransmissions based on a binomial distribution for the retry probability and the cross section for single message errors was developed in [LaBe-93c, and revisited in LaBe-98c]. A space experiment based on AS-1773 using a Boeing transceiver [Bone-96], has been developed, radiation tested in the lab, and flown on board the Microelectronics and Photonics TestBed (MPTB) [LaBe-96b, Jack-96]. The goal of the experiment was to fully simulate the AS-1773 spacecraft bus. We note that operating frequency was fixed at 20 Msps. The experiment was reconfigurable between two modes of operation: transmission could be reversed so that data flowed between two subsystems in either direction—but not simultaneously. The

II-65

60

Retries per day Average Retries 35, 96

Number of Retries

50 40 30 20 10

97

93

89

85

81

77

73

69

65

61

57

53

0

Julian Day of Year

Figure 5.20 Comparison of predicted retry rate to the measured value for SEDSII on HST. [LaBe-97c] Boeing transceiver utilized 1300 nm InGaAs photodetector and GaAs LED, the design employed several mitigation techniques: • A small volume photodetector was used to minimize charge collection. • The analog preamplifier circuitry had excess drive to minimize the pulse widths of SETs • It utilized a proprietary asynchronous digital data reconstruction to filter narrow-width SETs produced in the photodiode and the preamplifier. • The transceiver electronics were fabricated in a radiation hardened, 1.2 µm, thin epilayer CMOS process. • The system used a proprietary clock-recovery architecture. • The optical fiber used is radiation-tolerant. Ground testing showed that the SETs in the photodiode would be the most sensitive source of radiation-induced message-errors [LaBe-96b]. The data link has seen very few errors [Jack-99]. This is mainly due to the decreased size of the photodiode. (We should note that the link has been exposed to a relatively low proton fluence while operating in MPTB host spacecraft’s highly elliptical orbit.) These on-orbit results confirm the ground results, which demonstrated the effectiveness of the hardening solution. Although on-orbit bit errors are occurring during one of the modes of operation, these are hardware related and are not radiation induced. The on-orbit robustness of these digital optical data links—confirmed by ground testing and modeling—shows that intra-satellite digital communications can be achieved with little or no radiation-induced performance degradation.

II-66

6.0 Radiation Effects in Emerging Optical and Optoelectronic Technology This section will focus on recent radiation-effects testing of technology that could eventually be used in digital optical data transfer system in spaceflight applications. The intent is not to be exhaustive, but rather to give the reader a sense of radiation-effects issues for the state-of-the-art technology. A glance at these technologies and the associated radiation-effects issues will show that the future for high-speed (>1Gsps) optical data links in spaceflight application is very promising. However, the path toward this goal will not be without its challenges, from the perspective of design as well as that of radiation-effects issues. SEU and SET effects are certain to be among the more complex issues that engineers and radiation-effects researchers will face in implementing these technologies into spaceflight hardware. Ø [LaBe-98a] listed vertical cavity surface emitting lasers and Metal-Semiconductor-Metal (MSM) photodetectors as two technologies that have strong potential for use in spaceflight highspeed optical data links. VCSELs are high-speed, small-scale (< 10 µm diameter) top emitting lasers. These devices can be fabricated to form an array of light emitters with each individual device having very small beam divergence. An account of testing for proton-induced radiation effects in VCSELs can be found in [Paxt-97, Barn-99]. These data show that VCSELs demonstrate excellent performance for most spaceflight applications. More detail on VCSELs can be found in segment II of this Short Course. Ø MSMs are high-speed (10s-100s GHz) photodetectors. Fabrication techniques used in making MSMs vary greatly. They use Schottky barriers with semiconductor sandwiched between [Mars-97] the interdigitized electrodes, see Figure 6.1. Like VCSELs, they are smallscale and can be fabricated to form arrays. Proton-induced bit-error tests on MSMs [Mars-97] showed the same trends for SETs over angle and optical power as those shown in photodiodes (see Section 4.4). However, the acceptance angle for direct ionization effects is much narrower than that for p- i-n diodes, and the MSM cross section has a stronger dependence on optical power. This leads to the possibility that MSM could be fabricated to be much less sensitive to radiation induced increases in BER than are p- i-n diodes. The MSM used in this study proved to be robust to TID and DDD effects. Ø Several commercial vendors have developed fiber-optic transceivers. A world wide web search on May 25, 2000 for “fiber-optic transceiver” uncovered over 1200 pages. The list below is a subset of the list of vendors. • Telebyte, Inc—http://www.telebyteusa.com/index.htm • TC Communications—http://www.tccomm.com/LAN.htm • 3Com—http://www.3com.com • Agilent Technologies (formerly HP)—http://www.agilent.com/ • Lasermate Corporation— http://www.lasermate.com/ Some of these transceivers may use technologies that are very robust to the space radiation environment. Careful radiation effects characterization would need to be carried out to evaluate each transceiver’s susceptibility to TID, DDD, and SEE. Results from proton-induced SEE testing on two of these links show the trends similar to those described in Section 5 of this Short Course [OBry-00]. Ø Wavelength-Division Multiplexers (WDMs) combine the signals transmitted from several sources—each source operating at a different wavelength—pass the signal through a single fiber and then separate out the various signals by wavelength (see Figure 6.2). There are several II-67

methods of combining and separating an optical signal according to its wavelength. [See for example Agra-97.] The methods vary from those using passive components like lenses to those using a combination of active and passive components—such as those in Mach-Zehnder interferometers. (Recall that Mach-Zehnder interferometry applies an electric field to an optical material to shift its refractive index.) Radiation-effects testing on a fused biconical-taper WDM was described in [Guti-94]. They found that radiation had very little effect on optical power loss of the signal passing through the WDM. However, there was a small, but not insignificant increase in the individual channel signal loss to other channels. Polymer optical waveguides are another promising technology. Polymer traces can be placed on flexible sheets of materials such as Kapton. These can then be laminated onto multi-layer boards using standard lamination processes. Typical waveguide dimensions are 1 mil square. Polymer optical waveguides can be used to transfer light from one side of a printed circuit board to another (see Figure 6.3). One obvious application of this technology is in WDMs. Radiationeffects testing has been done for some polymer waveguides [VanE-96]. Ø Erbium-Doped Fiber Amplifiers (EDFAs) are passive optical power amplifiers. Optical power gain can be achieved as light travels through an Er doped silica-based fiber [Desu-93]. There are several potential applications for this technology in spaceflight hardware—use in WDMs being one. EDFAs are known to be much more sensitive to radiation than classical silica fibers. [Will-98, Hens-98 and references therein] offer excellent reviews of radiation effects in EDFAs. One issue of great importance for use of these fibers in space-based systems is whether one can predict their performance in the space radiation environment from ground data. [Will98] carefully addresses this issue. Ø High speed (> 1 GHz) support electronics are required for Gbps data rates and beyond. GaAs MESFET and bipolar emitter-coupled logic (ECL) are mature technologies that can operate in the GHz regime. Other III-V and Si semiconductor technologies are currently expected to reach speeds much greater than 1 GHz. TID testing on GaAs, ECL, SiGe, and CMOS SOI has proven these technologies to be sufficiently robust for the space environment TID effects. SEU testing at high data-rates (>1GHz) on ECL, GaAs, and SiGe technologies has shown them to be sensitive to radiation-induced bit errors, which would add to the BER. Growing GaAs with a buried layer at low temperatures has been shown to be effective at mitigating SEUs [Mars-95a]. We direct the reader to some of the more recent discussions of the radiation tolerance of these technologies: ECL [Reed-96, O’Bry-00, and references therein], CMOS SOI [Brot-97], SiGe [Mars-00], and GaAs [Mars-95c, Weat-97, and references therein]; these are just starting points, we suggest a the reader review more recent IEEE TNS for up to date information on the radiation tollerance of these emerging semiconductor microcircuits.

II-68

V

Metal

Semiconductor Figure 6.1 Cartoon of a MSM photodetector showing the interdigitized electrodes.

Tx

λ1

λ1

λ2

λ2

Tx

Rx Rx

Optical Fiber Tx

λN

λN

Multiplexer

Rx

Demultiplexer

Figure 6.2. Block diagram of a WDM. Individual subsystem’s transmit (Tx) data at different wavelengths. Each receiver (Rx) is tuned to accept data at a defined wavelength. Some WDM use wavelength tunable components.

Figure 6.3. A polymer waveguide trace laid on a PCB configured in a Mach-Zehnder interferometer.

II-69

7.0 Summary and Conclusions We began this segment of the Short Course with a brief overview of the space radiation environment, summarizing the basic space radiation effects important for microelectronics and photonics and giving an example of a typical mission’s radiation environment requirements. We then gave an overview of intra-satellite digital optical data link systems, including a general discussion of the digital optical data link, some link components, and a brief description of optical link power budget and bit error ratio. The subsequent discussion of radiation effects in optical and optoelectronic components focused on degradation of passive optical components and SETs in photodiodes. We briefly mentioned DDD degradation of source output power and laser turn-on threshold. We discussed in some detail the characterization of TID induced signal attenuation resulting from darkening in passive transmission medium. (Proper choice of fiber can largely resolve this issue.) Radiationinduced transients in photodetectors were seen to pose a significant concern for data link performance in space missions. DDD induced degradation of detector output current and TID and DDD related increases in background noise were seen to be significantly less of a problem. We noted that radiation effects in support circuitry—though not covered in this segment of the short course—could also threaten data-link performance—or even data-link survival. Significant threats included: TID and DDD parametric and functional failure, SEUs in logic, SEL in CMOS, and SETs in linear devices. Proper design can resolve most, if not all these issues. From device effects, we went on to consider the system response of optical data links to the space radiation environment. First a discussion of system-level SEE ground testing was given that stressed the importance of understanding the contributions of bit errors in all devices in the data path to the system BER. A clear understanding of the system response to radiation induced errors is critical. The system protocol and employed mitigation approaches determine when and how radiation induced errors impact the system functionality. Simple component level testing cannot illuminate these issues. A carefully designed test plan will simulate the way spacecraft subsystems use the optical data link. Note that this is also true for data links that utilize an electronic physical layer. SETs due to proton-induced direct ionization in p- i-n photodiodes and phototransistors were seen to pose specific challenges. Proper characterization was seen to demand that cross-section measurements be made of over the entire system phase space of beam angle, optical power and operating frequency. The potential frequency dependence of bit errors in other support circuitry was emphasized. Finally, we reemphasize that without in-situ testing to determine the bit-error ratio and the multiple-error ratio, one cannot determine how device level effects impact system-level performance (with protocol layer), that is, whether or not a particular device effect translates into a system effect. The discussion of ground-based, system-level testing was followed by a discussion of system-level assessment of data-link performance in the space-radiation environment, emphasizing the importance of managing TID and DDD effects in sources and passive transmission media. We emphasized that these degradation effects could be avoided with proper device selection and/or mitigated by proper management of the system’s optical power budget. We also saw that predicting on-orbit, radiation-induced BER in support electronics is possible using standard on-orbit-error-rate calculation tools, but that a modified approach is needed to account for the contribution by SETs in photodiodes.

II-70

Lastly, we looked at some state-of-the-art technologies and found that the prospects for highdata-rate (>1 GHz) fiber-optic data links looks to be excellent in the near term. The bottom line is: Designers must consider radiation effects when making decisions about optical data transfer system configuration and protocol, hardware design (especially the receiver) and device selection. A proper hardness assurance program at the component level, careful electrical, optical, and optoelectrical designs, and well thought-out system protocol will result in a qualified fiber optic data link for use in the space radiation environment.

II-71

8.0 Acknowledgments We would like to thank each person in the Radiation Effects and Analysis Group and the Radiation Physics Office at NASA Goddard Space Flight Center for their support. Specifically we would like to thank Paul Marshall, Cheryl Marshall, Janet Barth and Ken LaBel for their insightful technical discussions, guidance and careful review of this work. This course would not have been possible had they not been willing to let us tap into their vast knowledge of this subject matter. Thank you. We would also like to thank Martha O’Bryan for her support in formating the graphics and text in the written and oral presentation of this Short Course. We would also like to thank Janet Jew and Christian Povey for their excellent review of this segment of the Short Course. Like most of our adventures, we would not have completed this one had it not be for the forgiving support of our families.

II-72

(All References are unclassified)

9.0 References [A-1773] [Agra-97] [Barn-84] [Barn-86] [Barn-90]

[Barn-99]

[Bart-97] [Bart-98] [Bone-96]

[Bosc-99]

[Bris-93]

[Brot-97]

[Buch-93]

[Buch-95]

[Cart-97]

[Chap-93]

“Fiber optic mechanization of a time division multiplex data bus,” AS-1773, The Society of Automotive Engineers (SAE), 1995. G.P. Agrawal, Fiber-Optic Communication Systems, 2nd edition, New York, Wiley, c1997. C.E. Barnes and J.J. Wiczer, “Radiation effects in optoelectronic devices,” Sandia Report, SAND84-0771, 1984. C.E. Barnes, “Radiation hardened optoelectronic components: sources,” Proc. SPIE, vol. 616, pp. 248-252, 1986. C. Barnes, L. Dorsky, A. Johnston, L. Bergman, and E. Stassinopoulos, “Overview of fiber optics in the natural space environment,” in Fiber Optics Reliability: Benign and Adverse Environments IV, Proc. SPIE, 1990, vol. 1366, pp. 9-16. C. Barnes, G. Swift, S. Guertin, J. Schwank, M. Armendariz, G. Hash, K. Choquette, “Proton irradiation effects in oxide-confined certicle cavity surface emitting laser (VCSEL) diodes,” to be published in the Procs. of 1999 RADECS. J. Barth, “Modeling space radiation environments,” Notes from 1997 IEEE Nuclear and Space Radiation Effects Conference Short Course. Private comunications, 1998. R.K. Bonebright, J.H. Kim, R.A. Hughes, J.W. Clement, T.M. Bocek, E.Y. Chan, J.H. Nitardy, and C. Hong, “Development of dual-rate MIL-STD-1773A data bus transceiver,” ion 1773 DR SPIE D. Boscher, S. Bourdarie, A. Vacaresse, “Contribution of physical modeling to engineering models of the radiation belts,” to be published in the Procs. of 1999 RADECS. J. Bristow and J. Lehman, “Component tradeoffs and technology breakpoints for a 50 Mbps to 3.2 Gbps fiber optic data bus for space applications,” Proc. SPIE, vol. 1953, pp. 159-169, 1993. C. Brothers, R. Pugh, P. Duggan, J. Chavez, D. Schepis, D. Yee, S. Wu, “Totaldose and SEU characterization of 0.25 micron CMOS/SOI integrated circuit memory technologies,” IEEE Trans. on Nucl. Sci., vol. NS-44, pp. 2134 –2139, 1997. S. Buchner, K. Kang, D. Krening, G. Lannan, and R. Schneiderwind, “Dependence of the SEU window of vulnerability of a logic circuit on magnitude of deposited charge,” IEEE Trans. on Nucl. Sci., vol. NS-40, no. 6, pp. 1853-1857, 1993. S. Buchner, A. B. Campbell, D. McMorrow, J. Melinger, M. Masti, and Y. J. Chen, “Modification of single event upset cross section of an SRAM at high frequencies,” RADECS 95, pp. 326-332, 1995 M.A. Carts, P.W. Marshall, C.J. Marshall, K.A. LaBel, M. Flanegan, J. Bretthauer, “Single event test methodology and test results of commercial gigabit per second fiber channel hardware,” IEEE Trans. Nucl. Sci., Vol. 44, No. 6, pp. 1878-1884, 1997. M. de La Chapelle, A. W. Van Ausdale, and M. E. Fritz, “The STAR-FODB (fiber optic data bus) program,” in Proc. GOMAC ’93 Conf., pp. 395-397.

II-73

(All references are unclassified.)

[Cros-93]

[Dale-92]

[Dale-93a]

[Dale-93b]

[Dale-95]

[Dale-96]

[DeLa-94] [DeRu-93] [Desu-93] [Dodd-99] [Dres-98] [Dres-89] [Dyer-98] [Emil-96] [Endo] [Evan-93]

[Flee-95]

[Frie-84a].

D. Cross, M. Fritz, D. Haakenson, C. Hoeflein, R. Hodges, S. Kelnhofer, J. Lam, and M. Summerhays, “The Boeing photonics space experiment,” Photonics for Space Environments I, vol. SPIE-1953, pp. 116-26, 1993. C.J. Dale , and P.W. Marshall, “Radiation response of 1300 nm optoelectronic components in a natural space environment,” Proc. SPIE, vol. 1791, pp. 224-232, 1992. C.J. Dale, P.W. Marshall, B. Cummings, L. Shamey, and A. Holland, “Displacement damage effects in mixed particle environments for shielded spacecraft CCDs,” IEEE Trans. Nucl. Sci., Vol. 40, No. 6, pp. 1628-1637, 1993. C.J. Dale, and P.W. Marshall, “Candidate NRL space experiments for the Microelectronics and Photonics Test Bed,” SPIE Proc. on Photonics for Space Environments, vol. 1953, pp. 17-22, 1993. C.J. Dale, P.W. Marshall, K.A. Clark, M. de La Chapelle, M.E. Fritz, and K.A. LaBel, “Fiber optic data bus space experiment on board the microelectronics and photonics test bed (MPTB),” Photonics for Space Environments III, vol. SPIE2482, p. 285, 1995. C.J. Dale, P.W. Marshall, M.E. Fritz, M. de La Chapelle, M.A. Carts, and K.A. LaBel, “System radiation response of a high performance FODB,” IEEE Trans. on Nucl. Sci., vol. NS-43, no.3, p 1030, 1996. M. Delaus, “Radiaiton concerns in state-of-the-art processing technology,” Notes from 1994 IEEE Nuclear and Space Radiation Effects Conference Short Course. J.DeRuiter, “Survivable ring architecture for spaceborne applications,” Proc.SPIE,vol. 1953, pp. 128-135, 1993. E. Desuvire, Erbium-Doped Fiber Amplifiers, Principles and Application, New York: John Wiley and Sons, Inc., 1993. P. Dodd, “Basic mechanisms for single event effects,” Notes from 1999 IEEE Nuclear and Space Radiation Effects Conference Short Course. P. V. Dressendorfer, “Basic mechanisms for the new millenium,” Notes from 1998 IEEE Nuclear and Space Radiation Effects Conference Short Course. Dressendorfer, P. V., Ma, T. P., Ionizing Effects in MOS Devices and Circuits, New York: John Wiley and Sons, 1989. C. S. Dyer, “Space radiation environment dosimetry,” Notes from 1997 IEEE Nuclear and Space Radiation Effects Conference Short Course. D.W. Emily, “Total dose response of linear bipolar microcircuits,” Notes from 1996 IEEE Nuclear and Space Radiation Effects Conference Short Course. Nikkei Science, Inc. of Japan, image by K. Endo B.D. Evans, H.E. Hager, B.W Hughlock, “5.5-MeV proton irradiation of a strained quantum-well laser diode and a multiple quantum-well broadband LED,” IEEE Trans. on Nucl. Sci., vol NS-40, no. 6, pp. 1645 –1654, 1993 D.M. Fleetwood, “A first principles approach to total-dose hardness assurance,” Notes from 1995 IEEE Nuclear and Space Radiation Effects Conference Short Course. E.J. Friebele, C.G. Askins, M.E. Gingerich, and K.J. Long, “Optical fiber waveguides in radiation environments, II,” Nucl. Inst. Meth. in Phys. Res. B1, 355369 (1984).

II-74

(All references are unclassified.)

[Frie-84b]

[Frie-85]

[Frie-91] [Frit-93] [Frit-94]

[Frit-95]

[Frit-96] [Gall-96] [Gate-96]

[Gris-94]

[Gros-93] [Guti-94]

[Hash-94]

[Heid-99] [Hens-92]

[Hens-97]

[Hens-98]

E.J. Friebele, K.J. Long, C.G. Askins, and M.E. Gingerich, “Radiation response of optical fibers and Selfloc Microlenses at 1.3µm,” Fiber Optics in Adverse Environments II, Proc. SPIE, vol. 506, pp. 202-208, 1984. E.J. Friebele, K.J. Long, C.G. Askins, M.E. Gingerich, M.J. Marrone, and D.L. Griscom, “Overview of radiation effects in fiber optics,” Crit. Rev. Tech. :Opt. Materials in Radiation Environ., P. Levy and E.J. Friebele,Ed. Bellingham, WA: SPIE, 1985, vol. SPIE-541, pp. 70-88. E.J. Friebele, “Photonics in the space environments,” Notes from 1991 IEEE Nuclear Space Radiation Effects Conference Short Course. M.E. Fritz, et al., “The Boeing photonics space experiment,” SPIE Proceedings, vol. 1953, pp. 116-126, April 1993. M.E. Fritz, B.E. Daniels, M. de La Chapelle, D.A. Cross, A.W. Van Ausdale, “The STAR FODB program,” SPIE Proc. on Photonics for Space Environments II, vol. 2153, 1994. M.E. Fritz, M. de La Chapelle, and A.W. Van Ausdale, “Boeing’s STAR FODB test results,” Photonics for Space Environments, vol. SPIE-2482, pp. 226-235, 1995. M.E. Fritz, G. Berg, D.A. Cross, and M.C. Wilkinson, “Photonics space experiment on-orbit results,” SPIE Proceedings, vol. 2811, pp. 106-115, Aug. 1996. Galloway’s short course M. Gates, K.A. LaBel, J. Barth, A. Johnston, and P.W. Marshall, “Single event effects criticality analysis,” NASA Report, See NASA GSFC Radiation Effects & Analysis Home Page, http://flick.gsfc.nasa.gov/radhome.htm, 1996. D.L. Griscom, M.E. Gingerich, and E.J. Friebele, “Model for dose, dose-rate, and temperature dependence of radiation induced loss in optical fibers,” IEEE Trans. Nucl. Sci., vol. NS-41, no. 3, pp. 523-527, 1994. S. Gross, “ATM-based protocol for Gbps ring networks,” in Proc. Gomac ’93 Conf., p. 399. R.C. Gutierrez, G.M. Swift, S. Dubovitsky, R.K. Bartman, C.E. Barnes, and L. Dorsky, “Radiation effects on fused biconical taper wavelength division multiplexers,” IEEE Trans. Nucl. Sci., vol. NS-41, no. 6, p1950, 1994. S. G. L Hash et al. T Hash, G.L.; Schwanck, J.R.; Shaneyfelt, M.R.; Sandoval, C.E.; Conners, M.P.; Sheridan, T.J.; Sexton, F.W.; Slayton, E.M.; Heise, J.A.; Foster, C.C., “Proton irradiation effects on advanced digital and microwave III-V components,”, IEEE Trans. Nucl. Sci. vol. NS-41, pp. 2259-2266, 1994 W.F. Heidergott, “System level mitigation strategies,” Notes from 1999 IEEE Nuclear and Space Radiation Effects Conference Short Course. H. Henschel, O. KØhn, and H.U. Schmidt, “Radiation sensitivity of fibre optic couplers,” SPIE, Optical Materials Reliability and Testing, vol. 1791, pp. 151-163, 1992. H. Henschel, O. Kohn, W. Lennartz, S. Metzger, H.U. Schmidt, J. Rosenkranz, B. Glessner, B.R.L. Siebert, “Comparison between fast neutron and gamma irradiation of optical fibres,” Proc. RADECS 1997, pp. 430 –438, 1997 H. Henschel, O. KØhn, H.U. Schmidt, J. Kirchhof, and S. Unger, “Radiationinduced loss of Rare Earth doped silica fibres,” IEEE Trans. Nucl. Sci.,vol. NS-45, no.3, p 1552, 1998.

II-75

(All references are unclassified.)

[Hinr-98]

[Hous-98] [Hsie-81] [Hugh-90] [Jack-99]

[Jack-96]

[Jord-93] [John-92] [John-94]

[John-99a] [John-99b]

[Kinn-98] [Koga-94]

[LaBe-91a] [LaBe-92a]

[LaBe-92b]

[LaBe-93a]

P.F. Hinrichsen, A.J. Houdayer, A.L. Barry, J. Vincent, “Proton induced damage in SiC light emitting diodes,” IEEE Trans. Nucl. Sci. vol. NS-45, pp. 2808-2812, 1998. S.L. Houston and K.A. Pfitzer, “A new model for the low altitude trapped proton environment,” IEEE Trans. on Nucl. Sci., vol. NS-45, no.6, pp. 2972-2978, 1998. C.M. Hsieh, P.C. Murley, R.R. O’Brien, “A field-funneling effect on collection alpha-particle generated carriers,” IEEE Electron Device Letters 2, p. 103, 1981 B.W. Huglock, G.S. LaRue, and A.H. Johnston, “Single-Event upset in GaAs E/D MESFET,” IEEE Trans. Nucl. Sci.,vol. NS-37, no.6, p 1894, 1990. G.L Jackson, K.A. LaBel, C.J. Marshall, J.L. Barth, J. Kolasinski, C.M. Seidleck, P.W. Marshall, “Preliminary flight results of the Microelectronics and Photonics Test Bed NASA dual rate 1773 (DR1773) fiber optics data bus experiment,” Procs. 1999 GOMAC, pp. 340-344, 1999. G.L. Jackson, K.A. LaBel, M. Flanegan, C. Dale, P.W. Marshall, R.K. Bonebright, J.H. Kim, E. Y. Chan, T. M. Bocek, C. White, “The Microelectronics and Photonics Test Bed dual rate 1773 fiber optics data bus experiment,” SPIE Proc. for Photonics for Space Environments, Vol. 2811, pp.116-127, 1996. A.F. Jordan, “On the brink: Fiber optic LAN’s for avionics and space,” Defense Electronics, vol. 25, no. 11, pp.43-47, 1993 A.R. Johnston and E.W. Taylor, “A survey of the LDEF fiber optic experiments,” JPL Report D-10069, Nov. 10, 1992. A.H. Johnston, G.M. Swift, B.G. Rax, “Total dose effects in conventional bipolar transistors and linear integrated circuits,” IEEE Trans. Nucl. Sci. vol. NS-41, pp. 2427-2436, 1994. A.H. Johnston, B.G. Rax, L.E. Selva, C.E. Barnes, “Proton degradation of lightemitting diodes,” IEEE Trans. Nucl. Sci. vol. NS-46, pp. 1781-1789, 1999 A.H. Johnston, T. Miyahira, G.M. Swift, S.M. Guertin, L.D. Edmonds, “Angular and energy dependence of proton upset in optocouplers,” IEEE Trans. Nucl. Sci. vol. NS-46, pp. 1335-1341, 1994 J.D. Kinnison, ”Achieving reliable, affordable systems” Notes from 1998 IEEE Nuclear and Space Radiation Effects Conference Short Course. R. Koga, R.J. Ferro, D.J. Mabry, S.D. Pinkerton, D.E. Romeo, J.R. Scarpulla, T.K. Tsubota, M. Shoga, “Ion-induced sustained high current condition in a bipolar device,” IEEE Trans. Nucl. Sci. vol. NS-41, pp. 2172-2178, 1994 K. LaBel, E.G. Stassinopoulos, and G.J. Brucker, “Transient SEU’s in a Fiber Optic System for Space Application,” IEEE Trans. Nucl. Sci., vol. 38,no. 6,1991. K.A. LaBel, “A spacecraft fiber optic data system – radiation effects,” RADECS 91: IEEE Proceedings from, La-Grande Motte, France, Sept. 1991, pp. 412-415, vol. 15. K.A. LaBel, J.A. Cooley, E.G. Stassinopoulos, P. Marshall, C. Crabtree “Single event test methodology for integrated optoelectronics,” SPIE Proceedings of Integrated Optical Circuits II, Boston, MA, vol. 1794, Sept. 1992, pp. 225-233. K.A. LaBel, E.G. Stassinopoulos, P.W. Marshall, E.L. Petersen, C. Dale, C. Crabtree, and C. Stauffer, “Proton irradiation SEU test results for the SEDS MILSTD-1773 fiber optic data bus: integrated optoelectronics,” SPIE Proc. for Photonics for Space Environments, Vol. 1953, pp. 27-44, 1993.

II-76

(All references are unclassified.)

[LaBe-93b] K.A. LaBel, M. Flanegan, P. Marshall, C. Dale, and E.G. Stassinopoulos, “Space flight experience and lessons learned with NASA’s first fiber optic data bus,” Second European Conference: RADECS 1993 Proc., pp. 221-225, 1993. [LaBe-93c] K.A. LaBel, P. Marshall, C.Dale, C.M. Crabtree, E.G. Stassinopolous, J.T. Miller and M.M. Gates, “SEDS MIL-STD-1773 fiber optic data bus: Proton irradiation test results and spaceflight SEU data,” IEEE Trans. Nucl. Sci., vol. 40, no. 6, pp. 163844, 1993. [LaBe-94a] K.A. LaBel, D.K. Hawkins, J.A. Cooley, C.M. Seidleck, P. Marshall, C. Dale, M. M. Gates, H.S. Kim, and E.G. Stassinopoulos, “Single event effect ground test results for a fiber optic data interconnect and associated electronics,” IEEE Trans. Nucl. Sci., vol. 41, no. 6, pp. 1999-2004, 1994. [LaBe-94b] K.A. LaBel, P.W. Marshall, C.J. Dale, E.G. Stassinopoulos, A. Johnston, C.M. Crabtree, and H.S. Kim, “Single event effects on associated electronics for fiber optic systems,” Proc. SPIE, vol 2215, pp. 74-93, April 1994. [LaBe-95] K.A. LaBel, A.K. Moran, D.K. Hawkins, A.B. Sanders, E.G. Stassinopoulos, R.K. Barry, C.M. Seidlick, H.S. Kim, J. Forney, P. Marshall, and C. Dale, “Single-event effect proton and heavy-ion test results in support of candidate NASA programs,” 1995 IEEE Radiation Effects Data Workshop Record, pp. 16-32, July 1995. [LaBe-96a] K.A. LaBel, and M.M. Gates, “Single-event-effect mitigation from a system perspective,” IEEE Trans. Nucl. Sci., NS-42, no. 2, pp. 654-660, April 1996. [LaBe-96b] K.A. LaBel, M. Flanegan, G. Jackson, D. Hawkins, C.J. Marshall, P.W. Marshall. D. Johnson, C. Seidleck, R.K. Bonebright, J.H. Kim, E. Y. Chan, T. M. Bocek, and B. Bartholet, “Preliminary ground test radiation results on NASA’s MPTB dual-rate 1773 experiment,” SPIE Proc. for Photonics for Space Environments, Vol. 2811, pp.128-135, 1996. [LaBe-97] K.A. LaBel, P.W. Marshall, C.J. Marshall, M. D'Ordine, M. Carts, G. Lum, H.S. Kim, C.M. Seidleck, T. Powell, R. Abbott, J. Barth, E.G. Stassinopoulos, “Protoninduced transients in optocouplers: in-flight anomalies, ground irradiation test, mitigation and implications,” IEEE Trans. Nucl. Sci. vol. NS-44, pp. 1885-1892, 1997. [LaBe-98a] K.A. LaBel, “Applying state of the art commercial and emerging technologies to space systems,” Notes from 1998 IEEE Nuclear and Space Radiation Effects Conference Short Course. [LaBe-98b] K.A. LaBel, A.H. Johnston, J.L. Barth, R.A. Reed, C.E. Barnes, “Emerging Radiation Hardness Assurance (RHA) Issues: A Nasa Approach For Space Flight Programs,” IEEE Trans. Nucl. Sci. vol. NS-45, pp. 2727-2736, 1998 [LaBe-98c] K.A. LaBel, R.A. Reed, H. Leidecker, J. Barth, P.W. Marshall, C.J. Marshall, C. Seidleck, “Comparison of MIL-STD-1773 fiber optic data bus terminals: single event proton test irradiation, in-flight space performance, and prediction techniques,” IEEE Trans. on Nucl. Sci., NS-44, no. 3, 1998. [LaBe-98d] K.A. LaBel, C.J. Marshall, P.W. Marshall, M.O. Ott, C.M. Seidleck, D.J. Andrucyk, “On the suitability of Fiber Optic Data Links in the space radiation environment: a historical and scaling technology perspective,” 1998 IEEE Aerospace Engineering Conference Proc., V 4, pp. 421-434, 1998. [Lach-98] G. Lachs, Fiber Optic Communications : Systems, Analysis, and Enhancements, New York, McGraw-Hill, c1998.

II-77

(All references are unclassified.)

[Lee-99]

[Lema-91] [Lera-99] [Mars-92]

[Mars-93a]

[Mars-93b]

[Mars-93]

[Mars-94a]

[Mars-94b]

[Mars-95a]

[Mars-95b]

[Mars-95c]

[Mars-96]

[Mars-98]

S.C. Lee, Y.F. Zhao, R.D. Schrimpf, M.A. Neifeld, K.F. Galloway, “Comparison of lifetime and threshold current damage factors for multi-quantum-well (MQW) GaAs/GaAlAs laser diodes irradiated at different proton energies,” IEEE Trans. Nucl. Sci. vol. NS-46, pp. 1797-1803, 1999 P.J. Lemaire, “Reliability of optical fibers exposed to hydrogen: Prediction of long term loss increases,” Opt. Eng. 30, 780-789 (1991). J.L. Leray, “Total dose effects: modeling for the present and future,” Notes from 1999 IEEE Nuclear and Space Radiation Effects Conference Short Course. P.W. Marshall, C.J. Dale, and E.A. Burke, “Space radiation effects on optoelectronic materials and components for a 1300 nm fiber optic data bus,” IEEE Trans. Nucl. Sci., vol. 39, no. 6, pp. 1982-1989, 1992. P. Marshall, C. Dale, and K. LaBel, “Charged particle effects on optoelectronic devices and bit error rate measurements on 400 Mbps fiber based data links,” in Proc. RADECS Conf., Saint Malo, France, Sept. 13-16, 1993, pp. 266-271. P.W. Marshall, K.A. LaBe l, C.J. Dale, J.P. Bristow, E.L. Petersen, and E.G Stassinopoulos, “Physical interactions between charged particles and optoelectronic devices and the effects on fiber based data links,” Proc. SPIE, vol. 1953, pp. 104115, 1993. P. Marshall, J. Cutchin, and T. Weatherford, “Space radiation effects in a GaAs CHIGFET logic family suitable for satellite data transmission above 1 Gbps,” in Proc. GOMAC ’93 Conf., pp. 227-229. P.W. Marshall, C.J. Dale, M.A. Carts, and K.A. LaBel, “Particle induced bit errors in high performance data links for satellite data management,” IEEE Trans. Nucl. Sci.,vol. NS-41, no.6, pp. 1958-65, 1994. P.W. Marshall, C.J. Dale, E.J. Friebele, and K.A. LaBel, “Survivable fiber-based data links for satellite radiation environments,” SPIE Critical Review CR-14, Fiber Optics Reliability and Testing, pp. 189-231, 1994. P.W. Marshall, C.J Dale, T.R. Weatherford, M.A.Carts, D. McMorrow, A. Peczalski, S. Baier, J. Nohava, J. Skogen ”Heavy ion immunity of a GaAs complementary HIGFET circuit fabricated on a low temperature grown buffer layer,” IEEE Trans. Nucl. Sci.,vol. NS-42, no.6, p. 1850, 1995. P.W. Marshall, C.J. Dale, M.E. Fritz, M. de La Chapelle, M.A. Carts, and K.A. LaBel, “Total ionizing dose and single particle effects in a 200 Mbps star-coupled fiber optic data bus,” Proc. of SPIE Conference on Photonics for Space Environments III, Proc. #2482, 1995. P.W. Marshall, C.J Dale, T.R. Weatherford, M. La Macchia, and K.A. LaBel, ”Particle induced mitigation of SEU sensitivity in high data rate GaAs HIGFET technologies,” IEEE Trans. Nucl. Sci.,vol. NS-42, no.6, pp. 1844-1854, 1995. P.W. Marshall, C.J. Dale, and K.A. LaBel, “Space radiation effects in high performance fiber optic data links for satellite management,” IEEE Trans. Nucl. Sci.,vol. NS-43, no.3, pp. 645-653, April 1996. C.J. Marshall, P.W. Marshall, M.A. Carts, R.A. Reed, K.A. LaBel, “Proton-induced transient effects in a Metal-Semiconductor-Metal (MSM) photodetector for opticalbased data transfer,” IEEE Trans. on Nucl. Sci., vol. NS-45, pp. 2842-2848, 1994

II-78

(All references are unclassified.)

[Mars-99]

[Mess-86] [Mess-97] [McCl-94]

[McDo-00] [McMo-96]

[Miya-79] [OBry-99]

[OBry-00]

[Ott –97]

[Paxt-97]

[Pala-98] [Pete-97] [Powe-93] [Rax-99]

[Ream-95] [Reed-96]

P.W. Marshall and C.J. Marsahll, “Proton effect and test issues for satellite designers,” Notes from 1999 IEEE Nuclear and Space Radiation Effects Conference Short Course. G. C. Messenger, M. S. Ash, The Effects of Radiation on Electronic Systems, New York : Van Nostrand Reinhold Co., c1986. G. C. Messenger, M. S. Ash, Single Event Phenomena, New York: Chapman and Hall, 1997. S. McClure, R.L. Pease W. Will, G. Perry, “Dependence of total dose response of bipolar linear microcircuits on applied dose rate,” IEEE Trans. Nucl. Sci. vol. NS41, pp. 2544-2549, 1994 McDonald, P. T. et al. to be published in Proceedings of the GOMAC/HEART Conference, March, 2000. D. McMorrow, T.R. Weatherford, S. Buchner, A.R. Knudson, J.S. Melinger, L.H. Tran, and A.B. Campbell, “Single event effects in GaAs devices and circuits,” IEEE Trans. Nucl. Sci.,vol. NS-43, no.2, pp. 628-624, 1996. T. Miya, Y. Terunuma, T. Miyoshita, Electron. Lett. 15, 106, 1979 M.V. O'Bryan, K.A. LaBel, R.A. Reed, J.W. Howard, J.L. Barth, C.M Seidleck, P.W. Marshall, C.J. Marshall, H.S. Kim, D.K. Hawkins, M.A. Carts, K.E. Forslund,”Recent radiation damage and single event effect results for microelectronics,” 1999 IEEE Radiation Effects Data Workshop Record, pp. 1-14, July 1999. M.V. O’Bryan, C.M. Seidleck, M.A. Carts, K.A. LaBel, R.A. Reed, J.L. Barth, C.J. Marshall, D.K. Hawkins, A. Sanders, J. Forney, J.W. Howard, H.S. Kim, R. Ladbury, P. Marshall, D. Roth, E. Nhan, J. Kinnison, K. Sahu, S. Kniffin, to be published in the 2000 IEEE Radiation Effects Data Workshop Record, 2000 M. Ott, J. Plante, J. Shaw, M.A. Garrison Darrin, “Fiber optic cable assemblies for space slight: issues and remedies,” Paper number 975592 AIAA/SAE World Aviation Congress, Anaheim, CA. 1997. A.H. Paxton, R.F. Carson, H. Schone, E.W. Taylor, K.D. Choquette, H.Q. Hou, K.L. Lear, M.E. Warren, “Damage from proton irradiation of vertical-cavity surface-emitting lasers,” IEEE Trans. Nucl. Sci., vol NS-44, no. 6, pp. 1893 –1897, 1997. J.C. Palais, Fiber Optic Communications, 4th edition, Prentice Hall, Upper Saddle River, N.J., c1998. E.L. Petersen, “Single event analysis and prediction,” Notes from 1997 IEEE Nuclear and Space Radiation Effects Conference Short Course. J.P. Powers,.An Introduction to Fiber Optic Systems, Aksen Associates ; Homewood, IL : Irwin, c1993. B.G. Rax, A.H. Johnston, T. Miyahira, “Displacement damage in bipolar linear integrated circuits,” IEEE Trans. on Nucl. Sci., volume NS-46, pp. 1660 –1665, 1999. D. V. Reames, “Solar Energetic Particles: A Paradigm Shift,” Revs. Geophys. (Suppl.), 33,585, 1995. R.A. Reed, M.A. Carts, P.W. Marshall, C.J. Dale, S. Buchner, M. La Macchia, B. Mathes, D., McMorrow, “Single Event Upset cross sections at various data rates,” IEEE Trans. Nucl. Sci. vol. NS-43, pp. 2862 –2867, 1996.

II-79

(All references are unclassified.)

[Reed-98]

R.A Reed, P.W. Marshall, A.H. Johnston, J.L. Barth, C.J. Marshall, K.A. LaBel, M. D'Ordine, H.S. Kim, M.A. Carts, “Emerging Optocoupler Issues With Energetic Particle-Induced Transients And Permanent Radiation Degradation,” IEEE Trans. Nucl. Sci. vol. NS-45, pp. 2833 –2841, 1998. [Schn-92] R. Schneiderwind, D. Krening, S. Buchner, and K. Kang, “Laser confirmation of SEU experiments in GaAs MESFET combinational logic,” IEEE Trans. on Nucl. Sci., vol. NS-39, no. 6, pp. 1665-1670, 1992. [Shog-94] M. Shoga, K. Jobe, M. Glasgow, M. Bustamante, E. Smith, and R. Koga, “Single event upset at Gigahertz frequencies,” IEEE Trans. on Nucl. Sci., vol. NS-41, no. 6, pp. 2252-2266, 1994. [Stap-95] W.J. Stapor, “Single-event effects qualification,” Notes from 1995 IEEE Nuclear and Space Radiation Effects Conference Short Course. [Tayl-90] E.W. Taylor, “Behavior of coupled waveguide devices in adverse environments,” SPIE, Fibre Optics, vol. 1314, pp. 155-167, 1990. [Tayl-91a] E.W. Taylor, “Ionization-induced refractive index and polarization effects in LiNbO3 :Ti directional coupler waveguides,” J. Lightwave Tech., vol. 9, no. 3, pp. 335-340, 1991. [Tayl-91b] E.W. Taylor, J.N. Berry, A.D. Sanchez, R.J. Padden, and S.P. Chapman, “Preliminary analysis of PL experiment #701, space environment effects on operating fiber optic systems,” in LDEF-69 Months in Space- First Post Retrieval Symposium, Report No. NASA CP-3134, pp. 1257-1282, 1991. [Tayl-92a] E.W. Taylor, “Radiation effects in guided wave devices,” SPIE, Integrated Optical Circuits II, vol. 1794, pp. 54-61, 1992. [Tayl-92b] E.W. Taylor, J. Berry, A.D. Sanchez, R.J. Padden, S. DeWalt, and S. Chapman, “First operational space fiber optic data links orbited aboard the Long Duration Exposure Facility-lessons learned,"”1992 DOD Fiber Optics Conference Proceedings. [Teag-72] M.J. Teague,”A model for the starfish flux in the inner radiation zone,” X-602-72487, NASA/Goddard Space Flight Center, Greenbelt, MD, December 1972. [Thel-94] D. Thelen, S. Rankin, P.W. Marshall, K.A. LaBel, M.A. Krainak, “Dual-rate MILSTD-1773 fiber optic transceiver for satellite applications,” Proc. of SPIE Conference on Photonics for Space Environments II, Proc. #2215, pp.101-112, 1994. [Turf-90] T.L. Turflinger and M.V. Davey, “Understanding single event phenomena in complex analog and digital integrated circuits,” IEEE Trans. on Nucl. Sci., vol. NS37, no. 6, pp. 1832-1838, 1990. [Turf-94] T.L. Turflinger, M.V. Davey, and B.M. Mappes, “Single event effects in analog-todigital converters: device performance and system impact,” IEEE Trans. on Nucl. Sci., vol. NS-41, no. 6, pp. 2187-2194, 1994. [Tylk-96] A.J. Tylka, J.H. Adams, Jr., P.R. Boberg, B. Brownstein, W.F. Dietrich, E.O. Flueckiger, E.L. Petersen, M.A. Shea, D.F. Smart, E.C. Smith, :CREME96 a revision of the cosmic ray effects on microelectronics code,” IEEE Trans. Nucl. Sci.,vol. NS-43, no.6, pp. 2758-2766, 1996. [VanE-96] T.E. Van Eck, D.G. Girton, J.A. Marly, S.P. Ermer, W.W. Anderson, L.E. Robinette, G.K. Lum, and J.W. Garrett, “Space environment testing of polymer photonic modulators,” SPIE, vol. 2911, pp. 25-28, 1996.

II-80

(All references are unclassified.)

[Weat-97]

T.R. Weatherford, P.W. Marshall, C.J. Marshall, D.J. Fouts, B. Mathes, and M. La Macchia, “Effects of low temperature buffer layer thickness and growth temperature on the SEE sensitivity of GaAs HIGFET circuits,” IEEE Trans. Nucl. Sci.,vol. NS-44, no.6, 1997. [Weis-90] J.D. Weiss, “The radiation response of a Selfoc microlens,” J. Lightwave Technol.,vol. 8, no. 7, pp.1107-1109, 1990. [Wicz-86] J.J. Wiczer, “Radiation hardened optoelectronic components: Detectors,” Proc. SPIE, vol. 616, pp. 254-266, 1986. [Will-98] G.M. Williams and E.J. Friebele, “Space radiation effects on Erbium-doped fiber devices: sources, amplifiers, and passive measurements,” IEEE Trans. Nucl. Sci., vol. NS-45, no.3, p 1531, 1998. [Widm-87] A.X. Widmer, U.S. Patent 4,667,517, 12 May, 1987 [Wino-92] P.S. Winokur, “Total-dose radiation effects (from the perspective of the experimentalist),” Notes from 19XX IEEE Nuclear and Space Radiation Effects Conference Short Course. [Xaps-98] M.A. Xapsos, G.P. Summers and E.A. Burke, “Probability model for peak fluxes of solar proton events,” IEEE Trans. Nucl. Sci.,vol. NS-45, no.6, pp. 2948-2953, 1998. [Xaps-00] M.A. Xapsos, R.J. Walters, G.P. Summers, J.L. Barth, E.G. Stassinopoulos, R. Messenger, and E.A. Burke, “Characterizing solar proton energy spectra for radiation effects,” presented at the IEEE NSREC 2000. [Zhao98] Y.F. Zhao, R.D. Schrimpf, A.R. Patwary, M.A. Neifeld, A.W. Al- Johani, R.A. Weller, and K.F. Galloway, “Annealing Effects on Multi-Quantum Well Laser Diodes after Proton Irradiation,” IEEE Trans. Nucl. Sci., Vol. 44, pp. 2826-2832, 1997. [Zieg-84] J.F. Ziegler, J.P. Biersack, and U. Littmark, The Stopping and Range of Ions in Solids. New York: Pergamgon, 1884.

II-81

10.0 List of Acronyms AE8 AP8 APD BER BERT CME CRRESELE CRRESPRO DDD EDFA ELDRS EOL FBIU FODB Gbps GLAS GRC HST LED LET LDEF Mbps MPTB MSM NRZ NSREC p-i-n PRN RADECS RZ SAA SAMPEX SEB SEDS SEDSI SEDSII SEE SEGR SEL SET SEU SSR TID TNS VCSEL WDMs

model for trapped electron environment model for trapped proton environment Avalanche PhotoDiodes Bit Error Ratio Bit Error Ratio Tester Coronal Mass Ejections model for trapped electron environment model for trapped proton environment Displacement Damage Dose Erbium-Doped Fiber Amplifier Enhanced Low Dose Rate Sensitivity End Of Life Fiber Bus Interface Unit Fiber Optics Data Bus Giga-bits per second Guided Laser Altimeter System Galactic Cosmic Rays Hubble Space Telescope Light-Emitting Diode Linear Energy Transfer Long Duration Exposure Facility Mega-bits per second Microelectronics and Photonics TestBed Metal-Semicondictor-Metal NonReturn-to-Zero Nuclear and Space Radiation Effects Conference p-intrinsic-n PseudoRandom Number Proceedings from the Radiation Effects in Components and Systems Conference Return-to-Zero South Atlantic Anomaly Solar Anomalous Magnetospheric Particle Explorer Single-Event Burnout Small Explorer Data System first generation SEDS the second generation of SEDS Single-Event Effects Single-Event Gate Rupture Single-Event Latchup Single-Event Transient Single-Event Upset Solid State Recorder Total Ionizing Dose Transactions on Nuclear Science Vertical Cavity Surface Emitting Laser Wavelength-Division Multiplexers

II-82

2000 IEEE NSREC SHORT COURSE

Section III

Optoelectronic Devices with Complex Failure Modes

Allan Johnston/JPL

Approved for public release; distribution is unlimited

OPTOELECTRONIC DEVICES WITH COMPLEX FAILURE MODES Allan H. Johnston Jet Propulsion Laboratory California Institute of Technology Pasadena, California

1. Introduction 2. Space Environments A. Overview B. Solar Flares C. Earth Orbiting Environments D. Deep Space and Planetary Environments 3 . Charged Particle Interactions A. Basic Considerations B. High-Energy Protons C. Displacement Damage Effects D. Displacement Damage Effects in Semiconductors 4.

Key Properties of Semiconductors Used in Optoelectronic Devices A. Band Structure B. Optical Emission and Absorption C. Absorption Coefficient D. Quantum Efficiency E. Snell’s Law F. Optical Transmission

5.

Properties of III-V Semiconductors A. Electrical Properties of Basic Materials B. Ternary and Quaternary Materials with Variable Bandgap C. Heterojunctions D. Strained Lattices in Thin Layers F. Quantum Wells

6

Light Emitting Diodes A. Properties of Various LED Technologies C. Radiation Degradation D. Radiation Testing of Light Emitting Diodes

III - 1

7.

Laser Diodes A. Basic Features B. More Advanced laser Structures C. Radiation Degradation D. Vertical Cavity Semiconductor Lasers (VCSELs) E. Radiation Testing Considerations for Semiconductor Lasers

8.

Optical Detectors A. Basic Considerations B. Elementary Detectors Based on p-n Junctions C. Specialized Detector Technologies D. Noise and Figures of Merit

9.

Optocouplers A. Basic Features B. Radiation Degradation C. Testing Issues

10.

Solar Cells A. Construction and Electrical Properties B. Radiation Degradation

11.

Charge-Coupled Devices A. Conventional CCDs B. Active Pixel Sensors

12.

Examples of Complex Failure Modes A. Optocoupler Failures in Space B. Power Converter Failures During Ground Tests

13.

General Radiation Testing Issues for Optoelectronic Devices A. Radiation Sources B. Energy Selection C. Single-Event Upset Testing

14.

Summary

15.

Acknowledgements

16.

References

III - 2

1 – INTRODUCTION This part of the NSREC-2000 Short Course discusses radiation effects in basic photonic devices along with effects in more complex optoelectronic devices where the overall radiation response depends on several factors, with the possibility of multiple failure modes. Complex failure modes can occur either because of the way different types of components within an optoelectronic structure are related (including physical factors that affect light transmission), or because different types of radiation affect those components in different ways (for example, ionization and displacement effects). In addition, some types of responses are application dependent, introducing yet another level of complexity in interpreting radiation responses. Photonic devices can be used over a very wide range of wavelengths, and it is not possible to cover all aspects of optoelectronics in a course of this type. Many photonic devices are designed to be compatible with the three “windows” for fiber optics where SiO2-based optical fibers allow light transmission over very long distance with low absorption, nominally 850, 1300 and 1500 nm, as well as with compatibility with commonly used detector materials. Those wavelength ranges are shown in Figure 1-1. The course briefly discusses photonic devices for the medium infrared region where it is generally necessary to use cooled detectors in order to obtain suitable signal-to-noise ratios. The far-infrared region -- above 5 µm -- is discussed only briefly in this course.

Figure 1-1. Wavelengths of primary interest for photonic devices.

This segment of the short course begins with a very brief review of radiation environments (Section 2), stressing the fact that we generally have a mix of high-energy electrons and protons in space environments, and that although radiation tests are often done with cobalt-60 gamma rays, those tests simulate only ionization damage, not displacement damage effects. Generally speaking, tests with gamma rays are inadequate to characterize the performance of optoelectronic devices in space. Energies for testing optoelectronic devices with protons and electrons are recommended, along with other important factors that affect the way that radiation testing and device evaluations are performed. Sections 3-5 discuss fundamental effects in semiconductors that are important for photonic devices, including light emission and light absorption properties in direct and indirect semiconductors. The GaAs-AlGaAs system is used as an example of bandgap engineering to tailor devices towards specific wavelengths. Homostructure and heterostucture junctions are discussed, along with fundamental damage mechanisms.

III - 3

The sixth and seventh sections discuss light emission from LEDs and laser diodes. The LED discussion contrasts three different LED technologies, and also discusses measurement and characterization techniques for those classes of devices. Promising new laser technologies are included, particularly vertical cavity semiconductor lasers. The eighth section discusses optical detectors. It includes conventional silicon photodiodes and phototransistors as well as III-V detectors. A brief discussion of infrared detectors is also included, as well as a discussion of noise in detectors and electronics. The ninth section discusses the important topic of optocouplers, a good example of a device with multiple, complex failure modes. Optocoupler performance depends on physical characteristics of the emitter and detector and optical coupling compounds as well as on degradation of individual components. The tenth and eleventh sections discuss other optoelectronic devices, including solar cells and charge-coupled devices. This is followed by another section with two examples of complex failure modes. Section 13 discusses test techniques, including selection of proton energy, interpretation in the context of actual environments with a distribution of proton energies, and a limited discussion of SEE testing. The course is summarized in Section 14.

2 - SPACE ENVIRONMENTS A. Overview Environments in space consist of protons and electrons that cause permanent damage effects in optoelectronic devices because there are very large numbers of them. For the most part, damage in optoelectronic devices is due to the large number of interactions that result from exposure to relatively high fluences of these particles. There are also galactic cosmic rays that are relatively few in number compared to protons and electrons. For galactic cosmic rays the primary concern is effects from the interaction of a single particle with the device (such as single-event upset). We are primarily concerned with permanent damage effects from protons and electrons in this part of the 2000 Short Course, except for a brief discussion of single-event effects testing at the end, and therefore very little attention will be given to heavy ions or transient effects from proton recoils. Space environments are heavily influenced by trapped radiation belts around the earth (or trapped belts around other planets with magnetic poles) as well as by solar flares. This section is intended to provide only a brief summary of space radiation environments. A far more thorough treatment of radiation environments was given by J. L. Barth in Section 1 of the 1997 Short Course [Bart1], along with an earlier review paper by E. Stassinopoulous and J. Raymond [Stas1]. Readers are encouraged to use those references for a more accurate and complete description of space environments.

B. Solar Flares Solar flares are important for many environments that are encountered by spacecraft. Solar flare activity is periodic, increasing during periods of intense sunspot activity on an eleven-year cycle. Most solar flares have very low intensity, but statistical results from the last three solar cycles have shown that there is a high probability of getting at least one solar flare with relatively high intensity during each solar cycle [Feyn1]. Consequently, most spacecraft that operate within the 5-6 year period of more intense solar activity need to include an intense solar flare in their environmental requirements. However, basing the requirements on an extremely hard, intense flare (such as the flare that occurred in October, 1989 [Crol]) is overly conservative for most spacecraft because such intense flares occur very rarely. Xapsos, et al. have developed a statistical model that incorporates data from the last 30 years to predict the maximum expected fluence from solar flares in a less conservative manner [Xap1].

III - 4

Their model predicts upper bounds for various energies as shown in Table 1. Those values do not take geomagnetic and atmospheric shielding into account, but are applicable to geosynchronous or deep space environments. Table 1 shows the results of their model for various proton energies. Table 1. Worst-Case Solar Event Fluences from Statistical Models [Xaps1] Energy (MeV)

Worst-Case Proton Fluence (p/cm2) 4.4 x 1010 1.3 x 1010 6.1 x 109 1.7 x 109

10 30 50 100

These values represent the total fluence from a single intense flare at the earth, not the total fluence from all solar flares that occur during an extended mission. However, the results show that the maximum expected fluence from a single flare is on the order of a few times 1010 p/cm2. The fluence from solar flares also depends on the distance from the sun; the fluence falls off roughly as the inverse square of the distance. This is important for interplanetary missions.

C. Earth Orbiting Environments The trapped radiation belts that surround the Earth are the major contributors of radiation for low-Earth (LEO) and medium-Earth (MEO) orbits. Figure 2-1 shows a pictorial diagram of the proton belt and the outer electron belt (the inner electron belt is not shown).

Proton Belt N Outer Electron Belt

South Atlantic Anomaly

Figure 2-1. Pictorial diagram of the earth’s radiation belts.

III - 5

The proton environment that earth-orbiting spacecraft encounter depends on altitude and inclination. The intensity of the proton belts increases sharply with altitude, extending from about 800 to 15,000 km. The edges of the belt are not well defined, and the pictorial diagram of Figure 2-1 is somewhat misleading in that respect. Figure 2-2, after Stassinopoulos and Raymond [Stas1], shows contours of various proton energies for various altitudes along the equatorial plane (the belts are closer to the earth at higher latitudes). The figures also shows the L-shell parameter that is used to more accurately describe the position of the proton belts at other latitudes. Figure 2-2 shows that the edge of the proton belt actually begins at about 500 km, and that very high levels of proton radiation are encountered for altitudes above 2000 km. Most satellites that operate in low-earth orbits are constrained to altitudes below about 1500 km in order to avoid the intense radiation levels in the middle of the proton radiation belt.

Altitude at the Equator (thousands of km) Figure 2-2. Energy contours for protons at various altitudes in the equatorial plane (after Stassinopoulos and Raymond [Stas1]).

For very low orbits (such as the Space Shuttle, approximately 305 km), the orbit is below the edge of the proton belt, and most of the protons occur when the spacecraft passes through the South Atlantic anomaly (SAA). An asymmetry in the proton belt causes it to extend (locally) to much lower altitudes in this region. The main reason for the SAA is that the earth’s magnetic dipole is not exactly aligned with its rotational axis. Proton contributions from the SAA are also important at intermediate altitudes, but become less of a factor for spacecraft with high altitude and inclination. For high-inclination LEO and MEO orbits, protons from solar flares also make a significant contribution to the total proton fluence. However, at low inclination geomagnetic shielding keeps most of the solar flare particles out of the inner belt region, and the effect of solar flares can largely be ignored. Spacecraft environments usually take shielding into account, typically assuming a spherical shell of aluminum that is a reasonable first approximation of the amount of shielding that surrounds most of the electronics. Moderate amounts of shielding reduce the number of electrons and low energy protons, but have little effect on protons with higher energy. Figure 2-3 shows how the internal environment of one orbit, the 705 km/98º orbit used for many of the missions to monitor climatic changes around the earth, is affected by shielding. In this case the spacecraft is assumed to operate for 5 years. With 100 mils of shielding, protons contribute more than twice as much total dose as electrons, and protons are even more dominant for thicker amounts of shielding. Although small amounts of shielding are effective in reducing the total dose in the “raw” environment, once the low energy particles are removed by the thin shield one

III - 6

is left with relatively energetic protons that become very difficult to shield unless very large additional amounts of shielding can be tolerated. Thus the effects of shielding are highly nonlinear. Note the presence of low levels of Bremsstrahlung radiation that place a lower limit on the effectiveness of shielding even for very thick shields. Many different earth orbits are possible, and the number of protons expected during a mission depends on altitude, inclination and mission duration. Typical proton fluences (after moderate amounts of shielding) for LEO orbits range from about 2 x 108/cm2 for a shortduration space shuttle mission (305 km) to as much as 1011/cm2 for a 1300 km polar mission operating for five years. Higher proton fluence levels are encountered for higher orbits or geostationary transfer orbits (see [Bart1] for details).

Figure 2-3. Effect of shielding on total dose from electrons and protons for a 705 km, 98º earth orbit.

Proton fluences are far lower for satellites that operate at high altitudes because they are beyond the edge of the proton radiation belt. For geostationary satellites (approximately 36,000 km), the proton belts do not contribute, and most of the proton fluence comes from solar flares.

D. Deep Space and Planetary Environments Proton levels in deep space environments are similar to those of geostationary orbits, with most of the contribution coming from solar flares (correcting, however for the falloff of solar flare intensity with distance from the sun). However, interplanetary missions may have to pass through intense radiation belts associated with other planets. Jupiter and Saturn both have intense trapped radiation belts that extend over far greater distances than the trapped radiation belts surrounding the earth. Planetary missions that need to traverse those regions can encounter very high levels of radiation. More details are given in Section 2 of the 1993 Short Course, by H. Garrett [Garr1]. Not all planets have trapped radiation fields. For example Mars has a weak magnetic field that is too low to trap energetic particles and consequently radiation levels near Mars are dominated by galactic cosmic rays and solar flares. The Mars atmosphere is much thinner than that of earth, but still provides sufficient shielding to reduce the radiation level on the Martian surface to relatively low levels. III - 7

3 - CHARGED PARTICLE INTERACTIONS A. Basic Considerations When electrons and protons interact with materials the dominant energy loss process is by electromagnetic interaction with loosely bound electrons in the valence band. This process, known as ionization, elevates an electron from the valence band into the conduction band of the material, simultaneously creating an excess electron in the conduction band and a hole in the valence band (electron-hole pair). The process requires about 3.8 eV in silicon and 18 eV in silicon dioxide [McLe1], and thus only a very small amount of energy is absorbed by each ionization event. Ionization produces electron-hole pairs in insulators (such as silicon dioxide) as well as in semiconductors and the dominant damage effect from radiation in many semiconductor devices is due to the way that the excess carriers generated by ionization are trapped at critical interface regions between semiconductors and insulators (in particular, SiO2). Examples include the threshold shift of gate and field oxides in MOS devices [Win1, Dres1], and gain degradation due to increase in surface recombination in many types of bipolar transistor structures [Peas1]. Ionization damage is so dominant for many important devices that system specifications often include only the net deposited dose from ionization in their specifications, even when the environment is mainly due to high-energy protons. Another effect of ionization in insulators is the creation of color centers that affect the optical properties of the material. Color centers are created when an electron is trapped at an impurity site or vacancy in the material, or when the charge from ionization changes the valence state of an impurity atom. This effect is important in optical fibers, glass windows, and lenses. Absorption losses from color centers are wavelength dependent. In silicon dioxide absorption is much greater at short wavelengths. The degree to which color centers are created depends on the number and type of impurities within the material (for example, lead glass is extremely susceptible to darkening at relatively low levels of ionizing radiation). Some types of optical fibers are highly susceptible to formation of color centers (see Section 4 of the 1992 Short Course by E. J. Friebele [Frei1]). Color centers in quartz or other forms of pure silicon dioxide are generally only important for components with very long path length (such as optical fibers), and can usually be ignored for most space applications. However, ionization is not the only energy loss mechanism. Electrons and protons can transfer energy to bound nuclei within the lattice via interactions with atomic nuclei. If the energy transferred during the collision is high enough the atom can be moved from its stable position, creating a disordered damaged region within the material. This process, referred to as displacement damage, is usually the dominant radiation effect for optoelectronic devices. The threshold energy to create a displaced atom in silicon is about 13 eV, but the total energy transferred to the lattice can exceed 10,000 eV for complex damage cascades [Srou1]. On the average, each displacement event deposits much more energy than ionization events. Ionization loss processes still dominate from the standpoint of overall energy loss by the incident particle because the cross section for nuclear interactions (displacement) is about a factor of 104 lower than that due to ionization due to the small size of the nucleus. We will concentrate most of the discussion in this section of the course on displacement damage in various kinds of optoelectronic devices because it turns out to be the most important damage mechanism for most photonic devices. In addition to effects on semiconductor properties displacement damage can alter basic optical properties of materials (such as the index of refraction), but such effects are usually only important for very high fluences that are above those encountered in typical spacecraft and will not be considered further in this course. Single-event effects from galactic cosmic rays or proton-induced reaction are also important for many photonic devices, and are addressed in Part 1 of this year’s short course. Although single-event effects are generally omitted from this part (Part 2), some aspects of single-event radiation testing of optoelectronic devices are discussed in the next-to-last section. III - 8

B. High-Energy Protons Proton displacement damage depends on energy, and the energy dependence must be taken into account in order to relate laboratory test results -- usually done at a single proton energy -to equivalent effects on devices when they are exposed to real space environments with a distribution of proton energies. The energy dependence depends on the material as well as the specific property that affects the optoelectronic devices that are being considered, i.e., lifetime damage, mobility degradation, or carrier removal. During the last 15 years considerable progress has been made in calculating the energy dependence of the component of energy loss that goes into displacement damage. The term nonionizing energy loss (NIEL) is used to describe this. Figure 3-1 shows the energy dependence of NIEL for protons in silicon as derived by Summers, et al. [Summ1] and experimentally verified with measurements of lifetime damage in discrete transistors. The damage increases rapidly at low energies (with approximately a 1/E dependence) basically because it takes longer for lowenergy protons to traverse the target material, providing a longer time for the interaction to occur. Although this is effectively a collision between the proton and the nucleus, the interaction is actually electromagnetic at lower energies which is the reason for the increase in NIEL at low energies. The NIEL value for 1-MeV equivalent neutrons is also shown (for neutrons the interaction is kinematic rather than electromagnetic). Protons with energy of 200 MeV have about the same NIEL value as 1-MeV equivalent neutrons, which is of interest because of the body of data that exists for neutron degradation.

Figure 3-1. Dependence of NIEL on proton and electron energy for silicon.

Similar calculations have been done for NIEL in GaAs, but the results are less consistent with experimental results. Figure 3-2 shows the calculated energy dependence [Burk1, Summ2, Summ3] along with experimental results obtained by Barry, et al. for light-emitting diodes [Barr1]. The calculated values of NIEL agree reasonably well with older experimental data for JFETs, which change resistivity after proton irradiation because of carrier removal. However, the calculated values disagree at high energies compared to the more recent experimental results for LEDs, which (for the particular devices studied) degrade because of changes in minority carrier lifetime. The discrepancy between the NIEL calculations and LED experimental results is about a factor of three at 200 MeV. Note that there is reasonable agreement at energies below approximately 80 MeV.

III - 9

Figure 3-2. Dependence of NIEL on proton and electron energy for GaAs.

For space environments the increased effective NIEL at low energies makes the low energy component of the proton energy spectrum relatively more important from the standpoint of damage in devices. Figure 3-3 shows the “raw” proton spectrum for a high-inclination Earth orbit along with adjusted spectra for silicon and GaAs displacement damage that take the energy dependence into account by weighting each energy interval by the relative displacement damage effect. For GaAs, the effect is to approximately double the raw spectrum when the damage is normalized to 50 MeV. System specifications often include only the total number of protons as part of their requirements. There is less difference between the raw and adjusted spectra for silicon. These equivalence factors depend on the amount of shielding that is present, along with the initial spectrum. As the amount of shielding is increased, the mean energy becomes higher and the difference between the raw and adjusted spectrum becomes smaller.

Figure 3-3. Typical energy spectrum; example of “adjusted” spectrum and difference between “raw” and “effective” fluences normalized to equivalent damage at 50 MeV.

III - 10

Proton testing is usually done at only a single energy (except for special studies to investigate energy dependence) because it is too costly to include tests at several different energies on a routine basis. Although in principle any energy can be used (subject to range limitations), the best practice is to select an energy for testing that is near the peak in the differential energy spectrum, approximately 50 MeV for many systems (see Figure 3-3). If the energy is too low, the range of the protons may be insufficient to penetrate the package and dead layer of the device, leading to uncertainties in the energy of the protons that actually penetrate the active region and cause the degradation. Solar cell data is usually based on 10 MeV protons because the cells are generally shielded only by very thin cover slides, but higher energies are recommended for testing most other optoelectronic components because there is much more extraneous material in lenses, packaging and subsystem enclosures, leading to higher mean energies in the spectrum of protons that actually reaches the device. Using energies above 80 MeV is potentially a problem for III-V devices because of the discrepancy in NIEL between theoretical and experimental work discussed earlier. When high energies are used there is no ambiguity about the measured results, but the interpretation of how the damage affects devices with a continuous proton spectrum then depends on the assumed energy dependence for NIEL. If the NIEL calculations are revised at a later date (or inaccurate), then the interpretation of the effect of the actual protons in the environment will be incorrect. For example, 200 MeV protons are readily available at one high-energy proton facility, and have frequently been to test optoelectronic devices. If the NIEL values for liftetime damage are used, the damage for a typical spectrum of protons will be a factor of three higher than if the NIEL values for carrier removal (or the theoretical calculations of NIEL) are used. Testing with 50 MeV protons reduces such uncertainties to about 20-30%.

C. High-Energy Electrons Electrons also produce displacement damage, with an energy dependence that is quite different than for protons. As shown previously in Figures 3-1 and 3-2, NIEL for electrons increases with energy, in contrast to protons. With electrons there is a sharp energy threshold -approximately 150 keV for silicon -- below which displacement damage does not occur [Baur1, Dale1]. For GaAs the threshold energy is about 250 keV. Although electron displacement damage is important in some space environments, the NIEL values are two to three orders of magnitude lower than for protons. This, in combination with the reduction in electron fluence from moderate amounts of shielding usually makes electron displacement damage less important than proton displacement damage in earth-orbiting environment. Exceptions are cases where there is very little shielding (such as for solar cells), or interplanetary missions where the electron spectrum may be much harder than the spectrum of electrons in the earth’s radiation belts. For example, electrons in the Jovian radiation belts have energies up to 500 MeV [Garr1].

D. Displacement Damage Defects Displacement damage is a complex process, and the nature of the defects that are created depends on the energy that is actually transferred to the lattice site, which varies over a wide range. The displacement energy threshold is about 9 eV for GaAs and 13 eV for silicon. Displacement events that transfer relatively small amounts of energy create isolated vacancyinterstitial pairs along the particle track (Frenkel pairs). When larger amounts of energy are transferred to the “struck atom” the nature of the localized damage sites changes. In silicon damage clusters tend to form over distances of about 60 Å. The clusters are charged by the energy deposition process (the displaced atom at the site of the original collision creates the clustered damage site by electromagnetic interaction with other atoms in the lattice as it dissipates energy). The cluster damage sites are not completely stable, and some of the damage gradually anneals with time. Because the cluster is charged, annealing occurs much more rapidly when current passes through the region [Greg1]. This is termed injection-enhanced annealing. One would expect that when a lattice atom absorbs energy that is several orders of magnitude above the threshold energy for displacement that the result would be a very large localized

III - 11

damage region, spreading radially from the collision site. However, this is not the case. Figure 3-4 shows an example of how the damage progresses through the lattice, based on a computer model [Lint1]. In this case the initial recoil atom absorbs 60 keV from the collision. The effect of the higher recoil energy is to create a cascade of damaged cluster regions, each with about the same dimension as the clusters that are created by events that transfer more moderate amounts of energy to the lattice. Frenkel pairs are also produced along the track of the recoil. In this example clusters are produced nearly 1000 Å beyond the site of the initial collision.

Figure 3-4. Damage cascade structure, calculated from a theoretical model. Note that there are several different damaged regions along the track with about the same disordered extent, not a single region with large defect density.

Figure 3-4 illustrates the different character of the microscopic defects that are created by displacement damage. Based on these considerations, it is rather surprising that the concept of a general equivalence for displacement damage (NIEL) will work at all. Several aspects of damage are not treated directly by NIEL calculations, including the issue of damage stability and annealing. As noted by Summers [Summ1, Summ4] as well as by C. Marshall and P. Marshall in Part 3 of the 1999 Short Course [Mars1], it is surprising that the NIEL results agree as well as they do considering that they are applied to different types of particles over a wide range of energies. Thus, it is important not to expect too much from NIEL calculations. The concept of NIEL is extremely valuable when making damage comparisons, but there are differences in the damage mechanisms that are likely to lead to discrepancies if the concept is pushed too far. Table 4-1, after Dale, et al. [Dale1] shows the way that defects are distributed between clusters and isolated point defects for electrons and protons in silicon for different amounts of deposited energy.

III - 12

Table 3-1. Partition of Microscopic Damaged Regions Percent Of Defects Protons

Energy Interval Range of Primary Knock-on Atoms

Defect Structure

25 to 1000 eV

Point defects

97

44

34

12

1000 to 10,000 eV

Single Cascade

3

28

21

7

0

28

45

81

above 10,000 eV

Several subcascades

Electrons

4.1 MeV 53 MeV

8.3 MeV 60 MeV

E. Displacement Damage Effects in Semiconductors Displacement damage affects various properties of semiconductors. The importance of displacement effects and their interpretation is different for various optoelectronic devices. These effect are discussed below; it is necessary to have a reasonably good understanding of the underlying structure and principles of the operation of devices in order to put these effects in the proper context. Those issues are discussed in more detail in later sections of this part of the Short Course. Lifetime Damage The semiconductor property that is most sensitive to displacement damage effects is minority carrier lifetime. Lifetime damage can be described by the equation

Κ

(1)

where τo is the initial lifetime, τ is the lifetime after irradiation, Φ is the particle fluence, and K is the damage constant. The relationship between the reciprocal liftetime and the reciprocal of the damage constant is linear over a wide range of fluence values, until second-order effects (such as carrier removal) become important. The damage constant differs for n- and p-type material, and also depends on doping level. This equation is widely used to describe lifetime damage, but does not address damage stability or annealing. It can be applied to other device parameters that are sensitive to minority carrier lifetime, such as common-emitter current gain [Mess1] or light-emitting diode degradation [Rose1]. Figure 3-5 shows how protons affect minority carrier lifetime in n-type silicon for 50-MeV protons. In this example the silicon doping concentration is 1016 atoms/cm3. Devices that depend on long carrier lifetime for operation are degraded significantly at levels between 1010 and 1011 p/cm2, but devices that begin with shorter initial lifetimes (or are insensitive to lifetime changes, such as MOS transistors) will be unaffected until much higher levels of radiation. Thus, the lower bound for concern about proton displacement effects is on the order of 1010 to 1011 p/cm2, for 50-MeV protons.

III - 13

Figure 3-5. Effect of 50-MeV protons on minority carrier lifetime in silicon.

Similar results hold for GaAs and related materials, but the damage constants are different. The operation of many GaAs semiconductors (e.g., JFETs and MESFETs) does not depend directly on carrier lifetime, and for this reason relatively little data exists in the literature on lifetime damage in GaAs and other III-V materials. Data on some types of LEDs are an important exception, but in most LED studies the lifetime has not been measured, just the overall performance of the LED after irradiation. Carrier Removal Carrier removal occurs when radiation introduces deep impurity levels that can act as amphoteric impurities. Carrier removal is the result of compensation of the initial dopant atoms by the radiation-induced impurities [Peas2]. The carrier removal rate depends on doping level as well as whether the material is n- or p-type. The units for carrier removal are cm-1. Table 3-2 below shows approximate carrier removal rates for electrons and protons in silicon and GaAs as a rough guide to the magnitude of carrier removal effects. For material with a doping concentration of 1015 atoms/cm3 carrier removal will begin to become significant at proton fluences on the order of 1012 p/cm2, about two orders of magnitude higher than the “lower bound” fluences that are of concern for lifetime degradation. Higher fluences are required in order for carrier removal to be important in materials that are doped at higher levels. Table 3-2. Approximate Carrier Removal Rates for Protons and Electrons

III - 14

For small numbers of defects the deep-level impurities that are introduced by radiation act as amphoteric dopants. However, the impurities are not located exactly at the center of the bandgap and consequently the net doping concentration will shift towards n- or p-type when very large numbers of defects are present. For example, n-type silicon becomes p-type after high levels of neutron or proton irradiation . This will be discussed further in Section 11. Mobility Carrier mobility is also affected by displacement damage. Radiation-induced defects increase carrier scattering, reducing carrier mobility. Mobility degradation is important for MOS devices because mobility is directly related to transconductance, but mobility degradation generally requires much higher particle fluences than the levels required to produce large changes in minority carrier liftetime [Srou1]. Lifetime damage and carrier removal are usually the most important material parameters for optoelectronic devices, and mobility degradation will not be considered further in this course.

4 - KEY PROPERTIES OF SEMICONDUCTORS USED IN OPTOELECTRONIC DEVICES A. Band Structure Basic features of the interaction of light quanta (photons) with semiconductors can be explained with the band theory of solids, which describes the energy and momentum of holes and electrons within a semiconductor crystal (a periodic lattice). Allowable quantum states within a periodic structure are described in terms of crystal momentum, p, and wavevector, k, corresponding to the electron (or hole) states within the lattice. Note that p and k are both vectors; they are “natural” descriptions of solutions to the quantum-mechanical problem of stable states within an idealized periodic crystal (Bloch functions). As shown in Reference [Achc1], the wave vector k is related to lattice periodicity and is a basic characteristic of eigenvectors that are allowable quantum states within the lattice. Figure 4-1 shows the way that energy depends on the wave vector for a simple periodic lattice. Near the boundary of the potential change (effectively a lattice site) there is a region of forbidden energy that arises from the quantum properties of the crystal. This leads to an energy gap between different wave vector states, as shown by the periodic discontinuities in the figure.

Figure 4-1. Relationship between energy and wave vector for a simple periodic potential showing energy gaps between allowed wave vector states.

III - 15

With this formalism, the momentum of a hole or electron within the crystal is related to the wave vector that describes it by h k /2π (2) p = where h is Planck’s constant (6.62 x 10-27 erg-s). The momentum generally has a different relationship with k for holes and electrons. Note that the crystal momentum for either electrons or holes differs from the classical momentum of a free particle because the particle behaves differently under the influence of the crystal potential. This difference can be taken into account by using the concept of an effective mass, related to the curvature of the energy band structure. The band structure of real materials is far more complex than that of the simple model described by the E-k relationship in Figure 4-1 (see [Sze1] for more detail). For optoelectronic devices the key feature of the band structure is the location of the minimum energy transition point in k-space. Figure 4-2 shows examples of two different band structures. For the “direct” semiconductor, the minimum energy in the conduction band occurs at the same k-value as the maximum energy in the valence band. This allows transitions between the two energy bands for the same value of k that can be accomplished by absorption or emission of a single photon. GaAs and several other types of semiconductors have band structures that allow this type of direct transition. Figure 4-2(b) shows a different structure where the minimum conduction band energy occurs for a different value of k than for the k-value corresponding to the maximum valence band energy. This type of indirect transition requires additional energy, typically provided by a phonon from the crystal lattice, to provide the additional momentum that is required to change the wave vector, k. Although direct transitions can still occur, they have very low probability, and require more energy than indirect transitions. Silicon is an example of a material with a band structure that is dominated by indirect transitions. Materials with indirect bandgap structures are very inefficient at producing light, which prevents their use in LEDs and lasers.

Figure 4-2. Band structure of direct and indirect semiconductors showing allowed wave vector transitions.

The examples in Figure 4-2 apply for crystals with low impurity concentrations where the band edges are well defined. For high impurity concentrations (above 1018 cm-3), the band edges are “smeared” because of the presence of many different impurity atoms. The result is a lowering of the bandgap for high impurity concentrations. The presence of such bandtails affects emission and absorption properties because transitions can then occur at energies that are below the bandgap energy for the basic material when it is lightly doped.

III - 16

B. Optical Emission and Absorption Although many properties of light are the result of its behavior as a wave, absorption and emission in atoms and solids requires that light be viewed from the standpoint of quantized photons (a beam or pulse of light corresponds to a packet of photons). The energy of a photon with frequency ν and wavelength λ = c/ν (c is the velocity of light) is given by the relationship E = hc/λ

(3)

The practical formula E = 1.24 / λ is useful to remember, where E has units of electron-volts, and λ has units of µm. The simplest optical absorption process in a solid corresponds to direct absorption by a particle in the valence band that causes its energy to increase to (or beyond) the energy of the valence band. When this occurs, two particles are effectively created: an electron in the conduction band, and a hole in the valence band. In order for this process to occur the photon energy must exceed the band gap in the material. The corresponding process for photon emission occurs when an electron in the conduction band loses energy by emitting a photon, and falls into the valence band. A hole and electron are “destroyed” or annihilated when photon emission occurs. As discussed earlier, the band structure of a semiconductor is important in determining whether direct absorption of a photon is the dominant process for transitions between states. Although direct transitions are possible in materials with indirect bandgaps, the probability of direct transitions in indirect semiconductors is small. Instead, the dominant process involves a second particle (usually a phonon of energy hϖ that is present because of thermal motion of atoms within the crystal lattice). Because of the additional energy provided by the phonon, the energy associated with the transition is slightly different from the bandgap energy for an indirect semiconductor. Indirect transitions take much longer to occur than direct transitions, and cause the dominant recombination process to be strongly affected by nonradiative recombination centers in a typical semiconductor junction instead of the radiation recombination that dominates for direct materials. This effectively means that indirect semiconductors cannot be used for laser diodes or LEDs. Other transitions are possible besides direct transitions, but will not be discussed here.

C. Absorption Coefficient Absorption of light can be described from a macroscopic point of view by the equation I =

Io e-αx

(4)

where Io is the initial light intensity at the surface, I is the reduced light intensity at a distance x within the material, and α is defined as the absorption coefficient. The quantity 1/α (absorption depth) corresponds to the depth in the material where the intensity has fallen to 1/e times its initial value. The absorption coefficient depends on wavelength. The dependence is abrupt for semiconductors with direct bandgap, but it is much more gradual for indirect semiconductors such as silicon. Figure 4-3 shows the absorption coefficient of silicon, GaAs, and InGaAs [Dash1, Case1, Sze1]. The absorption coefficient falls to very low values once the energy of the photon is below the energy corresponding to the bandgap in the material.

III - 17

Figure 4-3. Absorption coefficients of GaAs, InGaAs and silicon. Note the more gradual dependence of α on wavelength for silicon, which has an indirect bandgap.

The absorption coefficient depends on temperature primarily because the bandgap is temperature dependent. For silicon, the absorption coefficient changes about -1%/ºC for temperatures near room temperature. Although this may appear to be a small dependence, it causes substantial differences in light absorption even over a moderate temperature range. Note that because α depends on wavelength, the absorption depth also depends on wavelength. The absorption depth is also affected by temperature. For photodetectors these factors cause the responsivity to change with wavelength. This is discussed in more detail in Section 8.

D. Quantum Efficiency When a photon is absorbed within a semiconductor, equal number of holes and electrons are created. For a steady-state process, this results in an excess carrier density. The excess carrier population can decrease either through radiative or non-radiative processes. Typical nonradiative processes include lattice vibrations (thermal energy) or recombination through impurity centers. The internal quantum efficiency is defined as the fractional number of non-radiative processes relative to radiative processes. Carrier lifetimes can be assigned to radiative and nonradiative processes, which allows the internal quantum efficiency η to be defined as η = (1 + τrad/τnr)-1

(5)

where τrad is the radiative lifetime, and τnr is the non-radiative lifetime. In order to keep the quantum efficiency high, the ratio τrad/τnr should be as small as possible. The internal quantum efficiency places an upper bound on the efficiency of absorption and radiative processes, but the net efficiency of an optical emitter or detector must also consider other processes that lower the effective efficiency when one considers the overall operation of the device. The external quantum efficiency includes additional factors such as absorption of light in highly doped or transition regions, and reflection from interfaces which limit light transmission.

III - 18

E. Snell’s Law Light is refracted when it passes through two surfaces of different refractive index. A light wave incident at an angle Θ1 in a material with refractive index n1 will emerge at an angle Θ2 after it passes through the interface. The two angles depend on the refractive indices of the two media as described by Snell’s law

n1 n2

=

sin Θ2 sin Θ1

(6)

where n1 and n2 are the refractive indices of the two materials. For the case where the second material has a higher index of refraction than the first, this equation can be satisfied for all angles. However, for the case where the second material has a lower refractive index than the first material, this equation can no longer be satisfied when the angle of the refracted wave exceeds 90 °. Once the angle of incidence exceeds the critical angle defined by Θc = sin-1 (n1 / n2)

(7)

the incident wave can no longer be transmitted through the lower index medium. Instead, it is reflected at the interface (total internal reflection), as shown in Figure 4-4. The angle Θc is known as Brewster’s angle. Total internal reflection is extremely important in determining how optical power within LEDs and laser diodes can be transmitted externally because the high refractive index of III-V compounds causes the Brewster angle to be low, about 16 degrees. This means that light that is incident at the surface at angles above 16 degrees will be totally reflected, and cannot be extracted from the material.

Figure 4-4. Refraction at an interface where n2 < n1.

It is also the principle by which light can be guided through extremely long distances in optical fibers, which are coated with a cladding layer of lower refractive index than the fiber core. Table 4-1 shows the refractive index for several semiconductor materials as well as glass.

III - 19

Table 4-1. Index of Refraction for Various Materials Type

Refractive Index

Si AlAs GaAs SiO2 air

3.45 2.97 3.59 1.46 1.00

The index of refraction depends on wavelength for all materials. For semiconductors, the doping level also affects n; it decreases with increasing doping density. For example, high-purity n-type GaAs has an index of refraction at 0.9 µm of 3.54. The index of refraction is lowered to 3.24 -- a 10% decrease -- for n-GaAs that is doped with a concentration of 6 x 1017 cm-3.

F. Optical Transmission Optical transmission through an interface of two materials depends on the refractive index. At normal incidence, optical transmission is given by the relationship

ρ =

n o - n1

(8)

no + n1 where ρ is the ratio of the amplitudes of the incident and reflected waves that describe the light path, and no and n1 are the refractive indices of the first and second materials, respectively. Note that optical power depends on ρ2. If ρ is negative, the reflected wave is opposite in phase to that of the incident wave. The above equation is a reasonable approximation for moderate angles of incidence. However, at more extreme angles the amplitude and phase of the reflected wave depend on angle in a complex manner (see Reference Huen1). For the case where the first material is air and the second material is glass, n1 = 1.5, the reflected optical power is 0.04; that is about 4% of the incident energy is reflected from the interface. For GaAs, n1 is 3.54 and about 33% of the incident energy is reflected from an airGaAs surface (the term Fresnel loss is often used to describe energy loss due to reflections at interfaces). Note that total internal reflection limits the angle over which light can be transmitted for the case where the light wave originates within a material with high index of refraction. The net amount of light that is transmitted depends on the compound effects of the limitations imposed by the Brewster angle and Fresnel losses. This is shown in Figure 4-6 for GaAs and the more frequently encountered condition of a glass-air interface. For GaAs, only about 2% of the internal light is transmitted from GaAs to air compared to silicon. The use of an optical coating or index matching medium can increase the amount of energy that is transmitted through such interfaces. Nearly all optoelectronic devices use coatings or matching compounds to increase optical transmission at interfaces. Index matching coatings are important in radiation testing because their transmission properties may be affected by ionizing radiation.

III - 20

Figure 4-6. Effects of Fresnel loss and total internal reflection on light transmission for GaAs and silicondioxide with an air interface.

5 - PROPERTIES OF III-V SEMICONDUCTORS A. Electrical Properties of Basic Materials Many III-V semiconductors are direct-bandgap materials, and therefore can be used to fabricate light-emitting diodes, laser diodes, and other optoelectronic devices that are not possible with silicon or other materials with indirect bandgaps. Gallium arsenide is an example of a typical III-V direct-bandgap semiconductor, but there are several others of interest, as shown in Table 5-1. The bandgap energy and wavelength corresponding to the absorption edge are shown in the table. Light is not absorbed beyond the absorption edge that corresponds to the bandgap energy, which provides an upper limit to the material. Table 5-1. Bandgap Energies of Some Direct-Bandgap Semiconductors

Material

Bandgap Energy (eV)

Absorption Edge (µm)

Typical Application

GaAs

1.42

0.87

LED/laser technology

InP

1.35

0.92

LED/laser technology

GaSb

0.72

1.72

Long wavelength LEDs

InAs

0.36

3.44

Long wavelength detectors

It is much more difficult to grow high-purity III-V devices compared to semiconductors with a single element (such as silicon and germanium) because of the difficulty of maintaining constant growth of both types of atoms in the crystal. A great deal of effort has been spent in developing advanced methods of growing single-crystal devices in such materials. The presence of two different types of atoms in the III-V lattice makes it possible to form admixtures of different constituents, changing the bandgap. Although this is a problem when attempting to grow pure GaAs, this feature of III-V devices can be used to advantage because the properties of the semiconductor change when the ratio of the constituents is altered. This makes it possible to tailor the wavelength of light-emitting diodes and laser diodes by varying the relative

III - 21

composition. Only two such material classes will be discussed in these notes; others are discussed in References [Chua1] and [Osb1].

B. Ternary and Quaternary Materials with Variable Bandgap Although one usually thinks of III-V semiconductors as having only two constituents, it is possible to form solid solutions that incorporate a third type of atom and still retain the singlecrystal characteristics of a semiconductor. For example, by introducing varying amounts of aluminum, a solid solution of aluminum-gallium-arsenide is formed. The bandgap of this solid solution can be adjusted over a relatively wide range by varying the amount of aluminum [Yabl1]. The resulting composition is designated AlxGa1-xAs, where the subscript x denotes the fractional concentration of aluminum. Figure 5-1 shows how the bandgap and nominal wavelength depend on the fractional aluminum concentration. Note, however that the band structure changes to that of an indirect semiconductor for aluminum concentrations above 30%.

Figure 5-1. Bandgap and wavelength of the AlGaAs system

AlGaAs can be tailored to operate over a wavelength range of approximately 630 to 870 nm. The index of refraction also changes with composition, increasing as the amount of aluminum increases. One unusual property of AlGaAs is that the lattice constant changes only slightly as the composition is varied, making it possible to grow crystals with low defect concentrations. Other ternary semiconductor systems are possible besides AlGaAs, with different wavelength ranges, but generally have lattice constants that are not closely matched to other material types except for specific compositional ratios. That restricts their use to specific wavelengths where adequate lattice matching occurs. To overcome this basic material limitation, quaternary compounds have been investigated for LEDs and lasers that operate at longer wavelengths with better lattice matching than most ternary materials. AlGaAsP is widely used because of excellent lattice matching to fabricate devices with wavelengths of about 1.3 µm [Suzu1].

C. Heterojunctions A heterojunction is formed when two semiconductors that have different bandgap energies are joined together. If the two materials are of different types (e.g., n-and p-type semiconductors), then the junction formed by the two materials functions much like the p-n III - 22

junction of a conventional semiconductor material (homojunction), providing a barrier to minority carrier injection with rectifying properties [Chua1]. If the two materials in the heterojunction are of the same type, there is no rectifying junction but the heterojunction helps to confine minority carriers within the region with lower bandgap. In addition to the electrical properties, the two materials in a heterojunction have different refractive indices which is an important feature for LEDs and laser diodes. Figure 5-2 shows the bandgap structure of a GaAs/AlGaAs heterojunction. The GaAs region is p-doped with a bandgap of 1.42 eV, and the Al0.7Ga0.7As region is n-doped with a bandgap energy of 1.82 eV. It is common practice to use upper-case prefixes for heterojunction materials (e.g., N-AlGaAs). The heterostructure introduces a discontinuity in the conduction and valence bands that aids in carrier confinement. The bandgap discontinuity makes it possible to get efficient carrier injection over a very short distance compared to conventional p-n junctions. Although not shown in the figure, the refractive indices of the two semiconductors are 3.59 and 3.39, respectively, which acts to confine photons to the GaAs region with higher refractive index. Materials used for heterojunctions must have lattice constants that are closely matched in order to minimize crystal defects. For example, a lattice spacing mismatch of 1% will introduce a dislocation at approximately every 100 lattice plains (about 500 Å). The requirement for close lattice matching severely limits the material choices for heterojunctions. As discussed earlier, AlGaAs has the unusual property that the lattice constant is nearly the same over a wide range of concentrations, which makes it a good choice for bandgap engineering.

Figure 5-2. Diagram of a GaAs/AlGaAs heterojunction structure.

D. Strained Lattices in Thin Layers Lattice strain is a major problem for conventional heterostructures because it generally introduces defects in the material that interfere with device operation, limiting yield and reliability. However, it is possible to deliberately produce strained lattices over a restricted distance that are stable, and do not result in defects. Strains within such structures change the properties of the crystal in a manner that can be advantageous for optoelectronic devices [Osb1]. For example, the effective mass of holes and electrons within III-V materials is affected by lattice strain; the effective mass of the holes is considerably reduced in strained lattices, increasing recombination efficiency compared to unstrained lattices of the same material. Auger recombination is also lower in strained lattices. These properties are deliberately exploited to

III - 23

fabricate laser diodes with improved operation at elevated temperature and lower threshold current [Ahn1]. Examples will be discussed in a later section. E. Quantum Wells Very compact layered structures can be used to fabricate dissimilar regions that effectively form quantum wells for confined carriers. Quantum effects become significant for layers with thicknesses below 200 A. Figure 5-3 shows the band diagram and energy state densities for a two-level quantum structure. The structure is formed by an AlGaAs/GaAs heterojunction. In this case the width of the structure is less than 100 A, causing the energy diagram in that direction to be restricted to discrete, quantized energy states. The hole states are split into two regions corresponding to “light” and ‘heavy” holes.

Figure 5-3. Energy band diagram and energy state density of a simple quantum structure.

The quantum properties of the structure modify the absorption spectrum and the effective gain of regenerative optical structures that are used in lasers, increasing the stability of the laser and decreasing threshold current, although it is much more difficult to fabricate lasers with such thin layers. The details are quite complicated, and beyond the scope of this course. Readers are referred to References [Chon1] and [Derr1] for details. Many new lasers are fabricated with quantum-well structures, and is important to have a basic understanding of how they are fabricated.

6 - LIGHT EMITTING DIODES A. Basic Features Light-emitting diodes can be fabricated with direct bandgap semiconductors such as GaAs. The key property is that carrier recombination within the junction formed by the semiconductor regions must have a high probability of producing a photon, along with a long enough carrier lifetime to allow the light to travel to the interface where it can be emitted. Even in diodes made with direct semiconductors most carrier recombination occurs at surface regions under very low injection conditions, and consequently there is very little light output at low current levels. However, when the current through the junction increases to a sufficiently high level the surface recombination saturates and the dominant recombination process begins to occur within the junction. This produces large numbers of photons in a direct bandgap material. The current-voltage and light power-voltage dependencies of a typical LED are shown in Figure 6-1 (in this figure, the light output was measured with a photodiode, and is measured in relative units). Note the change in slope that occurs in the forward voltage characteristics at the onset of light emission. Ideally the I-V characteristics can be described as the superposition of a

III - 24

non-radiative recombination term (with ideality factor = 2) and a diffusion term corresponding to the region where radiation recombination within the junction dominates (with ideality factor = 1): (9) I = Krec exp (qVF/2kT) + Kdiff exp(qVF/kT) where Krec and Kdiff are the factors for the recombination and diffusion terms, q is electronic charge, VF is the forward voltage, k is Boltzmann’s constant and T is absolute temperature.

Figure 6-1. I-V and optical power of an LED.

Some types of LEDs show behavior that deviates from the ideal relationship of the previous figure. An example is shown in Figure 6-2 where the slope at low injection changes only gradually as the injection level increases. This may be due to defects within the LED structure that are sensitive to injected current, and also contribute to nonradiative recombination just as the surface recombination component. As noted later in this section, proton damage causes a large increase in the recombination rate at low-to-moderate currents, effectively shunting much of the current into non-radiative processes and decreasing the forward voltage. This causes the slope at low and moderate injection to change in irradiated devices. Thus, even though LEDs are rarely used under low injection conditions, measuring their properties under low injection is an important characterization tool to evaluate non-radiative recombination after radiation damage has occurred. There are also unit-to-unit differences in the slope of the current-voltage characteristics prior to irradiation that may indicate the presence of internal defects in the crystalline structure of the device. The light output of an LED is affected by temperature, falling approximately 1% for each 1 ºC increase in temperature. This is a strong temperature effect, and it must be taken into account when performing radiation tests as well as in determining the degradation and margin in system applications. Many LEDs operate under conditions where the device temperature is 3040 ºC above the nominal temperature of heat sinks or base plates, which causes the light output to be significantly lower than the light output at nominal room temperature conditions.

III - 25

Figure 6-2. Abnormal behavior of I-V characteristics of an unirradiated LED.

LEDs also have a “wearout” mechanism that causes gradual degradation in the light output over time (there is no analogous mechanism for conventional silicon electronics). The amount of degradation depends on the LED technology and the amount of current that flows through the LED. Figure 6-3 shows typical degradation curves for wearout for a high-reliability LED that is rated for 100 mA maximum DC operating current. Note the pronounced difference in degradation for devices that are not subjected to initial burn-in conditions. The curves in Figure 6-3 are for typical devices. Wearout degradation can vary for different devices from the same batch or wafer lot. Most LEDs in space applications are operated at currents well below the maximum rated current in order to reduce the amount of degradation from “wearout.”

Figure 6-3. Decrease of light output of an LED with extended operation at maximum current.

The data in Figure 6-3 are normalized to values of light output before the stress time begins, and do not take into account possible reductions in light output due to burn-in for the devices that were initially subjected to burn-in. Thus part of the reason for the apparent improvement of performance of burned in devices may simply be due to decrease of lifetime during burn-in. III - 26

B. Properties of Various LED Technologies Many different materials and physical structures can be used to fabricate LEDs, and many advances in LED technology have been made during the last thirty years. In spite of this progress, many LEDs in the near infrared region are still fabricated with a very old process that relies on amphoteric doping, an inexpensive process that produces LEDs with very high output efficiency that are closely matched to the peak responsivity of silicon detectors. Amphoteric doping relies on a dopant (typical Si for GaAs and AlGaAs) that can act either as an n- or p-type impurity, depending on the growth temperature. With amphoteric doping, it is possible to create a p-n junction in a layer that is initially doped with only a single impurity by gradually altering the temperature during the growth of the epitaxial layer. The resulting structure, shown in the diagram of Figure 6-4, has a graded doping level, and is a compensated semiconductor (the net doping level depends on the difference between the background doping level and the impurities that are altered by the high-temperature growth process). The optical efficiency of amphoteric devices is very high for a number of reasons, including reduced free carrier absorption because of the compensation, and the existence of a complex about 0.1 eV below the valence band that effectively eliminates band-to-band absorption processes that would normally increase the amount of non-radiative recombination, decreasing efficiency [Kres1]. Amphoterically doped devices require relatively long minority carrier lifetimes to operate efficiently because the graded junction extends over a region that is 50-100 µm wide.

Figure 6-4. Diagram of an amphoterically doped LED.

It is also possible to fabricate LEDs with conventional diffusion processes where different types of impurities are diffused into the starting material to create layers with different doping levels. Most diffused LEDs are made in the visible region, and are generally of less interest for optoelectronic applications in space compared to those in the near infrared region. However, one manufacturer used diffused LEDs with a wavelength of 700 nm in their line of optocouplers, so there are cases where diffused LED performance is important.

III - 27

More advanced LEDs are manufactured with a multi-layer structure, sandwiching the active region between two layers of a different semiconductor type that confines the carriers to a narrow active region. A diagram of a double-heterostructure is shown in Figure 6-5.

Figure 6-5. Diagram of a double-heterojunction LED.

Many more fabrication steps are required to fabricate such devices, and the different materials used in the heterojunctions must have nearly identical lattice constants in order to avoid defects. The active region of a double-heterojunction LED is generally very thin, on the order of 1-4 µm. The thin layer allows pulsed or high frequency modulation at higher effective bandwidth than amphoterically doped LEDs, and also makes performance less dependent on lifetime.

C. Radiation Degradation Amphoterically Doped LEDs The effect of proton degradation on a typical amphoterically doped LED is shown in Figure 6-6. This type of LED is extremely sensitive to displacement damage mainly because the width of the junction – 50 to 100 µm – requires a long minority carrier lifetime. Note that the light output has been reduced to about 30% of initial value at a fluence of 3 x 1010 p/cm2, which is comparable to the total fluence from a single intense solar flare.

Figure 6-6. Proton degradation of an amphoterically doped LED

III - 28

The damage is higher at low currents, and the current dependence has to be taken into account when radiation testing is done in order to make sure that the actual use conditions are included in the measurement set. Operating the device at high current reduces radiation degradation, but it increases the amount of degradation due to aging and consequently most space applications of LEDs restrict the operating current to less than 1/3 of the maximum operating current. Older data on amphoterically doped LEDs also indicated that they were degraded at low levels [Rose1, Hum1, Dimi1, Lisc1], but the measurements were over a more restricted range of currents and operating conditions. Another way to evaluate LEDs is to measure light output as a function of forward voltage. Figure 6-7 shows a plot of the light output along with a plot of forward current in the diode vs. forward voltage. Note that although the light output is far lower after irradiation, the operating current at which light first begins to be produced is unaffected by the radiation damage. This is in sharp contrast to laser diodes (discussed in the next section) where the threshold current is strongly affected by radiation damage.

Figure 6-7. Dependence of forward voltage and light output on injection level for an amphoterically doped LED before and after irradiation with protons.

Damage in these types of LEDs is very stable, even over time periods of several months as long as the device remains unbiased. However, part of the damage can be readily annealed by applying a forward current through the device after irradiation [Lang1, Barn1, Barn2]. Figure 68 shows the recovery of three different types of amphoterically doped LEDs when a moderate current is applied after they are degraded by radiation. About 20% of the damage recovers if one waits for long time periods, although there are differences in how long one must apply the current to get recovery in LEDs from different processes and manufacturers. The degree of recovery depends on the total charge that flows through the device after irradiation, and is the same even in cases where the device remains unused for periods of many months after irradiation as for devices where current is passed through the device shortly after the irradiation has been completed.

III - 29

Figure 6-8. Recovery of damage in amphoterically doped LEDs under forward bias.

Diffused LEDs Although only a limited number of LEDs made with conventional diffused processes have been included in radiation effects studies, they appear to be less susceptible to displacement damage than amphoterically doped LEDs. Most diffused LEDs have shorter wavelengths, within the visible spectrum. Although it is possible to produce conventional diffused LEDs in the 850930 nm range that is near the peak in responsivity for silicon detectors, amphoterically doped LEDs are much more efficient, and dominate that range of wavelengths.

Figure 6-9. Proton degradation of GaAsP LEDs manufactured with a conventional diffused process.

Figure 6-9 shows the degradation of GaAsP devices with a wavelength of 700 nm when they are irradiated with 50-MeV protons. Although these devices are far less affected by radiation than amphoterically doped LEDs, the initial light output is also lower compared to that of amphoterically doped LEDs in the near infrared region. III - 30

Double-Heterojunction LEDs Double-heterojunction LEDs are, on average, much more resistant to displacement damage than amphoterically doped devices. Figure 6-10 shows representative results for doubleheterojunction LEDs with a wavelength of 820 nm that are intended for high-reliability applications. Unlike amphoterically doped devices, double-heterojunction LEDs do not exhibit injection-enhanced annealing [John1]. Although the increased radiation resistance would appear to make double-heterojunction devices a better choice for space applications, amphoterically doped devices have considerably more initial light output than DH LEDs in the 820-900 nm region, and this tradeoff has to be taken into account when devices are selected and characterized. Device uniformity is another potential issue for double-heterojunction LEDs. In some cases a small number -- 5 to 10% of the population -- exhibit large decreases in light output when they are operated at moderate currents, as shown by the lower set of curves in Figure 6-10. The abnormal devices exhibit a large increase in non-radiative recombination at moderate injection levels, which results in much lower forward voltage after irradiation. This behavior may be related to defects in the material used to fabricate the LED; similar effects have been observed in reliability studies of double-heterojunction LEDs that are subjected to operating stress [Lind1].

LIGHT OUTP UT (NORM ALIZED)

1

M ean of 15 parts

0.8

Worst of 17 devi ces 0.6

0.4

Solid lines : 5 x 10 10 p/cm 2 Das hed lines: 8 x 10 10 p/cm 2

0.2

0 0

20

40

60

80

100

120

140

160

LED C URRE NT (mA)

Figure 6-10. Proton degradation of a double-heterojunction LED.

A comparison of degradation of several different types of LEDs is shown in Figure 6-11. These curves represent mean devices from test lots of approximately 30 parts of each device type. They do not take unit-to-unit variability into account, which is typically about a factor of two for amphoterically doped devices, and somewhat greater (and far less predictable) for double-heterojunction LEDs. The results show that amphoterically doped LEDs degrade by significant amounts at relatively low radiation levels, which can cause severe problems in space applications unless additional design margin is included to take the degradation into account. Visible (diffused) and more advanced double-heterojunction devices – shown in the upper curves of Figure 6-11 -- can operate at radiation levels that are about two orders of magnitude higher than that of the amphoterically doped devices.

III - 31

Figure 6-11. Comparison of degradation of several different types of LEDs from various manufacturers.

Damage in light-emitting diodes depends on fluence in a nonlinear manner. The relative change in light output is actually greater at higher fluences compared to incremental changes at low fluences. This causes samples with more extreme radiation damage to fall well below the mean values of other devices from the same lot at high fluences. Degradation in amphoterically doped LEDs appears to be dominated by changes in minority carrier lifetime. The levels at which double-heterojunction LEDs degrade is sufficiently high for carrier removal affects to be important, which makes it more difficult to analytically determine how the device degrades. The complex structure of double-heterojunction LEDs, with thin layers of AlGaAs and GaAs material, add further complexity to this problem.

C. Radiation Testing of Light Emitting Diodes Proton testing is more difficult and costly than tests with passive sources (such as gamma rays), and considerably more planning is needed to do this type of testing. In addition to facility costs, test boards are activated by the radiation, making it necessary to avoid prolonged exposure to boards and equipment when tests are done between radiation steps. One approach that can be used for radiation testing is to mount devices in a pattern on a test board that places them within the uniform region of an accelerator beam (typcially a diameter of 5-10 cm). This allows several devices to be irradiated simultaneously, reducing the testing cost. At predetermined intervals the irradiation is stopped, and the devices are removed from the accelerator area and placed in a light-tight transition fixture that couples light from each LED to a corresponding photodiode. The assembly, shown in Figure 6-11 is designed to provide uniform physical spacing between the LED and the photodiodes used to measure the light output. It may be necessary to control the temperature of the LEDs during measurement to eliminate III - 32

interference from temperature effects that affect the LED light output. Peripheral electronics are connected to the LED and phototransistor assemblies through cable arrays, using special care to allow low level measurements.

Figure 6-11. Diagram of a test assembly used for irradiation and testing of an array of LEDs.

The light output of LEDs can vary substantially with the position and angle subtended by the detector. This makes it vitally important to have test fixtures that place the LED and detector in fixed, reproducible positions so that the measurements are consistent at different radiation levels (the assembly has to be disassembled for each irradiation because the photodetectors would be damaged by the radiation). Figure 6-12 shows the dependence of light output on angle for two types of LEDs. Note the large “dip” in light output for the LED that is packaged with an internal lens.

Figure 6-12. Angular dependence of the light output of LEDs in two different package configurations. The LED with the internal lens has a very strong variation in light output with angular position, increasing the difficulty of making reproducible measurements.

III - 33

The diagram in Figure 6-12 shows another potential issue for LED testing. In some cases the window or lens used as part of the LED assembly can be affected by total dose, which introduces color centers that darken the material. Consequently it is important to evaluate the lens or window, at least indirectly, to make sure that ionization damage in the window or lens is not contributing to the degradation. Alternatively, devices that are tested without an internal lens may underestimate the damage of equivalent devices with an internal lens that is sensitive to total dose darkening.

7 - LASER DIODES A. Basic Features Semiconductor lasers depend on the process of stimulated emission. Operation of the laser requires that internal quantum states within the semiconductor are “pumped” by an external or internal source so that their population is well above the thermal equilibrium level (inverted condition). Stimulated emission occurs when a photon -- produced as a result of normal recombination processes -- travels through the pumped laser cavity. The initial photon will cause an additional photon to be produced through recombination while producing a new photon that is an exact duplicate of the photon that initiated the recombination process. A property of stimulated emission is that the duplicate photon has the same energy, direction, and polarization as the initial photon [Lash1]. For an optical cavity with an inverted population, the probability of stimulated emission becomes very high. The result is a “flood” of photons with nearly constant energy and direction that are triggered by a small number of initial photons (produced by normal recombination processes) within the cavity. Steady-state laser operation requires that the rate of production of excited levels equals the rate that they are depleted by stimulated emission, leading to a minimum threshold current condition for operation. Laser cavities have an effective optical gain that depends on the material properties and the design of the laser cavity [Bern1]. The threshold current (or current density) required to establish operation of the cavity as a laser is inversely proportional to the optical gain. Most semiconductor lasers are internally pumped, relying on current that flows through a p-n junction within the laser structure to pump the internal laser cavity with photons. As discussed above, these photons are generated by recombination within the junction. Figure 7-1 shows a simplified structure of a double-heterojunction laser. In this example the active layer is GaAs, surrounded by p- and n-doped layers of AlGaAs that form a p-n junction in a transverse direction to the laser cavity. The AlGaAs layers have a lower refractive index, confining the photons that are emitted by stimulated emission to the GaAs region at the center. The AlGaAs layers form

Figure 7-1. Simplified diagram of a double-heterojunction semiconductor laser.

III - 34

heterojuctions that injects carriers very efficiently into the GaAs region. The laser is similar in construction to the double-heterojunction LED, but has cleaved facets to partially reflect the light in a precise transverse direction along the axis. The laser facets must be parallel and smooth, which is an additional constraint during manufacturing. It must also be designed to withstand higher current densities than typical LEDs. The dependence of light intensity on injected current is shown in Figure 7-2 for an older laser technology. At low injection, the slope of the forward current vs. VF characteristic is two. In this region most of the recombination occurs at surface regions or at non-radiative defect sites within the junction, and very little light is produced. As the current increases, the slope of the current-voltage curve decreases to one, and the structure begins to emit light because most of the recombination is now due to radiative recombination within the junction. In this region the laser operates very much like a light-emitting diode because the optical gain of the cavity is too low for laser operation. The spectral width is relatively broad, typically 6-8% of the peak wavelength. As the current increases, the slope begins to change once again, and it increases with a very steep slope when the injection level is high enough to cause the structure to lase (at approximately 100 mA in this example). Once this region is reached, there is an abrupt increase in light power along with a sharp decrease in spectral width to a very small fraction --0.2 to 1% -of the peak wavelength. At still higher current the slope decreases due to thermal effects and internal resistance. Under pulsed conditions the slope at high current is considerably steeper (heating is reduced), and it is possible to operate the device at much higher currents compared to steady-state conditions.

Figure 7-2. Light output vs. injection level for a laser diode and LED.

Threshold current is one of the most important parameters, and the threshold current of newer laser structures is considerably lower than the older example shown in Figure 7-2. Threshold current is highly sensitive to temperature, as shown in Figure 7-3, and for AlGaAs it also depends on the composition. Because of the extreme sensitivity to temperature it is usually necessary to precisely control the laser temperature, or to provide external feedback circuitry to control the current through the laser at fixed value above the threshold current. This can be done

III - 35

by monitoring the light output, and using the sampled light as the input to the feedback control network. Some lasers are fabricated with internal photodetectors that can be used for that purpose. Lasers operate at much higher operating current densities than LEDs, which can adversely affect their reliability. This has been a barrier to their application in space systems in the past, but laser technology has improved to the point where they can be used in undersea applications and long-haul communications networks with comparable reliability requirements [Fuji1]. Nevertheless, it is more difficult to use lasers than LEDs in most systems, partly because of the need to carefully control the current through the device and to limit the actual operating temperature range.

Figure 7-3. Dependence of threshold current on temperature for various fractions of aluminum.

Lasers operate at very high internal power levels, and they generally have more severe wearout and reliability problems compared to LEDs. Laser failures can be grouped into three different categories: (1) gradual increase in threshold current with extended operation (2) gradual degradation of light output due to the development of dark-line defects (3) catastrophic degradation or “steps” with lower light output, attributed to facet degradation because of the high power levels associated with laser operation. Lasers fabricated with AlGaAs/GaAs (with wavelengths between 940 and 990 nm) exhibit all three failure modes. However, lasers that are fabricated with the InGaAs/InP system (1300 1500 nm) appear to be less affected by dark line defect degradation than lasers fabricated with AlGaAs/GaAs. Changes in material technology have lowered the defect level, and those changes along with decreases in threshold current density have improved laser reliability considerably compared to reliability of earlier lasers.

B. More Advanced Laser Structures Numerous advances have been made in laser technology during the last 20 years [Mose1, Ghit1, Fu1, Shim1, Su1], and these changes have allowed special structures to be made with new materials, covering a much wider range of wavelengths compared to older laser diode structures. Threshold current has also been reduced by several orders of magnitude for newer types of lasers. The ability to grow extremely thin layers of material by molecular-beam epitaxy provides III - 36

many additional degrees of freedom in designing specific laser structures. However, this is very confusing when one examines the relatively limited data on radiation damage that is available for lasers, because the laser structures are often quite different. The first lasers were homojunction structures which required extremely high current densities, ≈ 100,000 A/cm2. They were only capable of operation for a few hours. Heterojunction structures were developed in the early 1970’s which provided a much more efficient way to inject carriers and decrease threshold current density. Table 7-1 shows how the threshold current has decreased during forty years of development. Note how much the threshold current has been reduced for advanced lasers. Table 7-1. Threshold Current Trends in Laser Development

Although it is not possible to cover laser technology in a great deal of depth in this course, we will briefly discuss strained layers, which are used in some types of lasers. Properly designed strained layers can be used to design lasers with improved operation. Some of the advantages of strained layers include reduced effective mass for holes (the high effective mass of holes in unstrained semiconductors limits efficiency), reduced Auger recombination, and the ability to operate the laser structure at higher temperatures [Chua1]. Figure 7-4 shows a calculation of the optical gain of an InGaAs laser structure (designed for a wavelength of 1.55 µm) for an unstrained lattice and one that is deliberately designed to induce a strain of 1.5% along one axis of the crystal [Liu1]. The optical gain is nearly four times higher in the strained layer, with a current density that is also much lower. This illustrates how strained lattices can be used to design lasers with lower threshold current and improved operation. Strained layers can be combined with quantum wells to design advanced lasers. Single and multiple quantum-well lasers are available that have far better performance than older laser technologies. However, these structures require very sophisticated processing steps along with the ability to grow extremely thin layers of material that are closely matched to the wavelength of the laser. The thickness and structure of the layers used in fabrication have to come very close to meeting the conditions for the expected laser wavelength in order for the structure to work properly, and this may cause larger unit-to-unit differences to occur in the properties of the lasers and their sensitivity to radiation damage compared to older structures used in LEDs and lasers.

III - 37

Figure 7-4. Optical gain of strained and unstrained InGaAs laser cavities.

C. Radiation Degradation Only a limited amount of data is available on radiation degradation of lasers, partly because earlier laser technologies did not have adequate reliability for applications in military or space systems. The operating life of early lasers was on the order of hours. The reliability of newer lasers is far better, but it is still more difficult to apply lasers in space because of the need to control operating current and temperature. Lasers are now being considered for space use, particularly in high-speed data links. Threshold current is one of the most critical parameters of semiconductor lasers. Displacement damage causes the threshold current of a typical laser to increase, as shown in Figure 7-5 for an older AlGaAs laser [Chow1]. In this example the device (tested in 1989) was

Figure 7-5. Increase in threshold current of a laser diode after irradiation with 1-MeV (equivalent) neutrons.

III - 38

irradiated with neutrons; it is shown because there is very little data available with protons for older laser structures. Note the very high threshold current. Moderate levels of radiation damage shift the threshold current to higher values, but cause only a very slight change in the slope of the light output characteristics (sometimes referred to as slope efficiency). At higher levels the nature of the damage changes, and the slope is severely degraded along with the threshold current. The increase in output power at high currents after the highest radiation level is due to annealing from the high power dissipation within the laser. Note particularly the increase in optical power that occurs at a current of 1 ampere. Although this data is for neutrons, NIEL calculations show that 1-MeV neutrons have nearly the same effective displacement damage effect as 200 MeV protons, allowing comparison with more recent laser structures with proton degradation. An example of degradation of a more advanced laser structure is shown in Figure 7-6 [Evan1]. This was an advanced strained quantum-well laser that was designed for operation over a wide temperature range. The original data was taken with 5.5 MeV protons because they were conveniently available to the research group that developed the laser. The data in the figure have been altered by applying a NIEL factor of 6.4 so that the results are equivalent to 50 MeV protons. This was done to allow a more intuitive comparison of the degradation of this device with other data in the course. The main effect of the radiation damage is to shift the threshold current; the magnitude of the change is almost exactly proportional to fluence. Note that the slope is essentially unchanged, even after the highest radiation level.

Figure 7-6. Degradation of a strained quantum-well laser after irradiation with protons (data reported in equivalent 50 MeV protons, applying a factor of 6.4 to allow for the increased effective damage of the 5.5 MeV protons used in the original work).

More recent data on a commercial multi-quantum well laser (operating at 780 nm) is shown in Figure 7-7 [Zhao1]. The results are quite similar to those obtained for the strained quantumwell laser in the previous figure, although the threshold current of the 780 nm device is slightly lower. Threshold current shifts by about the same relative amount after irradiation, and there is little change in the slope efficiency. In subsequent work they reported significant variability in the radiation degradation of different units from the same test lot [Zhao2], illustrating that unitto-unit variability can be an important issue for advanced lasers.

III - 39

Figure 7-7. Degradation of a multi quantum-well laser after irradiation with protons. Zhao et al. also found that the lasers that they studied were sensitive to injection-enhanced annealing. In some cases the threshold current that was initially degraded by radiation recovered to a lower value than that observed prior to irradiation. They attributed this to change in the refractive index of internal ridge waveguides used in the laser structure, although other effects may also contribute. Their results illustrate yet another complication in evaluating advanced types of lasers.

D. Vertical Cavity Semiconductor Lasers (VCSELs) A new approach for laser design has been developed during the last ten years that provides many advantages compared to conventional laser diodes. A diagram of a vertical-cavity semiconductor laser (VCSEL) is shown in Figure 7-8 [Jewe1, Choq1]. VCSELs have very low threshold currents compared to conventional laser structures because of the small dimensions of the laser cavity.

Figure 7-8. Diagram of a vertical-cavity semiconductor laser.

III - 40

VCSELs can be made with very low threshold current, a major advantage compared to conventional laser diodes. Their small size also reduces the dependence of threshold current on temperature, which not only reduces the power required for the device, but also eliminates the need for external temperature control or feedback circuitry that is usually required for more conventional lasers. VCSELs are usually operated over a relatively narrow range of currents because internal heating of the small laser cavity reduces output efficiency at currents that are more than about a factor of three above the threshold current. Proton damage in VCSELs causes the threshold current to increase, just as for conventional lasers. Figure 7-9 shows results from VCSEL devices that were fabricated at Sandia National Laboratories [Paxt1, Barn3]. Note that the slope efficiency of VCSELs is more affected by moderate levels of radiation than for conventional lasers. At high currents the small VCSEL cavity undergoes substantial heating, increasing the amount of non-radiative current in the device. That is the main reason for the large drop in light output at higher operating conditions after irradiaton. The limited radiation data that is available for VCSELs indicates that they are only degraded by very high levels of radiation, and appear to be a good choice for optical emitters in most space systems.

Figure 7-9. Effect of protons on the operating conditions of a VCSEL.

E. Radiation Testing Considerations for Semiconductor Lasers The basic approach used for characterizing lasers is similar to that of LEDs, with even more concern about physical alignment between the laser and detector because lasers typically have a much narrower beam angle. Special care needs to be taken to control the device temperature with high precision when measurements are made between irradiations. This is required for lasers because the threshold current is so strongly affected by temperature that the temperature dependence will interfere with measurements to determine the effects of radiation on threshold current except for very large changes. Special thermoelectric modules are available that provide a convenient way to control temperature for devices that are suitably mounted. Although controlling case temperature is effective when moderate currents are used for characterization, internal heating can be important at high currents and it may be necessary to restrict the operating duty cycle to keep from heating the device to the point where it affects measurements. The most important parameters for lasers are the threshold current and slope efficiency. Lasers are often intended for fiber-optic applications, and it may be necessary to extend the measurements to much higher current values using pulsed current sources (with careful attention given to controlling the power and duty cycle) in order to encompass the operating characteristics in the application within the measurements done to characterize radiation degradation. The data

III - 41

in Figures 7-5 and 7-9 shows that the optical power can be more degraded at high operating currents, and it is not possible to extrapolate data taken at lower currents to the very high currents that are sometimes used for pulsed operation. More thorough characterization testing can be done, including measurement of wavelength and spectral width. This requires an optical spectrometer, and such measurements are considerably more time consuming than indirect measurements of the optical power with a photodiode or other basic detector. At moderate radiation levels most lasers that have been tested in the past do not show significant changes in wavelength or spectral width, but this may not be valid for new types of lasers. The underlying mechanisms for degradation of advanced laser technologies are not well understood. Carrier removal and microscopic cluster damage, similar to that reported for silicon CCDs [Srou2] may play a role in damage in new types of lasers, and more thorough characterization is generally recommended.

8 - OPTICAL DETECTORS A. Basic Considerations Several different mechanisms can be used to detect photons, but the treatment here will be limited to two mechanisms that are useful with semicocondutor detectors: photoconductivity and the photovoltaic effect. Photoconductivity Light absorbed within a semiconductor increases electrical conductivity because excess carriers are generated (recall that conductivity, σ , is related to the carrier density by the equation σ = n q µ where n is the carrier density, q is electronic charge, and µ is the mobility). For intrinsic semiconductors, absorption depends on wavelength as discussed in Section 3, up to the bandgap edge (the photon energy must exceed the bandgap energy). Table 8-1 lists the properties of some semiconductors that are commonly used as detectors. Table 8-1. Properties of Some Semiconductors Used as Detectors

Material

Temperature (ºK)

CdS GaP GaAs Si InGaAs Ge PbS InAs PbSe InSb HgCdTe

295 295 295 295 295 295 295 195 195 77 77

Wavelength Limit (µm)

0.52 0.56 0.92 1.1 1.6 1.8 2.9 3.2 5.4 12 12

Normally semiconductor detectors rely on direct excitation of carriers from the valence band to the conduction band, and can only be used at wavelengths below the “bandgap edge.” However, if the semiconductor is doped with a very high number of impurities then there is a very high concentration of impurity levels within the bandgap. and it is possible to have transitions from the impurity states to the valence or conduction bands that correspond to much lower energies compared to direct transitions. This process is referred to as extrinsic photoconductivity. For example, indium can be used as a dopant to extend absorption in silicon

III - 42

to about 5 µm [Scla1]. This type of detector needs to be operated at temperatures below 77 K in order to improve the signal-to-noise level Measurements (or applications) of photoconductivity require a suitable electrical circuit, and thus photoconductivity actually consists of the combination of the excess carrier generation process along with the way that the excess carriers are transported. Photoconductivity is the mechanism that causes photocurrents in reverse-biased photodiodes, discussed in the next subsection. Photovoltaic Effect Excess carriers generated by absorbed light produce photocurrents that will develop a voltage across an initially unbiased p-n junction. One application of the photovoltaic effect is in solar cells, which are designed to provide relatively large currents to external sources. It is also possible to use the photovoltaic effect for low-level photodetectors. That mode of operation provides some improvement in signal-to-noise ratio compared to photoconductive detection, but is generally slower than photoconductive processes in semiconductors because the low electric field that is present causes most of the charge to be collected by diffusion.

B. Elementary Detectors Based on p-n Junctions Responsivity The responsivity of detectors is determined by the dependence of the absorption coefficient on wavelength, discussed in Section 4-C, along with the effective depth of charge collection in the p-n junction structure. The lightly doped material must extend well beyond the maximum absorption depth in order to collect light near the bandgap edge. For indirect materials, such as silicon, this requires a collection depth of 100 µm or more. However, photodiodes made with direct bandgap materials do not require this extended depth because the absorption coefficient changes very little until the wavelength reaches the wavelength corresponding to the bandgap edge. Figure 8-1 shows the responsivity of a typical deep p-n silicon detector along with that of an AlGaAs detector (the wavelength limit can be tailored by selecting different concentrations of aluminum).

Figure 8-1. Responsivity of silicon and AlGaAs detectors.

III - 43

p-n and p-i-n Photodiodes Photodiodes are one of the most commonly used photodetectors. When an elementary photodiode is used in the reverse-biased mode, excess carriers generated within the depletion region are collected very rapidly because of the strong electric field that is present. Carriers generated outside the depletion region are collected by diffusion. This is shown schematically in Figure 8-2(a). Although the p-i-n photodiode in Figure 8-2(b) looks nearly the same as the conventional photodiode, there is an important difference in its operation. When a sufficiently high reverse bias is applied, the entire i- region is depleted of carriers and consequently all the photocurrent generated within the i- region is collected by field-enhanced drift, not diffusion. This decreases the response time and also results in an improved signal-to-noise ratio. It is possible to manufacture p-i-n detectors from direct-bandgap semiconductors as well as silicon. As discussed in section 4-C, direct-bandgap detectors can be made with a very shallow collection volume. This allows high-speed detectors to be fabricated that can be fully depleted at lower voltages compared to silicon p-i-n detectors.

Figure 8-2. Excess carrier generation and collection in a biased p-n junction.

The depth over which light is collected in a conventional photodiode depends on the absorption coefficient, α (see Section 3-C). This is depicted for a silicon photodetector (for which α has a somewhat gradual dependence on wavelength) in Figure 8-3. Light at very short wavelengths is collected near surface, and much of the photocurrent produced in that region is lost because of recombination in the highly doped contact region. Intermediate wavelengths are collected mainly by drift, with diffusion playing some role if the absorption depth extends beyond the depletion layer. Light at longer wavelengths is absorbed deep within the material. For example, at 850 nm the absorption depth is about 40 µm, and the depth of the detector must extend beyond that value in order to efficiently collect light at longer wavelength. Charge collection at longer wavelengths are more affected by radiation damage because the diffusion length is reduced, as shown in the figure. Figure 8-3b illustrates the effect of the reduced diffusion length after radiation damage on the depth over which carriers are collected in the photodiode structure. Because longer wavelengths are absorbed over extended distances, displacement damage has a more severe impact on longer wavelengths.

III - 44

Figure 8-3. Diagram showing change in absorption depth for light of different wavelengths in a silicon detector.

Phototransistor It is possible to use photocurrent in the collector-base region of a transistor as a p-n detector and then use the gain of the transistor to amplify photocurrent. There are several advantages in such structures, including reduction of stray capacitance and improved response time. A physical diagram of a phototransistor is shown in Figure 8-4. The structure consists of an extended base region surrounded by a narrow emitter ring (the triangular region of the emitter is a contact region provided to attach a bonding lead). The extended open base region functions as an integrated photodetector. Although the physical diagram is for a single transistor, a compound (Darlington) transistor is often used that provides considerably more gain compared to that of a single transistor, as shown on the right side of Figure 8-4.

Figure 8-4. Physical structure of a typical phototransistor.

Radiation Damage in Conventional and p-i-n Detectors The photoresponse of silicon p-i-n detectors is much less affected by radiation damage than the photoresponse of conventional p-n photodiodes, as shown in Figure 8-5. This difference is particularly large at longer wavelengths near the maximum responsivity of silicon, and is due to the fact that light collection in p-i-n detectors does not depend on minority carrier lifetime (all

III - 45

charge is collected by drift in the fully depleted i-layer). However, carrier removal effects in the lightly doped i-region cause leakage current to increase, and these changes are noticeable at levels below 1010 p/cm2. Leakage current changes in conventional photodiodes are much smaller because the doping levels are higher.

Figure 8-5. Effects of proton radiation damage on the photoresponse of a conventional and p-i-n detector.

C. Specialized Detector Technologies Avalanche Photodiode Avalanche photodiodes are physically very similar to conventional p-n photodiodes. However, they are operated at electric fields that are within the avalanche breakdown of the p-n junction where avalanche multiplication acts to increase the photocurrent. It also affects leakage current. The avalanche factor can be a factor of ten or more, providing higher current and a significant improvement in signal-to-noise compared to conventional photodiodes. However, the bias conditions must be carefully controlled to keep the device in a stable operating mode. A diagram of an avalanche photodiode is shown in Figure 8-6. The solid line shows the I-V characterisitics without light, and the dashed line shows how photocurrent affects the characteristics.

Figure 8-6. Physical diagram and operating conditions of an avalanche photodiode.

III - 46

MSM Photodetectors Another type of detector, the metal-semiconductor-metal (MSM) detector, is also used for high-speed applications, such as fiber optic links. MSM detectors are essentially Schottky diodes. This type of detector is very resistant to radiation damage, and was discussed in detail in Section III of the 1999 Short Course by P. W. Marshall and C. J. Marshall [Mars2], and that reference should be consulted for more details.

D. Noise and Figures of Merit Basic Noise Concepts Noise occurs in resistive materials in the form of a fluctuating voltage or current. The mean square noise power PN of a measurement is given by PN = k T ∆f (9) where k is Boltzmann’s constant, T is absolute temperature, and ∆f is the bandwidth of the measurement. Real noise measurements usually result in measurement of voltage or current with a specific source resistance, R. The ideal voltage noise vN from an unbiased resistor is then given by vN = (4 k T R ∆f)1/2

(10)

which has the somewhat unusual units of volts/√Hz. However, this form of unit arises naturally when noise is discussed. Note that for an ideal resistor the noise power density is the same at all frequencies. Noise from a resistor arises from basic thermodynamic considerations, and it establishes the lower limit for noise with a given bandwidth and resistor value. It is sometimes referred to as Johnson noise. An equivalent form to Equation 10 for the noise current that arises from the noise power can be easily derived. A second form of noise is generation-recombination (G-R) noise. For an intrinsic photoconductor where G-R noise is dominated by only one type of carrier (typically electrons), the short-circuit current term that occurs because of G-R noise from a current IB is 2IB

P τ ∆f

N

1 + ω2τ2

½

(11)

IN =

where N is the number of free electrons, P is the number of free holes, B is the bandwidth, τ is the lifetime of free carriers, and ω is the angular frequency. A third form of noise, shot noise , occurs because of fluctuations in the flow of carriers within a semiconductor junction. The shot noise current IN due to a current I flowing through the junction is given by IN = [(2 q I + 4 q Io ) ∆f]1/2

(12)

where Io is the ideal reverse bias junction current in the diode equation. A fourth noise source also exists that is termed “1/f noise” because it rises rapidly at very low frequencies. III - 47

An example of how these noise sources affect noise in an actual low-noise amplifier is shown in Figure 8-7. The importance of the various noise contributions depends on the bandwidth and frequency range that is of interest. The first component, Johnson noise, dominates for high frequencies or for detector applications with wide bandwidth. For more limited bandwidth applications the other noise terms become dominant. At very low frequencies, 1/f noise is usually the dominant noise term. Note that all noise terms are proportional to the square root of the bandwidth, and thus it is necessary to restrict the bandwidth in order to make effective low level measurements in cases where the sensitivity limit approaches the ideal noise “floor”. This figure is a useful guide in determining which noise sources are likely to be the dominant problem in low-noise designs.

Figure 8-7. Relative contributions of the different noise terms for an actual detector.

Signal-to-Noise Ratio: D* Noise in infrared detectors is still more complicated because they are sensitive enough to detect blackbody radiation from passive structures that surround them as well as the more usual electrical contributions to noise. Thermal energy from the background (or from materials that are within the field of view) ultimately limits sensitivity. A different figure of merit D* has been developed for detectors that are intended to operate near maximum sensitivity limits, and that figure of merit is usually used for infrared systems instead of the equivalent circuit noise terms that were discussed above. D* is defined as the rms signal-to-noise ratio in a 1-Hz bandwidth per unit rms incident radiant power (per square root of detector area), with units of cm-Hz1/2/W. D* can be defined in two different ways, either in response to a monochromatic source, usually referred to as D*λ,, or in response to a reference source of blackbody radiation (a wide range of wavelengths). In the latter case, the reference temperature is typically 500 K. The detector field of view is assumed to be hemispherical. The equation below shows how D* is related to normal measurement parameters (Adet ∆f)1/2 vs D*

= P

vn

(13)

where Adet is the detector area, ∆f is the bandwidth, P is the incident radiant power, vs is the rms signal amplitude and vn is the rms noise amplitude. III - 48

Figure 8-8 shows D* for various wavelengths and fields of view, assuming a background temperature of 290 K. At short wavelengths, there is very little difference in D* for different fields of view, but there is a very large dependence on viewing angle for wavelength between 5 and 10 µm. Far better detector performance can be obtained in the infrared region by cooling the detector and any associated apertures or baffling. More details can be obtained from Section IV of the 1993 Short Course [Pick1], including how various types of infrared detectors are affected by radiation.

Figure 8-8. D* for various wavelengths and fields of view, assuming a background temperature of 290 K.

Circuit Issues One circuit technique that is frequently used for photodetectors is the transimpedance amplifier, shown in Figure 8-9. This circuit uses a high-gain, low noise operational amplifier as a current-to-voltage converter. It provides much higher bandwidth than conventional amplifiers that use a resistor in series with a detector to provide an input voltage, partly because to first order there is no voltage change at the amplifier input (virtual ground). That lowers the effective capacitance at the input. Special transimpedance amplifiers are available that provide bandwidths above 1 GHz for high speed applications.

Figure 8-9. Diagram of a transimpedance amplifier. Transimpedance amplifiers are frequently used because they provide higher bandwidth and lower noise compared to conventional amplifiers.

In most cases better performance can be obtained by integrating a preamplifier (or complete amplifier) with the photodetector, much like the phototransistor is integrated in Figure 8-4. One of the main advantages with integrated amplifiers is the marked reduction in capacitance between the detector and amplifier input. Noise in high frequency amplifiers increases with the 3/2 power of input capacitance [Smit1]. III - 49

9 - OPTOCOUPLERS A. Basic Features Optocouplers use a light emitter (typically an LED) to provide an internal optical signal to a photodetector and amplifier (or phototransistor). The optocoupler provides a very high degree of isolation between the electrical signal that drives the LED and the output of the amplifier because there is no direct electrical connection between them (other than very small stray capacitance), only the optical signal. Figure 9-1 shows two different construction techniques for optocouplers. Optocouplers are very simple hybrid devices, consisting of an LED assembly, mounted on a carrier, with a silicon integrated circuit containing a photodiode and transistor (or high-speed amplifier). Some manufacturers produce only the photodiode/amplifier, purchasing the LED from outside sources, while others fabricate -- and control -- the LED as well as the silicon-based part of the optocoupler. Different physical configurations are used to fabricate optocouplers. Direct coupling, shown in Figure 9-1(a), uses a surface-emitting LED that is inverted and placed directly over the photodiode in the silicon die. This approach is straightforward, providing highly efficient light coupling. A thin layer of optical coupling material (barely detectable in the figure) is usually placed between the LED and phototransistor in order to reduce Fresnel losses. The indirect method shown in Figure 9-1(b) uses a side-emitting LED. It relies on total internal reflection from a silicone compound that is placed over the LED and detector/amplifier. Optocouplers with indirect coupling are easier to fabricate compared to those with the direct coupling method shown in Figure 9-1(a). However, the amount of light that is transmitted depends on physical properties that are difficult to control -- the roughness of the cleaved edge of the LED and the presence of bubbles in the silicone -- as well as on the electrical properties and optical efficiency of the two materials. Many optocouplers are made with indirect coupling because of the ease of manufacturing them along with improved voltage isolation between the LED assembly and the silicon subassembly compared to optocouplers with direct coupling.

Figure 9-1. Diagram of two basic optocoupler configurations.

One of the most important parameters of an optocoupler is the current-transfer ratio (CTR) which is the ratio of the output current of the amplifier to the forward current in the LED. For a simple optocoupler with a transistor output, the CTR is closely analogous to the gain (hFE) of a bipolar transistor. However, the optical process is relatively inefficient, typically resulting in CTR values between 1 and 10, much lower than typical transistor gain values.

III - 50

Optocouplers can be designed for either digital or analog applications. The negative temperature dependence of the LED causes the CTR to vary over a relatively wide range. For digital optocouplers, the variability can be dealt with quite easily by requiring the LED to be driven well beyond the “active” CTR region (essentially specifying the device only for a saturated condition). Figure 9-2 shows the I-V characteristics of a simple digital optocoupler with a transistor output stage. The characteristics are very similar to that of a discrete transistor, with the LED forward current taking the place of base current to the transistor. The dashed curve for each input current show how the transfer characteristic change when the device is heated to a modest temperature, 36 ºC, compared to the lower curve at 25 ºC. The curves decrease at higher temperature because the LED output drops with temperature much faster than transistor gain increases with temperature.

Figure 9-2. I-V characteristics of a digital optocoupler with a simple transistor output stage showing effect of heating on characteristics.

Transfer characteristics of a more complex digital optocoupler with an integrated amplifier and digital output stage are shown in Figure 9-3. The “active” region where the device makes the logical transition is typically a factor of three or more lower than the conditions under which logical operation is guaranteed by the device specifications. The logical “low” specification drives the LED current to the point that the output is strongly saturated (note the break in the horizontal axis scale). The active switching point also depends on output loading, as shown by the dashed line, but this dependence is sublinear because of the high gain of the internal amplifier. Measurements of the digital output in saturation provide no information about where the device actually makes the logical transition, and effectively prevent one from determining how parametric changes in the LED output and amplifier photoresponse are affected by radiation. Digital measurements also “mask” unit-to-unit variability in the active switching region, which can be important. Measurements of the active switching region are recommended as auxiliary measurements for radiation characterization even though this is a special measurement that is not included in the standard device specifications. These measurements are easily made for most digital optocouplers, although some types of amplifiers contain internal comparators with hysteresis that require two sets of measurements, one for high-to-low, and the other for low-tohigh transitions.

III - 51

Figure 9-3. Transfer characteristics of a more complex digital optocoupler with an internal high-gain amplifier.

Design and operation of analog optocouplers is quite different from that of digital optocouplers. Analog optocouplers require that the CTR be held within very close limits over a broad temperature range (this is not an issue for digital optocouplers for which temperature effects are “swamped out” by simply overdriving the device into saturation). Analog optocouplers usually have a very basic phototransistor (or photodiode/transistor) output stage. One way to reduce the temperature dependence is to operate the transistor in the high-injection region, well above the maximum point in the hFE-IC curve of the phototransistor. In this region, the output current actually increases slightly when the (effective) base current provided by the photocurrent from the LED decreases, compensating for the drop in LED output at higher temperature. This design approach allows the current transfer ratio to be kept within very narrow limits, even over a wide range of temperatures. It also makes analog optocouplers inherently less sensitive to changes in LED output due to radiation until the changes are large enough so that the phototransistor operates at low injection instead of high injection.

B. Radiation Degradation Digital Optocouplers The 4N49 is an example of a basic digital optocoupler that has frequently been used in space or other high-reliability applications. This device uses a simple phototransistor as a detector/amplifier, along with an LED with unspecified characteristics (the manufacturer can select any wavelength or LED type that will enable the completed device to meet the overall electrical specifications of the optocoupler). The CTR is directly related to the transistor gain and the LED output. The three manufacturers that provide the 4N49 all use amphoterically doped LEDs; two manufacturers do not manufacture the LED, but obtain it from outside sources. The data in Figure 9-4 show how CTR of the 4N49 is degraded by 50-MeV protons. Note the extreme sensitivity of this device to displacement damage, which is mainly due to degradation of the LED. The proton fluence from a single intense solar flare (~ 1010 p/cm2) is sufficient to degrade the CTR by about a factor of three when it is operated at low forward current. The recommended forward current of the 4N49 for high-reliability applications is 1 mA because of concerns about LED wearout. That current is far below the maximum operating current, and inadvertently makes the device considerably more sensitive to radiation damage. The damage is lower for conditions where higher current is used for the LED, even when the device is irradiated without bias and with a low duty cycle to minimize annealing. The lower damage is the result of operating the phototransistor at higher injection levels. III - 52

Figure 9-4. Proton degradation of a widely used optocoupler with an amphoterically doped LED.

It is possible to examine the different factors that control CTR in the 4N49 separately by partially disassembling the device and measuring the light output of the LED, transistor gain, and photoresponse separately after each irradiation. An external light source of constant amplitude was used so that radiation degradation of the photoresponse is made at the same light injection level, not the lower injection level that occurs when the internal LED is the light source. The results are shown in Figure 9-5 [Rax1]. Note that the CTR degrades more severely at low currents than would be estimated from the product of the photoresponse and LED output. This is because the phototransistor operates at lower and lower current levels when the LED output degrades. The phototransistor operates less efficiently under those conditions, substantially increasing the overall degradation. Phototransistor gain contributed very little to the degradation. The optocoupler CTR degrades by more than two orders of magnitude at the highest radiation levels used in these tests.

Figure 9-5. Degradation of the LED, photoresponse and gain of the components within the 4N49 optocoupler.

III - 53

It is also useful to consider another basic type of optocoupler, the 6N140, that has also been widely used in space applications. The 6N140 uses an internal Darlington transistor. It requires a much higher input current at the LED in order to guarantee performance compared to the 4N49, but has faster response time. This device uses a 700 nm AlGaP LED that degrades far less than the amphoterically doped LED used in the 4N49. The shorter wavelength used in this optocoupler makes the phototransistor less sensitive to degradation, and helps to contribute to the improved radiation performance, even though this is a commercial part that is not designed to be hardened to radiation (Figure 9-6).

Figure 9-6 Degradation of the 6N140 optocoupler that uses a diffused LED with 700 nm wavelength.

Analog Optocouplers Figure 9-7 shows how CTR for various forward current conditions is affected by proton irradiation [John2]. The LED in this particular optocoupler is an amphoterically doped LED, which is highly sensitive to proton damage. Prior to irradiation the maximum CTR occurs at about 0.5 mA. The CTR is considerably reduced for currents above that value, and normal practice is to operate above the peak in order to reduce the sensitivity of the CTR to temperature, gain and other variables, which is usually required of analog optocouplers. At low forward current, the CTR is strongly degraded by protons. This occurs because of two factors: the LED output drops, and the phototransistor gain depends on current. Thus, operating this type of optocoupler at low currents results in more degradation than expected from the LED. If the optocoupler is operated well above the peak current, then the current dependence of the transistor gain reduces the sensitivity to LED drive, thus reducing the relative degradation. For example, if the optocoupler is operated at 2 mA the CTR decreases by about 30% at a fluence of 2 x 1010 p/cm2. At 0.2 mA, the CTR decreases by about a factor of three at the same fluence.

III - 54

Figure 9-7. Proton degradation of an analog optocoupler with an amphoterically doped LED.

Just as for digital optocouplers considerable improvement in radiation performance can be achieved by using a different LED technology for analog optocouplers. Figure 9-8 shows proton degradation of an analog optocoupler with a double-heterojunction LED. It is degraded far less than the other optocoupler shown in Figure 9-7.

Figure 9-8. Proton degradation of an analog optocoupler that uses a double-heterojunction LED

III - 55

C. Testing Issues Optocouplers generally use a combination of a III-V LED with silicon detectors and amplifiers. This creates an additional layer of confusion because the failure mechanism can occur due to either component, or for a combination of the two. For this reason it is particularly important to select an energy where the interpretation of NIEL is unambiguous, and 50 MeV protons are a good choice because of the uncertainty in GaAs NIEL for higher proton energies. The construction of optocouplers must also be taken into account when selecting appropriate test energies. For devices with a sandwich construction care must be taken to ensure that the protons used for testing have sufficient range to penetrate the LED, ceramic substrate and other packaging material without degrading the energy to the point where the NIEL for either component is affected because of energy loss in layered materials. Careful attention needs to be given to the way that electrical measurements are done between radiation levels. Injection-enhanced annealing can have a large impact on test results, particularly if high forward currents are used for some of the electrical tests. Pulsed measurements may be required to minimize annealing, as well as to keep from overheating the LED, which is affected by temperature (see Figure 9-2). As discussed earlier, special measurements of the transfer characteristics should be added to the normal set of measurements in order to measure the threshold behavior. Physical factors are present in optocouplers that can also affect the radiation response and the operating margin. For example, the silicone coupling material in lateral optocouplers often contains bubbles that affect light transmission. The output of edge-emitting LEDs depends partly on the way that the edges are cleaved because Fresnel losses are increased if the surface is rough instead of smoothly cleaved. These factors can lead to larger unit-to-unit variability in radiation response compared to conventional silicon semiconductors. Test sample sizes should be large enough to determine device variability for these classes of components.

10 - SOLAR CELLS A. Construction and Electrical Properties Solar cells are essentially photovoltaic detectors that operate with no external bias. Unlike conventional detectors, they are designed to produce relatively high currents, with high efficiency. They are required to absorb light over a wide range of wavelengths -- from approximately 0.2 to 2 µm. Figure 10-1 shows a physical diagram of a basic crystalline silicon solar cell. It consists of a shallow n-region, diffused into a lightly doped p-substrate. The psubstrate has long lifetime, allowing light at long wavelength to be absorbed deep within the substrate. The diffusion length L = √ D τ, where D is the diffusion constant (≈ 30) and τ is the minority carrier lifetime.

Figure 10-1. Diagram of a typical solar cell.

III - 56

A low-resistance contact region is diffused on the top surface to provide a low-resistance path for current at the top of the cell and to allow ohmic contact to the metallization. The metallization is deposited as a lattice to minimize occlusion of the incident light. An antireflection coating is used at the top surface, designed to reduce reflection at the peak of the solar spectrum (about 0.6 µm). Although not shown in the figure, a thin cover glass is used on solar cells for space application to protect the cells. The I-V characteristics of a typical solar cell are shown in the inset in Figure 9-2, along with the solar spectrum. The cell develops a voltage above 0.7 V with no load, but the voltage drops as the load current increases. The cell voltage and short-circuit current are basic parameters used to characterize the cell. Designs of basic silicon solar cells involve many tradeoffs. Low internal resistance is needed to reduce internal losses at high current, but that conflicts with the requirement to have high minority carrier lifetime and extended charge collection from deep regions in the cell. The solar spectrum extends from about 0.2 to 2 µm, a very wide range. Silicon does not absorb light beyond the absorption edge, approximately 1 µm, and consequently about 1/2 the energy of the solar spectrum is beyond the operating range of silicon solar cells. Considering the overall responsivity of silicon, the maximum efficiency of a basic silicon solar cell is about 29%.

Figure 10-2. I-V characteristics of a typical solar cell There are many design features that can be used to increase solar cell efficiency, including (1) designing special lens assemblies, (2) using tandem cell designs that consist of a sandwich of two different types of cells, with the lower cell designed to efficiently absorb light at longer wavelengths that are beyond the absorption edge of the material in the top cell, and (3) using special concentrator assemblies that focus larger amounts of power on the cell. The cell efficiency is higher with increased power levels. Further details are provided in Reference [Hove1].

B. Radiation Degradation Degradation of silicon solar cells is usually dominated by the decrease in minority carrier lifetime from displacement damage. The diffusion length is reduced, lowering the amount of energy that is collected at longer wavelengths. Figure 10-3 shows how the short circuit current of a typical n-on-p silicon solar cell is degraded by 10-MeV protons (the de facto standard energy for damage comparison in solar cell work because of the minimal amount of shielding). Degradation of this type of cell has been studied extensively, and the results agree closely with experimental results. Damage in solar cells is affected by annealing, and tests of solar cells must

III - 57

be done using special solar simulators that provide the operating power and total thermal heating of the cell that is expected in the actual environment. Because so little shielding is present, the mean energy of the proton distribution in typical space environments is significantly lower than 10 MeV. Consequently, the low energy protons from a single intense flare can degrade solar cell performance more than 30%.

Figure 10-3. Degradation of solar cells from 10-MeV protons

Extensive work has been done to model solar cell damage, and closed-form equations to describe the damage have been developed for both silicon and GaAs cells [Tada1, Tayl1]. Computer programs are available that can calculate the net degradation for a spectrum of electron and/or proton energies [Tada1]. Even though solar cell degradation has been studied extensively, there are limitations in the existing models, which basically consider only degradation from the reduced minority carrier lifetime . Those limitations became painfully obvious when solar cell panels failed abruptly on the NASA ETS-VI satellite [Yama1]. The reason for the failure was that the radiation level -dominated by electrons in the earth’s radiation belts -- was high enough so that carrier removal was a significant factor in cell degradation. Carrier removal effects were severe enough so that the p-region was converted to n-type material (recall that at high radiation levels the high levels of impurities “pin” silicon to n-type material because the impurity levels are not located at the center of the bandgap), with the result that the solar cells were no longer operational. Just before this catastrophic cell failure occurred the cell efficiency actually increased somewhat. Figure 104 shows how this degradation was modeled by Yamaguchi, et al. [Yamu2] for a lightly doped non-p cell that is 50 µm thick. The smooth line shows the predicted cell behavior when only lifetime degradation is considered. The other two curves show how carrier removal and changes in internal cell resistance affect the results.

III - 58

Figure 10-4. Anomalous degradation of lightly doped solar cells at high radiation levels

11 - CHARGE-COUPLED DEVICES A. Conventional CCDs Charge-coupled devices are frequently used as imagers in spacecraft. The operation of a CCD is shown in the simplified diagram of Figure 11-1. Charge “packets” are induced by absorbed photons within each pixel. Information within the high-density array is transferred to the output by operating the device as a simple shift register, transferring the contents of each pixel laterally with a three-phase clock. This readout technique results in a very simple, open structure with high pixel density. However, its operation depends on the ability of the charge

Figure 11-1. Simplified diagram of a basic CCD that uses a three-phase clock to transfer information.

III - 59

within each pixel to be shifted multiple times with minimum signal loss. This requires a high minority carrier lifetime within the array. One of the key CCD parameters is the charge transfer ratio (CTR), which is typically on the order of 0.9999 to 0.99999 for high-quality unirradiated cells. The charge transfer inefficiency, CTI, defined as (1 - CTR), is often used instead of CTR. CCDs are strongly affected by minority carrier lifetime degradation. Recent work has investigated an alternative p-channel CCD technology in contrast to the usual n-channel CCD. By changing the material type, the minority carriers are electrons, not holes, which are less affected by radiation damage. However, it is more difficult to fabricate p-channel CCDs, and consequently the initial CTR is not as high as for n-channel devices. A comparison of radiation damage in n- and p-channel CCDs is shown in Figure 11-2, after Hopkinson, et al. [Hopk1]. The upper curve shows CTI for various internal signals for an unirradiated n-channel CCD. The lower set of curves show the performance of experimental pchannel devices after they are irradiated to 10 krad(Si); at that level the performance of typical nchannel devices is severely degraded. This illustrates the degree of improvement that is possible with different CCD technologies.

Figure 11-2. Performance of irradiated p-channel CCDs.

CCD pixels are small enough so that microscopic damage from the interaction of one or more protons can cause a small number of pixels to be severely degraded [Srou3]. Characterization of CCDs must take this into account, along with degradation of noise and sensitivity. Micro-dose damage effects in CCDs were reviewed by P.W. Marshall and C.J. Marshall in the 1999 Short Course [Mars1], as well as by Hopkinson, et al. [Hopk1]. Numerous papers on CCD damage have been published within the last five years, and readers are referred to those references for more detail on CCD radiation effects.

B. Active Pixel Sensors An alternative approach for fabricating CCDs is integration of the electronics required for readout within the array. Such devices are called active pixel sensors (APS). The APS technology allows individual pixels to be addressed, eliminating the sequential series of chargetransfer events that are necessary for operation of conventional CCDs [Wong1]. The resulting structure is far more efficient, and no longer depends on maintaining a very high minority carrier lifetime within the array. Thus, active pixel sensors are expected to perform far better in a

III - 60

radiation environment compared to conventional CCDs. They are also likely to be less affected by microscopic damage effects because of the “local” way in which information is accessed. Figure 11-3 shows a diagram of an active pixel sensor.

Figure 11-3. Diagram of an active pixel sensor.

Active pixel sensors are compatible with high-density CMOS processes, which are now available with very small feature sizes. This allows the array to be small enough to provide high resolution. A key requirement is advanced isolation (such as shallow trench isolation) that allows active pixels to be spaced close together. Active pixel sensors are far more complex than conventional CCDs, and this is an active area of development. Two papers in this year’s conference will discuss radiation effects in active pixel sensors [Hopk2, Cohe1], and readers should consult those papers for up-to-date information about radiation performance of this technology. Active pixel sensors have also been developed that rely on direct heat absorption (bolometry) using a micromechanical structure [Ho1]. They can be used at longer wavelengths, and are a promising new detector technology, but have much lower sensitivity compared to active pixel sensors with conventional photodetectors. One advantage of this type of sensor is a broad spectral response, particularly at longer wavelengths.

12 - EXAMPLES OF COMPLEX FAILURE MODES A. Optocoupler Failures in Space The first example of a complex failure mode occurred on the Topex-Posdeidon spacecraft, operating in a 1300 km five-year high-inclination orbit that is well within the inner portion of the Van Allen proton belt. A basic optocoupler was used in several different applications, including direct control of engine thrusters used to maneuver the spacecraft. No radiation tests had been done on the optocoupler, and it was assumed that there were no significant radiation failure issues because the expected total dose level during the mission was less than 10 krad(Si). Although detailed information about the distribution of the total dose between electrons and protons in the radiation belts was available, it was not provided as part of the abbreviated description of mission requirements and could only be obtained with considerable effort. Several failures occurred in optocouplers after 2 1/2 years of operation. The estimated proton fluence was approximately 2 x 1010 p/cm2. The first failures that occurred were in circuits that transmitted the status of thrusters -- used to maneuver the spacecraft -- back to ground controllers during thruster operation. Fortunately the optocouplers used in the circuit that actually controlled thruster operation continued to operate. This was later found to be due to a much more conservative design practice for the optocouplers used in that particular circuit,

III - 61

which allowed the current transfer ration to degrade by a factor of ten before thruster operation would be affected. The issues in this example are first that the designers of the spacecraft were unaware of the extreme sensitivity of the optocouplers that they used to displacement damage from protons. The second issue is that the environment was only specified as an overall total dose requirement, with inadequate awareness of the actual proton fluences that the spacecraft would encounter. The final issue is that the impact of the severe damage to the optocouplers was negligible in one application simply because the designers were ultra-conservative in their design implementation. This was a lucky happenstance. Because the potential of failure from the optocouplers was not recognized, it was very difficult to deal with the failures that actually occurred during the critical real-time operations that were involved in maneuvering the spacecraft.

B. Power Converter Failures During Ground Tests The second example is a hybrid power converter that was procured from a high-reliability manufacturer that had provided power modules for many previous spacecraft. This manufacturer, like many other hybrid manufacturers, considered all aspects of the design proprietary, and provided very limited information about either the design of the converter or the specific parts that were used to fabricate it. The initial parts list failed to include a linear optocoupler that was used to provide feedback from the output of the converter to the control electronics. This issue was made even more complicated because the optocoupler (in itself a simple hybrid part) was procured from yet another manufacturer by the company that designed and built the hybrid power converter. Thus, we have “nested” hybrid manufacturers in the tree of parts used in the final product with limited awareness or control of either. Test results for several hybrid converters made with this design are shown in Figure 12-1. The failure mode is lack of control of the output, with the output voltage gradually increasing towards the “raw” supply voltage. Note the very low levels of proton fluence at which these devices begin to degrade.

Figure 12-1. Failures of hybrid power converters during laboratory tests due to CTR degradation of linear optocouplers contained within the device.

III - 62

The optocoupler manufacturer procured LEDs from an outside vendor, and at one point changed the LED from a shorter wavelength double-heterojunction device (relatively tolerant to displacement damage) to an amphoterically doped LED that had better overall performance. The result was a decrease in the radiation hardness of more than one order of magnitude. It is fortunate that this problem was identified before these devices were deployed. Normally proton tests would not be required for this type of part, particularly because the optocouplers were not included in the initial parts list provided by the power converter manufacturer.

13 - GENERAL RADIATION TESTING ISSUES FOR OPTOELECTRONIC DEVICES A. Radiation Sources Several different sources are available that provide high-energy protons. The highest energies are produced at cyclotron facilities. Van de Graaff accelerators are also available that can produce protons with energies up to 20 MeV, and may be less costly to operate. Table 13-1 lists three commonly used cyclotron facilities. All three provide steady-state proton beams with a circular beam area that is typically 5-10 cm in diameter. Operating costs for these facilities are nominally $500-$600 per hour. Table 13-1. Comparison of Proton Accelerator Facilities Facility Univ. of California, Davis Univ. of California, Berkeley Univ. of Indiana

Energy Range (MeV) 15 - 65 50 65 - 200

High energy protons activate certain materials that are used in test hardware (particularly gold, copper and solder). The induced radioactivity can present a hazard to workers during experiments, particularly if tests are done at high fluences (> 1013 p/cm2). Activation also affects shipping; it may not be possible to transport irradiated material to other facilities without waiting for a week or more for the induced radioactivity to die down. A number of linear accelerators and Van de Graaff machines are available that can provide high-energy electrons, but they will not be discussed in detail because electrons are generally of less interest for optoelectronic devices than high-energy protons (except for solar cells or other cases where very little shielding is present). When electron testing is needed energies of 3-5 MeV are typically used. Cobalt-60 facilities (passive sources) are usually used to evaluate ionization damage. They are widely available at many aerospace companies, universities and government laboratories, but only simulate ionization damage, not displacement damage effects†. We have not spent much time discussing ionization damage because displacement damage usually is the dominant effect for optoelectronic devices. Protons produce ionization as well as displacement effects, and there may be cases where ionization damage is really the dominant mechanism. However, tests with protons (or electrons where the environment is mainly from electrons) produce both ionization and displacement effects, so separate tests with gamma ray facilities are not required. -------†The Compton electrons that are produced by gamma rays when they interact with material produce a slight amount of

displacement damage, but can generally be ignored.

III - 63

B. Energy Selection Selection of energies for testing is a somewhat complicated problem. Usually the purpose of the test is to evaluate damage (essentially displacement damage) at a single energy to minimize cost and testing complexity, and then determine how the damage that is measured at that energy can be related to the damage produced in an actual space environment, where there is a wide distribution of proton energies. Usually this is done using the non-ionizing energy loss concept (NIEL), which was discussed at some length in Section 3. One key point that was made is that NIEL calculations for GaAs do not always agree with experimental results for energies above approximately 80 MeV, and consequently energies used for proton testing should be below that value in order to avoid possible errors in interpretation of damage. For protons, energies of 50 MeV are recommended because this is close to the mean proton energy of many earth-orbiting spacecraft, and the NIEL calculations agree reasonably well with experimental results for both silicon and GaAs. However, a lower energy is recommended for solar cells. This is appropriate because (1) there is much less shielding (at least on the front surface), and (2) nearly all of the archival results were done with 10 MeV protons. There are always special situations that may require different energies. Note particularly that some interplanetary missions may have very different energy ranges than earth-orbiting spacecraft. Another factor in proton testing is the range of the protons. Some optoelectronic devices have considerable amounts of other material surrounding the critical devices, and the energy of the protons must be sufficient to allow them to penetrate to the active regions that are important for device operation without seriously degrading the energy. Note particularly the optocouplers with “sandwich” configurations where a thick III-V device and its associated substrate are placed over the silicon photodiode (see Figure 9-1). Table 13-2 below shows the range for several proton energies assuming silicon (with density of 2.33; the range in other materials scales inversely with density). Note that GaAs has a density (5 .32) that is more than twice as large as that of silicon, so that the range of protons is considerably less in GaAs and other III-V materials. Table 13-2. Range of Protons with Selected Energy in Silicon

Energy MeV 200 100 65 50 30 20 15

Range in Silicon (µm) >20,000 >20,000 18,000 8,610 5,220 2,580 1,585

Although we have not spent much time discussing testing with electrons, there are cases where electrons are a key factor in the environment. The solar cell results discussed in Section 11 are an example. Some interplanetary missions involve electrons with energies above 100 MeV. For earth-orbiting environments electron energies of 3-5 MeV are typically used. It is also possible to use 2-MeV electrons (conveniently available at some low energy accelerators). The energy of electrons is generally less important than energies selected for protons because NIEL for electrons depends much more gradually on energy, provided the energy is above the displacement damage threshold (about 150 keV). Table 13-3 shows the range of electrons of various energies in silicon. III - 64

Table 13-3. Range of Electrons at Selected Energies in Silicon Energy MeV 1 2 3 10 100

Range in Silicon (µm) 2,800 4,700 6,900 17,200 138,000

Most laboratory tests of optoelectronic devices are done with the device normally incident to the beam direction. This makes it relatively straightforward to make sure that the range of the particles that are being used is adequate to go through packaging material, lenses, coatings and any other material that is present in the structure. It may be necessary to do complex transport calculations to determine the effective energy in the real environment for cases where a great deal of extra material is present. For example, solar cells have very little shielding on the top surface, but the energy of electrons and protons that go through the device at angle (or from the back of the cell) will be different. Similar issues are important for other applications, including CCDs or detectors that are behind baffles and shields. This does not necessarily affect the way that radiation tests are done, but it does affect the interpretation of the test results.

C. Single-Event Upset Testing Although single-event effects are not addressed in this part of the course, a brief discussion of single-event testing has been included for completeness that examines some of the special issues that have to be dealt with when optoelectronic devices are subjected to single-event testing. Before beginning this discussion, recall that SEE effects are usually described in terms of linear energy transfer (LET), with units of MeV-cm2/mg. Galactic cosmic rays and solar flares have a continuous distribution of LET values, up to an LET of about 100 MeV-cm2/mg [Bart1], but the number of particles at high LET falls rapidly as the LET increases. There is an abrupt drop in the LET distribution at about 30 MeV-cm2/mg (the “iron” threshold) that is of great practical interest. Because relatively few particles are present with LET’s above 30 MeVcm2/mg, devices with threshold LET values above that value† are relatively immune to singleevent upset. On the other hand, some devices (such as DRAMs) have threshold LET values of 12 MeV-cm2/mg, and are extremely sensitive to single-event upset effects. Two types of single-event tests will be considered, as depicted in the simple diagram in Figure 13-1: (1) Tests with heavy-ions that have very limited range (typically 30-50 µm) and must be done in a vacuum chamber; and (2) Tests with high-energy protons that have much longer range, but can affect devices either directly from proton ionization, or indirectly, via a nuclear or kinematic collision with a lattice atom. Direct ionization from protons produces ionization tracks with low charge density (LET approximately 0.1 MeV-cm2/mg), but some optoelectronic devices are affected by direct ionization. The indirect process produces a recoil atom with higher effective LET, but a relatively short track length (a few µm). ---------†The number of particles falls rapidly for LET>30 MeV-cm2/mg. However, the charge produced in devices increases as the secant of the incident angle. Consequently, the effective “iron threshold” for most devices is above 60 MeV-cm2/mg.

III - 65

Each particle produces an ionization track

M ost protons pass through the device with little effect A few protons cau se nuclear reactions

+ n -+ ++ + + - p-substrate ++ - + Short-range

-+ +-

n p-substrate

recoil produces ionization

a) Heavy Ions (ionization by each particle)

b) Protons (nuclear reaction needed to produce recoil)

Figure 13-1. Diagram showing direct interactions from high-energy cosmic rays and indirect interactions from protons through an intermediate nuclear reaction.

Physical Factors The first problem is that of recognizing that unlike silicon integrated circuits, the active region of many optoelectronic devices is located far below the surface of the device. For example, the active region of a typical LED is 50-100 µm below the top region. This presents a major problem for heavy ion tests because it is difficult to find a source with sufficient penetration depth. Most facilities are limited to ranges below 100 µm for ions with LET values above approximately 20 MeV-cm2/mg [Koga1]. Fortunately, single-event effects in LEDs and laser diodes are generally of minor importance, and usually one is more concerned about SEU effects in amplifiers and detectors where the active region is close to the surface. However the package must be removed in order for the heavy ions to reach the device. Optocouplers are an example where physical construction is very important. For optocouplers with the LED array placed over the photodiode it is generally not possible to do SEE testing with heavy ions because of occlusion by the LED. One way to deal with this is to disassemble the device, removing the LED array, allowing unobstructed access to the photodiode and amplifier chip. Even after the LED assembly is removed, a layer of optical coupling compound is still present on the top of the silicon chip that has to be removed with a solvent. The silicon chip that remains cannot be tested as an optocoupler, but is essentially equivalent to the optocoupler in the “off” state. Electrical Requirements Many optoelectronic components (including optocouplers) are relatively slow devices that are designed to work with load resistances of several kilohms or more. They cannot drive terminated 50-ohm cables, and it is generally necessary to use an active line driver in order to monitor output signals. The output response of optocouplers is strongly affected by resistive and capacitive loading, and the test conditions must closely mimic the actual application.

III - 66

Example: Heavy Ion Results for a Digital Optocoupler An example of test results for heavy-ion test of a digital optocoupler is shown in Figure 13-2 (the LED assembly was removed before the tests were done). There are several subtleties that are not obvious at first glance. With heavy ions, this device is extremely sensitive to upset effects. The threshold LET of approximately 0.3 MeV-cm2/mg is well below the level that even highly sensitive dynamic memories exhibit upset. However, unlike a digital circuit there is a wide variation in the pulse amplitude and pulse width. The nature of the output depends where the ion happens to strike the device. Thus, the cross section has to be defined in terms of specific criteria for output amplitude and pulse width, as well as loading conditions.

Figure 13-2. Variation of mean pulse width from a digital optocoupler at various LET values.

Near the threshold the pulse width is extremely narrow, gradually increasing as the LET of the ions used for testing is increased. The pulse width gradually increases to about 130 ns, and most of the pulses remain at about that pulse width until the LET exceeds 10 MeV-cm2/mg. However, prior to the point where the pulse width begins to increase a small number of pulses are apparent with considerably lower amplitude. This causes the mean pulse width to decrease slightly, and that is due to the gradual contribution of a second mechanism, the response of the high-gain amplifier. Once the amplifier is fully turned on the cross section and the nature of the pulse widths that are observed change radically, as shown in the figure. Direct Ionization Effects in Optocouplers The photodiodes used in optocouplers have a large diameter compared to that of most components, and this makes it possible for direct ionization from protons to be a factor in their response. Because of the large diameter, far more charge is collected when the proton passes diagonally through the photodiode compared to the charge generated at normal incidence. This was first observed by LaBel, et al. [Labe1]; the cross section increased by nearly an order of magnitude when they carried out tests of optocouplers at high angles.

III - 67

A more thorough investigation of the effect of direct ionization was done for high angles that used a much wider range of proton energies [John3]. In this case it was not only necessary to remove the LED assembly, but also to grind down part of the side of the package in order to allow protons with lower energies to reach the surface of the die, which was recessed somewhat in the package. The results of these tests are shown in Figure 13-3. Note that at lower proton energies direct ionization has increased the cross section by about three orders of magnitude. This is a sufficiently large increase to affect the upset rate in the real environment.

Figure 13-3. Dependence of upset cross section on incident angle for an optocoupler for various proton energies.

Note that even though low energy protons in an experiment are partially shielded by the device package, protons with the continuous distribution of energies in the actual environment are simply shifted down in energy. Thus, although 15 MeV protons will not penetrate the package during a test at a single energy, substantial numbers of protons in that energy range will reach the active part of the device in the real environment because protons with somewhat higher energy will lose part of their energy when they pass through the package. Although one might initially think that interference in this type of experiment from self-shielding by the package makes the effect inconsequential, that is not the case when one considers the net effect of shielding on the distribution of proton energies within the spacecraft (and within the device). Particles with higher energy are shifted to lower energies by the scattering process and consequently there are still significant numbers of low energy particles impinging on the active device. The data in Figure 13-3 required an extensive effort that is well beyond that normally expected for routine characterization. of this type of part. An alternative way to estimate whether indirect ionization is important was developed in Reference [John3] using relatively straightforward tests with laboratory alpha particle sources. That method can be used as an initial screen to determine whether the costs and difficulties of additional detailed tests at angle are needed for specific devices.

III - 68

14 - SUMMARY This part of the 2000 NSREC Short Course has discussed many aspects of optoelectronic devices, which are not as well understood as more conventional semiconductor devices. The radiation response of optoelectronic devices is usually dominated by displacement damage effects that can become important at relatively low radiation levels. The importance of displacement damage for optoelectronics is often overlooked because most other types of electronic components are relatively unaffected by displacement effects until much higher radiation levels are reached, and can be adequately characterized by testing only with gamma rays. A considerable amount of material was presented on the structure and design of optoelectronic devices. This was done for several reasons, in particular because some types of new optoelectronic structures are very different from their older counterparts. It is usually necessary to have a basic understanding of these devices in order to evaluate radiation test data, or to plan radiation tests. LEDs and laser diodes are examples where device technology has evolved in several different directions, making it particularly difficult to select devices for use in space or to evaluate new devices from the relatively sparse data on older LED and laser diode devices. Optocouplers are good examples of the way that different effects interact to produce a complex interdependence of failure modes, requirements, and use conditions. Some optocouplers are extremely sensitive to proton displacement damage and have actually failed operationally in space. Optocouplers are sensitive to transients from protons and heavy cosmic particles as well as permanent damage, and particular care should be taken to ensure that these effects are accounted for when optocouplers are selected for use in space. There are many evolving optoelectronic devices -- vertical cavity semiconductor lasers, active pixel sensors, and others we have not discussed -- that are likely to be seriously considered for future space applications, particularly because of the need to decrease the size, weight and cost of spacecraft. New ways of using optoelectronics for optical interconnects or as special dedicated integrated optic devices are being developed, and it is likely that radiation effects in these structures will be an interesting topic during the next decade. There are a number of more exotic structures -- including the use of porous silicon light emitters [Arra1] or avalanche emission from silicon [Fauc1] -- that have been considered to allow direct integration of light sources with silicon-based electronics. The intent of this part of the course is to provide the necessary background to understand and appreciate the mechanisms and principles that affect optoelectronics.

ACKNOWLEDGEMENTS The author wishes to thank Dr. Peter Winokur of Sandia National Laboratories for helpful suggestions and careful reading of the manuscript, and Mr. Peter Schrock for invaluable assistance in preparing the material.

III - 69

REFERENCES (All references are unclassified) [Ahn] [Arra1]

[Ashc1] [Barn1] [Barn2] [Barn3]

[Bart1] [Barr1] [Baur1] [Bern1] [Burk1] [Case1] [Chon1] [Choq1] [Chow1] [Chua1] [Cohe1] [Crol1]

[Dale1] [Dash1] [Derr1] [Dimi1] [DOrd1] [Dres1] [Duz1]

D. Ahn and S. L. Chuang, “Optical Gain in a Strained Layer Quantum Well Laser,” IEEE J. Quantum Elect., 24, pp. 2400-2406 (1988). H. F. Arrand, T. M. Benson, P. Sewell, A. Loni, R. J. Bozeat, R. Arens-Fischer, M. Kruger, M. Thonissen and H. Luth, “The Application of Porous Silicon to Optical Waveguiding Technology,” IEEE J. of Selected Topics in Quantum Electronics, 4(6), pp. 975-982 (1998). N. W. Ashcroft and N. D. Mermin, Solid State Physics, Holt Rinehart Winston: New York, 1976. C. E. Barnes, “A Comparison of Gamma-Irradiation-Induced Degradation in Amphoterically Si-Doped GaAs LEDs and Zn-Diffused GaAs LEDs,” IEEE Trans. Elect. Dev., 26, pp. 739-745 (1979). C. E. Barnes and J. J. Wiczer, “Radiation Effects in Optical Devices,” Sandia National Laboratories Report SAND84-0771, May 1984. C. E. Barnes, J. R. Schwank, G. M. Swift, M. G Armendariz, S. Guertin, G. L. Hash and K. D. Choquette, “Proton Irradiation Effects in Oxide-Confined Vertical Cavity Surface Emitting Laser Diodes,” presented at the RADECS Conference, Sept. 13-17, 1999, Fontevraud, France. J. L. Barth, “”Modeling Space Radiation Environments,” Section 1 of the 1997 NSREC Short Course notes. A. L. Barry, A. J. Houdayer, P. F. Hinrischsen, W. G. Letourneau and J.Vincent, “The Energy Dependence of Lifetime Damage Constants in GaAs LEDs for 1-500 MeV Protons,” IEEE Trans. Nucl. Sci., 42, 2104 (1995). R. Baurlein, “Displacement Threshold in Semiconductors,” in Radiation Damage in Solids, edited by D. S. Billington, Academic Press: New York, 1962. M. G. A. Bernard and G. Duraffourg, “Laser Conditions in Semiconductors,” Phys. Status Solidi, 1, pp. 699-703 (1961). E. A. Burke, C. J. Dale, A.B. Campbell, G. P. Summers, W. J. Stapor, M. A. Xapsos, T. Palmer and R. Zuleeg, “Energy Dependence of Proton-Induced Damage in GaAs,” IEEE Trans. Nucl. Sci., 34, 1220 (1987) H. C. Casey, Jr., D. D. Sell and K. W. Wecht, “Concentration Dependence of the Absorption Coefficient for n- and pType GaAs between 1.3 and 1.6 eV,” J. Appl. Phys., 46(1), pp. 250-257 (1975). T. C. Chong and C. G. Fonstad, “Theoretical Gain of Strained-Layer Semiconductors in the Large Strain Regime,” IEEE J. Quantum Elect., 25, pp. 171-178 (1989). K. D. Choquette and H. Q. Hou, “Vertical-Cavity Surface Emitting Lasers: Moving from Research to Manufacturing,” Proc. IEEE, 85, 1730 (1997). W. W. Chow and R. F. Carson, “Neutron Effects in High-Power Laser Diodes,” IEEE Trans. Nucl. Sci., 36, 2076 (1989). S. L. Chuang, Physics of Optoelectronic Devices, Wiley Interscience: New York, 1995. M. Cohena and J. P. David, “Radiation-Induced Dark Current in CMOS Active Pixel Sensors,” to be presented at the 2000 Nuclear and Space Radiation Effects Conference, Reno, Nevada, July 24-28, 2000. D. R. Croley, H. B. Garrett, G. B. Murphy and T. L. Garrard, “Solar Particle Induced Upsets in the TDRS-1 Attitude Control System RAM During the October 1989 Solar Particle Events,” IEEE Trans. Nucl. Sci., 42, pp. 1489-1496 (1995). C. J. Dale, P. W. Marshall, E. A. Burke, G. P. Summers and E. A. Wolicki, “High Energy Electron Displacement Damage in Silicon,” IEEE Trans. Nucl. Sci., 35, pp. 1208-1214 (1988). W. C. Dash and R. Newman, “Intrinsic Optical Absorption in Single-Crystal Germanium and Silicon at 77 K and 300 K,” Phys. Rev., 99(4), pp. 1151-1155, 1955. P. J. Derry, R. J. Fu, C. S. Hong, E. Y. Chan and L. Figuoroa, “Analysis of the High Temperature Characteristics of InGaAs-AlGaAs Strained Quantum Well Lasers,” IEEE J. Quant. Elect., 28, pp. 2698-2705 (1992). K. C. Dimiduk, C. Q. Ness and J. K. Foley, “Electron Irradiation of GaAsP LEDs,” IEEE Trans. Nucl. Sci., 32, pp. 4010-4015 (1985). M. D’Ordine, “”Proton Displacement Damage in Optoisolators,” 1997 IEEE Radiation Effects Data Workshop, IEEE Document 97TH8293, p. 122. P. B. Dressendorfer, “Radiation Effects on MOS Devices and Circuits,” in Ionizing Radiation Effects in MOS Devices and Circuits, P.V. Dressendorfer and T. P. Ma, editors, John Wiley: New York, 1989. S. Duzellier and R. Ecoffet, “Recent Trends in Single-Event Ground Testing,” IEEE Trans. Nucl. Sci., 43, pp. 671-677 (1996).

III - 70

[Evan1] B. D. Evans, H. E. Hager and B. W. Hughlock, “5.5 MeV Proton Irradiation of a Strained Quantum-Well Laser Diode and a Multiple Quantum-Well Broad-Band LED,” IEEE Trans. Nucl. Sci., 40, pp. 1645-1654 (1993). [Fauc1] P. M. Fauchet, “Progress Towards Nanoscale Silicon Light Emitters,” IEEE J. of Selected Optics in Quantum Electronics, 4(6), pp. 1020-1028 (1998). [Feyn1] J. Feynman and S. B. Gabriel, “High-Energy Charged Particles in Space at One Astronomical Unit,” IEEE Trans. Nucl. Sci., 43, pp. 344-352 (1996). [Frie1] E. J. Friebele, “Photonics in the Space Environment,” Section 3 of the 1991 NSREC Short Course notes. [Fu1] R. J. Fu, C. S. Hong, E. Y. Chan, D. J. Booher and L. Figueroa, “High Temperature Operation of InGaAs Strained Quantum-Well Lasers,” IEEE Phot. Tech. Lett., 3, No. 4, 308 (1991). [Fuji1] O. Fujita, Y. Nakano and G. Iwane, “Reliability of Semiconductor Lasers for Undersea Optical Transmission Systems,” IEEE Trans. Elect. Dev., 32(12), pp. 2603-2607 (1985). [Garr1] H. B. Garrett, “Radiation Environments within Satellites,” Section 2 of the 1993 NSREC Short Course notes. [Ghit1] A. Ghiti, M. Silver and E. P. O’Reilly, “Low Threshold Current and High Differential Gain in Ideal Tensile and Compressive Strained Quantum Well Lasers,” J. Appl. Phys., 71, pp. 4626-4628 (1992). [Greg1] B. L. Gregory and H. H. Sander, “Transient Annealing of Defects in Irradiated Silicon Devices,” Proc. IEEE 58(9), pp. 1328-1341, 1970. [Ho1] J. Ho, Y. K. Fang, W. J. Lee, F. Y. Chan, W. T. Hsieh, S. F. Ting, M. S. Ju, S. B. Huang,, K. Wu, and C. Y. Chen, “Dynamic Response Analysis of a Pyroelectric Thin-Film Infrared Sensor with Thermal Isolation Improvement Structure,” IEEE Trans. Elect. Dev., 26(12), pp. 2289-2297 (1999). [Hopk1] G. R. Hopkinson, “Proton Damage Effects on P-Channel CCDs,” IEEE Trans. Nucl. Sci., 46, pp. 1790-1796 (1999). [Hopk2] G. R. Hopkinson, “Radiation Effects in a CMOS Active Pixel Sensor,” to be presented at the 2000 Nuclear and Space Radiation Effects Conference, Reno, Nevada, July 24-28, 2000. [Hove1] H. J. Hovel, “Solar Cells,” in Semiconductors and Semimetals, Vol. 11, R. K. Willardson and A. C. Beer, editors, Academic Press: New York, 1975. [Huen1] T. Huen, “Reflectance of Thinly Oxidized Silicon at Normal Incidence,” Applied Optics, 18(12), pp.1927-1932 (1979). [Hum1] R. H. Hum and A. L. Barry, “Radiation Damage Constants of Light-Emitting Diodes by a Low-Current Evaluation Method,” IEEE Trans. Nucl. Sci., 22, pp. 2482-2487 (1975). [Jewe1] J. L. Jewell, J. P. Harbison, A. Scherer, Y. H. Lee and L. T. Florez, “Vertical-Cavity Surface-Emitting Lasers: Design, Growth, Fabrication, Characterization,” IEEE J. Quantum Electron., 27, 1332 (1991). [John1] A. H. Johnston, B. G. Rax, L. E. Selva and C. E. Barnes, “Proton Degradation of Light-Emitting Diodes,” IEEE Trans. Nucl. Sci., Dec. 1999. [John2] A. H. Johnston and B. G. Rax, “Proton Damage in Linear and Digital Optocouplers,” presented at the RADECS Conference, Sept. 13-17, 1999, Fontevraud, France; accepted for publication in the conference proceedings. [John3] A. H. Johnston, T. Miyahira, G. M. Swift, S. M Guertin and L. D. Edmonds, “Angular and Energy Dependence of Proton Upset in Optocouplers,” IEEE Trans. Nucl. Sci., 46, pp. 1335-1341 (1999). [Koba1] K. Kobayashi, “High Light Output-Power Single-Longitudinal Mode Semiconductor Laser Diodes,” IEEE Trans. Elect. Dev., 32, 12, pp. 2594-2602 (1985). [Koga1] R. Koga, “Single-Event Ground Test Issues,” IEEE Trans. Nucl. Sci., 43, pp. 661-670 (1996). [Kres1] H. Kressel and J. K. Butler, Semiconductor Lasers and LEDs, Academic Press: New York, 1977. [Labe1] K. LaBel, P. W. Marshall, C. J. Marshall, M. D’Ordine, M. Carts, G. Lum, H. S. Kim, C. M. Seidleck, T. Powell, R. Abbot, J. Barth and E. G. Stassinopoulos, “”Proton-Induced Transients in Optocouplers: In-Flight Anomalies, Ground Irradiation Test, Mitigation and Implications,” IEEE Trans. Nucl. Sci., 44, 1895 (1997). [Lang1] D. V. Lang and L. C. Kimerling, “Observation of Recombination-Enhanced Defect Reactions in Semiconductors,” Phys. Rev. Lett., 33(8), pp. 489-492 (1974). [Lash1] G. Lasher and F. Stern, “Spontaneous and Stimulated Recombination Radiation in Semiconductors,” Phys. Rev., 133, A553-A563 (1964). [Lind1] P. F. Lindquist, “A New Model for Light Output Degradation of Direct Bandgap Semiconductors,” 1980 International Reliability Physics Proceedings, p. 145. [Lint1] V. A. J. van Lint, R. E. Leadon and J. F. Colwell, “Energy Dependence of Displacement Effects in Semiconductor,” IEEE Trans. Nucl. Sci., 19, No. 6, 181 (1972). [Lisc1] H. Lischka, H. Henschel, O. Kohn, W. Lennartz and H. U. Schmidt, “Radiation Effects in Light Emitting Diodes, Photodiodes, and Optocouplers," Proceedings of the 1993 RADECS Conference, pp. 226-231. [Lin1] C. H. Lin and Y. H. Lo, “Empirical Formulas for Design and Optimization of 1.55 um InGaAs/InGaAsP StrainedQuantum-Well Lasers,” IEEE Phot. Tech. Lett., 5, 268 (1993).

III - 71

[McLe1] F. B. McLean, H. E. Boesch, Jr., and T. R. Oldham, “Electron-Hole Generation, Transport and Trapping in SiO2,” in Ionizing Radiation Effects in MOS Devices and Circuits, P.V. Dressendorfer and T. P. Ma, editors, John Wiley: New York, 1989. [Mars1] P. W. Marshall and C. J. Marshall, “Proton Effects and Test Issues for Satellite Designers,” Section IV of the 1999 NSREC Short Course notes. [Mars2] P. W. Marshall, C. J. Dale and E. W. Burke, “Space Radiation Effects on Optical Materials and Components for a 1300 nm Fiber Optic Data Bus,” IEEE Trans. Nucl. Sci., 39, 1982 (1992). [Mess1] G. C. Messenger, “A Two Level Model for Lifetime Reduction Processes in Neutron Irradiated Silicon and Germanium,” IEEE Trans. Nucl. Sci., 14(6), 88 (1967). [Mose1] A. Moser, A. Oosenbrug, E. E. Latta, Th. Forster and M. Gasser, ‘High Power Operation of Strained InGaAs/AlGaAs Single Quantum Well Lasers,” Appl. Phys. Lett., 59, [Obry1] M. V. O’Bryan, K. A. LaBel, R. A. Reed, J. L. Barth, C. M. Seidleck, P. Marshall, C. Marshall and M. Carts, “Single Event Effect and Radiation Damage Results for Candidate Spacecraft Electronics,” 1988 IEEE Radiation Effects Data Workshop, IEEE Doc. 98TH8385, pp. 51-57. [Osb1] G. C. Osbourn, “InGaAs Strained Superlattices: A Proposal for Useful, New Electronic Materials,” Phys. Rev. B, 27, pp. 5126-5128 (1983). [Paxt1] A. H. Paxton, R. F. Carson, Harald Schöne, E. W. Taylor, K. D. Choquette, H. Q. Hou, K. L. Lear and M. E. Warren, “Damage from Proton Irradiation of Vertical-Cavity Surface-Emitting Lasers,” IEEE Trans. Nucl. Sci., 44, 1893 (1997). [Peas1] R. L. Pease, “Total-Dose Issues for Microelectronics in Space Systems,” IEEE Trans. Nucl. Sci., 43, pp. 442-452 (1996). [Peas2] R. L. Pease, E. W. Enlow and G. L Dinger, “Comparison of Proton and Neutron Carrier Removal Rates,” IEEE Trans. Nucl. Sci., 34, 1140 (1987). [Pick1] J. C. Pickel, “Novel Devices and Sensors,” Section 4 of the 1993 NSREC Short Course notes. [Rax1] B. G. Rax, C. I. Lee, A. J. Johnston and C. E. Barnes, “Total Dose and Proton Damage in Optocouplers,” IEEE Trans. Nucl. Sci., 43, pp. 3167-3173 (1996). [Rose1] B. H. Rose and C. E. Barnes, “Proton Damage Effects on Light Emitting Diodes,” J. Appl. Phys., 53(3), pp. 1772-1780 (1982). [Scla1] N. Sclar, “Extrinsic Photoconductive Infrared Detectors,” Infrared Physics, 16, 457 (1976). [Shim1] A. Shima, M. Mayashita, T. Miura, T. Kadowaki, N. Hayafuji, M. Aiga and W. Susaki, “Uniform and High-Power Characteristics of 780-nm AlGaAs TQW Laser Diodes Fabricated by Large-Scale MOCVD,” IEEE J. Quant. Elect., 30, 24 (1994). [Smit1] R. G. Smith and S. D. Personick, “Receiver Design for Optical Communication Systems,” in Semiconductor Devices for Optical Communication, Vol. 39 of Topics in Applied Physics, H. Kressel, editor, Springer-Verlag: Berlin Heidelberg New York, 1980. [Srou1] J. R. Srour and J. M. McGarrity, “ Radiation Effects on Microelectronics in Space,” Proc. IEEE, 76(11), pp. 1443-1469, November 1988. [Srou2] J. R. Srour, S. Othmer and K. Y. Chiu, “Electron and Proton Damage Coefficients in Low-Resistivity Silicon,” IEEE Trans. Nucl. Sci., 22, 2656 (1975). [Srou3] J. R. Srour, R. A. Hartmann and K. S. Kitazaki, “Permanent Damage Produced by Single Proton Interactions in Silicon Devices,” IEEE Trans. Nucl. Sci., 36, pp. 1597-1604 (1986). [Srou4] J. R. Srour, G. J. Vendura, Jr., D. H. Lo, C. M. C. Toporow, M. Dooley, R. P. Nakano and E. E. King, “Damage Mechanisms in Radiation-Tolerant Silicon Solar Cells,” IEEE Trans. Nucl. Sci., 45, 2624 (1998). [Stas1] E. G. Stassinopoulos and J. P. Raymond, “The Space Radiation Environment for Electronics,” Proc. IEEE, 76(11), pp. 1423-1442, November 1988. [Su1] Y.-K. Su, W.-L. Li, S..-J. Chang, C. S. Chang and C.-Y. Tsai, “High-Performance 670-nm AlGaInP/GaInP Visible Strained Quantum Well Laser,” IEEE Trans. Elect. Dev., 45, 763 (1998). [Summ1] G. P. Summers, E. A. Burke, C. J. Dale, E. A. Wolicki, P. W. Marshall and M. A. Gelhausen, “Correlation of ParticleInduced Displacement Damage in Silicon,” IEEE Trans. Nucl. Sci., 34, 1134 (1987). [Summ2] G. P. Summers, E. A. Burke, M. A. Xapsos, C. J. Dale, P. W. Marshall and E. L. Petersen, “Displacement Damage in GaAs Structures,” IEEE Trans. Nucl. Sci., 35, 1221 (1988). [Summ3] G. P. Summers, E. A. Burke, P. Shapiro, S. R. Messenger and R. J. Walters, “Damage Correlations in Semiconductors Exposed to Gamma, Electron and Proton Irradiation,” IEEE Trans. Nucl. Sci., 40, pp. 1327-1379 (1993). [Summ4] G. P. Summers, “”Displacement Damage Mechanisms and Measurements,” Section IV of the 1992 NSREC Short Course notes.

III - 72

[Suzu1] A. Suzuki, T. Uji, Y. Inomoto, J. Hayashi, Y. Isoda and H. Nomura, “InGaAsP/InP 1.3 um Surface-Emitting LEDs for High Speed Short Haul Optical Communication Systems,” IEEE Trans. Elect. Dev., 32, 2609 (1985). [Sze1] S. M. Sze, Physics of Semiconductor Devices, John Wiley: New York, 1981. [Tada1] H. Y. Tada, J. R. Carter, B. E. Anspaugh and R. G. Downing, Solar Cell Radiation Handbook, 3rd ed., Jet Propulsion Laboratory Document 82-69, November, 1982. [Tayl1] S. J. Taylor, M. Yamaguchi, T. Yamguchi, S. Watanabe, K. Ando, S. Matsuda and T. Hisamatsu, “Comparison of he Effects of Electron and Proton Irradiation on n+-p-p+ Silicon Diodes,” J. Appl. Phys., 83, 4620 (1998). [Vand1] D. Vanderwater, I.-H. Tan, G. E. Höfler, D. C. Defevere and F. A. Kish, “High-Brightness AlGaInP Light Emitting Diodes,” Proc. of the IEEE, 85, No. 11, 1752 (1997). [Win1] P. Winokur, “Radiation-Induced Interface Traps,” in Ionizing Radiation Effects in MOS Devices and Circuits, P.V. Dressendorfer and T. P. Ma, editors, John Wiley: New York, 1989. [Wong1] H.-S. Wong, “Technology and Device Scaling Considerations for CMOS Imagers,” IEEE Trans. Elect. Dev., 43, 2131 (1996). [Wong2] H.-S. Wong, R. T. Chang, E. Crabbe and P. D. Agnello, “CMOS Active Pixel Sensors Using a 1.8-V, 0.25 µm CMOS Technology,” IEEE Trans. Elect. Dev., 45, pp. 889-893 (1998). [Xaps1] M. A. Xapsos, G. P. Summers, J. L. Barth, E. G. Stassinopoulos and E. A. Burke, “Probability Model for Worst-Case Solar Proton Event Fluences,” IEEE Trans. Nucl. Sci., 46, pp. 1481-1485 (1999). [Yabl1] E. Yablonovitch and E. O. Kane, “Band Structure Engineering of Semiconductor Lasers for Optical Communications,” J. Lightwave Tech., 6, pp. 1292-1299 (1988). [Yama1] T. Yamaguchi, S. J. Taylor, W. Watanabe, K. Ando, M. Yamaguchi, T. Hisamatsu and S. Matsuda, “Explanation for the Carrier Removal and Type Conversion in Irradiated Silicon Solar Cells,” Appl. Phys. Lett., 72, 1226 (1998). [Yama2] M. Yamguchi, A. Khan, S. J. Taylor, M. Imaizumi, T. Hisamatsu and S. Matsuda, “A Detailed Model to Improve the Radiation Resistance of Si Space Solar Cells,” IEEE Trans. Elect. Dev., 46, 2133 (1999). [Zhao1] Y. F. Zhao, A. R. Patwary, R. D. Schrimpf, M. A. Neifeld and K. F. Galloway, “200 MeV Proton Damage Effects on Multi-Quantum Well Laser Diodes,” IEEE Trans. Nucl. Sci., 44, 1898 (1997). [Zhao2] Y. F. Zhao, R. D. Schrimpf, A. R. Patwary, M. A. Neifeld, A. W. Al-Johani, R. A. Weller and K. F. Galloway, “Annealing Effects on Multi-Quantum Well Laser Diodes after Proton Irradiation,” IEEE Trans. Nucl. Sci., 45, 2826 (1998). ---------------------------

This work was carried out by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Adminstration (NASA), Code AE.

III - 73

2000 IEEE NSREC SHORT COURSE

Section IV

Radiation Effects Testing of Mixed Signal Microelectronics

H. Jake Tausch and Dave Alexander/MRC

Approved for public release; distribution is unlimited

Radiation Effects Testing of Mixed Signal Microelectronics H. Jake Tausch, Jr. and David R. Alexander Mission Research Corporation 5001 Indian School Road NE Albuquerque, NM 87110-3946

4.1

Introduction

Mixed signal microcircuits (MSMs) are increasingly found in applications of space and military systems that may be subjected to natural and/or nuclear radiation. The purpose of this presentation is to review the unique aspects of radiation effects in mixed signal microcircuits and prepare the reader to perform testing in appropriate environments and interpret the test results. As a rule, modern MSMs employ both MOS and bipolar circuit elements in their design. This is true both for devices fabricated in CMOS technologies as well as those fabricated in BiCMOS processes. CMOS technologies have bipolar elements as an inherent part of their structure in the form of lateral and substrate PNPs (in the case of N-well technologies). Designers frequently use these “parasitic” devices for voltage references, current mirrors, and biasing elements. Therefore, radiation testing and data analysis must consider effects on both bipolar and MOS devices in planning the characterization of MSMs. Furthermore, circuit designs used in the analog sections of MSMs are far different from those used in the digital sections. The test engineer must plan his test to be sensitive to radiation induced changes in all sections of the device. He must have an appreciation for how the analog and digital sections interact, and how those interactions may be affected by the mechanisms associated with radiation exposure. The presentation will initially focus on the impact of radiation mechanisms on mixed signal devices and circuits in order to alert the test engineer to radiation induced changes in performance. The discussion will then proceed to consideration of test procedures designed to ensure that worst case failure mechanisms are excited and observed. A key element in achieving a valid test is the development of a test plan. The quality of the test plan is the single most important factor in determining the quality of the test results. The plan must consider measurement and data management procedures before, during, and after radiation exposure. Once the plan has been formulated, it must be implemented carefully. Implementation issues include development of test fixtures, instrumentation selection, and management of the test sequence to ensure that data are not compromised by excessive time between exposure and measurement. In summary, the presentation guides the engineer in “What to look for; How to find it; and How to ensure it is correct.” 4.2

Encountering Mixed Signal Microcircuits

In planning for any test of a modern microcircuit, the test engineer is well advised to consider issues associated with mixed signal circuits. Many so-called digital ICs have mixed signal circuitry buried in their functions. Of course, there are microcircuits such as digital-toanalog converters (DACs) and analog-to-digital converters (ADCs) that are clearly mixed signal devices. However, they come in a wide variety of architectural variants some of which are dominated by the analog portion of the design (e.g., successive approximation ADCs) while others are dominated by the digital portion of the design (e.g., delta sigma converters). The test engineer is must consider the type of architecture and its impact on potential failure mechanisms

IV - 1

when he plans his test. In short, mixed signal devices are highly individualized and one test approach will not be appropriate over the range of architectures and technologies represented by this class of devices. The mixed signal nature of other types of microcircuits are less obvious, yet the test engineer must consider analog circuit effects as well as digital in preparing for a test. Memories are a prime example of this class of circuits. Because these devices store such densely packed information, they should really be thought of as complex charge management circuits. For example, in the simplified schematic of DRAM (dynamic random access memory) shown in Figure 1, we can identify several purely analog design elements that are particularly sensitive to radiation effects and which determine the radiation hardness of the device. First, the information (logic 1 or 0) is stored on a capacitor, typically on the order of 50 fF. An ion strike through the drain of the pass transistor can easily discharge the capacitor to a level where its charge can no longer be identified as a 1. Similarly, total ionizing dose induced leakage current in the pass transistor can discharge the capacitor. Finally, total dose induced changes in the offset voltage of the sense amp can render it incapable of distinguishing a 1 from a 0. All of these radiation effects lie more in the province of mechanisms typically associated with analog electronics, and they are only a few examples.

Figure 1. Conceptual DRAM Schematic. Another important example can be found in flash memories. Those devices typically use an analog charge pump to boost the voltage supplied to the chip to a level needed to charge/discharge a floating gate by either Fowler-Norheim tunneling or hot electron injection. The charge pump boosts the voltage to 12 to 20 volts, which is typically held on a depletion region capacitor of a substrate diode. Charge trapping in the field oxide can invert the surface around the cathode and produce leakage paths that bleed off the charge and reduce the programming voltage. This is the dominant failure mechanism observed in two popular, commercial flash memories [1]. Even I/O (input/output) circuits for modern, high-speed digital microcircuits are employing analog techniques for data communication. The increasingly popular LVDS (low voltage differential signaling) standard (EIA-644) is a full analog design with a differential IV - 2

driver and sense amp for the output and input respectively [2]. Many implementations use extensive analog temperature compensation to ensure specified performance over an extended temperature range. The analog designs have the advantage of reducing power and noise and increasing transmission rate. In the case of LVDS, logic 1 and 0 are represented by ±100 mV across a nominal load resistance of 100 ohms. Total ionizing dose effects that change the input offset voltage of the receiver or imbalance the output drive capability of the transmitter can degrade the I/O performance. Although most engineers recognize the analog nature of sense amps, charge pumps, and differential I/O, there are many applications of mixed signal technology that are unfamiliar even to experienced test personnel. For example, the clock on many modern, high speed microprocessors and ASICs is often buffered by a phase locked loop (PLL). The PLL may be used to synchronize the internal chip clock with the external clock and/or to boost the internal clock frequency to a higher rate than the external [3]. The circuitry to implement a PLL includes a phase detector, a charge pump, a filter, a voltage controlled oscillator, and a frequency divider. Together they constitute a highly non-linear, analog circuit that is often very susceptible to degradation from total ionizing dose. As technology advances to smaller and smaller feature sizes, PLLs will be used in most microcircuits and may become the dominant failure point. 4.3 Unique Aspects of Mixed Signal Microcircuits There are several unique aspects of mixed signal microcircuits, which make them challenging test subjects both for normal electrical characterization and especially for radiation testing. The category of mixed signal devices encompasses a huge variety of circuit types, design architectures, process technologies, and applications. This variety makes generalization quite difficult. However, the most important step in developing a test approach for any mixed signal device is to develop an appreciation of the potential effects of radiation on key elements in the design. As an example of this approach, we have selected a hypothetical mixed signal ASIC which is typical of space applications. Its block diagram is shown in Figure 2. Several properties of its design are good examples of the types of circuits and potential radiation effects that are encountered in testing mixed signal devices. A trade-off study of fabrication cost versus radiation hardness study would be required early in its development to determine if the design should be fabricated in a radiation hardened or commercial process. In the following discussion, we identify several effects that would be encountered if a commercial process were used. In that sense, the discussion is appropriate for many commercial mixed signal devices that might be tested for space application. The hypothetical mixed signal ASIC includes an analog multiplexer, an analog-to-digital converter, digital-to-analog converters, a voltage reference, control circuitry, and data registers. We will briefly review the characteristics of each of these circuit blocks and then consider the impact that space radiation environments can have on their operation.

IV - 3

Anlg 1 Anlg 2 CS

Analogto-Digital Converter

Anlg 4

MUX Anlg 5 Anlg 6 Anlg 7 Anlg 8

Analogto-Digital AnalogConverter to-Digital AnalogConverter to-Digital Digital to Converter Analog Converter

Aout 1 Aout 2 Aout 3

Voltage Reference

Control & Data Registers

Anlg 3

RD WRT CLK RST BSY EOC

Data Bus

Aout 4

Figure 2. Mixed Signal ASIC Block Diagram. 4.3.1

Analog Multiplexer

External analog inputs may be connected to one of the 8 input channels. The multiplexer is implemented with 8, CMOS transmission gates. A particular channel is selected by writing its selection code to the multiplexer address and command register. A conversion sequence begins at the completion of the address write. 4.3.2

Analog to Digital Converter

The ADC design is based on a successive approximation, switched capacitor architecture. A 12 bit ADC is within the capability of such an architecture. In a 5 volt system, the full-scale range would be 0 to 4.096 volts, thus making the least significant bit (LSB) equivalent to 1 mV. 4.3.3

Digital to Analog Converter

The DACs are programmed through the registers in the digital interface. Each DAC output is a current source that drives a load ohm resistor to provide a voltage output. If a 10 bit DAC is assumed, the LSB represents 4 mV. Since commercial CMOS technologies typically do not have high resistivity layers, the load resistors would be located off-chip.

IV - 4

4.3.4

Voltage Reference

The reference voltage is based on a bandgap circuit that is illustrated in the conceptual schematic shown in Figure 3 [5]. The value of the reference is proportional to the emitter base voltage (Vbe) of transistor Q1 plus the closed loop gain of the amplifier times the difference in Vbe between Q1 and Q2. Since the delta Vbe is proportional to the logarithm of the emitter current densities in the two transistors, the reference voltage can be described by Equation 1. Vref = Veb1 +

R3 kT  J 2  ln  R2 q  J1 

(1)

If J2=10*J1, then the negative temperature coefficient of Vbe (-2 mV/°C) is almost exactly cancelled by the temperature coefficient (0.198 mV/°C ) of the second term of the equation. The second term is referred to as the PTAT (proportional to absolute temperature). With appropriate calibration, it can act as a thermometer, and its buffered voltage is sometimes made available on an output terminal to permit it to be digitized by the ADC. Errors in the reference are the result of mismatched values of Vbe, input offset in the amplifier, and the temperature coefficient of the resistors. As indicated in Figure 3b, the PNP transistors are constructed from the P-plus source/drain implant, acting as the emitter, the N-well, acting as the base, and the substrate, acting as the collector. The transistors are diode connected so they require little or no current gain. However, it is important for them to have matching Vbe. The effects of radiation on the reference will be discussed later.

IV - 5

Figure 3a. Conceptual schematic of bandgap voltage reference.

Figure 3b. Cross section of vertical PNP in CMOS N-well technology.

4.3.5

Control Circuits and Data Registers

The mixed signal ASIC has several modes of operation, which are typically managed by a microcontroller. The specific operational mode selection and the data transfer are performed by writing and reading the data registers that provide the microcontroller interface. The data registers are composed of D-type latches and directly interact with the ADC and DAC functions. In the case of the ADC, the data registers contain the digital value of the conversions. For the DACs, the registers contain the digital code that directly determines the analog output values.

IV - 6

4.4

Radiation Effects on Mixed Signal Microcircuits

The hazards that mixed signal microcircuits encounter in a space environment include electrons, protons, and heavy ions in the form of galactic cosmic rays. The interaction of these environments with the silicon material change the electrical characteristics of circuit elements and affect the performance of the mixed signal microcircuit. The radiation effects are typically categorized under the headings of total ionizing dose effects (TID), single event effects, and displacement effects. Electron and proton exposures produce total ionizing dose effects. Strikes from galactic cosmic rays and protons produce single event effects. Proton exposure also produces displacement damage. 4.5

Total Ionizing Dose Effects on Mixed Signal Microcircuit Elements

Total ionizing dose effects include positive charge trapped in gate and field oxides and the generation of interface states at SiO2/Si boundaries. In MOS devices, the trapped charge results in threshold voltage (Vt) shifts. In N-channel devices, the Vt shifts operation toward depletion mode, while P-channel devices shift toward enhancement mode. Interface states shift both types of devices toward enhancement mode operation and decrease the subthreshold slope of drain current versus gate voltage (Id versus Vg). They also decrease the carrier mobility in both types of devices with the net result of decreasing drive strength. Figure 4 is the Id versus Vg characteristic of a P-channel transistor as a function of total dose exposure under two different bias conditions. Note that there is a difference in threshold voltage shift depending on the bias during radiation The amount of threshold voltage shift caused by oxide trapped charge is a function of the oxide thickness squared as expressed in Equation 2. ∆Vt ∝ tox2

(2)

Most mixed signal CMOS technologies are based on a recessed field oxide process such as the one depicted in Figure 5. The transistor gate covers not only the thin gate oxide, but also the oxide in the transition from gate oxide to field oxide. This transition oxide is referred to as the “bird’s beak” and is typically a region of high mechanical stress. Consequently, the threshold voltage shift in this region can be quite large resulting in a “turn-on” of the parasitic transistor at either end of the N-channel transistors. Figure 6 shows an Id versus Vg characteristic of a commercial N-channel transistor as a function of radiation. At relatively low level of total dose exposure, the leakage current from the parasitic edge transistors is a significant portion of the conductance. Total dose induced positive charge trapping also occurs in the field oxide that isolates transistors from each other and from the wells. As depicted in Figure 7, the trapped charge can invert the surface of P-type material and result in a leakage path between adjacent N-channel devices and between N-channel transistors and the N-well. Figure 8 shows the field oxide leakage current measured in a commercial N-well technology.

IV - 7

Figure 4a. Changes in PMOS transistor characterics as a function of total ionizing dose with “on-site” bias condition (Vg=gnd, Vd=5V, Vs=5V, Vsub=5V).

Figure 4b. Changes in PMOS transistor characteristics as a function of total ionizing dose with “off-state” bias condition (Vg=5, Vd=gnd, Vs=5V, Vsub=5V).

IV - 8

Figure 5. Charge trapping in recessed field oxide technologies.

Figure 6. Edge leakage effects on N-channel tranistor Idversus Vg characteristics with “on-state” bias condition (Vg=5, Vd=gnd, Vs=gnd, Vsub=gnd).

IV - 9

Vdd

N+ contact

Vss

(+ )

N-WELL

polysilicon

) (+ (+ N+ source ) ) (+) (+) (+) (+) (+) (+) (+) (+) (+)(+ Leakage Path P-epitaxial layer P+ Substrate

F ield O xide T ransistor D rain C urrent (A m

Figure 7. Total dose induced leakage paths in N-well CMOS technology.

C o m m ercial S u b m icro n F O X I/V C h a racteristics 1e-4

30 K rad(S i) 10 K rad(S i)

1e-5

6 K rad(S i)

1e-6

3 K rad(S i) P re-rad

1e-7 1e-8 1e-9 1e-10 1e-11 1e-12 1e-13

-5

0

5

10

15

20

25

F ield O xide T ransistor G ate V olta ge (V olts) Figure 8. Parasitic leakage under field oxide as a function to total dose.

IV - 10

4.5.1

Total Ionizing Dose Effects on Analog Multiplexer Block

Having identified the major device level effects of total dose we can now relate them to some expected changes in the performance of mixed signal microcircuits. Beginning at the analog multiplexer (shown conceptually in Figure 9), we find that the transmission gate implementation of the analog multiplexer could be susceptible to edge leakage in the N-channel transistors. This is especially the case if one channel is predominantly “on” during irradiation. The positive bias on the N-channel gate will enhance charge trapping. The result will be that the transmission gate will be difficult to turn “off”, and the signal on that channel will leak through to the ADC even when other channels are selected.

PMOS 15.2/1.2 5.0 V IN

OUT

LEAK PATH GND

NMOS 7.6/1.2

Figure 9. Conceptual diagram of a transmission gate analog multiplexer showing potential leakage paths.

4.5.2

Total Ionizing Dose Effects on a Successive Approximation Register (SAR)

The ADC is also potentially susceptible to total dose degradation at several points in the design. These can be appreciated by examining the conceptual diagram of a charge redistribution, switched capacitor SAR design as shown in Figure 10 [6]. The switches shown in the figure are implemented with N-channel transistors. Edge leakage currents can cause the input signal to bleed off of the capacitors and be degraded. Furthermore, any input offset voltage resulting from unequal threshold voltage shifts in the differential input to the comparator will produce an equivalent error in the conversion. For that reason, special care must be taken in the design of the comparator to compensate for unequal Vt shifts.

IV - 11

Figure 10. Conceptual diagram of a successive approximation, charge redistribution, switched capacitor ADC.

4.5.3

Total Ionizing Dose Effects on a DAC

The DAC is a current mode converter similar in concept to the diagram shown in Figure 11 [7]. The current source values are set from the reference derived from the bandgap. The currents are converted to an output voltage by the external resistor Rf. Total dose sensitivity can result from leakage around the switches, if they are implemented with N-channel transistors, and errors in the reference voltage for setting current values.

Figure 11. Conceptual diagram of current mode DAC.

IV - 12

4.5.4

Total Ionizing Dose Effects on a Voltage Reference

The reference segment of the mixed signal ASIC is the heart of the design. Any radiation effects on it will be reflected in all other sections. Referring back to Figure 3, any change in the input offset voltage of the amplifier will clearly change the reference voltage. Therefore, it must be carefully designed to compensate for differences in threshold voltage shift in the transistors in the two legs of the differential input. Another concern for the reference is the radiation effects on the bipolar transistors used to generate the reference. Total dose typically affects PNP transistors by changing the surface recombination velocity in the base region producing an increase in base current. The temperature compensation reference depends on the difference in Veb’s resulting from the different current densities through the PNP transistors (see figure 3a). In Figure 12, the emitter base current of a PNP transistor has been plotted as a function of emitter base voltage before irradiation and after several total dose exposures [21]. The Veb values for typical current values of both Q1 and Q2 from Figure 3 have been indicated before and after irradiation to 100krad. The post-irradiation value of ∆Veb is significantly larger than its pre-irradiation value. Since the bandgap reference design assumes that all ∆Veb change is due to temperature, the radiation will cause the reference to be in error and have a direct impact on the accuracy of the converters.

Pre –rad ∆Veb

Post-rad ∆Veb

Figure 12. Total ionizing dose effects on PNP emitter characteristics. IV - 13

In the last 5 years, researchers have found that total dose effects in some transistors (particularly lateral and substrate PNPs) are much more pronounced if the dose is accumulated at a low rate (i.e., 0.001 Rad(Si)/s) that is typical of space exposures. Figure 13, shows the increase in base currents of a PNP transistor irradiated at various dose rates from 294 Rad(Si)/s down to 0.001 Rad(Si)/s [8]. These currents are reflected in the emitter currents (Ie = Ib + Ic) and produce differences in the ∆Veb for the bandgap reference which are a function of both total dose and the rate the dose was delivered.

Figure 13. Excess base current in PNP transistors following 0.001 Rad(Si)/s irradiation. Unfortunately low dose rate irradiations are extremely time consuming. Slightly more than 7.5 months is required to accumulate 20 Krad(Si) at a rate of 0.001 Rad(Si)/s compared to 3.33 minutes for the same dose accumulated at 100 Rad(Si)/s. Fortunately not all devices show this phenomena and much recent work has been done to develop less time consuming ways to screen parts for this sensitivity. This will be discussed more in the section on testing. 4.5.5

Total Ionizing Dose Effects on Control Circuits and Registers

The final circuit blocks to be considered for the effects of total ionizing dose are the control circuits and data registers. Typically the digital elements take up a relatively small IV - 14

portion of the layout. However, they are important, because they are the interface to the digital system. Edge leakage currents are of greatest concern. Referring to Figure 6, we see that after 160 Krad(Si), the edge leakage is only about a factor of 4 less than the “on” current of the intrinsic transistor. If the designer uses NOR type circuits as shown in Figure 14, the leakage in the parallel N-channel transistors will significantly load the series P-channel transistors when the output is in a “1” state. If the output is not able to maintain a solid 1 state, logic errors may be generated. Furthermore, since the P-channels are driving current into the output node and the N-channels are leaking it to Vss, there will be a significant increase in power dissipation. The digital logic blocks can also be responsible for a large portion of the field oxide leakage current. To make the layout of digital gates most efficient, rows of N-channel and P-channel devices are often placed side by side. This places the N-channel sources, connected to Vss, within a minimum design rule space of the N-well, connected to Vdd. Such an arrangement provides a potential leakage path from Vdd to Vss. In many commercial, mixed signal microcircuits, a sharp increase in Idd is the first indication that the devices is about to fail from total dose effects. Therefore, monitoring Idd while the device is being irradiated and during post-irradiation characterization is extremely important.

Vdd IN A IN B IN C IN D

OUT Leak Paths

Figure 14. TID induced leakage currents in a NOR gate. 4.6

Single Event Effects on Mixed Signal Microcircuits

Single event effects in mixed signal microcircuits include single event upset (SEU) in logic registers, single event transients (SET) in analog circuits, and single event latchup (SEL). All of these effects are the result of charged particles passing through the semiconductor material. In the case of galactic cosmic rays, the particles are ions of elements with the lighter elements being most abundantly represented as shown in Figure 15 [9]. In the case of protons, the ionization is the result of secondary particles from nuclear reactions and scattering.

IV - 15

Figure 15. Cosmic ray abundance for most important varieties. If the ionization track intersects a PN diode junction as depicted in Figure 16, a current spike is generated. The spike can charge or discharge the node capacitance associated with the diode terminal and produce a voltage transient. The transient may be of sufficient amplitude and duration to change the logic state of a latch, or it may propagate as a signal through the circuit. Alternatively, the current spike may excite parasitic circuit elements into a high conduction state known as latchup.

Figure 16. Ionization track intersecting a PN junction. IV - 16

4.6.1

Single Event Latchup

In mixed signal microcircuits, all of these effects are of concern. However, SEL is the most potentially catastrophic of the phenomena. The potential for latchup is inherent in CMOS technology. As shown in Figure 17, a PNPN structure consisting of the P-channel source, the N-well, the P-substrate, and the N-source constitute a coupled transistor pair that is similar to the model for a silicon controlled rectifier. An ionization path that intersects the N-well/P-substrate junction will generate a localized photocurrent that tends to turn-on both the NPN and PNP transistors. The resulting collector current from each transistor will increase the base current drive in its companion, and lead to a regenerative state in which both transistors are saturated. This produces a low impedance path between Vdd (connected to the P-channel source) and Vss (connected to the N-channel source). The electrical I/V characteristic of the latchup elements are shown in Figure 18 [18]. Once latchup is initiated, only interrupting the power supply can break the latch. The greatest danger from latchup is a catastrophic burnout of metallization, which would permanently damage the device. Although latchup is possible in either the digital or analog portions of the microcircuit, the analog portions are often most susceptible because their layouts are more irregular and often do not have the frequency of well and substrate contacts found in digital layouts.

Vdd Vdd P+ source N+ contact Anode

Vss N+ source Cathode

Vss P+ contact

N-well P-epi P-substrate

Figure 17. CMOS structure showing potential latchup paths. Latchup may be eliminated by carefully employing process and design techniques such as dual wells (N-well and P-well), retrograde wells, epi-layer processes, guardbanding, and frequent well and substrate contacts. However, all of these techniques must be employed with great care to ensure they are effective. In many commercial devices, the designers take great care to eliminate latchup in I/O circuits, because they are concerned about electrical transients initiating the latch condition. However, they give less attention to the interior cells. Since particle ionization tracks occur in the interior as readily as the I/O, SEL may be a problem. The test engineer should always include a latchup test in his single event characterization plan.

IV - 17

The conditions that enhance the likelihood of latchup are high voltage and high temperature. High voltage increases the likelihood that the anode to cathode voltage of the PNPN element will exceed the holding voltage (see Figure 18). High temperature increases the gain in the parasitic bipolar transistors that constitute the latch path. This makes it more likely for the gain product to exceed 1. Typically, SEL tests are performed with the power supply voltage 10% greater than the nominal supply, and at 100°C.

Figure 18. I/V characteristic of a latch path. 4.6.2

Single Event Transients

The transients associated with an ion strike typically have a rise time of approximately 10 ps and a decay time constant of 100 ps. Analog circuits respond to the transients with their unit impulse response, that is their gain-bandwidth characteristic convolved with the frequency content of the impulse. The net result is attenuation and broadening of the response waveform. The waveform is also affected by the slew rate capability of the circuit. Figure 19 [16], shows the response of an operational amplifier to an ionized particle strike. The response time of the op amp is over 8000 times longer than the duration of the initial SET. If such a transient occurs at the output of an analog multiplexer, in the voltage reference amplifiers, or the DAC amplifiers, the circuits will be disrupted for at least the duration of the transient. Those disruptions will cause erroneous outputs. However, the errors will be flushed by the next conversion sequence. The importance of the error depends on the system requirements, but most systems are sufficiently robust to survive uncorrelated instances of transient errors. However, the test engineer should organize his test to be sensitive to transient errors generated in the analog portion of the mixed signal microcircuit. IV - 18

Figure 19. Operational amplifier transient response to a heavy ion strike. Single event transients can also occur in digital, combinational logic. When technologies were based on transistor feature sizes of 0.5 microns and above, these transients were of little concern. The digital logic gates were not fast enough to transmit 100 to 200 ps spikes. The spikes were essentially filtered out by the first logic gate they encountered. However, as transistor features shrink to 0.35 microns and below, the logic gates become fast enough to permit a 100 to 200 ps transient to be passed through an infinite number of combinational gates. When these transients encounter a latch input, they may be clocked into the latch if their arrival coincides with the latch clock. At that point, the transient becomes a logical error. Since the clock frequency is increasing with reduced feature size, the probability of the latching in an erroneous value can be significant in small feature size microcircuits. Currently, only state-ofthe-art converters are fabricated in deep submicron technologies. However, the advanced technologies are becoming increasingly available, and errors caused by SET are expected to increase sharply. The net result will be a significant increase in the number of errors produced since the entire area devoted to digital logic (combinational and sequential) participates in the error generation process. To fully characterize the mixed signal device error rate, the device should be operated at it maximum clock frequency to increase the probability of transients in the combinational logic being latched in as errors. 4.6.3

Single Event Upset

If a single event transient occurs at a sensitive node of a sequential circuit (memory cell, latch, flip flop, register), it can cause the state of the cell to change (upset) resulting in a static error. In order to cause an upset, the charge from the transient must exceed the critical charge on the node required to change its state for long enough for the regeneration loop in the cell to establish a new stable state in the latch. Figures 20 and 21 are the layout and schematic of a typical D latch cell. If an ionization track intersects the node U1OUT and injects a charge of .58 pC, the latch will upset as illustrated in Figure 22. In a Mixed Signal ASIC, all of the latches for the registers containing the results of the ADC conversions, the DAC inputs, and the calibration values would use a similar design. Therefore, it is easy to configure an SEU test that will determine the bit error rate of the device. Since all the registers can be written or read, a pattern is loaded into the register array, and readout periodically to check for SEU.

IV - 19

Figure 20. Layout of the D-Latch Cell in the Mixed Signal ASIC.

IV - 20

Figure 21. D-Latch Schematic.

IV - 21

Figure 22. Simulation of a single event upset (SEU) in the Mixed Signal ASIC D-Latch. Special attention should be given to SEU in registers controlling DACs. The registers directly control the current sources that are ultimately summed to provide the DAC output. If an upset occurs in a DAC register, the result will be reflected in the analog output. Unfortunate responses could be generated if the DAC controlled critical electromagnetic actuators. 4.7

Proton Effects on Mixed Signal Microcircuits

As discussed above, protons encountered in the earth’s radiation belts or as a component of the solar wind can cause total ionizing dose damage or single event effects (through nuclear reactions and scattering). In addition, protons cause displacement damage in the semiconductor crystal. The resulting damage clusters act as recombination/generation centers and decrease the minority carrier lifetime in the silicon. Unless the displacement damage is particularly severe, the MOS transistors are not significantly affected, because they are majority carrier devices. However, bipolar transistors, such as those used in the bandgap reference, can be seriously degraded. For example, Figure 23 is a plot of bipolar gain degradation as a function of proton fluence [10]. Note that the gain change from protons occurs at all current amplitudes as opposed to total ionizing dose degradation that occurs primarily at low currents. In the case of the bandgap reference, additional error will be introduced from the compounding of total ionizing dose degradation and proton damage. The test engineer should be careful to monitor reference voltage changes as well upsets and total dose effects during proton testing.

IV - 22

Figure 23. Proton irradiation effects on bipolar transistor gain.

4.8

Mixed Signal Microcircuit Performance Degradation from Radiation Exposure

In the material above, radiation effects have been related to changes in performance of mixed signal cells and building blocks for ADCs and DACs. However, the test engineer will be observing changes in performance at a macro level. For example, in the Mixed Signal ASIC that we have used as an example, the engineer will be observing a change in the ADC and DAC conversion characteristics rather than changes in amplifier, comparator, latch, or other subcircuit performance. While insight into the reasons for the changes in performance is invaluable in selecting irradiation conditions, determining test procedures, and analyzing data, the applications engineers are primarily interested in how the integrated performance changes. In Figure 24, the ADC conversion characteristic for a commercial A to D converter is shown before and after total ionizing dose exposure. The figure plots the value of the digital output count against the analog input to the converter. Prior to irradiation, there is close to a 1 to 1 relationship between the two values over the entire input voltage range (0 to 4.096 volts / 0 to 4096 counts). This indicates a high quality conversion. However, following irradiation, the form of the conversion characteristic has changed drastically. The conversion slope (referred to as the gain) has been drastically changed, and the characteristic saturates before the entire input range is covered. In the remainder of the course, we discuss the macro measurements to be used in testing mixed signal microcircuits and provide guidelines for their interpretation.

IV - 23

4500 4000

Output Counts (Counts)

3500

78.9 KRad

3000 2500 2000 1500

PreRad

1000 500 0 0

1

2

3

4

Input Voltage (Volts)

Figure 24. ADC Conversion characteristic for a commercial ADC as a function of total ionizing dose. 4.9

Testing/Characterizing Mixed Signal Microcircuits

Each type of mixed signal microcircuit has unique test requirements. For this discussion, we will use analog to digital (A/D) converters to illustrate radiation effects and associated test techniques. Figure 25 is a conceptual drawing showing the operation of an A/D converter. Note that an A/D converter quantizes an analog input voltage and converts it to a digital output. If the analog input is slowly varied at the same time the digital output is examined, the conversion characteristics of the A/D can be determined. In general, this will be a series of steps where the transitions between steps occur when an input voltage has a 50% chance of being converted to either one code or its nearest neighbor (e.g. multiple conversions might generate equal numbers of readings 0×2F3 and 0×2F4). Many devices and test set-ups have sufficient noise that a given input voltage will be converted into a range of digital readings covering 2, 3, or even more adjacent values. Nonetheless, the conversion curve between analog inputs and digital outputs can be mapped using averaging techniques. This curve is useful because it indicates how well internal devices (e.g. r-2r resistor ladders) match. The transfer curve in conjunction with noise measurements will indicate how well the A/D will perform in an actual application.

IV - 24

Figure 25. Basic (DC) functionality of A/D Converters. Several parameters can be extracted from the conversion curve. First, a best fit line can be calculated to the curve, and the average gain and offset of the line can be compared to an ideal curve. Gain and offset can typically be trimmed as desired or software corrected when the A/D is used in an actual circuit. Consequently, these are not usually critical parameters. Deviations of the transfer curve from the best fit line are very important and are generally given as Differential Non-Linearity (DNL) and Integral Non-Linearity (INL). DNL is a measure of each step width and is given as a ratio of each step to the average of all steps. INL is a measure of the variation of the transfer curve from the best fit line. The conversion curve will also indicate if there are any missing bits. Figure 26 shows examples of DNL and INL plots from a typical A/D. Note that the errors are all less than one bit. This is typical of most modern ADCs. Also note that there is a fine structure to the INL that appears as a series of saw-tooth curves in which the error builds up over several codes and then resets over one transition. This large transition is also seen in the DNL plot as a large spike. For this ADC, these steps occur regularly every 32 codes and are a result of the way this device was designed. The overall “S” shape of the INL curve may indicate bias sensitivity of internal analog circuitry such as buffer amplifiers.

IV - 25

Differential Non-Linearity is the Difference in Successive Errors

Integral Non-Linearity is the Error Between Measured and Expected Values

Figure 26. Examples of INL and DNL in an ADC. Other key parameters come from the time domain response of the A/D. Figure 27 [13] illustrates this. First note that internal analog circuitry of the A/D may not respond to a large step input as quickly as desired (i.e. one sample interval); therefore, it may take several conversions before a reading is stable to the desired resolution (figure 27a). This can be particularly important in applications where an A/D input is switched between signals that have widely different values. Quite often one A/D will be used to sample many signals (sometimes called scanning) so this settling time may be critical to proper system performance. Another important time domain parameter is the measure of cross talk between channels. This is illustrated in figure 27b where a digitized output is shown in which the A/D is switched alternately between a DC input and a sine wave input. Actual digitized waveforms of such a test are shown on the right. The top right plot of figure 27b shows a record of digitized signals made by alternating between two signals as described above, and the bottom trace shows only the digitized values associated with the DC input. This cross talk between channels may be a result of actual signal coupling between signal paths (e.g. capacitive coupling) or may result from long settling times on the output of the switching circuitry. A third set of parameters are associated with the response of the A/D to sine wave inputs. Figure 28 shows conceptually how this is measured. A sine wave signal is digitized and recorded in a sequential data file, and this file is analyzed using Fast Fourier Transform (FFT) techniques to generate the frequency domain representation of the data. FFT data is typically presented in terms of decibels (db) of power with respect to either the total A/D range or the fundamental signal. Once this FFT calculation is done, parameters such as signal-to-noise ratio (SNR), total harmonic distortion (THD) and effective number of bits (ENOB) can be easily calculated. SNR is calculated by measuring the energy in the sine wave signal and comparing this with the energy contained in all the other frequency components. THD is calculated my measuring the energy in the sine wave component and comparing it with the energy contained in the frequency components associated with all harmonics of the sine wave.

IV - 26

500

31

26

21

16

11

6

0 1

Output Word

1000

Sam ple Number

1000 500

241

211

181

151

121

91

61

31

0 1

Output Word

Figure 27a.

Figure 27b.

544 539 113

97

81

65

49

33

17

534 1

Output Word

Sam ple Num ber

Sam ple Num ber

Figure 27. Time domain parameters for A/D converters.

Figure 28. Frequency domain parameters for A/D converters. ENOB is a term that was derived by a mathematical analysis of the effect of quantizing an ideal sine wave into different numbers of discrete levels. For example an ideal 12 bit A/D would quantize a sine wave into 4096 discrete levels. If a sine wave was quantized in this way then there would be errors resulting from round-off of the actual voltages as they were converted to numbers representing each bin of the A/D. If an FFT were calculated from the resultant data set it would have a noise floor caused by this round-off error and some resultant SNR. When this is done, it can be shown mathematically [20] that the SNR will be: IV - 27

SNR{ideal)} = (6.02n + 1.76)dB where: n is number of bits If this equation is re-arranged, the effective number of bits for any given SNR will be: ENOB = (SNR{meas} - 1.76)/6.02 Figure 29 shows an example of an FFT plot of data from an A/D and the associated figures of merit. Note that the ENOB of 8.82597 can be calculated directly from the SNR using the above equation. This was a 12 bit A/D. It was later found that the loss of 3.2 bits of effective range was due to test fixture noise and lack of sine wave fidelity.

Figure 29. Example of frequency domain parameters. Each type of characterization has benefits and limitations. INL and DNL give a good indication of the internal circuit matching but no indication of noise, frequency or time response. Time domain tests give a good indication of dynamic response, but no indication of linearity, missing bits, etc. Frequency domain measurements give good information on noise, but limited information on linearity and time domain step response. Accurate measurement of any of the above mentioned parameters requires good measurement equipment and careful design of test fixtures. This is easy to see when you consider that a 16 bit A/D with a 1 volt dynamic range will have a resolution of 15 micro volts. Signals provided to the part should have less noise than this, and other noise sources on the fixture should be carefully controlled. Generally, all power supply inputs should be well filtered, possibly using “PI” or “T” filters capable of rejecting noise from 60Hz on the low end to 100MHz on the high end (to reject radio broadcast noise picked up on power cabling).

IV - 28

Frequency domain characterization has additional requirements. The input sine wave must have high signal purity with a very low FM component so that non-linearities in the signal waveform will not contaminate the A/D measurements. The digitizing clock of the A/D must have very low phase noise (i.e. the A/D sampling must be very precisely controlled so the time between each sample is exactly the same) as any jitter in sample intervals will artificially distort the results. Frequency domain testing also requires careful planning to maximize the coverage of the test. Consider the case of a 12 bit A/D characterized by taking 1024 samples and doing the above analysis. Since the A/D has 4096 possible digital output codes then, in the best case, the digitized sample will contain only ¼ of all possible codes. If the digitizing frequency and input sine wave frequency are not carefully chosen then the actual number of output codes may be considerably less. According to the Nyquist theorem the minimum number of samples that must be taken per cycle is 2. In an extreme case, where the sampling rate was exactly twice the sine wave input frequency, then it would be possible to record a sample where only two A/D output values was recorded. The best sine wave frequency would be one that would result in an exact number of waveforms being digitized over the sample interval, but also one in which every sample would be taken at a unique output value. In cases where an exact number of sine waves are not digitized, then the sample data set must be multiplied by a “windowing” function before an FFT calculation is performed. This is because the FFT calculation is based on the assumption that the sample set represents one portion of a continuous waveform that would be formed if some number of the sample sets were placed “end-to-end”. Thus, if the sample set were of 1.5 cycles of a waveform then the FFT would show frequency components of a waveform which would look like 1.5 cycles followed by an abrupt transition and 1.5 more cycles, etc. The windowing function is a mathematical weighting multiplier that insures the signal amplitude of the sample set goes to zero at each end and is unchanged in the middle. Various standard windowing functions are used and have known effects on the frequency domain results, typically spreading the input signal and each of its harmonics so their energy covers several adjacent frequency bins. 4.10

General Radiation Testing of Mixed Signal Microcircuits

For any type of radiation testing there are several preliminary steps that should be taken to insure good results. Some of these follow: Define Failures: In many applications, some of the spec sheet parameters are not of prime importance (e.g. open loop gain in an op amp) and defining part failure based on spec sheet values could result in parts failing at unrealistically low radiation levels. Instead, failure criteria should be selected from the application requirements. Select Key Parameters to Monitor: Some parameters will be chosen because they are critical to the application. Others may be chosen because they may provide insight into device failure mechanisms for later analysis. Test Sample Disposition: Plan ahead for the number of environments to be tested and the conditions in each environment. Create a parts disposition tree to insure you have enough

IV - 29

devices. Figure 30 shows an example of a parts disposition tree. Some radiation effects are dominated by the electrical design and layout and show little part to part variation (e.g. SEU). Other effects are dominated by process variations and may show significant part to part variation (e.g. total dose effects). A small sample size may be used for radiation testing of design dominated effects. A larger sample size should be tested for process dominated effects.

ADC Test Sample 20 Units Pre-test Electrical Units 1-20 Control Units 1-2

TID Units 3-10 Step Stress Irradiate Units 3-6 Estimate Spec 1.5x Spec Units 7-10 Anneal Units 3-10

Dose Rate SEE Units 11 - 13 Units 14-16 Step Stress for Upset Threshold Units 11-13

SEU TEST Units 14-15

Latchup Test Units 11-13 SEL TEST Unit 16

Neutrons Units 17-20 1E13 n/cm 2 Unit 17 2E13 n/cm 2 Unit 18

4E13 n/cm2 Unit 19 1E14 n/cm 2 Unit 20

Figure 30. General Test Planning – Device Allocation. Test Implementation Description: For each test type document what will be measured and how. It is a good idea to document the forcing functions (e.g. sine wave, lab power supply) along with the range of expected values. Also, document the data recording formats that will be used. Quite often formalizing the tests through a written description will disclose test features that must be resolved before good results can be expected. Physical Characterization: As a minimum, the date code of the devices to be tested should be recorded. It is good practice to take a photomicrograph of the die and also to do crosssection and spreading resistance probe measurements. This documents the version of the device being tested, which is important because process upgrades and die shrinks can have major impacts on radiation response. It is also useful when comparing with results from other experimenters or data taken on similar devices. Predictions: In many cases, it is strongly recommended that you have an idea of the general response the part will have to the radiation so you can plan the correct exposure regimen. For example, in SEU testing, every exposure to heavy ion irradiation will result in collateral accumulation of total ionizing dose. If ion species are initially chosen that are much too high or too low in LET from the upset threshold, then the part may receive a lot of total dose exposure IV - 30

with little or no interesting SEU behaviour. Consequently, by the time the correct ion energies are found the part response may be compromised. In some cases a detailed circuit analysis may be performed to predict the threshold exposure level that will produce a significant effect. Alternatively, you may review existing literature for testing of similar devices, etc Pre- and Post-Rad Characterization: Devices to be tested should be thoroughly tested and understood before going to a radiation facility. Doing this insures your test set-up is properly functional and also gives confidence that you are measuring parameters correctly. Quite often a more complete set of tests will be performed before and after a radiation test than is possible at the facility. Keep Reference Parts: The question often arises whether measured changes were truly introduced by radiation stress or whether there was a change in instrumentation (e.g. bad connections, calibration drift, etc). The best way to insure it was not the instrumentation is to keep reference parts and measure them periodically. Figure 31 shows standby current measured on several devices along with a post rad measurement of a reference part.

Figure 31. Measurement of a reference part ensures data integrity.

IV - 31

Keep a Good Test Log: Record critical events as they occur. This will include things such as radiation levels, part serial numbers, observed part response, etc. A good log can help resolve many apparent anomalies later when the data are being analyzed in detail. Figure 32 shows an example of a test log.

Test Log for 12 Bit A/D 26-May-99 60 Co Irradiation Time

DUT SN 900 905 910 920 930 1000 1020 1040 1050

Comments Rad Level Idd (mA) INL (Bits) DNL (Bits) ENOB 1 Pre 25.4 1.2 0.8 10.5 2 Pre 25.2 1.5 0.75 10.6 3 Pre 24.9 1.3 0.9 10.4 1 1000 25.4 1.25 0.75 10.45 2000 25.4 1.15 0.8 10.5 5000 25.4 1.2 0.85 10.5 Note ENOB Increase .. Repeat test Duplicated 10000 27 1.2 0.9 9.8 Results, make next rad step smaller 12500 30 1.8 1.2 8.5 Sudden increase in Idd, Device non-functional, Stopped run 14000 150 *** *** ***

Figure 32. Example of test log. Monitor DUTs While They’re Being Irradiated: Especially when doing initial testing on a new part type, the onset of radiation damage may occur at a level or manifest itself in a way that is unexpected. At a minimum, the supply current should be monitored while a test is in progress. In total dose environments, the supply currents will typically (but not always) increase at about the same level as other parameters start to change. In heavy ion testing, there is always the possibility of devices latching-up and drawing excessive current. Design Good Test Fixtures: Many facilities have constricted space, so test fixtures must be carefully designed. It may also be necessary to have considerable distance between the DUT and test equipment. This may adversely affect device performance, unless it considered early in the test/fixture design. Make sure the DUTs can be easily inserted/removed in the fixture. DUTs are often awkward to reach and change. 4.11

Total Ionizing Dose Testing (TID) of Mixed Signal Microcircuits

Most TID testing is done using a step-stress process in which the parts are measured, irradiated, then measured again. In characterization tests, irradiations are generally done in a 1-2-5 sequence so that devices will be characterized at doses of 1,000 R(Si), 2000 R(Si), etc. until they reach a preset limit or until they fail. In production screening tests, devices are generally irradiated to a preset level and tested at that level. Some characterization testing is done “in-flux” so that certain behaviors/parameters can be monitored to see if they recover rapidly after the radiation is removed. TID testing should be performed with the DUTs placed in a lead-aluminum box to filter out low energy radiation. Figure 33 shows a cross section of a typical shield box [11]. If

IV - 32

supporting circuitry is present on a test card, it is generally shielded to insignificant levels (e.g. below 10 R(Si)).

Figure 33. Total Dose enclosure per ASTM F1892-98. TID testing is different for MOS devices as compared to bipolar devices because of different physical phenomena that affect their performance. MOS devices are affected by oxide trapped charge and also by interface trap formation at the surface of the gate region. Trapped charge occurs immediately with radiation and may anneal with time. Interface traps form through processes that are logarithmic with time after irradiation. Consequently, some devices may fail immediately after irradiation and then recover if their damage was due to oxide trapped charge. Other devices may initially be unaffected by radiation and afterward degrade over time if their damage is dominated by interface state build-up. Annealing of trapped charge and interface trap formation are both accelerated by heating parts under bias and this “Rebound Testing” is quite often done to determine if the part is susceptible to interface state induced degradation. Bipolar devices don’t exhibit this rebound phenomena. On the other hand, certain lateral PNP transistors show a strong enhancement to their radiation response if the radiation is delivered slowly. This phenomenon is called Enhanced Low Dose Rate Sensitivity (ELDRS). Figure 34 illustrates this phenomena by showing the variation in response of various bipolar devices at low dose rates compared to the response at 50 rad(Si)/sec [12]. Because of these differences the two families of devices are generally tested using different standards to define the test sequence.

IV - 33

Figure 34. Low dose rate sensitivity can cause significant dose enhancement in bipolar devices.

MIL-STD, 883, Method 1019.5 Differences shown in parentheses

Figure 35. TID MOS Test Flow per ASTM F1892-98.

IV - 34

Figure 35 shows a flow diagram describing the test procedure for MOS devices as defined in ASTM F 1892, “Standard Guide for Ionizing Radiation (Total Dose) effects Testing of Semiconductor Devices” [11]. This test flow is similar to that described in Mil Standard 883b, Method 1019.5 with certain small differences. The figure shows the differences in the Mil Standard flow by placing the differences in parentheses. Note in the figure that devices are irradiated to spec level and tested. If the parts fail immediately after irradiation, they may, in certain cases, be annealed under bias at room temperature and tested again later. If failed parameters recover significantly with time under a room temperature bias then the belief is that they would not have failed at all if the radiation had been delivered over a longer time at a lower rate. In space applications dose is accumulated much more slowly (0.001 Rad(Si)/S) than in terrestrial tests (50-300 Rad(Si)/S). An example of such a recovery phenomena can be seen in Figure 36. In this example an SRAM exhibited a large number of bit failures when the radiation was turned off, but the number of failures began to recover rapidly once the radiation was removed. This was probably due to changes in the internal sense amps or reference biases.

y

Figure 36. Room temperature anneal accommodates parameters which anneal quickly.

IV - 35

DNL, LSBs

A second example is shown in Figure 37 where data are presented from two types of A/Ds that failed after some initial exposure to radiation [13]. In one case, the device type recovered after a 1 day anneal. In the other, there was very little change even after a two week anneal.

20 15 10 5 0

10 krad 20 krad 40 krad P re

Post

1 Day

3 D a ys

1 W eek

2 W eek s

Signal Processing Technologies SPT7855

DNL, LSBs

20 15

80 krad

10

160 krad

5 0 Pre

Po s t

1 Day

3 Days 1 Week

2 Weeks

Analog Device AD9200 Figure 37. Examples of A/D converter room temperature annealing.

The test flow also shows the provision for the accelerated aging (rebound) test described above. One other interesting feature of the test flow is that it recommends devices be tested within one hour of irradiation and that parts be transported (if required) between the irradiation source and test facility under no bias and with leads shorted. Transporting devices with no applied bias will minimize any annealing that might otherwise take place.

IV - 36

Figure 38 shows the TID test flow for bipolar devices also from ASTM F 1892, “Standard Guide for Ionizing Radiation (Total Dose) effects Testing of Semiconductor Devices”. Note that, if there is no evidence of ELDRS the parts can be tested using 50-300 R(Si)/S and room temperature conditions. If ELDRS is known or suspected, then they must be irradiated at the dose rate of the intended use, irradiated at 10mR(Si)/S and tested with a design margin of 2, o or irradiated at 100 C and 10R(Si)/S and tested with a design margin of 3. The design margins used for the two test conditions are based on the compromise between exposure time and expected additional degradation if the part had been exposed at space environment dose rates.

Figure 38. TID bipolar test flow per ASTM F1892-98. TID radiation effects vary widely between manufacturers and are not necessarily the same for any given circuit type. For example, Figure 39 shows examples of TID effects on two A/Ds from different manufacturers [13]. In one case, the supply current drifted out of spec and the DNL stayed in spec to a quite high level. In the other case, the DNL changed drastically at a fairly low dose while the supply current hardly changed. TID is perhaps the radiation environment where the effects vary the most between manufacturers, because the phenomena are so dependent on manufacturing processes, doping profiles, oxide chemistry, etc. 4.12

Heavy ION Testing of Mixed Signal Microcircuits (SEU and SEL)

Most heavy ion testing is performed in a vacuum chamber with delidded parts because of the limited range of the ions available for testing. A drawing of a typical test chamber is shown in Figure 40 [14]. Because of the time involved in pumping down the vacuum, several DUTs are IV - 37

generally placed within the chamber at the same time and then tested/placed in front of the beam one at a time. The test chambers generally have a motorized platform for moving the DUTs into the beam and for rotating them to adjust the effective LET. 2.000 DNL, LSBs

DNL, LSBs

2.000 1.500 1.000 0.500

1.500 1.000 0.500 0.000

0.000 0

20

40

60

80

0

100

5

Dose, krad

15

20

40.000 Current, mA

35.000 Current, mA

10 Dose , krad

30.000 25.000 20.000 15.000

35.000 30.000 25.000 20.000

0

20

40

60

80

100

0

Dose, krad

EXAR XRD64L15

5

10

15

20

Dos e , k rad

Signal Processing Technologies SPT7855

Figure 39. Examples of ADC performance versus TID. Heavy ion testing by its very nature is statistical and requires that a DUT be tested in each condition for a sufficient time to allow all possible effects to occur. In practice, this means a DUT will be exposed to a beam of some ion species at a certain flux with the device rotated at a given angle for a time sufficient to generate a statistically significant number of events or until the probability of an event occurring is low enough to be insignificant. A DUT is characterized by exposing it to enough ion/angle combinations to determine the threshold of upset and also the saturated cross-section. Precautions need to be taken to insure the flux is sufficiently low so photocurrent effects are negligible but sufficiently high so exposure times will be manageable. Generally, an exposure time of 2-10 minutes is a good target and, over that time, at least 30 errors/events should be observed to provide statistically meaningful results. A close track must be maintained on the total deposited dose so that TID effects do not compromise the tests. If a large number of errors are observed, then care should be taken to refresh the DUT often enough to prevent multiple bit flips (i.e. a bit being flipped to an error state and then flipped back to appear undisturbed). Alternately, the flux can be reduced.

IV - 38

Figure 40. Overview of SEU vacuum chamber from ASTM F1192M-95. For mixed signal devices heavy ion testing falls into two broad categories: errors and latchups. Errors can be the result of either a change of state of a bit (SEU) or a transient signal that is clocked into some portion of the circuit if it occurs coincident with a clock signal (SET). Preferred test conditions for upset testing are the lowest rated voltage for the part type and room temperature. For latchup the conditions should be the highest rated voltage and elevated temperature. Figure 41 shows the test flow for heavy ion testing from ASTM F 1192M-95, “Standard Guide for the Measurement of Single Event Phenomena (SEP) Induced by Heavy ION Irradiation of Semiconductor Devices”. Of special interest in this test flow is the fact that it recommends a “dry run” test of the part to insure no errors are detected before the IONs are turned on. Note that the flow is designed to reduce the number of times the vacuum chamber is cycled and also to reduce the number of times the ion species is changed. In some cases, it can take as long as ½ hour to change ion species and re-tune the source to the desired flux. The other interesting feature of the flow is that raw measurements are corrected for the effects of surface layers over the sensitive portions of the circuits. The types and energies of ions used for these tests can be significantly slowed be these layers, and their effects have to be corrected before the data is truly representative [19].

IV - 39

(

)

Figure 41. SEP Test flow per ASTM F1192M-95 (see also EIA/JEDEC Standard 57). An example of a simplified instrumentation diagram for performing an SEU test is shown in Figure 42. This is for a mixed signal ASIC which contains 4 DACs, an analog MUX and an A/D. In this test the DACs are programmed every 100mS (a process that takes ~1mS) and then read back at the end of the next 100mS wait time to detect digital bit flips. The A/D values of their analog outputs are also read at the end of the 100mS wait interval to detect any analog transients. Digital and analog errors are recorded for every 100mS cycle. This is typical of sequences used for SEU testing.

IV - 40

Every 100mS Set DAC Values For I=1 to 4 Switch input I to ADC Read ADC Wait 80mS For I=1 to 4 Read DAC(I) Control Register Compare ADC Readings & DAC register values with expected record errors

Figure 42. Simplified instrumentation diagram. Most mixed signal microcircuits are composed of distinct subcircuits with widely different responses to heavy IONs. Figure 43 shows a typical successive approximation A/D with a gain-difference amplifier, an A/D circuit or comparator, and a feed back DAC.

Typical Successive Approximation A/D Convertor

Figure 43. An A/D converter is composed of several sub-circuits.

IV - 41

In such complex circuits, it can be expected that each circuit element will have distinct response to heavy ions. This concept is shown in Figure 44, which compares the capture crosssection of a simple device with one composed of elements with distinctly different responses. A/D circuits are further complicated because they can have both analog and digital type upsets. This is shown in Figure 45, which shows a histogram of many conversions from an A/D with a DC voltage on its input [15]. In this figure the normal spread of readings due to device and fixture noise was about 10 counts. After these count bins are removed, the remaining signature is as shown. First note that there is a Gaussian distribution of digitized measurements around the nominal readings. This set of errors can be attributed to noise induced by ion strikes on analog circuits. Further note that there are groups of digitized measurements significantly separated from the normal range of readings. These “offset errors” correspond to upsets in digital circuitry which caused subrange errors.  Curve A: Simple Device z One Upset Mechanism

 Curve B: Complex Device z Several Distinct Upset Mechanisms z Total Response = Sum of Each

Figure 44. SEU Effects in A/Ds are sum of sub-circuit responses.

Histogram of ADC Readings with DC Input While Being Irradiated with Heavy IONs Figure 45. SEU errors in ADCs can be grouped as Noise (Gaussian) and Offset (Digital).

IV - 42

Turflinger, et al [15] performed SEU testing on a hybrid A/D where it was possible to separately expose the gain-difference amp, A/D and DAC circuits. When they did this, they noticed that they detected both noise and offset errors for each portion of the system. Cross section plots of offset and noise for all three components is shown in Figure 46. Note the wide variation in capture cross sections for the noise and offset errors of the different components.

 GDA = Gain Difference Amp  ADC = Single Bit A/D Converter  DAC = Digital to Analog Converter

Figure 46. Noise and Offset cross sections for various portions of Hybrid ADC. To further illustrate some of the variations and complexities that can confound A/D testing, Figure 47 shows capture cross section plots for analog transients out of an operational amplifier [16]. In this test, pulses were detected and analyzed based on a variety of pulse heights. Note that there is a difference in cross section performance based on pulse height.

Transient Events for Op Amp

Variation of Cross Section vs LET and Pulse Height Discrimination

Figure 47. Typical SEU effects on op amps. IV - 43

Some A/Ds use over sampling and error correction to improve their performance. An illustrative example of a sub-ranging A/D using this principle is shown in Figure 48. Note that the 4 stages of A/D conversion generate a total of 15 bits, but that the part outputs 12. Typically this type of A/D performs a self calibration and stores correction information in its correction logic.

Figure 48. Functional diagram of sub-ranging ADC. With this architecture, it is possible to get an error in the correction logic that will persist until the part is re-calibrated. An example of this is shown in Figure 49 which is a sequential record of A/D errors (readings where the expected and read values differed by more than 15 counts) on an AD676 hybrid circuit [17]. In this test, only the digital die performing the correction was irradiated so only persistent offset errors were generated. The input was a sawtooth waveform that was being digitized rapidly, and the part would be re-calibrated after every 10 errors. Because of the rapid digitization rate and (relatively) slowly varying input voltage, every time there was a persistent error several samples would be taken resulting in the step-like appearance of the plot.

IV - 44

Irradiation of Digital Correction DIE in Hybrid AD676, Saw-Tooth Input, Re-Calibrate every 10 Errors Figure 49. Example of persistent errors in ADC with AutoCalibration. 4.13

Summary

Looking into the future, one may expect mixed signal microcircuits to become increasingly important. Not only will the analog elements in so-called digital microcircuits continue to be important parts of the design, but conventional mixed signal devices will be incorporated on the same die as digital circuits. The great majority of silicon technologies with feature sizes below 0.35 micron include provisions for dual polysilicon layers to permit the construction of high quality capacitors. They also permit silicide blocks for at least one polysilicon layer to permit poly resistors to be designed with resistivities around 100 ohms per square. Such technologies are directed toward implementation of SOC (system on a chip) designs which will include ADCs, DACs, general purpose microprocessors, and digital signal processors all on the same die. Interfaces among these functional blocks will most likely be inaccessible further complicating the test process. Radiation testing of such systems could well be impossible unless careful thought is given during the design process to testability or built-in self test features. Such devices will certainly represent a challenge to radiation effects test engineers.

IV - 45

REFERECNCES All references are unclassified. 1.

Nguyen, D., et al., "Total Ionizing Dose Effects on Flash Memories," Workshop Record 1998 IEEE Radiation Effects Data Workshop., Vol. 98TH8385, July 24, 1998, pp. 100-103.

2.

Telecommunications Industry Association, Electrical Characteristics of Low Voltage Differential Signaling (LVDS) Interface Circuits, TIA/EIA-644, March, 1996.

3.

Weste, Neil H. E. and Kamran Eshraghian, Principles of CMOS VLSI Design. Addison-Wesley Publishing Company, New York, N.Y. 1993, pp. 334-336.

4.

Williams, Michael K., “Analog ASIC - OMC212,” MRC/ABQ-R-1944.

5.

Johns, David A. and Ken Matin, Analog Integrated Circuit Design. John Wiley & Sons, Inc. New York, N.Y., 1997, pp. 357-364.

6.

Johns, David A. and Ken Matin, Analog Integrated Circuit Design. John Wiley & Sons, Inc. New York, N.Y., 1997, pp. 492-496.

7.

Johns, David A. and Ken Matin, Analog Integrated Circuit Design. John Wiley & Sons, Inc. New York, N.Y., 1997, pp. 474-478.

8.

Witczak, S.C. et al, “Moderated Degradation Enhancement of Lateral PNP Transistors Due to Measurement Bias,” IEEE Transactions on Nuclear Science, Vol. 45, No. 6, December 1998, pp. 2644-2648.

9.

Petersen, Edward, "Single Event Analysis and Prediction," 1997 IEEE Nuclear and Space Radiation Effects Conference Short Course, p. 15.

10.

Butcher, Daryl T., (briefing on) Radiation Hard Linear IC Design.

11.

1999 Annual Book of ASTM Standards, Volume 10.04 Electronics (I), F 1892-98 “Standard Guide for Ionizing Radiation (Total Dose) Effects Testing of Semiconductor Devices,” pp. 422-456.

12.

Johnston, Allen et al., IEEE Transactions on Nuclear Science, 1994, p. 2432.

13.

Black, J.D. et al., “Total Dose Evaluation of State-Of-The-Art Commercial Analog to Digital Converters for Space-Based Imaging Applications,” 1998 IEEE Radiation Effects Data Workshop Record, pp. 121-126.

14.

1999 Annual Book of ASTM Standards, Volume 10.04 Electronics (I), F 1192-95 “Standard Guide for the Measurement of Single Event Phenomena (SEP) Induced by Heavy ION Irradiation of Semiconductor Devices [Metric],” pp. 297-307.

IV - 46

15.

Turflinger, Thomas L. and Davey, Martin V., “Understanding Single Event Phenomena in Complex Analog and Digital Integrated Circuits,” IEEE Transactions on Nuclear Science, December 1990, Vol. 37, No. 6, pp. 1832-1838.

16.

Ecoffet, R. et al., “Observation of Heavy ION Induced Transients in Linear Circuits,” IEEE Nuclear Plasma Sciences Society, 1994 IEEE Radiation Effects Data Workshop Record, pp. 72-77.

17.

Bee, S. et al., “Heavy-ion Study of Single Event Effects in 12- and 16-Bit ADCs,” 1998 IEEE Radiation Effects Data Workshop Record, pp. 58-67.

18.

“Latchup in CMOS Technology,” Ronald R. Troutman, Kluwer Academic Publishers, 1996, p. 12.

19.

Tylka, A.J., J.H. Adams, Jr., P.R. Boberg, B. Brownstein, W.F. Dietrich, E.O. Fluockiger, E.L. Petersen, M.A. Shea, D.F. Smart, and E.C. Smith, “CREME96: A Revision of the Cosmic Ray Effects on Micro-Electronics Code,” 1997 IEEE NSREC.

20.

T. Girard, “Understanding Effective Bits,” Application Note AN95091, Signatec, 1995.

21.

X. Montagner, P. Fouillat, R. Briand, R.D. Schrimpf, A Touboul, K.F. Galloway, M.C. Calvet, P. Calvel, “Implementation of Total Dose Effects in the Bipolar Junction Transistor Gummel-Poon Model,” 1997 IEEE TNS, December 1997, Vol. 44, No. 6, pp. 1922-1929.

IV - 47

2000 IEEE NSREC SHORT COURSE

Section V

Radiation Testing and Characterization of Programmable Logic Devices (PLDs) Lee Hoffmann & R.C. DiBari Honeywell International Rich Katz/NASA-GSFC Lew Cohn/DTRA Approved for public release; distribution is unlimited

Section V Radiation Testing and Characterization of Programmable Logic Devices L. Hoffmann & R. Dibari (Honeywell International) R. Katz (NASA-GSFC) L. Cohn (DTRA)

1.0 2.0 3.0

4.0

5.0

6

INTRODUCTION .............................................................................................. V-1 BACKGROUND: PLD TECHNOLOGY: PAST, PRESENT AND FUTURE. V-3 RADIATION EFFECTS – GENERAL DISCUSSION ..................................... V-6 3.1 Steady-State Total Ionizing Dose Effects................................................. V-6 3.2 Transient Ionizing Radiation Effects (SEE) ............................................. V-7 3.3 Use of Unhardened Microelectronics in Radiation Environments........... V-9 THE IMPACT OF RADIATION EFFECTS ON PLD TECHNOLOGIES..... V-12 4.1 Introduction ............................................................................................ V-12 4.2 PLD Configuration Issues ...................................................................... V-12 4.3 PLD Circuit Operational Issues.............................................................. V-23 TEST AND EVALUATION ............................................................................ V-34 5.1 Ionizing Dose Testing............................................................................. V-34 5.2 Single Event Effects (SEE) Testing ....................................................... V-43 CONCLUSIONS AND FUTURE TRENDS.................................................... V-58 REFERENCE.................................................................................................... V-60 APPENDIX 1 (Table of Acronyms .................................................................. V-63 1.0 INTRODUCTION

The past decade has seen dramatic advances in Programmable Logic Device (PLD) technology where the state-of-the-art (SOTA) has migrated from simple logic arrays of ~ 1K in size, available in the early 1970s, to presently available devices that contain upwards of 500,000 + million usable gates and can operate at clock frequencies > 100 MHz. Moreover, these devices now are manufactured to contain large drop-in macro cells to further increase both their complexity and capability. While these advances have vastly increased the usefulness of this technology it has not been without a price, which includes the issues of circuit verification and validation. The challenge to be discussed in this short segment is: How can one implement a simultaneously cost-effective and comprehensive approach to quantitatively determine if a programmed chip will maintain it’s configuration and perform as expected in the harsh environment of space?

V-1

While this technology has made significant penetration in many commercial terrestrial areas, the use of PLDs for space applications has lagged considerably due to a variety of issues. These issues include, but are not limited to, a lack of radiation hardened and tolerant PLDs, the testing and characterization of complex PLDs and the potential for inadvertent reprogrammation of these devices due to radiation phenomenon such as Single Event Effects (SEE), and Total Ionizing Dose (TID). Additionally, as we migrate from application specific integrated circuit (ASIC) to system-on-achip (SOC) based solutions, to meet electronics equipment and system needs, the issues involved with programmable microelectronics will become more profound (e.g. user controlled or autonomous circuit/system reconfiguration, programmable substrates, and other variations on this theme) especially in the areas of verification, validation and radiation tolerance. A graphic depiction of the advances in PLD technology and the current state-of-the-art are shown in Figure 1. Thus, to address these issues this short-course will: •

Provide a summary of the various types of PLD technologies

•

Discuss the specific radiation sensitivities of the different device types

•

Investigate the issues associated with PLD test and characterization

•

Provide some recommendations with respect to the issues associated with testing and characterization.

A word on the contents of this course: Although a variety of programmable electronics devices exist, this course will focus on programmable logic devices (PLD) and Field Programmable Gate Array (FPGA) technologies. However, it should be noted that many of the concepts and test methods to be discussed apply to other programmable technologies such as EEPROM and Flash Memory. Finally, in this section the following definitions will be provided to facilitate the ensuing discussions of PLD technology. 8,000 Gate FPGAs:1990 → 1998

ProASIC 500k Family (ACTEL & Gatefield) Figure 1a. FPGA Evolution

Figure 1b. Programmable ASIC Top View

V-2

Programmable Logic. A logic circuit/device or element of a device whose configuration can be modified to implement different functions by changing the internal and/or external interconnect scheme either through the use of antifuses or transistor switches. This term shall include field programmable gate arrays (FPGA), programmable logic arrays (PLA), simple and complex logic devices (SPLD/CPLD), and other programmable types of semiconductor integrated circuits. Note that for the purposes of this short course the generic term Programmable Logic Device (PLD) will be used throughout the discussion to identify any or all of the following types of programmable logic technology. Programmable Logic Array (PLA). A programmable logic device that uses an AND/OR/ Invert architecture with <3 levels of antifuse connectivity. A subset of this type of PLD is the Programmable Logic Sequence (PLS) that contains only AND/OR planes. Programmed Array Logic (PAL). A PLD technology with a programmable AND plane and a fixed OR plane. Sometimes referred to as a Simple PLD (SPLD). Field Programmable Gate Array (FPGA). A PLD technology similar in configuration to a gate array provided with general purpose pre-fabricated metallization to support the inter- and intra- connection of logic blocks, macro cell, memory blocks, input/output blocks, etc. through either antifuse or semiconductor switches. Complex PLD (CPLD). A high density PLD generally based on PAL or SPLD architecture. The CPLD can contain logic, specific macro and storage blocks, similar to an FPGA, however this technology type is typified by a more regular and predictable routing scheme than an FPGA. 2.0 BACKGROUND: PLD TECHNOLOGY: PAST, PRESENT AND FUTURE Since the inception of PLD technology in the early 1970’s, as a 1K PLA by National Semiconductor Corp., this sector of the semiconductor market has seen relatively slow acceptance and growth until the early-1990s. However, the very rapid advances in semiconductor technology accomplished during the 1990’s has either significantly reduced or eliminated the three major issues that were affecting universal acceptance (PSI Magazine, October 1999, Focus Report: Programmable Logic) through improvements in the following areas: 1. Size. The largest FPGAs now hold up to a million system gates with product roadmaps indicating over two million gates by the year 2000. 2. Speed. The fastest devices are now operating at clock rates over 200 MHz (system clock). 3. Cost. As critical feature size continues to decrease mask and non-recurring engineering costs for ASICs have grown dramatically making state-of-the-art (SOTA) FPGA technology more affordable for higher volume purchases. Thus, a technology that was once relegated to the replacement of “glue” logic and prototyping has now been identified for a wide number of applications in the areas of signal and data processing, system control, operational system reconfiguration, etc. Indeed, the FPGA market is now anticipated to account for about $22.2M of the total $211.2M 2000 space electronics market and grow at a compound annual growth rate of ~18 percent (Johari 25 October 1999). Moreover, commercial, NASA, and DoD projections indicate significant increases in satellite payload, guidance and navigation (G&N), tracking, telemetry and command (TT&C), and command and data handling (C&DH) system requirements for PLD technologies of all types and performance capabilities. A summary of some of the space systems now using PLD technology is found in Table 1.

V-3

Table 1. Launch and Satellites Systems Using PLD Technology (Courtesy Actel) Launch Vehicles Atlas II SeaLaunch EELV Ariene V

Military Satellites Mighty Sat P81 P59 FSED Clementine SBIRS-High SBIRS-Low

Commercial Satellites Globalstar FAISat Intelsat IX GE-1,2,3,6,7&8 Echostar Telstar Orbcomm Orbview Superbird

Civilian and Scientific Satellites Deep Space I TIROS Mars Pathfinder Landsat VII Mars Surveyor EOS-AM1 Mars 98 EOS-PM1 Mars’01,’03,’05 EOS-CHEM1 Seawind Cassini SIRTF TDRS HIRDLS Space Shuttle Lunar Prospector HST GALEX GOES Genesis Mars Climate Orbiter

In addition to the above noted systems the following international missions also employ PLD technologies: EnviSat, Cluster II, METOP, Rosetta, Champollion, Stentor, ETS VII, MTSat, ACeS, N-Star, L-Star, SOHO, SILEX, Integral, Intnil, Space Station, ATV and Poseidon. Moreover, since PLD technology has now transitioned from mainly “breadboarding” to flight hardware applications these devices must now meet system radiation requirements. As an indication of the growing popularity of PLD technology an informal survey of the manufacturers of PLD technology early in 2000 revealed more than twelve competitors. A partial list of these manufacturers and the programming technologies used is shown in Table 2. Table 2. PLD Manufacturers Manufacturer

PLD Technology

Actel

Flash ONO antifuse M2M SRAM EEPROM SRAM EEPROM UVPROM LPGA Flash EEPROM UVPROM SRAM EEPROM α:Si antifuse SRAM Flash ONO antifuse

Altera Atmel

Chip Express Cypress Semiconductor

Dynachip Corp. Lattice Semiconductor Quicklogic Corp. Xilinx, Inc.

Lockheed Martin (RH version of the Actel A1020 & 1280 FPGA technologies) Pico Systems (Programmable substrates) α:Si antifuse UTMC α:Si antifuse Note: ONO refers to Oxide-Nitride-Oxide antifuse technology and α:Si refers to amorphous: silicon antifuse technology.

V-4

Finally, we conclude this section with a brief summary of the advantages and disadvantages of the basic types of programming technologies as shown in Table 3 below: Table 3. A Comparison of PLD Technologies [McCollum 99] Technology

Advantages

Disadvantages

EEPROM

Reprogrammable (slow) Non-volatile SEU immune 100 percent testable

High voltage required SEGR & TID sensitivity Limited number of P/E cycles Low density

SRAM (CSRAM)

Reprogrammable (fast) Unlimited number of P/E cycles 100 percent testable

SEU & TID sensitivity Volatile Medium density

Antifuse

Non-volatile High density Fast

Non-programmable SEDR sensitive Not 100 percent testable

Flash (similar to EEPROM)

Reprogrammable (slow) Low power Highest density SEU immune

SEGR & TID sensitivity High voltage required Limited number of P/E cycles

Note: The radiation sensitivity comparisons refer to configuration control and will be explained in more detail in Section 3 and P/E refers to the number of program and erase cycles the technology is capable of performing.

V-5

3.0 RADIATION EFFECTS – GENERAL DISCUSSION In this section a general discussion of the effects of radiation, both natural and man-made (e.g. nuclear weapon related, nuclear reactor environment, etc.), on semiconductor microelectronics with emphasis on metal oxide semiconductor (MOS) technology will be provided. A more detailed and comprehensive discussion of ionizing radiation effects on MOS semiconductor devices can be found in Messenger 86 and Dressendorfer 89. An abbreviated listing of these effects and the sources of radiation are summarized in Table 4. Table 4. Radiation Effects on Microelectronics [Courtesy Aerospace Corp.] Effect

Source

Circuit Impact

Total Ionizing Dose (TID)

Trapped electrons Trapped protons Solar flares Nuclear Weapons

Parametric shifts Gain degradation Leakage current Speed reduction

Single Event Effects (SEE)

Cosmic particles Trapped protons Solar flares

Single event upset (SEU) Single event latchup (SEL) Single event gate rupture (SEGR)

Prompt Dose

Nuclear Weapon

Rail span collapse Transient upset Latchup Transient burn-out

Neutrons

Nuclear Weapon

Gain degradation Leakage current

Note that for MOS technology (except in the area of Charge Controlled Devices) displacement damage is not an issue and will not be discussed. In the next section of this short course a more focused discussion of the impact of radiation on PLD technologies will be provided. The effects of ionizing radiation on MOS technology can be delineated into two general categories, which are: • Steady state ionizing radiation or as it sometime called – total ionizing dose (TID) radiation, that denotes the buildup of charge in the oxide and SiO2/Si interface regions of an MOS structure. • Transient radiation effects that can be further devolved into Single-Event-Effects (SEE) and Dose-Rate-Effects (DRE) sometimes known as prompt dose or photocurrent effects. These two phenomena will be discussed in more detail in a subsequent part of this section. 3.1 Steady-State Total Ionizing Dose Effects As indicated in Table 4 steady state radiation, or TID, can be caused by either the natural environment (e.g. electrons or protons trapped in the Van Allan Belts) or by x-rays and gamma rays, engendered by a nuclear weapon detonation, impinging on a semiconductor device.

V-6

The impact of TID on MOS technology can be summarized as follows: 1. Changes in the threshold voltage of a transistor due to the trapping of charge in the gate oxide region that can result in changes to the device operating point. More specifically, for an N-channel transistor the device will initially begin to turn-on (i.e. enter the depletion mode of operation) as positive charge builds-up and then eventually recover to some degree or turn-off (i.e. Super-recovery) as interface states begin to build-up and offset the effects of positive charge [Messenger 86 and Dressendorfer 89]. The first and most obvious impact of the operating point change is a rapid increase in static leakage current as the N-channel device turns-on, that can be followed by a decline or perhaps turn-off of the device depending on the magnitude of the interface state build-up. Concerning P-channel devices the effect of the trapped charge and subsequent interface state build-up is to turn-off the transistor (i.e. the device enters the accumulation mode) and decrease leakage current. The overall impact on some types of CMOS ICs is that there is an initial increase in static leakage current – as the N-channel transistors turn-on, followed by a decrease in current as the P-channel devices turn-off. However, as we will see in the next paragraph this it not always the case. Another effect of trapped charge in the gate region for P-channel devices used as pass-gate transistors is an apparent speed-up in operation. This is caused by the trapping of charge in the spacer region that serves to shorten the electrical length of the gate. This effect can result in signal timing or “race” problems. Finally, changes in bias set point can adversely impact other critical operating parameters such as throughput delay, low and high input voltage and current, etc. 2. Increased device and IC leakage current due to the turn-on of various parasitic paths within the IC/transistor structure caused by surface inversion (i.e. trapped charge changing a P-doped surface to a N-doped surface) [Messenger 86 and Dressendorfer 89]. These parasitic paths can lead to both inter- and intra-transistor leakage and cause a circuit to malfunction as shown in Figure 2. 3. A decrease in transistor saturation drive current (IDDsat) and an increase in throughput delay (tPD) as a result of decreased mobility due to ionizing radiation per the Sun-Plummer Relationship. In Section 4 the specific impact of total ionizing dose irradiation on the various types of PLD technology will be discussed in detail and examples provided. 3.2

Transient Ionizing Radiation Effects (SEE)

3.2.1 Single-Event-Effects Transient radiation effects that fall into the category of SEE are caused by the impact of either heavy ions or energetic protons and neutrons, that occur naturally in space or the atmosphere, on sensitive areas in microcircuits as shown in Figure 3. These particles deposit ionizing energy into the circuit that can cause either a “soft” or non-permanent error or, in some cases, permanent damage to the circuit. SEE can be engendered in two ways; (1) through direct ionization caused by the strike of a heavy ion or (2) through a nuclear reaction initiated by the strike of an energetic proton or neutron. SEE can manifest themselves in a variety of different ways as denoted and defined in Table 5.

V-7

Total Ionizing Dose Issues -- Evolution of Oxide Leakage -Each new semiconductor generation presents new research challenges Device Evolution

p + DRAW

• • • •

POLY-GATE GATE OXIDE - 200Å

p+ SOURCE

n+ SOURCE n+ DRAIN n-W

EL L

I IL1 L1

BORAN FIELD IMPLANT COVERS SIDE WELL

Smaller geometry/higher packing density Lower current to represent signal Less noise tolerance Increased interference between devices

++++ +++++ ++++ ++ ++ IIL3

L3 IL2 BORON FIELD I L2 IMPLANT

p SUBSTRATE

I L Gate I L Sidewall Gate Oxide Field Oxide

1980s 1 µm devices

IL kag e

I L Field Oxide

CVD OX Gate Poly 1

Back channel

Lea

IGate Edge IGate Oxide N-Channel Si

BURIED OXIDE

So u

Si Substrate

rce s

I L Buried Oxide

Gate Oxide Field Oxide Gate Edge Buried Oxide

Gate Oxide Sidewall Back Channel

Early 1990s

Late 1990s

0.7 µm devices

0.35 µ m devices

Figure 2. TID Engendered Current Leakage Paths in Several Types of CMOS Technologies [DTRA 99]

A “SINGLE” energetic heavy particle (proton, neutron, Fe, O, etc.) creates a dense localized plasma causing • • • •

Upset Latchup Burnout Gate rupture (actual hole in gate material)

As device dimensions shrink so does material thickness and the critical charge representing information; problem worsens with each new generation and required unique solutions Multiple Bit Upsets

Single Event History Satellite Package NMOS SRAM digital Transuranics Rams Latchup flip-flop DRAMS

Gate Ruptures

Neutron Upsets Linear Circuits Enhance in Combined Microprocessor Neutron Upsets in Environment? Errors Optocouplers Avionics

?

none 1975

1980

1985

1990

1995

Figure 3. Single Event Effects [DTRA 99]

V-8

2000

Table 5. Single Event Effects Failure Modes Acronym

Definition

Description

SEU

Single Event Upset

Change of stored information

SED

Single Event Disturb

Momentary disturb of information stored in memory bit

SET

Single Event Transient

Current transient induced by passage of a particle, can propagate to cause output error in combinational logic

SEDR

Single Event Dielectric Rupture

Essentially antifuse rupture

SEGR

Single Event Gate Rupture

Rupture of gate dielectric caused by a high current flow

SEL

Single Event Latchup

High current regenerative state induced in 4-layer device (latchup)

SES

Single Event Snapback

High current regenerative state induced in NMOS device(snapback)

MBU

Multiple Bit Upset

Several memory bits upset by passage of the same particle

SEFI

Single Event Functional Interrupt

Corruption of control path by an upset

Concerning PLD technology the most important SEEs are SEU, SEDR, SEFI, SEL and SEGR. The significance of each of these effects with respect to PLD technology will be discussed in the Section 4 of this short course. 3.2.2 Dose-Rate-Effects (DRE) As previously discussed, DRE are the result of a nuclear weapon detonation and can cause a circuit to either upset, latch-up in a non-destructive mode, or be destroyed. The mechanism has to do with the generation of photocurrents within a device as a result of the weapons created x-rays and gamma rays that impinge on the circuit as shown Figure 4. The specific impact of dose-rate on PLD technology will not be discussed in this short course except to note that the upset response caused by the photocurrents can be thought of as a “global” SEU event. 3.3 Use of Unhardened Microelectronics in Radiation Environments As a general disclaimer the use of unhardened microelectronics and especially those high volume products that are sold without any support from the manufacturer, e.g. product technical change notification, engenders significant risk regardless of the testing and qualification efforts provided by a user. The risk encountered, as a result of the use of this class of devices, is due to the fact that one can never be absolutely sure that the flight hardware is the same, with respect to radiation performance, as those devices previously tested and qualified. This problem is caused by various commercial manufacturing practices that include: Specific manufacturing steps or material properties, that effect radiation response, are often not placed under stringent statistical process control since they due not impact electrical or reliability performance. This practice can result in significant lot-to-lot or possibly die-to-die radiation performance variations.

V-9

Commonly called γ (gamma dot) [rad/s]

•

– High energy x-rays (keV) and gamma rays generate enormous amounts of free electrons and holes in a homogenous and isotropic manner • Transient photocurrents flow across junctions – Orders of magnitude larger than normal signal

V SS p+

V DD n+

p+

n+

+ - +- +- +- ++ +- +- +- - +

++ - + +-+-++-+-+-++-+-+ n- epi +-+-++-+p-well

p+

n+

+-+-++-++ - + - + ++ +- +-- + -+-++-+-+ +-+-++-+-

= photocurrent flows

n-substrate

– Results • Upset (Temporary loss of information) • Latchup (Locked in high current mode) • Burnout (Destruction by Joule heating)

Figure 4a. Dose Rate Upset Mechanisms

• Voltage across each memory cell is known as "railspan" voltage • "Railspan" across each cell must be sufficient to retain data Photocurrent (mA)

0.6

0.4

0.2

0 0

0.2

0.4

0.6

0.8

Time (µs)

Photocurrent Spike

Pre-Irradiation

Memory “Model” Memory During Irradiation

Figure 4b. Dose Rate Upset Failure in a Memory

V-10

Data Loss

Changes are frequently made to process, design and layout to enhance electrical performance or reduce manufacturing cost without any consideration of the impact on radiation response, e.g. die shrink, epitaxial layer thickness change including removal, etc. Moreover, these changes are typically not reported, i.e. no change control. This practice can compromise or possibly invalidate prior test results. Similar device types, with same date code can, and often do, use die from more then one fabrication facility. This practice can totally invalidate any prior test results. Computer aided design rules are established to enhance parameters such as speed, power and die utilization without any consideration of the impact on radiation performance. This topic will be discussed in more detail in Section 4.2.3.1. Graphic example of the impact of these commercial practices can be found in Section 4.3.1, Figures 21 and 23. In Figure 21 the TID response for an Actel 1020, manufactured using three different processes, is shown. The significant difference in device response between 1.0 micron and the other two processes should be noted. In Figure 23, the TID responses for four different devices manufactured at several foundries are provided. The difference in response for the two 3.3 V, 0.35 micron devices produced by two different fabrication facilities should be noted. Moreover, other examples concerning SEE response exist. However, it should suffice to say that the use of unhardened microelectronics that are not subject to manufacturer support can engender significant problems when adopted for use in applications that involve operating in a radiation environment.

V-11

4.0 THE IMPACT OF RADIATION EFFECTS ON PLD TECHNOLOGIES 4.1

Introduction

For the purposes of the following discussion the impact of radiation on PLD technology will be delineated into two categories that are: 1. The effect on PLD configuration, either permanent or temporary, including TID, SEU, SET, SEGR, SEDR and SEFI will be discussed in Section 4.2. 2. The impact on PLD operation concerning parametric response (e.g. leakage current, VIN, VOUT, etc. specifications) and the functionality of the logic and storage elements of the device including TID, SEU, SED, SET, SEL, SEDR and SEFI will be discussed in Section 4.3. Concerning the test and characterization issues associated with device functionality and parametric response we can borrow, to a significant degree, on the standard radiation test methods previously developed for ASICs. However, for configuration issues no standard test methods or comprehensive body of knowledge exists. 4.2

PLD Configuration Issues

4.2.1 Total Ionizing Dose Effects The primary impact of TID on PLD configuration is associated with EEPROM reprogrammable FPGA technology (e.g. Actel ProASIC) wherein the effect of total ionizing dose irradiation is to reduce the stored charge in the floating gate. Thus, if the charge is reduced below some critical level the configuration will be lost. Other PLD technologies that also can have TID induced configuration issues are those types of devices that are SRAM programmable (e.g. Altera EPF10K100A and E series, and Xilinx XC4062XL series). Note that SRAM programmable technology will be hereafter referred to as CSRAM. Concerning this technology the impact of TID would be to render the program SRAM cells inoperative and thus compromise the configuration of the device. In Figure 5 a graphic depiction of the effect of TID on either EEPROM or Flash memory technology and in Figure 6 (left panel) the potential impacts on the configuration of the associated programmable device are shown. As can be seen from the two figures the effect of TID on these memory technologies is to reduce stored charge, which in-turn, will affect the state of the associated switches (e.g. L0, L1, etc.) that are responsible for the configuration of the PLD. One strategy to mitigate this effect would be to periodically refresh the memory to restore the stored charge and maintain the voltage at the appropriate level for proper switch operation. A schematic diagram of an EEPROM controlled switch is shown in Figure 7. Note that for programming and to ensure the ability to pass a full magnitude signal a voltage greater than VDD must be generated and applied to the storage cell and the gate transistor. Note the use of high voltage also makes this technology sensitive to SEGR. This issue will be discussed in more detail in Section 4.2.2.4.

V-12

Ionizing radiation will discharge the floating gate.. TID performance the worst case of CMOS and FLASH effects.

2.5 V in

out

N+

N+ P-Well 0V

0.25um CMOS >50krad(Si)

On State FLASH effects include Discharge, Vt shift and RiLC

2.5 V in

Discharge effect can be counteracted by periodically refreshing the device.

?

N+

N+ P-Well 0V

Off State

Figure 5. Impact of TID on EEPROM PLD Technology. Loss of charge may affect configuration. [Speers 99]

Transients in the switch elements can cause false clock pulses or assertions of asynchronous resets. SRAM Switch FLASH Switch

0

0

1

1

Q CSRAM

QB

I5 (X2) Pin 3 CLK

I5 (X2) Pin 3 CLK

L4 L5

Q

L2 L0 CSRAM

I2 (X1) Pin 2 Set/Reset

I2 (X1) Pin 2 Set/Reset

QB

Figure 6. Comparison of Impact on EEPROM and CSRAM PLD Technologies. No upset with EEPROM, however CSRAM upset can affect operation through the inadvertent make/break of clock, signal, or control line. [Speers 99]

V-13

The FLASH Switch The switch is comprised of two transistors which share a gate. PRG/SEN

SWITCH WORD LINE

Additional source/drain implant in PRG/SEN device enables tunneling to overlapping floating gate. ONO inter-poly dielectric

SEL 1 SEL 2

Figure 7a. Flash Switch Schematic [Speers 99] Program and Erase Program and erase are accomplished with F-N Tunneling where the floating gate overlaps the source and drain. 5V 5V Gate voltage is ramped to minimize field.

-11.5V 0V 0V

16.5

Figure 7b. Flash Switch Program/Erase Operation [Speers 99] Operation During operation, the floating gate is effectively 1.5x VCC in the on state and slightly negative in the off state. 1.25V 1.25V

1.5xVCC

2.5V

1.25V 1.25V

<0V

2.5V

Figure 7c. Flash Switch Normal Operation [Speers 99]

V-14

4.2.2 Single Event Effects on PLD Configuration In this section the impact of Single-Event-Effects, including Single-Event-Upset, SingleEvent-Disturb, Single-Event-Transients, Single-Event-Gate-Rupture, Single-Event-DielectricRupture and Single-Event-Functional-Interrupt on PLD configuration will be discussed. 4.2.2.1 Single Event Upset (SEU) Effects The impact of SEU on PLD technology is most apparent in SRAM or CSRAM controlled devices wherein the state of a CSRAM storage cell can be flipped (i.e. “1” state changed to a “0” state) and result in the inadvertent reprogramming (i.e. change in configuration) of a PLD. The operating scenario associated with a CSRAM SEU fault can be envisioned as follows: •

In Figure 8 a typical unhardened CSRAM memory cell is shown. For the purposes of this discussion the n-channel transistor controlling the state of the configuration switch is assumed to be OFF (switch ON). If a heavy ion or an energetic proton or neutron strikes the drain of the OFF n-channel transistor and imparts sufficient ionizing energy (either through direct ionization or a nuclear reaction) the OFF transistor will switch ON and the switch will be deenergized. Thus, the configuration of the PLD will be altered (See Figure 6 right panel).

Some additional issues associated with the above-described event are described in the following two paragraphs: •

Bus Contention: Figure 9 depicts a bus contention situation caused by an SEU where inadvertent reprogramming of the lower switch has “connected” the upper and lower busses such that one bus is now attempting to “drive” the other circuit. Such a fault mode can result in the maloperation and possibly the destruction of a device

•

Data or signal path modification: Figure 10 depicts the situation where a SEU has caused a switch controlling a data/signal path to change state and “open” the path. Alternatively a path may be inadvertently “closed”. In either situation the PLD will not perform as anticipated. CSRAM

VCC

GND Write Port VCC Switch

GND

Figure 8. A CSRAM Control Circuit and Switch [Wang 99]

V-15

CSRAM 1

VCC

Switch 1 Normally On A

‘High” Node

Contention Current

GND

OUT

CSRAM 2

VCC

B

Switch 2 Normally Off Upset to ON

‘Low” Node GND

Figure 9. An SEU/SED Caused Bus Contention Fault [Wang 99] Heavy Ion Hit Hitting Leading Edge

CSRAM

IN

Hitting Trailing Edge

OUT

(a)

Voltage (V) 2

NODE Q

1 0 2

IN

1 0 Narrowed

Widened

2

OUT

1 0 0

5n

10n

15n

20n

25n

30n

(b)

Figure 10. SEU Data Path Interruption Fault [Wang 99]

V-16

35n

Figure11 depicts an alternative data/signal path fault where an SEU caused a CSRAM controlled gate to reprogram such that the data stream now emanates from a different source than expected.

Heavy Ion Hit QB

These types of faults can, in addition to causing device maloperation, lead to the overstress (lifetime shortening) of a PLD and/or external failures to associated components or the system itself. It has been demonstrated that loading an incorrect configuration into a Xilinx SRAM based PLD can result in the destruction of that device and this may also apply to other manufacturers.

Q Normally ‘High’ Q

(a)

Other types of CSRAM PLD SEU issues can include: 1) bus contention fights on internal tri-state busses leading to overstress and possibly device destruction, 2) the isolation of pull-up resistors on tri-state busses that can result in floating inputs and oscillations, 3) changes in output slew-rate leading to system timing errors, 4) changes in input delays resulting in timing failures or metastable states and lastly, but not finally, 5) the switch of an input to an output node.

CSRAM

A ‘Low’ QB

OUT

QB A ‘High’ Q Voltage (V)

(b)

2 1 0

NODE Q

2 1 0

NODE QB

2 1 0

OUT

0

2n

4n

6n

8n

10n

Figure 11. SEU Data Path Fault Stream [Wang 99]

Several other factors worthy of mention concerning SEU in a CSRAM PLD technology include: • •

•

The number of CSRAM memory cells required for programming and device architecture preclude the use of error detection and correction (EDAC). The process technologies used to support the fabrication of the PLD to-date, do not have the capability to implement a cross-coupled polysilicon resistor in the cell and thus, cannot avail themselves of this type of radiation hardening technique. In addition the number of cells make it difficult to implement any type of transistor redundancy design to mitigate SEU. However as feature size continues to shrink this mode of hardening against SEU may become more prevalent in order to combat terrestrial or atmospheric SEU effects.

An SEU configuration cross-section per bit vs. LET plot is shown in Figure 12. In addition a SEU configuration response signature (Icc vs. Fluence) is shown in the right panel of Figure 18. In closing it should be noted that SEU events should not affect the configuration of EEPROM PLD technology. 4.2.2.2

Single Event Disturb (SED) Effects

An SED may be thought of as a transient SEU event, however for this situation a latch type circuit momentarily changes state (i.e. “1” transitions to a “0” state, stays there momentarily, and then recovers back to the original “1” state) but then recovers to its original state. An event such as this can occur if the transient pulse caused by the heavy ion strike produces a pulse of sufficient magnitude to cause a state change but is of inadequate width to sustain the change until the circuit reaches steady state operation. This is sometimes referred to as a metastable state of operation. Examples of SED faults in CSRAM PLD technology are shown in Figures 9 through 11 and the associated waveforms.

V-17

Clay-31 Configuration Error Cross Section per bit (8kB total)

Cross section per bit in cm2

1.00E-06

1.00E-07

1.00E-08

No Configuration Errors Observed

1.00E-09

Configuration Errors per bit 1.00E-10

1.00E-11 0

10

20

30

40

50

60

70

80

90

100

LET in MeV*cm2/mg

Figure 12. SEU Cross Section Detected by Function Failure (Curve serves as a lower bound) [Katz 97]

In Figure 9 an SED engendered bus contention fault is shown. The description of this situation follows that provided in Section 4.2.2.2 Single Event Upset Effects, with the exception that the propensity to device damage will be a function of the metastable recovery time. Figure10 depicts two manifestations of an SED fault: •

In the first situation a strike corresponding in time to the leading edge of the incoming data pulse causes a momentary turn-off of the switch to occur which results in a shortening of the information pulse. The consequence of such an event is somewhat dependent on the specific circuit application. However, if a pulse is drastically reduced in width it may fail to propagate through subsequent stages of the circuit and thus, result in the maloperation of a system.

•

In the second situation the strike occurs coincident with the trailing edge of the data pulse and results in an elongation of the pulse. Here again, the impact of such a fault will depend on the circuit’s mode of operation, however one can envision a situation where a “race” condition could be engendered and result in system maloperation.

A somewhat similar situation is shown in Figure 11, where the data stream is corrupted by the transient reprogramming of the data path, from Input A to Input B, by a strike to the CSRAM controlling the multiplexor. The metastable or disturb time period will depend on the design of the CSRAM cell, however it has been demonstrated that memory cells designed to be less sensitive to SEU events (e.g. use of cross-coupled resistors in the latch portion of the circuit) tend to have longer metastable recovery times. As previously stated the impact of an SED on device or system operation will depend on items ranging from the PLD design and configuration, the location of the strike in the PLD to the actual application of the device in the system. However, it should suffice to say that the use of CSRAM controlled PLD technology sensitive to SEU/SED poses an awesome radiation test and evaluation challenge.

V-18

Single Event Transients (SET) Effects

A SET fault occurs when a heavy ion or energetic particle strike, in a combinational circuit, produces a spurious pulse of sufficient amplitude and width that it propagates in a manner that results in the maloperation of the device or system (See Figure 13). The actual fault may occur in several ways including (1) the pulse propagates until it is 'latched” and then assumes the identity of a “valid” control signal or (2) the pulse strike impacts the clock bus and is interpreted as another clock pulse (See Figure 14) causing the propagation of invalid data.

Critical Transient Width vs Feature Size for Unattenuated Propogation Heavy Ion Hit

Data In

Transient Pulse Passed One Buffer

1000

Critical Transient Width (ps)

4.2.2.3

100

10

1 1000

100

10

Feature Size (nm)

Figure 13. Single Event Transient in Combinational Logic [Wang 99]

The actual sensitivity of modern microelectronic circuits to SET has been studied by Mavis [Mavis 98] and others. Figure 15 provides the results of a study concerning the sensitivity of CMOS technology to SET as a function of critical feature size. As can be seen from the figure as feature size shrinks (i.e. circuit speed increases) devices will become increasingly sensitize to this effect. Concerning the impact of a SET fault on PLD configuration, if the spurious pulse becomes latched in a configuration control circuit the effects can range from benign to catastrophic. However, since in most PLD designs the logic circuitry associated with the configuration control is deactivated post-programming the impact of SET in this area should not be a major problem for some designs. However, for the situation where an SET can result in an inadvertent reset (e.g. UT22VP10 for a high LET strike) or otherwise reprogram the device this failure mode can be of significant concern. The impact of SET on circuit operation will be discussed in Section 4.3. 4.2.2.4

Single Event Gate Rupture (SEGR) Effects

SEGR is an issue associated with EEPROM and Flash reprogrammable PLD technologies since high voltage must be applied to support write and erase cycles. It has been demonstrated that a heavy ion strike during the application of high voltage can result in the rupture of the gate oxide and destruction of the circuit [Wrobel 87 and Titus 98]. An SEGR scenario is depicted in Figure 16 and the potential impact on device configuration shown in Figure 6, left panel. As can be seen from Figure 6, a SEGR of the EEPROM cell associated with configuration control switches (e.g. L0, etc.) will result in a loss of control voltage and the subsequent failure of the switch. The ramifications of such an event are discussed in Section 4.2.2.2 SEU Effects and will not be repeated in this section. However, it should be noted that SEGR is a permanent failure mode unlike SEU. Double Clocking As a Result of Heavy Ion Induced Pulse 1

output

Heavy ion induced negative pulse

1

Output waveform Cartoon of clock/logic upset. The device is most sensitive during the transition.

Figure 14a. SET Fault Mode (Courtesy Aerospace)

Figure 14b. SET Double Clock Fault [Wang 99]

V-19

Figure 15. SET Pulse Propagation Parameters. Critical Transient Width vs. Feature Size for Unattenuated Propagation [Mavis 98]

-11.5 V 5V

5V

N+

N+ P-Well 16.5 V

0V

0V

N+

N+

Configuration required conventional application Peripheral voltage circuits susceptible to up.

P-Well

Figure 16. SEGR During Configuration – High voltage across thin gate oxide may impact configurable payloads [Speers 99]

A Flash memory switch schematic and circuit configuration along with the applied voltages for the various modes of operation are shown in Figures 7a-7c. Note the “High Voltage” applied to the gate results in a susceptibility to SEGR. However, since the application of high voltage is anticipated to occur infrequently and, in addition, must happen in conjunction with the strike of a very energetic heavy ion, this type of failure mode is not considered to be a major problem for some applications. 4.2.2.5 Single Event Dielectric Rupture (SEDR) Effects SEDR is a fault mode that occurs when the passage of a heavy ion through an antifuse results in the rupture of that circuit element. The sensitivity of an antifuse to this failure mode will depend on the electric field across the element and the composition (stacked layers) that comprise the structure. V-20

Two major types of antifuse technology exist that are: •

Oxide-Nitride-Oxide (ONO) based elements used by Actel and others as depicted in Figure 17 (left panel). It has been demonstrated that ONO technology can be susceptible to SEDR as shown in Figure 18 (left panel). It should be noted that this type of failure has only been noted to occur during the application of high voltage, resulting in a significant electric field across the antifuse structure (e.g. > 6 MV/cm), and a high LET ion strike (e.g. > 37 MeV-cm2/mg).

•

Amorphous Silicon (α:Si) based elements used by Actel, UTMC, QuickLogic and others as depicted in Figure 17 (right panel) have also been demonstrated to be susceptible to SEDR failure. However, α:Si elements are generally thicker than ONO antifuses and thus, should be less sensitive to SEDR.

Based on the available data the issue of SEDR is not considered to be of dramatic importance for certain applications. However, screening and characterization for new technologies is appropriate and lot sampling for radiation hardness assurance is considered prudent. 4.2.2.6

Single Event Functional Interrupt (SEFI)

This fault mode refers to the situation where an upset, caused by a heavy ion or energetic particle strike, results in the transfer of a device from normal operation into some other mode of operation, e.g. test, shutdown, etc. SEFI faults are not unique to PLD technology and to date have been recorded in devices such as EEPROMs [Katz 98] and DRAMS. However, due to the complex nature of PLD a discussion of this fault type in included for completeness in this short course. A dramatic demonstration of this fault mode is provided in Katz 98 where an SEU event associated with the built in test circuit (IEEE JTAG 1149.1) of an RH54SX16 PLD, results in a configuration error and a massive current draw as shown in Figure 19. Specifically, it was determined that an SEU event caused the JTAG TAP controller (Figure 20) to malfunction and load the indeterminate contents of the instruction update register into the actual instruction register, resulting in a loss of configuration. As previously stated such a failure could either damage the device and/or the associated external circuit. Polysilicon

ONO

Metal - 3 Top Electrode Amorphous Silicon

N++

FOX

Dielectric Metal - 2 Bottom Electrode

thermal oxide CVD nitride thermal oxide

ONO Antifuse

Poly/ONO/N++ Heavy as doped Poly/N++ Thickness controlled by CVD nitride Programs ~ 18V Typical Toxono ~ 85 Å RH1280 Toxono = 99 Å R = 200 - 500 ohms

TD Amorphous Silicon Antifuse

‘Pancake’ Stack Between Metal 2 and 3 Designed for 3.6V Operation in Sea Of Gates FPGA ‘Logic’ Devices Program at ~ 10V ‘Substrate’ Devices Program at ~ 30V Thickness ~ 500 - 1000 Å R = 20 - 100 ohms

Figure 17. Antifuse Technology [Katz 99a]

V-21

RH1280 S/N 063 Antifuse Rupture

CLAy-31 Heavy Ion Performance

45

Icc (mA)

40

35

30 20x10 6

0

40x10 6

60x10 6

80x10 6

100x10 6

Fluence (ions/cm 2 )

V CC = 4.7 VDC; LET = 53; Angle = 0 Degrees

SRAM

ONO Antifuse

Figure 18. SEDR vs. SEU Configuration Response Signature Configuration Errors in Routing Network [Katz 97 and 99a]

JTAG Upset Effect - Step Load TCK and TMS=1 Not Guaranteed Solution 700

Large Step Load

ICC (mA)

600 500

Brand X SEE Test BNL 02/98 NASA/GSFC BB Pattern/ 2 µm Epi X1B3 Bromine

400 300 200 100 0

0

5

10

15

20

Time (Sec) Figure 19. PLD SEFI Response Signature [Katz 98]

V-22

25

IEEE JTAG 1149.1 TCK

Shift Register is undefined in TestLOGIC-RESET State

TAP Controller (State Machine) Shift CLK

TDI

Shift Register

TDO

Parallel Latch

Chip Control

Reset Latch

Figure 20. SEFI Example – Built-In Self Test (BIST) Circuit Turn-on [Katz 98]

The test and characterization of a device for SEFI faults engenders a number of problems including: •

The need to develop test methods to accurately identify and differentiate the occurrence of SEFI faults from other types of faults that can engender the same failure signature, e.g. SEL.

•

Characterization of proton induced SEFI faults without engendering other types of failures such as TID due to high fluence requirements because of the generally low cross-sections involved.

Additionally, SEFI-type faults can also impact the operation of PLD technology without affecting the configuration and thus, this failure mode must also be considered. However, this type of failure will not be discussed separately in section 4.3, where operational vice configuration failure modes are discussed. 4.3.

PLD Circuit Operational Issues

4.3.1 Total Ionizing Dose In general the most significant TID issues associated with PLD technology are similar to those identified with ASICs and include changes (increases) in static leakage current (ICC), increased propagation delay and changes to VIH and VOL. This type of degradation has been recorded for PLDs from various manufacturers [Katz 95]. Figures 21 and 22 depict the impact of TID on several generations of Actel A1020 FPGAs. It should be noted that large lot-to-lot and within lot variations in response were identified during this testing. The salient points demonstrated by these data are: •

The effect of feature size on device leakage.

•

A significant increase in propagation delay is shown during the anneal (post-irradiation) period. This response suggests that the technology has a strong “rebound” response and thus may not be suitable for low dose-rate (e.g. ~ 10 mrd/s or less) space applications where one could expect a large anneal response to occur.

V-23

300 A1020 Dose Rate: 10 rad(Si)/sec)

250

ICC (mA)

1.0µm 200 150 100

2.0µm 1.2µm

50 0 0

20

40

60

80

100

Dose, krad(Si) Figure 21. Impact of Feature Size on TID Sensitivity [Swift and Katz 99]

Pre- and Post Anneal

Propagation Delay (ns)

90

A1020 10 rad (Si)/s 80 Control device 70

60

Pre Rad

10

-10

0

Dose (krad(Si))

100

101

Time (hr)

Figure 22. Propagation Delay vs. TID – Pre and Post-anneal [Katz 95]

V-24

102

Figure 23, shows the TID response of several types of FPGA technologies available from three manufacturers. Although not shown in Figure 23, the basic trend is that as scaling increases (i.e. feature size and voltage decrease) the TID response tends to improve. Concerning ONO antifuse technology devices (e.g. Actel 1020, etc.) at least two other significant TID issues must be addressed that are (1) parasitic leakage currents in the ONO fuse array and (2) charge pump degradation. It should be noted that the thicker oxides associated with the high voltage programming circuits can also result in higher leakage currents. Concerning charge pump circuit degradation the following applies: •

Charge pump (CP) circuits are used on Actel and other types of PLD technologies that use “pass gate” FETs for isolation as shown in Figure 24. The increased voltage (VPUMP) produced by the CP (i.e.VCC + VDS) allows the isolation transistor to pass a robust signal ~ VCC. If the CP is degraded and VPUMP is reduced then the magnitude of the voltage signal to the inverter buffer circuit will not be sufficient to bias the inverter circuit totem pole configuration off. Thus, the inverter will constantly draw current. Moreover, since 1000’s of these inverter circuits exist within a device such as an Actel A1020 the increase in operating current will be significant.

•

Another significant issue with serious test and characterization implications has to do with the turn-on response of PLD technology using CP circuits. Specifically, it has been noted that the effect of TID accumulation is to delay the operation of the CP. Thus, the start-up of the PLD will be delayed and significant current drawn until the CP circuit begins to function normally. The start-up response of an Actel A1020B is shown in Figure 25 and, as can be seen, the device took over 35 ms to begin normal operation with 12 krad(SiO2) applied. Other samples from the same lot, tested by Katz and co-workers, took as long as 150 ms to reach normal operation. The negative implication of drawing large magnitudes of current for extended periods of time or significant start-up delay for applications such as satellite or missiles is obvious. Moreover, this type of response makes it mandatory to accomplish parametric and functional testing immediately after irradiation or insitu, which is often not the case for device test and characterization. Submicron FPGA TID Tolerance 0.35 µm to 0.6 µm

50

RT54SX16 Proto, 0.6 µm, 3.3V, MEC A54SX16 Proto, 0.35 µm, 3.3V, CSM A42MX09, 0.45 µm, 5.0V, CSM QL3025, 0.35 µm, 3.3V, TSMC

40

XQR4000XL Proto, 0.35 µm, 3.3V, 60 kRads (Si) RH54SX16 Proto, 0.6 µm, 3.3V, > 200 kRads (Si)

30

A42MX09 QL3025 A54SX16

20

RT54SX16

10

0 10

20

30

40

50

60

70

80

90

kRads (Si)

Figure 23. Submicron FPGA TID Performance [Katz 99a]

V-25

100

Isolation Device Signal

Charge Pump Input Buffer Figure 24. Charge Pump (CP) and Isolation FETs Circuit [Wang and Katz 95]

1200 A1280A 10 rad(Si)/s (S/N 265)

1000 200

ICC (mA)

800

Current drops abruptly when charge pump reaches equilibrium

150 100

600

50 400

4 krad (Si)

0 12 krad (Si)

200 4 krad (Si) 0 -200 0

5

10

15

20

25

30

35

40

45

50

Time After Power Supply Voltage is Applied (ms) Figure 25. Impact of TID on Start-up Current for a FPGA with a Charge Pump [Katz 95]

4.3.2 SEE As previously stated the impact of SEE on PLD device operation and functionality, with the exception of SEDR, can be viewed and dealt with in a manner similar to that of ASIC technology. Clearly, all of the issues associated with modern microelectronics technology that result in increased sensitivity to SEE effects (e.g. SEU, SET, SEFI, MBU, etc.) manifest themselves in PLDs. In the following subsections of Section 4.3.2 the impact of SEU/MBU, SED, SET, SEL, and SEFI on PLD operation will be discussed.

V-26

4.3.2.1 Single Event Upset and Multiple Bit Upset Event (SEU and MBU) Effects Both SEU and MBU faults have been demonstrated in various PLD technologies and probably constitute one of the primary challenges to the use of these devices in a space environment, second only to SEU configuration fault issues. The basic approach to the test, evaluation and characterization of this phenomenon is similar to the methods employed for non-programmable ASIC and memory technologies. Recent SEU test results, from a variety of sources, indicate that many of the devices identified for space applications are very sensitive to upset and can even be upset by energetic protons and neutrons. (See Table 6). Table 6. Proton Test Results (at 195 MeV) Summary [Katz 98] Est. X-Sec (cm2 / flip-flop)

Device Type

Size/Voltage (Nominal core)

A1280A

1.0 µm / 5.0

RH1020

1.0 µm / 5.0

RH1280

0.8 µm / 5.0

~ 400 x 10

S-Module

QYH500

0.8 µm / 3.3

< 0.5 x 10-15

No upsets detected

RT54SX16

0.6 µm / 3.3

~ 6 x 10

QL3025

0.35 µm / 3.3

< 4 x 10-15

A54SX16

0.35 µm / 3.3

-15

JT22VP10

? µm / 5.0

-15

~ 137 x 10

< 2 x 10-15 -15

~ 3 x 10

Comments 19 Parts Tested No upsets detected

-15

~ 2 x 10-11

No upsets detected

Cypress die

An illustration of an SEU operational fault in a PLD technology is shown in Figure 26. For this situation the upset response of an Actel Al020 is shown. The figure in the right panel depicts the error counts at three locations in the circuit: 1. At the output of a hardened register chain – DOH 2. At the output of an unhardened register chain – DOS 3. At the output of a hardened triple-modular-redundant (TMR) voting circuit – TMR. There are several items to be noted from the figure that include: •

The error count at the output of the unhardened register chain increases linearly with time, indicating the susceptibility of this PLD to SEU and or MBU. In addition there are jumps in the output of the unhardened register chain output caused by the SEU of unhardened memory devices (flip-flops) in the clock circuit.

•

The error count at the output of the hardened register chain is flat, since the registers are hardened. However, errors in the output occur that are caused by the upset of these same memory devices in the clock circuit

•

The error count at the output of the TMR circuit increases linearly with time as a function of the upset rate of the flip-flop element.

The right panel of Figure 26 depicts the differential error count and permits the clock circuit upsets to be identified. This step is necessary to allow one to accurately assess what portion of the PLD (or ASIC) is responsible for the SEU rate. Based on this a cross-section vs. LET description of the clock circuit can be obtained as shown in Figure 27.

V-27

TAMU Run 14 S/N 3740 1 MHz/Checkerboard LET = 34.4

250 18

DOS DOH TMR Mon

16

Errors per Event

Error Counters

200

150

100

14 DOS’ 12

DOH’ Mon’

10 8 6 4

50

2 0

0 0

3

100x10

3

3

200x10

300x10

3

400x10

20

3

500x10

40

60

80

100

120

140

Error Event Number

Sample Number

Cumulative Error Counts

Differential Error Counts

Figure 26. SEU Clock Upset Example [Katz 98]

1.00E-06 P6-1

2

Cross Section (cm /device)

1.00E-05

P6-2

1.00E-07

P6-3 P6-4

1.00E-08

P6-5 Weibull Fit 1.00E-09

1.00E-10 10

20

30

40

50

60

70

80

2

LET (MeV-cm /mg)

Figure 27. SEU Clock Upset Example – Upset Cross Section [Katz 98]

A mitigating factor is that as feature size continues to decrease the level of integration density will permit significantly more functionality to be placed on each chip. Thus, SEU/MBU mitigation methods such as TMR, EDAC and multiple transistor latch and memory cell designs (insensitive to SEU) can be implemented with reduced area and power penalties. However, the need to perform comprehensive testing on complex designs as a function of clock rate and input vector sets will continue to be an increasingly vexing issue. Moreover, the addition of redundant structures to mitigate SEU will make testing and verification more difficult.

V-28

A factor not directly related to SEU but of significant importance to the design of combinational circuits is related to the computer aided engineering (CAE) design methods now coming into use. Specifically, these CAE tools begin with a hardware description language (HDL) description of the circuit to be synthesized (in a language such as Verilog or VHDL) and produce a circuit generally optimized for speed, based on the description. Moreover, a circuit designer may never actually see the final circuit design. Unfortunately these tools can produce unreliable circuit designs when subjected to a radiation environment. These circuit design problems can manifest themselves in a variety of ways including: •

The CAE automated tools often substitutes flip-flops for buffers to reduce propagation delay (See Figure 28). However, the use of a flip-flop can introduce other unintended states in the operation of a circuit. As can be seen from the figure a SEU can result in a situation where Q = QN and result in a system maloperation. Since these two circuit renditions are Boolean equivalent, the only way to identify this type of problem, at present, would be to scrutinize the final circuit design. Figure 29 also depicts another example of such a problem.

•

Another issue has to do with the use of “one hot” sequencer designs for efficiency, where only a single flip-flop is energized at any one time in a circuit. Although this type of design will use less combinational logic the use of more flip-flops introduces additional states in the structure. Unfortunately an SEU can drive the circuit into an illegal state from which it cannot recover.

•

Finally an issue with automated CAE routing occurs where routing rules do not take advantage of the parasitic circuit elements to improve SEU performance of flip-flop devices. The use of these parasitics can make a significant difference in performance as demonstrated in Figures 30 and 31 where the directly hardwired design provides significantly worse SEU immunity compared to the design that takes advantage of the parasitic circuit resistance and capacitance. D

D DF1

CLK

D

Q

Q A

CLK

D

Q

Y

QN

Q

DF1

CLK

D

QN

QN

DF1A

CLK

CLK

Example CAE tool speed optimization on a portion of a space-flight design. The two circuits are logically equivalent when analyzed with Boolean logic equations with lower, CAE-optimized circuit, permitting higher device speeds. An SEU analysis shows the addition of a second state variable with an upset resulting in the “optimized” circuit containing a state where Q = QN, violating the system equations and causing a failure.

Figure 28. Example: CAE Tool Optimization Resulting in SEU Failure [Katz 99b]

V-29

Block A

D

SIGNAL

Q

A

Y

A

Y

A

Y

Block B

Block C Block D

DF1

CLK

CLK

Block A D

Q

Block B

DF1

CLK

CLK D

Q

Block C

DF1

CLK D

SIGNAL

D

Q DF1

CLK

CLK

CLK Q

Block D

DF1

CLK

CLK

Two methods of signal distribution. The top version shows a signal distributed to multiple blocks with bufferings driving multiple loads. The bottom version replicated flip-flops, resulting in higher system speeds. Routing delays are significant. Recovery from SEUs with multiple flip-flops are not considered by current computer-aided engineering tools.

Figure 29. Example: CAE Software Ignores SEU Recovery [Katz 99b]

Cross-section (sq. cm/flip-flop)

10-5 C-Mod F-F (2 modules/F-F) K-Mod F-F (4 modules/F-F), mirror topology 10-6

10-7

-8

10

Device = A1020Z VCC = 5.5 VDC 10-9 30

40

50

60

70

80

LET ( MeV-cm2/mg)

Improvement in SEU performance by exploiting parasitic circuit elements and inserting them into the feedback loop of flip-flops. A mirror topology is used for these Act 1 devices, where each buffer in the feedback loop inverts, as the SEU responsive is not symmetrical The LETTH for this 1.2 µm has improved from ~18 to ~40 MeV-cm2/mg. Data is available from devices with 2.0 µm to 0.25 µm feature sizes.

Figure 30. Improvement in SEU Hardening by Exploiting Parasitic Circuit Elements [Katz 99b]

These above noted items demand that a detailed knowledge of the CAE tool, the latest revision of the tool and it’s operating characteristics be known and a comprehensive verification of all circuit designs be accomplished. 4.3.2.2 Single Event Disturb (SED) Effects The basic commentary contained in Section 4.2.2.2 applies to the impact of SED in the operational portion of the PLD device. Thus, no further discussion will be provided in this fault mode.

V-30

Figure 31. Results of Flip-flop Design Alternatives [Katz 97]

4.3.2.3 Single Event Transient (SET) Effects

Clock Upsets

In Section 4.2.2.3 the concept of SET was 100 introduced and a brief discussion concerning the impact of this type of fault on PLD configuration and operation was discussed. In this section we 10 will discuss the impact of SET on PLD operation in more detail. The crucial aspect of SET is p1B0 shown in Figure 14b where a “double clock” 1 p1B60 p2B05 pulse is generated from a heavy ion or energetic p2B604 particle strike in a clock circuit. In this situation the SET generates a “runt” pulse in the clock dis0.1 1.00E+03 1.00E+04 1.00E+05 1.00E+06 1.00E+07 tribution circuit. Moreover, if these pulses occur Clock Frequency on or near a clock transition they will propagate as a real signal as shown in Figure 14a and disFigure 32. SET Frequency Dependence cussed in Section 4.2.2.3. Additionally, as shown of Clock Upset [Katz 98] in Figure 32, the upset rate will vary as a function of clock frequency (i.e. the more clock transitions the greater the opportunity for a false signal). 4.3.2.4 Single Event Latchup (SEL) Effects Latchup is a classic failure mode associated with CMOS technology and is discussed in detail in [Messenger 86 and Dressendorfer 89]. A latchup failure results from the activation of a four layer (PNPN) parasitic structure that occurs in bulk CMOS technology that emulates a discrete silicon-controlled-rectifier (SCR) device as shown in Figure 33. Moreover, as with an SCR the parasitic structure can only be disabled by removing power from the integrated circuit. Latchup can be engendered in a number of ways including electrical overstress, a strike by a heavy ion or energetic particle at a critical node or a nuclear weapon detonation. This last one is denoted as dose-rate latchup and will not be discussed in this short course. Figure 4, previously discussed, depicts such an occurrence. The most basic result of a latch fault is the destruction of the device caused by burnout, if VDD (from a “stiff rail”) is inadvertently connected to either ground or VSS through a sufficiently low impedance path.

V-31

•

Radiation induced current spike or photocurrent places device in an anomalous state where it does not respond to input signals

Anode P Q2

N

Results from "turn on" of inherent npnp or pnpn path not activated by normal current flow

After Irradiation

Pre-Irradiation VIN

n-

VCC (or Ground)

VDD

VOUT n-

Cathode

Electrical Performance

VIN

VCC (or Ground)

p-

N P

Q1

p-

p-

n-

p-

n-

n-

p-

n-well

Normal Signal flow

VDD

VOUT

n-

n-well

p- substrate

High resistance path (~ an open switch)

p-

Time

p- substrate Low resistance path (~ a short)

Radiation Induced Switching Point

Time

Current

•

normal operation

Time

Voltage

If power is turned off in timely manner, device can be reset, if not, joule heating can melt materials = burnout Figure 33. Dose Rate Induced Latchup [DTRA 99]

A more insidious manifestation of latchup is denoted as “microlatch” and involves the latch of a small portion of an integrated circuit without a failure or any external indication such as significant increase in static input current. In this case the latch condition can go unnoticed, unless a circuit is exercised periodically. Concerning modern PLD technology the following statements about SEL can be made: •

Reductions in feature size and increases in integration density coupled with the move away from the use of epitaxial substrates will increase the sensitivity of modern integrated circuits to this failure mode. However, it should be noted that the use of epitaxial substrates do not guarantee the elimination of SEL since design rules (e.g. well spacing, well contacts, etc.) and physical layout play a crucial role concerning this phenomenon. Moreover, in some cases devices fabricated without epitaxial substrates (e.g. Chip Express QY55xx series) performed better then those fabricated using epitaxial substrates (e.g. Chip Express CX20xx and CX30xx series).

•

The use of insulating substrate materials, e.g. silicon-on-insulator, should eliminate or greatly reduce SEL

•

Planned reductions in operating voltage concurrent with reductions in feature size should ultimately preclude the occurrence of latchup as VDD reaches the < 1 volt range.

Test challenges for SEL include: • • •

Identification of proton sensitive SEL devices without inducing TID failure. Ensuring that actual worst-case structures or circuit implementations are used for test and characterization due to the dependence of SEL on layout. Establishment of cost-effective hardness assurance methods to address device-to-device, wafer-to-wafer, lot-to-lot or other variations in design and layout associated with commercial microelectronics that could modify the latchup response of a device.

V-32

Fortunately a significant body of information and established test procedures exist to support latchup testing and characterization. As previously stated SEL has been demonstrated in a number of PLD technologies as listed in Table 7. In addition a SEL response is shown in Figure 34. 4.3.2.5 Single-Event-Functional-Interrupt (SEFI) The mechanisms and impact of a SEFI fault are similar to those described in Section 4.2.2.6 concerning configuration control and no additional discussion will be provided. In the following sections specific test and characterization strategies will be discussed. Table 7. Latchup Summary [Katz 98] Device Type * Pre-prod.

Size/Voltage (nominal core)

Threshold (MeV-cm2/mg)

RH1020

1.0 µm / 5.0

> 74

QL24X32B

0.65 µm / 5.0

< 18

RT54SX16/32*

0.8 µm / 3.3

> 120

A54SX32A*

0.25 µm / 2.5

High

QYH530

0.8 µm / 5.0

52

One-Mask

CX2041

0.6 µm / 2.5

> 37

LPGA

CX3001

0.35 µm / 3.3

Low

A54SX16*

0.35 µm / 3.3

> 74

QL3025

0.35 µm / 3.3

< 11

XQR4062XL*

0.35 µm / 3.3

> 100

Comments Destructive

Destructive

500

400

ICC (mA)

300

200

100

0 0

5

10

15

20

Time (Sec)

Figure 34. QL3025/0.35 micron CMOS PLD SEL Response at VBIAS = 5.0V, 3.3V for LET = 18.8 MeV-cm2/mg (Titanium at 0 degrees). Note for 3.3 volt VDD, no latchup was demonstrated. [Katz 99a]

V-33

5.0 TEST AND EVALUATION STRATEGIES The evolution of PLD test strategy has attempted to keep pace with the increases in device complexity. “Coverage”, or how exhaustive a test strategy may be, has suffered due to the recent dramatic increase in PLD complexity. Separate from fiscal and schedule constraints, the basic capability to access each PLD logic module with sufficient observation capability to characterize is often beyond the capacity of available test equipment. Because the first PLDs consisted of many combinational logic blocks, the test engineer could choose a representative logic path to model behavior of the remaining available combinational logic paths. The addition of sequential logic in the data path did not appreciably complicate testing, as the PLD design still consisted of replications of “standard” logic blocks. Today, the required evaluation must include common combinational and sequential logic blocks, phase (or delay)-locked loops, specialized input/output structures, and normally-inaccessible registers used for die-level functional verification. Radiation testing should be performed under standardized conditions, as outlined in Method 1019 of MIL-STD-883 for steady-state total dose irradiation, and ASTM-F-1192, “Standard Guide for the testing of Single Event Phenomena Induced by Heavy Ion Irradiation of Semiconductor Devices”, and the EIA/JEDEC Standard 517, “Test Procedures for the Measurement of Single-Event Effects in Semiconductor Devices from Heavy Ion Irradiation” outline methods of performing heavy-ion testing. The dose rate applied during steady-state total dose irradiation has received particular attention lately, and can significantly affect a device’s response. The “use” environment must be carefully considered when planning tests. Temperature, voltage, operating speed, packaging, etc. all influence test results and will be discussed in the following sections. In the following paragraphs, we will review the basic concepts of ionizing dose testing, considerations specific to ionizing dose testing of PLDs, and review test results of representative PLDs. The Single Event Effects section follows an equivalent outline. 5.1

Ionizing Dose Testing

5.1.1 The Basics of Ionizing Dose Testing It is important to structure tests which meet the basic requirements of proper radiation testing. Historically, all ionizing dose exposure, for example, was performed on powered test articles. TM1019.5 still accurately specifies the need for powered, ambient-temperature exposure, with accelerated aging testing included for MOS microcircuits. But, unpowered exposures may be used to supplement powered exposure results when particular device applications are considered, as explained in the following paragraphs. Powered Exposures. Exposures should be accomplished with a voltage applied to the device which is selected to produce the greatest radiation-induced damage, or worst-case damage for the intended application, if known. This typically translates to an exposure at 105 percent to 110 percent of the nominal supply voltage used in the application. Care must be taken not to overload the device outputs during the exposure; the associated rise in junction temperature may cause premature device annealing effects. Control pins should be configured such that normal integrated circuit operation is emulated; i.e., chip and outputs enabled, device in a “functional mode” as compared to a “test mode”, etc.

V-34

Because a device input connects to at least one FET gate, this connection is important, and all inputs should be electrically connected and not left floating during exposure. Note, however, some device families have internal pull-up/pull-down resistors, which will do an adequate job of electrically connecting the inputs, and will make the bias board design smaller and simpler. Nchannel structures which are biased “on” will show a greater threshold voltage shift (VGS(TH)) than one which is in an “off” state. This is because the positive gate voltage pushes the positive charge closer to the SiO2-Si interface where it has more influence on the threshold. Therefore, if inputs are expected to be driven “high” during normal operation (this should be a common occurrence!), some inputs should be connected to VDD, turning those input n-channel structures “on” during irradiation. In the case where a complimentary MOS structure is outputting a logic “1” and the n-channel device threshold voltage falls below 0V, both the n-channel and p-channel devices in the complimentary MOS structure will be “on”, when normal operation requires the n-channel to be “off”. When this leakage through the partially “on” n-channel device is multiplied by the numerous CMOS structures experiencing similar radiation-induced n-channel device threshold shift, the standby current of the device is seen to increase substantially, as shown in Figure 35. Irradiation of p-channel structures causes the same negative threshold voltage shift, but pchannel threshold shift does not contribute to leakage current increases. However, the p-FET threshold shifts can induce changes in other parameters, including propagation delay, which cannot be ignored during post-exposure characterization. For bidirectional pins, several situations are possible: 1) the bidirectional pins may always default either to an input or an output mode; 2) the bidirectional pins may be commanded into either an input or an output state; 3) the bidirectional pins may power up in an indeterminate state, and the exposure fixture may lack the proper stimulus to change bus states. The first two situations are addressed by treating the pin either as an input or an output, depending upon the preferred or commanded state, provided there is no contention in status at power-up. When the pin status is indeterminate, a large-value series resistor can be used to provide a gate potential for 3 50 3 00

Icc (mA)

2 50 2 00 1 50 1 00 50 0 0

10

20

30

40

k rad (S i)

Figure 35. Total Dose Failure of an Experimental A1020 PLD. The first major increase in supply current is related to charge pump failure, while the second corresponds to the increased leakage and loss of functionality caused by n-channel gate threshold shift. [Wang 97]

V-35

an input pin, and/or a small load if presented to a pin configured as an output. When a choice can be made, bidirectional configuration as an input is preferred, as this effectively increases the sample size of input structures, whose threshold shifts contribute to worst-case test conditions. The electrical state (gate voltages) of an output structure is determined by the integrated circuit. Electrical connections to a device output will not change the device susceptibility to radiation due to bias conditions, and may cause increased power dissipation and possible thermal annealing effects. For ionizing dose testing, then, no output connection is necessary, unless for a monitoring condition. A second recommendation discovered during research for this course segment involves tri-stating outputs (when possible) and connecting an arbitrary number of outputs to VDD or VSS through pull-up or pull-down resistors [Johnston 00]. This is probably a preferred method, as most FPGAs possess bidirectional input/output pins, and this method will provide electrical connections to the input cells. Unpowered Exposures. Additional characterization may be performed when an application specifies that the device will be completely unpowered during the majority of its useful life. Honeywell experience suggests a factor of 4 or greater improvement in failure threshold for certain components. A specific example would be an EEPROM which stores program code; the device is powered at processor startup, and then power is removed once the information is transferred to RAM. In the actual application, care must be taken to completely isolate the device, including control, data and address pins, and to clamp the device power to ground to prevent charge buildup. Otherwise, bus activity can power the device to within a diode drop of an output logic high voltage, via current flow in the ESD protection structures connected to the VDD metalization. If a PLD is being considered for a similar low-duty-cycle application, powered and unpowered exposures are required to determine the radiation tolerance under both circumstances, and to quantify any benefit of unpowered conditions during a mission. Programming. Virtually any type of configuration lends itself well to TID testing, and unless critical timing parameters are an issue in the application which the PLD is being evaluated for, a configuration which allows easy characterization may be used. As is discussed in the SEE sections, large shift registers allow some input and output parameters to be measured, but for ac parameters, including propagation delay, large combinational logic chains provide the best conditions, as the delays from clock to flip-flop output are typically small. For demanding PLD applications, we have had to specify that the actual flight configuration be used for radiation lot acceptance testing as subtle changes in the timing attributes of the device defined failure long before significant changes in the dc parameters occurred. Lead/Aluminum (Pb/Al) Container. TM1019 discusses the use of a filter for low-energy, scattered radiation in 60Co sources. This container has become a common appliance for most ionizing dose tests at the Honeywell, NASA GSFC and NASA JPL facilities. The military standard allows omission of the container if it can be shown that the low energy scattered radiation is small enough not to cause dosimetry errors, but it may be easier to simply incorporate the container. Exposure Methods. For either the in-flux or remote test, TM1019 requires post-irradiation electrical measurements to begin within an hour of the end of an irradiation (sooner is better), and the device should not remain out of the source for more than two hours, including the time required for removal, characterization and return of the device to the irradiation source. Ambient temperature testing usually does not pose problems in typical 60Co setups.

V-36

Accelerated Aging. Accelerated aging is also referred to as the “rebound effect”, or “time dependent effect (TDE)”. TDE was briefly explained in section 3.1, and the importance of the applied dose rate’s effect on device degradation is depicted in Figure 36. TM1019 provides criteria for exemption from TDE characterization, but most of the criteria fail for a PLD targeted for a long-term satellite application, including the provision that the test can be carried out at the dose rate of the intended application. One possible criteria which could minimize recurring cost in production is to demonstrate that a device does not exhibit TDE during the development phase, thereby eliminating that criteria from a lot acceptance test program. The rationale for the accelerated anneal procedure is that this test will eliminate devices that could fail in space due to trapped hole buildup (revealed by the characterization at 100 percent the specified dose), while the additional 50 percent exposure and anneal is used to eliminate devices that could fail in space due to interface trap buildup [Fleetwood 89]. The Actel 1020A FPGA with MEC die is a good example of the rebound effect. Of the date code 9530 parts tested at Honeywell in 1996/97, several parameters showed worst-case performance post-anneal. These included input threshold voltages (which showed rebound) and output signal propagation delays (actually complete timing path prop delays). Some of these parts exhibited test vector functional errors after the anneal that were not present pre-anneal. Note that of about four parts that had functional errors occur after the anneal, three recovered after a few weeks unbiased at room temperature and one did not [Lintz 00]. The results of this test are shown in Figure 36a through 36c, with the input logic low voltage, input logic high voltage, and an example timing path delay graphed, respectively. Otherwise, ionizing dose testing of PLDs consists either of a series of powered exposures with interim electrical characterization, or of in-situ testing with selected parameters measured periodically during the exposure, optionally with a more complete characterization after the desired cumulative radiation dose is obtained. Test Sample “Pedigree.” Be sure that your PLD test sample is representative of what you are planning to fly. Some vendors procure PLD die from several vendors, and it is very possible that samples for total dose testing which may have been obtained in exchange for providing the manufacturer with radiation data may not possess the same vendor’s die when the flight lot is procured. Parameter: VIL for Clk(H) @ VCC=4.5V

VIL, Volts

1.80 1.60

s/n 54

1.40

s/n 55

1.20

s/n 56

1.00

s/n 59

0.80

s/n 60

0.60

s/n 61 s/n 74

0.40

s/n 75

0.20

s/n 76

0.00 PRE-RAD

POST 50K

POST 100K POST 150K

POST ANNEAL

s/n 77

Figure 36a. A1020 with MEC Die, Date Code 9530, Input Logic Low Voltage vs. Ionizing Dose

V-37

Parameter: VIH for Clk(H) @ VCC=4.5V 2.00 1.80

s/n 54

1.60

s/n 55

VIH, Volts

1.40

s/n 56

1.20

s/n 59

1.00

s/n 60

0.80

s/n 61

0.60

s/n 74

0.40

s/n 75

0.20

s/n 76

0.00 PRE-RAD

POST 50K

POST 100K POST 150K

POST ANNEAL

s/n 77

Figure 36b. A1020 with MEC Die, Date Code 9530, Input Logic High Voltage vs. Ionizing Dose Parameter: EB(45) MIN @ VCC=4.5V

Minimum Propagation Delay, ns

50 45 40

s/n 54

35

s/n 55

30

s/n 56

25

s/n 74

20

s/n 75

15

s/n 76

10

s/n 77

5 0 PRE-RAD

POST 50K

POST 100K

POST 150K

POST ANNEAL

Figure 36c. A1020 with MEC Die, Date Code 9530, Timing Path Propagation Time vs. Ionizing Dose

5.1.2 PLDs Incorporating ONO Antifuse Technology: Device-Specific Ionizing Dose Concerns Charge Pumps. The Oxide-Nitride-Oxide, or ONO antifuse technology requires high voltages to program the antifuses. To ensure the high voltage (typically 18V) used during programming does not overstress the component, high voltage transistors are included in the architecture to isolate lower-voltage internal circuitry from overstress by the programming process. Some of this high voltage circuitry consists of high voltage isolation devices (transistors) in series with the inputs and outputs of the logic module. After programming these transistors are not required and could potentially degrade the speed of the internal signals, due to the channel resistance. To minimize their effect there is a charge pump in the device that produces approximately 10V. This raised voltage is used to drive the gate inputs of all the isolation transistors to saturate them and minimize their “ON” resistance [Biddle 99]. This topology is shown in Figure 37.

V-38

Logic Module CP CP

Antifuse

}

Routing

Figure 37. Isolation FETs Protect the Logic Modules From the High Voltages Used to Program the Antifuses. The FETs are turned off during programming, and kept fully “on” during normal operation via the higher charge pump (CP) voltage. [Biddle 99]

Note that this 10V gate voltage is present during normal device operation. When evaluating charge pump operation in the ionizing dose environment, it is important to include propagation delay characterization so that any increased resistance in the high-voltage pass transistors due to charge pump (and/ or transistor) degradation is characterized. Furthermore, degradation of the charge pump causes the higher voltage transistors to cause longer start times and larger turn-on transients. Investigations into the turn-on transients implicate the limited current capability of the charge pump and its inability to rapidly charge the isolation transistor gates after total dose has increased their gate leakage. When the bias voltage on the isolation transistors is insufficient, then the logic voltages reaching logic module inputs will bias both n and p-channel transistors of the CMOS pair and significant totem pole currents result, creating the transient. An additional effect is the operation of the I/O logic during power-up. This characteristic can cause the device to take significant time to meet its truth table and for particular mission-critical situations may be unnacceptable [Katz 97]. At least one situation where a satellite mission has failed due to the indeterminate start-up state of an Actel 1020 (albeit not radiation-induced) has been documented [GibbonsAmes 99]. If the duration of the indeterminate state is important after ionizing radiation absorption of the device, perhaps well into a long-term space mission, the degradation of the start-up period must be included in a worst-case analysis or test of the circuitry. The associated effects which should be evaluated as part of an ONO-based PLD ionizing dose characterization include rebound, monitoring supply current vs. time during the exposure (with additional supply current measurements part of interim characterizations), input and output dc parameter shifts, and propagation time and startup time measurements to assess contributions from the charge pump circuitry and high-voltage isolation network. The effects and test specifics are summarized in Table 10.

V-39

Table 8 compiles failure levels in several ONO-based PLDs due to ionizing dose from proton irradiation [Katz 97]. Table 8. Recent ONO-Based PLD Proton (Ionizing Dose) Test Results Part Number

Feature Size

Die Manufacturer

Failure Threshold, krad (Si)

A1280A

1.0 µm

Matsushita

~7

A1280XL

0.8 µm

Winbond

<3

RH1280

0.8 µm

Lockheed-Martin

>300

A1280XL

0.6 µm

Chartered

<3

Act 3

0.8 µm

Matsushita

15 – 50+

Act 3

0.8 µm

Winbond

<5

A32140DX

0.6 µm

Chartered

<3

MKJ911

0.6 µm

Matsushita

30 – 50

KJ911

0.6 µm

Lockheed-Martin

>200

QYH580

0.8 µm

Yamaha

~15

CX2041

0.6 µm

Tower

~7 - 10

5.1.3 PLDs Incorporating α-Silicon Metal-to-Metal Antifuse Technology: DeviceSpecific Ionizing Dose Concerns Amorphous Silicon-based devices hold significant performance advantages over the ONObased PLD family. The programmed resistance of a metal-to-metal antifuse is typically 20 to 50 ohms, where the ONO antifuse resistance is 300 to 500 ohms. ONO-based antifuses typically require a programming voltage of 20V to 30V, while the metal-to-metal antifuse programs with only about 10V. This lower voltage permits α-Si PLDs to be fabricated on substrates with thin (2 µm) epitaxial layers, improving single event performance. Although the programming voltage is lower for the α-Si technologies, charge pumps and isolation transistors are still used, and the test engineer should plan an ionizing dose test to include characterization of these elements. The UTMC 22VP10 1 Mrad(Si) radiation hardness assured (RHA) device is an excellent example of the achievable hardness of α-Si technology PLDs. Actel’s RT54SX-series, another αSi technology, exhibits improved hardness when compared to most of the ONO-based PLD family; supply current for three experimental lot splits is plotted against dose in Figure 38. In the discussions on total dose testing in this text and on several websites, the common (and still recommended) practice is to plot supply current as a function of total dose. While this is important characterization data, it should not be interpreted as an accepted pass/fail method. For some parts, functional failure occurs just before the small, sudden rise in current (a good example of this sudden current rise can be seen in Figure 35). A second example is shown in Figure 39, where there is essentially no change in supply current for the 5V exposure, however post-exposure characterization revealed functional failure. [Katz 99]

V-40

Figure 38. Summary of Total Dose Experiments on RT54SX16 Prototypes, 3 Lot Splits, for the Data Shown Above. The worst lot split showed > 50 kRad (Si) performance with the best having > 100 krad (Si) tolerance. [Katz 99c]

5.1.4 EEPROM-Based Reconfigurable FPGAs The family of PLDs includes non-volatile, reconfigurable FPGAs, which use either EEPROM or flash memory devices to store configuration data, such as Actel’s Pro-ASIC series. Ionizing dose testing of non-volatile memory devices poses interesting situations depending upon whether on-orbit reconfiguration is required. During ionizing dose testing, an obvious method of verifying full device functionality would be to reconfigure the device by re-writing the memory. However, performing this memory write access between exposure has the effect of refreshing the stored charge in the memory cells; something that will not happen during a long-term space application where no reconfiguration is planned. As a result, an artificially high ionizing dose failure level may be achieved since the partial charge loss caused by each ionizing dose exposure step is replenished between exposures. The recommendation, then, is to assure that interim characterizations do not write to the non-volatile memory. Note, however, that it may be possible to extend the hardness of a device if the memory can be refreshed periodically on-orbit. This hardening technique assumes that the non-volatile memory is responsible for the inadequate ionizing dose hardness. To characterize this situation, or if onorbit reconfiguration is a possible device application, one approach is to use a double sample; rewriting one sample between exposures, but no reconfiguration of the second sample, to determine any benefit the refreshing may have.

V-41

Figure 39. Change in current as a function of dose for the RT54SX16 tested with VDD = 5V, and a separate device exposed with VDD = 3V. While the onset of failure is evident for the VDD = 3.3V exposure, no change in current is observed for the device exposed with 5V bias, but the device was non-functional after the test. [Katz 99]

5.1.5 Reconfigurable SRAM-Based FPGAs Reconfigurable SRAM-based FPGAs, like their EEPROM-based counterparts, also appear to possess the charge pump and high-voltage FETs required in ONO antifuse devices, but the applied voltages (on the order of 1.5VDD; reference Figure 7c) are lower, possibly improving the response of the higher voltage circuitry to ionizing dose. For this reason, characterization should involve similar test techniques as for ONO antifuse devices, namely startup time, propagation time, etc. Test complexity is increased by the addition of the configuration SRAM; from an ionizing dose perspective, SRAM degradation and/or failure must also be considered. Although the FPGA configuration is held “statically” by the memory locations, most FPGAs presently available also apportion to the user some fraction of the configuration memory to be used as application memory. If the SRAM is planned to be used as application memory, the characterization schedule must include memory propagation, input output dc parameters, etc. as part of an ionizing dose test. The apportioned memory usually cannot be accessed via an I/O and address bus; a best general recommendation is to program logic paths within the FPGA which propagate through memory at the desired application speed, and to verify that the device performs properly at the ionizing dose level of interest. A second solution may be to test the memory as an entity, using the readback function bus as an access port. In particular, Xilinx’s Virtex series provides multiple access ports for the purpose of reading and writing data to/from the configuration memory array [Carmichael 99].

V-42

Table 9 provides ionizing dose thresholds for some of Xilinx’s SRAM-based FPGA products; Xilinx defines the ionizing dose threshold as a supply current increase of a factor of two, with all ac parameters characterized to be within specification. [Xilinx 00] Table 9. Ionizing Dose Hardness of Xilinx SRAM-based FPGAs Part Number

XC4036XL XC4036XLA XQR4000XL XQR4036XL XQVR300 (Virtex)

Ionizing Dose Capability

40 to 60 krad (Si) 16 to 42 krad (Si) 60 krad (Si) 60 krad (Si) >50 krad (Si)

5.1.6 Ionizing Dose Effects Summary and Testing Information for PLDs Table 10 provides summary information on the various effects for consideration when designing ionizing dose radiation tests on PLDs. While this list attempts to be fairly comprehensive, PLD evolution in the areas of device scaling, process, operating voltage, design and architecture is very rapid. The complex interplay between these attributes cannot be generalized; many rules of thumb fail and detailed examination, analysis, and comprehensive testing is required to properly evaluate PLDs for use in an ionizing dose environment. 5.2 Single Event Effects (SEE) Testing 5.2.1 Basics of SEE Testing Historically, heavy ion SEE tests follow the standard method described in ASTM-F-1192 “Standard Guide for the Testing of Single Event Phenomena (SEP) Induced by Heavy Ion Irradiation of Semiconductor Device”. This standard has been superseded by EIA/JEDEC Standard 57 “Test Procedures for the Measurement of Single-Event Effects in Semiconductor Devices from Heavy Ion Irradiation.” [Petersen 97] The standard covers the specific topics of terminology, procedures and overviews of experimental setup, beam diagnostics, vacuum chamber diagrams, recommended fluence levels, etc. It also provides sections which specifically discuss performance of single event gate rupture testing. Some of the salient points are briefly described below: Supply Voltage. In general, SEU, SER, SEFI and other non-destructive single event effect testing is performed at 90 percent to 95 percent of the devices’ nominal supply voltage rating, because the lower voltage promotes SEU sensitivity. Today, it is more common to see prime power specifications of five percent regulation or less; so it may be overly conservative to perform tests at 90 percent of the nominal rating (but for a “generic” device characterization, 90 percent is a more comprehensive condition). For SEL, SEDR and other potentially destructive effects, 105 percent to 110 percent of the nominal voltage rating is used, since the higher voltage promotes latchup and/or places a larger electric field across the antifuses in a device so equipped. Physically, elevated voltage and temperature are used for latchup testing as both have the effect of reducing the “breakover voltage,” enhancing conditions favorable to latchup.

V-43

Table 10. Ionizing Dose Effects and Testing Information Summary Device Type FPGA, ONO Antifuse

Effect

Testing Information

• Rebound • Increased supply current • I/O dc parametric shifts • Charge pump; high-voltage FET degradation; and logic FET VTH shifts

• Use accelerated aging test methods from TM1019.4 • Chart supply current vs. time, include device supply current measurements in interim characterizations • Include dc parameter measurements in interim characterizations, if possible • Include propagation delay and startup time measurements in interim characterizations

FPGA, α-Si Metal-toMetal Anitfuse

• Rebound • Increased supply current • I/O dc parametric shifts • Charge pump; high-voltage FET degradation; and Logic FET VTH shifts

• Use accelerated aging test methods from TM1019.4 • Chart supply current vs. time, include device supply current measurements in interim characterizations • Include dc parameter measurements in interim characterizations, if possible • Include propagation delay and startup time measurements in interim characterizations, if critical

EEPROM-based Reconfigurable FPGA

• Rebound • Memory loss; EEPROM • Increased supply current • I/O dc parametric shifts • Logic FET VTH shifts

• Use accelerated aging test methods from TM1019.4 • Do not re-write EEPROM contents prior to interim characterizations • Chart supply current vs. time, include device supply current measurements in interim characterizations • Include dc parameter measurements in interim characterizations, if critical • Include propagation delay measurements in interim characterizations

SRAM-based Reconfigurable FPGA

• Rebound • Increased supply current • I/O dc parametric shifts • Logic FET VTH shifts

• Use accelerated aging test methods from TM1019.4 • Chart supply current vs. time, include device supply current measurements in interim characterizations • Include dc parameter measurements in interim characterizations, if critical • Include propagation delay measurements in interim characterizations

V-44

One exception to the rules of thumb regarding test voltage is for characterization of Single Event Transients, which may be enhanced by testing at 105 percent to 110 percent of nominal voltage. Temperature. Non-destructive single event effects tests are usually performed at device ambient temperature, while latchup testing is performed at approximately 100 to 125 degrees Celsius. Honeywell has used the electrostatic discharge (ESD) protection diodes in the input structure as a means of monitoring die temperature: Identify an input which can be held in a logic “low” state throughout the test, and use a source measurement unit (SMU) to sink ~100µA out of the input (see Figure 40). The SMU reports the voltage across the ESD diode, allowing the die temperature to be accurately read without any thermal resistance issues. Physically, the voltage change as a function of temperature, dV/dt, is approximately equal to -2.5mV per degree Celsius, but the magnitude of dV/dt decreases with increasing temperature. An oven can be used to establish equilibrium at a temperature of interest, and the corresponding diode voltage characterized for any particular temperature. Very long test runs using heavily populated test fixuring and high clock speeds can tend to cause fixture self-heating to the upper limits of the device’s temperature specifications, given the loss of convection in the vacuum chamber. Even if no elevated temperature testing is planned, it is good practice to provide a temperature monitoring capability when large power dissipation or lengthy operational periods in the vacuum chamber are planned. Clock Frequency. Clock distribution and combinational logic circuitry upset rates are both potentially sensitive to clock frequency. If possible, dynamic testing should be performed as close to the application speed as possible, and the clock frequency should be varied to investigate any frequency dependent error rates. Delidding. Unless we are able to receive test articles without sealed lids, Honeywell Failure Analysis Lab personnel usually perform the device delidding. The procedures for various packages are briefly described below [Fayad 00]. •

Ceramic Flat Package. An X-Acto or other sharp knife is used to carefully scrape away the sealing material (solder or weld) between the seal ring and the kovar cover. Then, the lid is carefully pried away so as not to upset the bond wires. Melting the seal by application of heat is never used, as there is too great a chance that die damage may result. Input Struct ure

V =Vf(T) aT

+ -

~100µA ~100 mA

PLD

Figure 40. Monitoring Die Temperature During SEU Testing

V-45

•

Plastic Encapsulated Modules (PEM). The device is first x-rayed so that the die location and void are understood, then a drill bit is used to create a depression in the packaging above the die surface. Yellow fuming nitric acid is then applied one drop at a time until the plastic encapsulation over the die surface is dissolved. While the nitric acid will not harm the die (the glassivation protects it), it will eventually dissolve aluminum, so care must be taken in deciding when the acid applications should cease.

•

Ceramic Dual Inline Package. Only the uppermost ceramic package layer is placed in a vise, with a fairly high compression force applied by tightening the jaws evenly along the longer package sides. Then, a blunt object (a file, perhaps) is used to firmly tap the uppermost ceramic package layer. The remainder of the device package containing the die will separate and drop from the upper package layer. Usually, the leadframe remains intact.

After delidding, the die surface should be inspected for evidence of a polyimide coating which may be applied to minimize device sensitivity to alpha particle-induced upset, caused by impurities in the packaging material. With respect to the die active region, this coating is very thick, and will hinder ion passage to the die during SEE test. When testing devices with the polyimide coating, we have seen a cross-section reduction as the effective LET is increased by impingement angle, as the path through the coating is lengthened. Other Considerations. Sometimes when devices are obtained in a “cavity down” package, SEE testing can still be accomplished by providing holes in both the socket and the printed wiring board to which the device will be secured. The board is then mounted “backwards” away from the beam so that the ion beam enters through the bottom of the board, through the socket and onto the die surface. Particularly in this case, but a general concern for all tests, is to ascertain the maximum angle (for adjustment of the effective LET) through which the test fixture can rotate before beam shadowing occurs. 5.2.2 Single Event Effects The various building blocks found in most complex PLDs, such as specialized input/output structures, sequential and combinational logic blocks, etc. can each possess a different response to heavy ion or proton fluence. Each of these elements may have a unique threshold LET, resulting in step increases in the device cross section as additional PLD elements become sensitive to the higher energy deposition with increasing LET. If the PLD consisted of one storage element, that element would function properly until adequate energy were deposited into the sensitive volume causing a bit flip, due to increasing LET. More often, the slope of the cross-section curve will temporarily increase as groups of PLD elements are affected by the charge deposition. In Figure 41, curve A suggests a device with only one type of upset mechanism with a single threshold LET and saturation cross-section. Curve B represents a more typical response from a complex device with several upset mechanisms, and therefore having several threshold LETs and saturation cross-sections [Pease 90]. Initially, analysis of SEU data consisted of “converting” the actual response of a complex device into a cross-section plot similar to Curve A, providing one cross-section (the saturated value) and one LET threshold, which was usually taken at 10 percent of the saturated cross-section. Other percentages were used by various experimenters, including use of the “zero upset” LET, where error onset was observed.

V-46

63520

Curve A

σ L (A)

Curve B

σ L (B) Cross Section

σ1+σ2+σ3

σ1+σ2 σ1 L C (A)

L C (B)

Figure 41. Device Characteristic LET vs. Cross Section for Two Devices with Different Upset Characteristics

Note that this method provides a very conservative rate prediction, as it assumes every sensitive element in the test article possesses a threshold LET which is the lowest observed during test. Inspection of the cross-section curve shows that this is clearly not the case, as the cross-section associated with the onset LET can be several orders-of-magnitude smaller than the saturated cross-section. This LET value (which is the energy deposited over a path length, or dE/dz) is converted into a critical charge (using the thickness, dz, of the sensitive volume), and the saturated cross-section value converted into two dimensions of the sensitive volume, typically assuming a square sensitive area by taking dx and dy to be the square root of the saturated cross-section (unless the actual dimensions are known). The critical charge, dx, dy, and dz values are inputted into the CREME code, along with the appropriate environmental description, to return an upset rate prediction. Analysts with a need for less conservative predictions (and more time on their hands) began to “slice” the cross-section curve into an arbitrary quantity of discrete cross-section values at their associated LET value. After the appropriate critical charge and sensitive area conversions, the CREME code was run repeatedly for each of the resulting critical charge, dx, dy and dz values, and all results summed to return an upset rate prediction. The use of the Weibull function to describe the cross-section curve to the analytical software provides a continuous set of critical charge, dx, dy and dz values for the upset rate prediction calculation. 5.2.3 Single Event Effects Testing The goal of any Single Event Effects (SEE) test is to provide a user with the information needed to accurately predict the SEE rate; the basis of any SEE rate calculation is the cross-section (σ) as a function of Linear Energy Transfer (LET). The cross-section is given by σ = (N/F)secθ and the LET is the LET0secθ , where N is the number of errors, F is the fluence, θ is the angle of incidence of the particle beam, and LET0 is defined as the LET at normal incidence to the device die. Figure 42 provides a sample response curve as well as tips for obtaining good data [Petersen 97]. For a detailed discussion on SEE calculation from test data see the NSREC 1997 Short Course module by Ed Petersen.

V-47

Figure 42. Experimental heavy ion cross section measurements, with representative ions for the Brookhaven accelerator. The solid and dashed curves are for Weibull and lognormal curves. [Petersen 97]

In order for the cross-section curve for an SEE test to be meaningful, the test must be structured in a way which allows accurate calculation of the number of elements involved in the test. It is fairly easy to generate shift registers which utilize most or all of the sequential logic in a PLD, but involving all custom blocks in a complex PLD may not be possible as part of a “reasonable” test. Upset rate prediction accuracy, for example, will suffer if the number of bits is unknown; trying to obtain a device-level upset rate by treating the device as one bit with a large cross-section will cause a serious overstatement of the device upset rate. The idea is not necessarily to utilize 100 percent of the building blocks, but to maximize the effective bit sample in a given device by using a relatively large percentage of them, and to know how many bits are involved in the test. Unless tested using the application configuration for which the device is a candidate, obtaining per-bit SEU rates, as well as the actual number of each type of structure used in the application’s configuration, is desirable in order to generate useful upset predictions. 5.2.4 SEE Test Strategies Test strategies range from very simple to somewhat complex, but all strategies which were researched will provide thorough results. The basic strategies include oscilloscope monitoring, golden chip and virtual golden chip methods. “Basic” Test Approach. A simple but very effective test approach has been adopted by Goddard Space Flight Center (GSFC) for some of their PLD testing. The PLD is configured as one large, or several small, serial shift registers. A constant square wave is inputted to the shift register, and logic circuits residing on the test card perform the comparison function, which are

V-48

counted by counters resident in a personal computer. This setup is enhanced by capturing the device current over the exposure time using strip chart recorders. GSFC has been able to glean a significant quantity of information using this setup. When output bits are observed to be in error (i.e. changes in data pattern), they are counted as data errors. If device output ceases or a dramatic shift in the output is recognized, a device reconfiguration error (i.e., SEU in configurable RAM portion of a device) has occurred. This latter error is dubbed a Single Event Reconfiguration (SER). As part of data reduction, clock upsets can be distinguished by logging errors either cumulatively or differentially (Figure 26), as clock errors will cause “bursts” of errors with a subsequent return to normal operation. Because the clock error causes several data errors, it is helpful to identify them and separate them from the actual data error, so that independent calculations of the on-orbit clock upset rate and data upset rate can be made. The number of clock errors for a given particle fluence is then used to establish a cross-section for the clock upset rate prediction. [Katz 97] The current versus time chart reveals several other attributes of a device’s SEE performance. A significant increase in the device current may be representative of a latchup. However, since a SER could reconfigure an internal node causing two output drivers to be in conflict, (Figure 9), an attempt to compare the occurrence of the current spike against the error signature captured by the DSO should be made. This situation is made easier to analyze by controlling the particle flux to allow adequate time between upsets to allow the experimenter to observe the DSO and strip chart signatures in real time before the next error occurs. Additionally, the magnitude of the current spike may be higher for a latchup occurrence when compared to the current increase caused by a reconfiguration-induced contention. And, latchup occurrences are usually sensitive to increases in voltage and temperature, whereas reconfiguration errors will probably be reduced at higher voltages. Note that when testing an antifuse device, the step increase in current contains information, which can be used to determine the antifuse impedance after a dielectric rupture (SEDR) event (Figure 49). Golden Chip Approach. The golden chip test approach configures two samples of the same device type identically, provides a unique stimulus to both simultaneously, and compares the outputs. When the device outputs disagree, an error is logged and the experiment reset. For smallscale integrated devices, this comparison can be accomplished with a simple exclusive-OR gate (which usually must be AND-ed with the inverted clock to filter out transients). The XOR output can be used to reset the experiment, and buffered to provide an output capable of driving a counter for error logging. As experimental complexity increases, many outputs may require comparison, requiring a larger XOR tree; PLDs lend themselves well to this application. A block diagram depicting a golden chip test is shown in Figure 43. During the reset interval, there are usually several clock cycles for which the device output is indeterminate; the compare signal should not be asserted during this time so that device output comparison does not occur until the reset instructions have been completed and the test articles have returned to “normal” functionality. For the best experiment accuracy, the reset interval should either be kept very short, or considered when calculation the device cross-section, as the ion beam fluence continues to accumulate during the reset interval, while the experiment is not actually reporting errors. One way of minimizing this error is to keep the ion beam flux low enough so that device errors are occurring far apart in time, with respect to the duration of the reset. If particle beam diameter and device physical spacing on the test fixture are considered, two delidded devices co-located on the test fixture can function as the golden chip for each other, by simply comparing the outputs of both devices while only one is exposed to the particle flux at a time.

V-49

Control

DUT (Exposed to Ion Beam )

PATTERN GENERATION & CLOCKING

CONTROL

SEQUENCER

Clock

Reference Device (Shielded from Beam ) Error Counte

Com pare

Reset

Figure 43. Golden Chip Test Block Diagram

Virtual Golden Chip Approach. For virtual golden chip testing, Goddard Space Flight Center incorporates Omnilab and VXI pc-based testers (as well as custom fixture designs). Both testers are capable of providing test patterns to the test boards and are capable of capturing output when errors occur. The VXI enhances the error capture by using an intrinsic compare and a custom-built FIFO buffer board. [GSFC 00] As noted in Figure 44, additional Actel A1280 devices (shielded from the ion beam) are used to analyze the behavior of the test articles and to categorize the error type. Honeywell has chosen Integrated Measurement Systems’ Logic Master as a preferred test host. The Logic Master is designed as an application specific integrated circuit tester; it is a pc-driven system capable of providing test patterns to test articles in vector form, and comparing the acquired data from the test article to a “compare” vector also stored in the pc. This feature is usually all that is required for a comprehensive test, but the setup is sometimes enhanced by incorporating a logic analyzer set up to trigger on the experiment’s error pulse. We adopted the Logic Master HS1000 as a standard tester in the component characterization laboratory. A customdesigned long cable interface consisting of level shifting logic on either side of an emitter-coupled logic (ECL) bus was designed and fabricated for the purposes of performing dose rate upset tests. Additional modifications included the addition of analog line drivers for monitoring signals during the radiation pulse, as well as lead shielding in the test head to improve the instrumentation's dose rate upset threshold. Recently, we have been successful in implementing a test concept where only one multilayer fixture board is designed for both dose rate and SEE tests. The long cable interface is not used for SEE testing, but is replaced by a shorter vacuum interface. For higher-speed (several megahertz) experiments, the clock signal is buffered at the test fixture, but all other forced and acquired signals travel over the vacuum interface (typically twisted-pair ribbon cable) with adequate fidelity. Because of the success we have had in using this unit, we recently upgraded the system by replacing the tester with a Logic Master XL60, and designing a new long cable interface, using low-voltage differential signal (LVDS) RS-485 devices and category 5 twisted pair data communication cable for our digital interface.

V-50

Figure 44. Example: Block Diagram of SEU Test Circuit Employing a Virtual Golden Chip Approach

A typical virtual golden chip SEU/SEL test block diagram is shown in Figure 45, and an electronic test assembly configuration schematic in figure 46. For heavy ion testing, the Device Under Test (DUT) is placed in a vacuum chamber. When performing proton tests, the vacuum chamber and device delidding are unnecessary, for proton tests, because protons have a longer range than heavy ions. For example, a 100MeV proton has a range of approximately 70.3m in gaseous oxygen, and approximately 3.1cm in Aluminum [Zieger 80]. A closed-loop heater is available for performing tests at elevated temperature. The voltage supplies are software configurable to limit supply current to a predetermined value or shut down on a high current IC latchup. Note, too, that the power is supplied separately to the DUT from any support electronics resident on the exposure board, so that any changes in device supply current are more easily observed and controlled. The test fixture outputs an error pulse to a counter and the logic analyzer (when used) is triggered each time a miscompare occurs. 5.2.5 Single Event Functional Interrupt A Single Event Functional Interrupt (SEFI) is a mode where the device loses it’s functionality because of one or more flip-flops being upset. Simply put, a SEFI is an SEU in any flip-flop whose output affects the mode or configuration of the device. SEFIs manifest themselves as unexpected resets, JTAG TAP controller upsets, and configuration control logic upsets [Fuller 99]. Often, a manu-

V-51

SEE Test Assembly Power Supplies

40 pin cable RDC SEU Data

Digital Controls Power Sense

Digital Controls

63517

Logic Analizer

Power Supplies Cable Interfaces Function Generators

DUT Test Board

Coax Clock

Ion Beam

Coax Error Pulse

Counters

25 pin Heater/Sense Controlling Computers

Vacuum Chamber

Heater Control

GPIB Interface

Figure 45. Typical SEE Test Setup 63518

SEU Vacuum Chamber Digital Tester (With Device-Specific Test Pattern)

Vacuum Interface Cable Type D-50

Inputs Data IEEE 488 Data Port

D-50 D-50

I/O I/O Clock1 Clock2 Clock3

Test Controllers

SEU Exposure Filter

DUT Inputs & DUT Selection DUT Outputs

D-50

DUT I/O DUT I/O

BNC BNC BNC

DUT Clock1 DUT Clock2 DUT Clock3 Heater Pwr, Temp Sense

Pattern Line# Out (F)

DUT 1

DUT 2

DUT 3

Pattern Line# In (A) DUT +5V and Sense

(486 PC) SEU Pulse Out BNC

Data Port

SEU Counter In SEU & SEL Counter In

Multiple Voltage Suppliers SEL Pulse Out Heater

SEL Counter In

BNC

BNC (2)

DUT Support Logic

D-25

Temperature Controllers

Voltage Supply and Sense

BNC (2)

Figure 46. Typical SEU Electronic Test Assembly Configuration

V-52

Logic +5V and Sense

facturer will install configuration registers in a device, which are used for die-level functional verification, but are not accessible once the die is packaged. To return this test register to its “normal,” functional mode requires power cycling, making this type of SEFI as catastrophic as latchup, from an availability perspective. Some believe the term SEFI is inaccurate, arguing that an upset is an upset, but it is useful to plan for recovery and mitigation strategies of these types of disastrous upset modes. 5.2.6 Single Event Transient Mavis and Eaton have observed heavy ion-induced Single Event Transient (SET) in combinational logic, global clock and control lines. They point out that with current feature sizes, not all of these upsets are propagated, due to the pulse width of the transients (100 to 200 ps). However, as feature sizes decrease and clock speeds increase, these transients will propagate as a typical circuit signal. They have observed that latch SEU in sequential circuits are independent of the clock's frequency, where as the combinatorial logic SET rate is proportional to the clock's frequency. Hardening at the system level can be accomplished by using redundant computations, but it can take several clock cycles for these errors to be detected. Therefore, returning the system to a previously known “good” state is advised [Mavis 99]. A false clock pulse in the global clock due to a SET can cause an error in every memory cell that employs the global clock [Wang 00]. Figure 47 provides examples of a digital SET capture circuit and clock pulse capture circuit. Example Digital SET Capture Circuit

Timing Waveforms (clocked capture of an SET resulting in output upset)

SET

Data

Clock

D-type Flip-flop D(in)

1 0

Data

Q(out)

1 0

D(in)

SET

1

(at inverter output)

0

> CLOCK IN (rising edge triggered)

Q(out) 1 0

Output Upset

Example Clock Pulse Upset Circuit (SET occurs at buffer driving clock inputs)

SET (false clock) 1

Clock 0 In

D-type Flip-flop Data SET

D(in)

1 Data Out (Q) 0

Q(out) Q(out)

> CLOCK IN

Expected 1 Data Out 0

(rising edge triggered)

To other Clock inputs

Error Detected

1 0 Error condition persists due to the

Figure 47. Examples of a Digital SET Capture Circuit and Clock Pulse Capture Circuit

V-53

5.2.7 Single Event Latchup Single Event Latchup (SEL) may be a destructive event where a parasitic silicon controlled rectifier structure (SCR) is activated by an ion strike [Petersen 97]. This state may lead to device burnout, but at minimum will stress the integrated circuit, if the device is not powered down [Messenger 97]. For latchup testing, the test circuitry is usually designed to limit current to the device to prevent burnout and to remove power momentarily after each latch to restore device operation. These tests are performed at elevated temperature and bias voltage [Pease 90]. Both high temperature and voltage have the effect of minimizing the breakover voltage of the SCR structure. However, the condition known as a “micro-latch” can occur which may appear as a stuck bit, or some other form of non-functionality, or as a step increase in current. Often in this situation, the latchup holding current cannot be resolved from the normal device supply current. For SEL testing, the test circuit board and test configurations are, generally, the same used in SEU testing. When latchup is a possible test article response, the experimental configuration should include some functionality monitoring, as opposed to simply applying power to the device, providing an ion flux, and monitoring the supply current. It may be difficult to distinguish between a microlatch condition and SEFI. Both events will probably result in some form of nonfunctionality and/or change in supply current, and will be remedied by cycling device power. Knowledge of detailed device architecture, if available, may be the only way to differentiate the two occurrences. Although not a very scientific approach, a device which has experienced SEFI, left in the ion beam long enough, may revert back to functionality by a second ion interaction with the responsible configuration bit. 5.2.8 Single Event Dielectric Rupture Single Event Dielectric Rupture (SEDR) is a heavy ion-induced breakdown of a biased (unprogrammed) antifuse [Cronquist 98]. Swift & Katz noted that, at least specific to the A1280A, a pattern of all zeros was slightly less susceptible to SEDR than a pattern of all ones or an alternating pattern [Swift 95]. Swift & Katz also noted dependencies on bias voltage and normal-incidence LET [Swift 95]. Katz et al. observed that biased unprogrammed ONO antifuses are vulnerable to heavy ion-induced rupture [Katz 98]. Cronquist et al. have observed that after a SEDR the ONO antifuse is left partially programmed and has a resistance of 3kohm to 10kohm, resulting in slight increase in the supply current [Cronquist 98]. The relationship between the bias voltage and LET required to cause SEDR for a radiation-hardened, ONO device and 3 preproduction amorphous silicon devices is Figure 49 [Katz 98]. Some tips for SEDR testing [Katz 00]: •

Control the particle flux. Control the particle flux, because if the flux is too high several of the antifuses will rupture at once, and destroy the device. When a device has a dielectric rupture, a spike in the current is observed, which will give you valuable information about the impedance of the antifuse.

V-54

63519

7

RH1020 - Pass 6, 90 Angstrom A54SX16 (Proto) RH54SX16 (Tech Development) RT54SX16 (Proto) - Lower Bound

Critical Bias (Volts)

6

5

4

3

2 15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

LET MeV-cm2/mg)

Figure 48. Showing the Relationship Between the Bias Voltage and LET Required to Cause SEDR [Katz 98].

Figure 49: Antifuse rupture currents for an RH1280, with an ONO antifuse and a developmental device with an amorphous silicon antifuse [Katz 00]. The developmental device with an amorphous silicon antifuse is a prototyped Actel SX series device with a 3.3V bias. The size of the current jump gives detailed information on the impedance of the antifuse.

V-55

•

Bring up the voltage from below. If you are trying to obtain antifuse impedance data, start testing with a bias just below an estimated critical bias, then gradually increase the bias as you test. This is time-consuming, but will keep the DUT from being quickly destroyed. The size of the current jump will give detailed information on the impedance of the antifuse. If the test purpose is simply to determine either if SEDR occurs or to calculate it’s occurrence for a specific application, testing within the manufacturer’s specified supply voltage range should suffice, and will shorten the test time.

•

Consider the resolution and speed of the current monitoring. During a SEDR test the current needs to be monitored, the spikes in current will signal a dielectric rupture. If the resolution is too coarse or the speed is too slow, data will be missed. A chart recorder can be used to accomplish this monitoring.

•

Record and monitor the size of the current jump. When a dielectric rupture occurs there is a noticeable spike in the current, the size of this current jump is a function of the construction. Actel ONO antifuses generally draw 1 to 10mA, where as the M2M antifuses, designed for lower impedance will draw quite a bit more current, 20mA or more. If possible, try to limit the current based on the type of antifuse being tested, this will prevent the device from being destroyed in testing.

5.2.9 Single Event Effects Summary and Testing Information for PLDs Table 11 provides summary information on the various effects for consideration when designing single event effects tests for PLDs. The rapid PLD evolution in the areas of device scaling, process, operating voltage, design and architecture limits the utility of this table to, perhaps, a checklist of items which should be considered as part of characterization of the various types of PLDs. The important aspects for proper SEE characterization of PLDs will certainly change as PLD performance is enhanced.

V-56

Table 11. SEE Summary Effect

Description

Testing Information

SEU (Single Event Upset)

Temporary change of memory or control bit [Petersen 97]

• Typically performed at minimum-specified supply voltage. • Perform a static measurement at room temperature, with a static test pattern (i.e. all zeros). Measures the characteristic of the devices storage latches [Fuller 99] • Vary the operating frequency between the device minimum and maximum during dynamic testing. • Have errors pulse to a counter and have a logic analyzer trigger each time an SEU (miscompute between the actual device output and the expected output stored in a Logic Master vector file) occurs

SEFI (Single Event Functional Interrupt)

Control path corrupted by an upset [Petersen 97]

• Monitor the upset signature as part of SEU testing. • Look for abrupt nonfunctionality, unexpected device resets as well as upsets in the control logic, the JTAG TAP controller, and the configuration control logic [Fuller 99].

SET (Single Event Transient)

Combinational logic state is temporarily “disturbed” by ion strike; self-recovers. But, transient can be captured by latching circuitry “downstream” of SET source, causing a SEU.

• Vary the clock frequency • Include combinational logic, and elements controlled by global clock & control lines when desiging the experiment.

SEDR (Single Event Dielectric Rupture)

So far, observed only in ground tests. Heavy Ion induced breakdown of a biased (unprogrammed) antifuse [Cronquist 98].

• Only a concern for antifuse devices. • Vary the test pattern (i.e. all1s, all 0s, and alternating) • Vary the bias voltage • Monitor the supply current • Control the flux [Katz 00]. • Bring up the voltage from below • Consider the speed of the current monitoring [Katz 00]. • Monitor the size of the current jump, which is also a function of antifuse construction [Katz 00]. • Monitor the sensitivity of the current as a function of bias across the antifuse [Katz 00].

SEL (Single Event Latchup)

A potentially destructive event where the device may latch into a high current state [Petersen 97]

• Perform at high temperature & voltage • Monitor the supply current • Look for increases in supply current, and loss of device functionality

V-57

6.0 CONCLUSIONS AND FUTURE TRENDS Two dominant trends now exist that will define the future use of PLD technology for space system applications. These trends are: 1. The increasing complexity of onboard data and signal processing requirements for commercial, scientific and military space systems. Some examples include: • The proposed Teledesic/ICO system that will embody complex antenna beam-forming requirements, high speed and density data routing and storage needs and the capacity for decision making concerning optimized data routing. • Proposed Earth science missions that will use hyperspectral imaging capability and robust and flexible onboard data processing to fully utilize this imaging technology. 2. The very rapid increases in semiconductor processing/manufacturing technology and CAE that will continue to lower the three main barriers (i.e. device size (integration density), operating speed, and cost) to the use of PLD technology. Examples of these trends include: • The recent announcement by Altera to adopt a 0.13-micron, all copper interconnect based, technology by the end of 2000 to support their PLD product line. It is anticipated that the use of the most advanced semiconductor processing technology should allow devices of >10 million gates, 10 levels of metal and operation at 1 GHz to be fabricated by 2005. Moreover, it is anticipated that other PLD technology providers will also follow the move by Altera to use SOTA semiconductor processing technology to support future product lines. As an additional note, Xilinx has just announced a product line with a chip >2 million gates in size. • The focus on system level integration (SLI) methods to solve technology problems which now, and will increasingly, involve the use of drop-in PLD blocks in complex ASIC devices or, conversely, the use of high complexity drop-in macro cells in PLD devices. A new type or classification of PLD has been recently defined by Atmel that includes a microcontroller, SRAM, and programmable logic on a chip and is, known as a Field Programmable System Level IC (FPSLIC) device. Indeed the idea of a System-on-a-Programmable-Chip is now under study. • The increasing availability of IP to ensure the availability of complex macro circuits for drop-in applications including DSP’s, microprocessors, ASSP, etc. • The recent announcement of electrically programmable analog integrated circuits (Programmable Analog ICs, W. Schweber, EDN, April 13,2000, page 72) Based on these two trends, one can anticipate a significant increase in the use of PLD technology in space systems. Moreover, as the use of these advanced technologies and more complex PLDs proliferate in space systems the issues involved with their test, characterization and flight qualification profound. Additionally, the issue of qualifying different semiconductor processing methods and materials (e.g. high atomic number (Z) material such as copper, alternatives to silicon-dioxide gate oxide material, etc.) will further exacerbate an already difficult issue. The various concerns for current PLD technologies has been discussed in this course section. With the existing antifuse designs, dielectric rupture is a destructive mechanism common to both the ONO and metal-to-metal technologies for SEE, but differences exist between the two designs which may provide different characterization requirements for ionizing dose tests. For reconfigurable devices, new architectures are being introduced by Xilinx (and possibly others) which

V-58

allow configuration memory readback and partial reconfiguration to be accomplished on-orbit – features which have the potential to greatly enhance the SEU hardness of these devices through fault-tolerant system design [Carmichael 99]. Throughout the PLD families, some single event effects may be difficult to isolate, and reliably distinguishing, say, a SEFI from a microlatch condition may not be possible. As semiconductor device technology evolves, issues such as the dose rate used for ionizing dose testing (with the application considered, of course) have increased in importance; the list of single event effects appears to be enjoying boundless growth, as additional subtleties of energy deposition in complex integrated circuits are better understood. In researching the material for this short course segment, it became obvious that there is an overwhelming amount of device knowledge required to effectively radiation-characterize PLDs and other complex integrated circuits. While a significant quantity of information exists on the subject, many “gray” areas exist where solutions to the test obstacles caused by device complexities are not well-described. Despite the large amount of information found on the subject, much more information would benefit the radiation effects community in the qualification of these devices for space applications.

V-59

REFERENCES All References are Unclassified Biddle 99: A. Biddle, et al, “Flight Qualification of Actel FPGAs,” Military and Aerospace Applications of Programmable Devices and Technologies Conference, September 28-30, 1999 Carmichael 99: C. Carmichael, et al, “SEU Mitigation Techniques for Virtex FPGAs in Space Applications,” Military and Aerospace Applications of Programmable Devices and Technologies Conference, September 28-30, 1999 Cronquist 98: - B Cronquist, M. Sarpa, J.J. Wang, J. McCollum, R. Katz “Modification of COTS FPGA Devices for Space Applications,” accepted for publication in the 1998 Military and Aerospace Programmable Logic Device Conference Proceedings, September 15-16, 1998, Greenbelt, MD. DTRA 99: A. Costantine, et-al, Unpublished DTRA tutorial on radiation effects in semiconductor circuits and devices, September 1999 Dressendorfer 89: P. Dressendorfer and T. Ma, “Ionizing Radiation Effects in MOS Devices and Circuits,” John Wiley, NY, 1989 Fayad 00: C. Fayad, Honeywell Failure Analysis Laboratory, private communication Fleetwood 89: D. Fleetwood, et al, “An Improved Standard Total Dose Test for CMOS Space Electronics,” IEEE TNS, 1989, p. 1967 Fuller 99 - E. Fuller, M. Caffrey, P. Blain, C. Carmichael, N. Khalsa, A. Salazar “Reconfiguration test results of the Vertex FPGA and ZBT SRAM for Space Based Reconfigurable Computing,” accepted for publication in the 1999 Military and Aerospace Programmable Logic Device Conference Proceedings, September 28-30, 1999, Laurel, MD. Gibbons-Games 99: “Use of FPGAs in Critical Space Flight Applications – A Hard Lesson,” Military and Aerospace Applications of Programmable Devices and Technologies Conference, September 28-30, 1999 GSFC 00: Various Goddard Space Flight Center Test Reports, from the Website, http:// flick.gsfc.nasa.gov/radhome.htm Johnston 00: A. Johnston, Jet Propulsion Laboratory, private communication Katz 95: R. Katz, etal, “Total Dose Response of Actel 1020B 1280A Field Programmable Gate Arrays,” IEEE TNS, RADECS 95, pop. 412-419 Katz- 97: R. Katz, etal, “Radiation Effects on Current Field Programmable Technologies,” IEEE TNS 1997, Vol. 44, No. 6, December 1997, page 1945 Katz 97a - R. Katz Programmable Logic Application Notes, March 1997 NASA, Military, and Aerospace Applications, URL http://rk.gsfc.nasa.gov/richcontent/eeelinks/eee_links/ 9703_eee.htm.

V-60

Katz 98 - R. Katz, J.J. Wang, K.A. LaBel, J. McCollum, R. Brown, B. Cronquist, S. Crain, T. Scott, W. Paolini, and B. Sin “Current Radiation Issues for Programmable Elements and Devices,” IEEE Trans. Nuc. Sci., NS-45, 2600, (1998). Katz- 99a: R. Katz, “FPGA in space Environment and Design Techniques (C0),” Military and Aerospace Applications of Programmable Devices and Technologies Conference, September 2830, 1999 Katz- 99b: R. Katz, et-al, “The Impact of Software and CAE Tools on SEU in Field Programmable Gate Arrays,” IEEE TNS 1999, Vol. 46, No. 6, December 1999, page 1461 Katz 99c: From the NASA/GSFC Programmable Logic Website, http://rk.gsfc.nasa.gov/ Katz 00: - R. Katz, NASA GSFC, Personal Correspondence Fri 2/11/00 Koga 98 - R. Koga, “Single Event Functional Interrupt (SEFI) Sensitivity in EEPROMs,” accepted for publication in the 1998 Military and Aerospace Programmable Logic Device Conference Proceedings, September 15-16, 1998, Greenbelt, MD. Lintz 00: J. Lintz, Honeywell Systems Survivability, private communication Mavis-99: D. Mavis and P. Eaton, “Temporally Redundant Latch for Preventing Single Event Disruption in Sequential Integrated Circuits,” Technical Report P8111.29 published on 8 Sep. 1998 and revised on 8 Oct. 1998. McCollum-99: J. McCollum, “Programmable Elements and Their Impact on FPGA Architecture, Performance and Radiation Hardness (B0),” Military and Aerospace Applications of Programmable Devices and Technologies Conference, September 28-30, 1999 Messenger- 86: G. Messenger and M. Ash, “The Effects of Radiation on Electronic Systems,” Van Norstrand, NY, 1986 Messenger 97: G. Messenger and M.S. Ash, “Single Event Phenomena,” Chapman & Hall, New York, 1997 Pease 90 - NSREC Short Course 1990, page 4.4-13. Petersen 97 - NSREC Short Course 1997, page III-105. Speers- 99: T. Speers, et-al, “0.25 micron Flash Memory Based FPGA for Space Applications (B6),” Military and Aerospace Applications of Programmable Devices and Technologies Conference, September 28-30, 1999 Swift 95 - G. Swift & R. Katz “An Experimental Survey of Heavy Ion Induced Dielectric Rupture in Actel Field Programmable Gate Arrays,” RADECS’95. Titus 98: J. Titus, et-al, “Effects of Ion Energy Upon Dielectric Breakdown of the Capacitor Response in Vertical Power MOSFETS,“IEEE TNS 1998,Vol. 45, No.6, December 1998, page 2492 Wang - 98: J. Wang, et-al, “Development of Total Dose Hardened Antifuse FPGA,” MAPLD 1998. Wang-99: J. Wang, et-al, “SRAM Based Reprogrammable FPGA for Space Applications,” IEEE TNS 1999, Vol. 46, No. 6, December 1999, page 1728

V-61

Wang 00 - J.J. Wang, F. Dhaoui, J. McCollum, W. Wong, B. Cronquist, R. Lambertson, E. Hamdy, R. Katz, I. Klener “Clock Frequency Dependence Of Single-Event-Transient Induced Memory Upsets,” Submitted to NSREC 2000. Wrobel 87: T. Wrobel, “On Heavy ion Induced Hard Errors in Dielectric Structures,” IEEE TNS 1987, Vol. 34, No. 6, December 1987, page 1262 Zieger 80 - “Handbook of Range distributions for Energetic Ions in All Elements”, J.F. Zieger, editor; vol. 6, Pergamon Press, New York, 1980 Acknowledgements The authors of this short course would like to acknowledge the efforts of Dr. Al Costantine and Mr. Glenn Kweder of RDA/Logicon and Ms. Martha O'Bryan/Raytheon for their assistance in the preparation of this course

V-62

APPENDIX 1 Table of Acronyms α:Si

Amorphous Silicon (fuse)

ASIC

Application Specific Integrated Circuit

ASSP

Application Specific Signal Processor

BIST

Built-in-Self-Test

CAD

Computer Aided Design

CAE

Computer Aided Engineering

CMOS

Complementary Metal Oxide Semiconductor

CP

Charge Pump

CPLD

Complex Programmable Logic Device

CSRAM

Control Static Random Access Memory

DRE

Dose-Rate-Effects

DRU

Dose-Rate-Upset

DSP

Digital Signal Processor

EDAC

Error Detection and Correction

EELV

Evolved Expendable Launch Vehicle

EEPROM

Electrically Erasable Programmable Read only Memory

EPAD

Electrically Programmable Analog Device

F/F

Flip-Flop (circuit)

FET

Field Effect Transistor

FPGA

Field Programmable Gate Array

FPSLIC

Field Programmable System Level Integrated Circuit

GOES

Geosynchronous Orbiting Environmental Satellite

GSFC

Goddard Space Flight Center

HDL

Hardware Description Language

HESSI

High Energy Solar Spectroscopic Imager

HST

Hubble Space Telescope

IC

Integrated Circuit

IC

Integrated Circuit

IP

Intellectual Property

JTAG

Joint Test Action Group

V-63

Table of Acronyms (Continued) LET

Linear Energy Transfer

LETT

Threshold LET

LPGA

Laser Programmable Gate Array

MBU

Multiple-Bit-Upset

METOP

Meteorological Operational Weather Satellite

M2M

Metal-to-Metal (fuse)

MOS

Metal Oxide Semiconductor

NASA

National Aeronautical and Space Administration

ONO

Oxide-Nitride-Oxide (fuse)

P/E

Program/Erase (cycles)

PAL

Programmable Array Logic

PLA

Programmable Logic Array

PLD

Programmable Logic Device

PROM

Programmable Read Only Memory

RILC

Radiation Induced Leakage Current

ROM

Read Only Memory

SBIRS

Space Based Infrared System

SEB

Single-Event-Burnout

SED

Single-Event-Disturb

SEDR

Single-Event-Dielectric-Rupture

SEE

Single-Event-Effects

SEFI

Single-Event-Functional-Interrupt

SEGR

Single-Event-Gate-Rupture

SEL

Single-Event-Latchup

SES

Single-Event-Snapback

SET

Single-Event-Transient

SEU

Single-Event-Upset

SIRTF

Space Infrared Telescope Facility

SOHO

Solar and Heliospheric Observatory

SoPC

System-on-a-Programmable-Chip

SOTA

State–of-the-Art

SPLD

Simple PLD

V-64

Table of Acronyms (Continued) SRAM

Static Random Access Memory

TDRS

Tracking and Data Relay Satellite

T&E

Test and Evaluation

TID

Total Ionizing Dose

TIROS

Television and Infrared Observation Satellite

TMR

Triple-Modular-Redundancy

Tox

Oxide thickness

UVPROM

Ultra-violet PROM

VHDL

VHSIC Hardware Description Language

VHSIC

Very High Speed Integrated Circuit

VIH

Input voltage-high

VIL

Input voltage-low

VOH

Output voltage-high

VOL

Output voltage-low

V-65