2004 IEEE NSREC Nuclear and Space Radiation Effects Conference Short Course Notebook
Hardness Assurance and Photonics Challenges for Space Systems
July 19, 2004 Sponsored by: IEEE/NPSS Radiation Effects Committee Supported by: Defense Threat Reduction Agency Sandia National Laboratories Air Force Research Laboratory NASA Electronic Parts and Packaging Program Jet Propulsion Laboratory Approved for public release; distribution is unlimited
2004 IEEE Nuclear and Space Radiation Effects Conference
Short Course Notebook
Hardness Assurance and Photonics Challenges for Space Systems
July 19, 2004 Atlanta, Georgia
Copyright© 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 08855-1331.
CONTENTS Introduction Biographies Section
Page
I.
Hardness Assurance for Space Systems……………………….….I-1 Gary K. Lum, Lockheed Martin Space Systems
II.
Microelectronic Piece Part Radiation Hardness Assurance for Space Systems………………………………………………….II-1 Ronald L. Pease, RLP Research
III.
Optical Sources, Fibers, and Photonic Subsystems…………….III-1 Allan H. Johnston, Jet Propulsion Laboratory
IV.
Optical Detectors and Imaging Arrays………………………….IV-1 Terrence S. Lomheim, The Aerospace Corporation
V.
Solar Cell Technologies, Modeling, and Testing………………...V-1 Robert J. Walters, Naval Research Laboratory
INTRODUCTION This Short Course Notebook contains the material prepared by the instructors, and their coauthors, for the 2004 IEEE Nuclear and Space Radiation Effects Conference (NSREC) Short Course. The course was held on July 19, 2004, in Atlanta, Georgia, and was the 25th time the NSREC provided such a course. The present notebook complements the oral presentations made by the instructors in Atlanta, and serves as a valuable technical reference and archival record for the radiation effects community. The 2004 Short Course, “Hardness Assurance and Photonics Challenges for Space Systems”, addresses two topics of primary importance for present and future space systems. Two of the course instructors discuss hardness assurance for space systems, and three instructors address radiation effects on photonics in space. Hardness assurance approaches and methodologies are implemented to assure that space systems not only survive the natural space radiation environment but also perform within specifications during the entire mission. Assuring system radiation hardness involves many technical considerations. Effective hardness assurance methods are an integral part of designing space systems to satisfy mission performance goals. The Short Course includes lectures on hardness assurance at the system level and at the electronic component level. Nearly all present and envisioned space systems include photonic devices and subsystems. Thus, it is necessary to understand space radiation effects on such components and to develop effective radiation hardening approaches as needed. Examples of photonics commonly employed are solar cells and arrays, discrete optical detectors, optical emitters, optical fibers, visible and infrared imaging arrays, and passive optical elements. The Short Course addresses radiation effects on most of those photonic components. Photonic subsystems are also considered. This Short Course Notebook is divided into five sections. A brief summary of each of the lectures presented at the 2004 Short Course follows. Biographies for the instructors are given in the next section. In Section I, “Hardness Assurance for Space Systems”, Gary Lum, Lockheed Martin Space Systems, presents a comprehensive review of the approaches and methodologies used to assure the radiation hardness of space systems. He discusses the key phases of a hardness program for space systems and describes the hardness assurance management plan. He discusses the space radiation environment and gives an overview of the key effects of that environment on electronics and on systems. System hardening approaches are addressed, and radiation testing considerations are described. Dr. Lum discusses hardness assurance implementation during the production and deployment phases of a system. He also describes emerging issues and challenges and their potential solutions for hardened space systems. In Section II, “Microelectronic Piece Part Radiation Hardness Assurance for Space Systems”, Ron Pease, RLP Research, describes hardness assurance as applied at the electronic piece-
part level. He defines key concepts and terminology and gives an overview of traditional parts hardness assurance methods and their modifications. Key hardness assurance documentation for users is identified. Parts qualification and radiation lot acceptance testing are addressed as well as exceptions and limitations in practice. Parts hardness assurance challenges for space systems are discussed and recommendations are given. In Section III, “Optical Sources, Fibers, and Photonic Subsystems”, Allan Johnston, Jet Propulsion Laboratory, addresses basic and applied aspects of space radiation effects on photonics. He discusses the physics of how photonic devices work. Radiation environments and effects of interest for photonics are summarized. He includes radiation effects on lightemitting and laser diodes and annealing behavior. Absorption effects in optical fibers are addressed as well as fiber comparisons. He also covers radiation effects on photonic subsystems for space applications, including digital optocouplers and optical receivers. In Section IV, “Optical Detectors and Imaging Arrays”, Terry Lomheim, Aerospace Corporation, discusses radiation effects on visible and infrared detectors and arrays. He describes today’s leading technologies and discusses key effects of the space radiation environment. He includes displacement damage effects, ionizing radiation effects, and radiation-induced noise in arrays, plus gives overviews of hardening approaches and technology trends. Dr. Lomheim also addresses radiation effects on readout integrated circuits. In Section V, “Solar Cell Technologies, Modeling, and Testing”, Rob Walters, Naval Research Laboratory, addresses basic and applied aspects of radiation effects on solar cells. He discusses solar-cell device physics and the mechanisms of radiation-induced degradation, and describes cell technologies for present and future applications. Dr. Walters describes and compares the leading modeling techniques used to predict solar-cell degradation in space. Simulation testing approaches are discussed, and on-orbit performance predictions are addressed. On behalf of the 2004 NSREC Committee and the greater radiation effects community, I sincerely thank the five Short Course instructors for their sustained efforts in preparing and presenting the course material. They contributed a great deal of their time and expertise to ensure the success of the 2004 Short Course. Their efforts, and the efforts of their co-authors, are greatly appreciated and will continue to be of significant benefit to the community through the publication of this notebook. The comprehensive material contained herein is expected to serve as a valuable resource for radiation effects engineers and scientists and for space system designers. I also thank Lew Cohn of DTRA for his efforts in reviewing the Short Course and ensuring that the Short Course Notebooks were printed on schedule. In addition, Dale Platteter of NAVSEA Crane deserves our thanks for making all NSREC Short Courses available on an archival disk. Joe Srour Short Course Chairman 2004 IEEE NSREC
BIOGRAPHIES Joseph R. Srour is employed in a senior engineering position at the Aerospace Corporation. Prior to joining Aerospace, he worked for TRW where he managed the Radiation and Survivability Engineering organization. Before TRW, he worked for the Northrop Corporation where he held various technical and managerial positions. Much of his technical work has focused on nuclear and space radiation effects on materials, devices, circuits, and systems. He has also made technical contributions in the areas of optical detectors, semiconductor device physics, and microelectronics. Joe is a Fellow of the IEEE and is a member of Sigma Xi and Tau Beta Pi. He is the author of one technical book and 49 articles published in refereed technical journals. He received the Outstanding Paper Award six times for papers presented at the IEEE Nuclear and Space Radiation Effects Conference, and received the Meritorious Paper Award twice for papers presented at that same conference. He holds two U.S. patents. Joe received bachelors, masters, and Ph.D. degrees in electrical engineering from the Catholic University of America, Washington, DC. Gary K. Lum received a B.A. in physics at the University of California, Berkeley and M.S and Ph.D. in physics at the University of Oregon. He was a graduate student under Dr. C. Wiegand and Prof. E. Segré (Nobel Laureate) at Lawrence Berkeley National Laboratory. After joining Lockheed Missiles System Division in 1980, Gary headed the radiation effects analysis group. He joined Intel Corporation in 1984 to work as a device physicist. In 1986, he returned to Lockheed where his areas of study included IC fabrication processes, modeling of CMOS and bipolar technologies, and radiation effects in semiconductor devices. In 1988, he received the AIAA award for Best Design Engineer. Gary has published over 20 technical papers. He has served in various technical and management positions for the IEEE Nuclear and Space Radiation Effects Conference and the Hardened Electronics and Radiation Technology Conference, and serves as a technical paper reviewer for both conferences. At Lockheed Martin, he provides recommendations and technical guidance to designers, program managers and customers. Presently, he is an Engineering Fellow supporting space programs by providing training and technical guidance in parts selection and in the design of hardened systems. He lectures at Stanford University and also conducts studies to understand radiation effects on electronics and to mitigate those effects in satellite and missile applications. Ronald L. Pease received the B. S. in Physics from Indiana University in 1965 and pursued graduate studies in Physics at the University of Washington in 1966. He has been active in radiation effects characterization, modeling, analysis and hardness assurance for 38 years, having worked at NAVSEA Crane (1966-1977), BDM (1977-1979), and Mission Research Corp. (1979-1993). Mr. Ron Pease is the president and sole employee of RLP Research, which was formed in 1993. He is a technical advisor and senior scientist on several DoD contracts that address radiation response and hardness assurance, the most recent being in the areas of Enhanced Low Dose Rate Sensitivity and Single Event Transients in bipolar linear circuits. Mr. Pease is very active in the IEEE NPSS having held every technical position for the Nuclear and Space Radiation Effects Conference including serving as Conference Chairman in 2000. He has served as a Short Course Instructor and Short Course Chairman
for the NSREC, as well as a short course instructor at the Nuclear Science Symposium, the Commercialization of Military and Space Electronics and Vanderbilt University. He has over 90 technical publications in the area of radiation effects in electronics and has received several NSREC Outstanding and Meritorious Paper Awards, the most recent being the Outstanding Conference Paper Awards in 2002 and 2003. Allan H. Johnston received B.S. and M.S. degrees in physics from the University of Washington. He began his career at Boeing Aerospace Corporation, performing research studies on radiation effects in microelectronics and optoelectronics. He joined the Jet Propulsion Laboratory in 1992 where he supervises applied research on radiation effects in microelectronics for space applications. His technical interests include ionization and singleevent upset effects in semiconductor devices, with emphasis on low-dose-rate effects, latchup, and space applications of advanced technologies. Related interests include determining how new device technologies and device scaling will influence their radiation performance and reliability in space as well as radiation effects on optoelectronics. He has authored more than 80 papers in refereed journals. He received the Outstanding Paper Award at the IEEE Nuclear and Space Radiation Effects Conference (NSREC) in 1999, Meritorious Paper Awards in 1995 and 1996, and the Distinguished Poster Paper Award in 1987. He has been active in the IEEE NSREC, serving as Short Course Instructor for four conferences, Local Arrangements Chairman, Short Course Chairman, and Awards Chairman. He was Technical Program Chairman for the 1997 NSREC and General Chairman for the 2003 NSREC. He is a Fellow of the IEEE. Terrence S. Lomheim is a Distinguished Engineer in the Sensor Systems Subdivision, The Aerospace Corporation, El Segundo, California, where he has worked since 1978. He received a Ph.D. in Physics from the University of Southern California in 1978. He has performed detailed experimental evaluation of the electro-optical properties, imaging performance capabilities, and radiation effects sensitivities of visible scanning and staring CCD and CMOS devices and hybrid infrared focal planes for a variety of DoD and Civil Programs. He has also been involved in the design, performance assessment, modeling and diagnostics of point-source detection, broadband, multispectral, and hyperspectral imaging electro-optical sensor systems in the visible through longwave spectral regions. Dr. Lomheim has authored and co-authored 38 publications in the areas of applied optics, focal plane technology, and imaging sensor performance and has been a part-time instructor in the physics department at the California State University, Dominquez Hills since 1981. He is a Fellow of The International Society for Optical Engineering (SPIE) and is a member of the Optical Society of American and the American Physical Society. Robert J. Walters received his Ph.D. in Applied Physics from the University of Maryland Baltimore County in 1994. He has worked at the US Naval Research Laboratory since 1991. His area of expertise is in radiation effects in semiconductor materials and devices, and his primary area of focus is radiation effects in solar cells for space applications. His research group has produced a new technique for modeling the effect of irradiation on semiconductor devices, which has gained international acceptance. His group has also produced groundbreaking work on new space solar cell technologies, and they are currently building a solar cell space experiment to be flown on the International Space Station. In addition to
space solar cell research, Dr. Walters is also directing a project to develop advanced photovoltaic devices for micro-power systems. The end product will be a self-powered optical data link for use in a distributed autonomous sensor system. Dr. Walters lives in Alexandria, VA with his beautiful wife, PJ, and two wonderful daughters, Sarah who is 13, and Molly who is 8.
2004 IEEE NSREC Short Course
Section I
Hardness Assurance for Space Systems
Gary K. Lum Lockheed Martin Space Systems
Hardness Assurance for Space Systems Gary Lum, PhD., Lockheed Martin Space Systems Company, Sunnyvale, CA Marion Rose, Titan Corp., San Diego, CA Table of Contents 1.0 Introduction 1.1 Perspective 2.0 System Hardness Program 2.1 Hardness Assurance Management Plan 3.0 Requirements Generation 4.0 Program Initiation Phase 4.1 Space Radiation Environments 4.1.1 Composition of the Space Environment 4.1.2 Trapped Protons Dose 4.1.3 Trapped Electron Dose 4.1.4 Solar Cycles 4.1.5 Shielding Against Solar & Galactic Environments 4.1.6 Environment Highlights 4.2 Radiation Requirements 4.3 System Radiation Effects 4.4 System Strategies for Reducing Radiation Effects 4.5 System Hardening Approaches 4.6 Space Environment Modeling Tools 4.7 Shielding Example 4.8 Upset Rate / Risk Example 5.0 Hardening Against Protons 6.0 Key Integrated Circuit / Semiconductor Degraded Parameters & Effects 7.0 Total Ionizing Dose Effects 7.1 MOS Devices 7.2 Bipolar Devices 8.0 Proton Displacement Damage 8.1 Gain Degradation 9.0 Displacement Damage in Other Technologies and System Solutions 10.0 Comparison of Commercial versus Hardened Technologies 11.0 Radiation Test Facilities 12.0 Production / Deployment Phase 12.1 System Hardness Assurance 12.2 Maintenance / Surveillance Program 12.3 Types of Hardness Degradation 12.4 Implementing Hardness Assurance 12.5 Classifying Parts in a Space Program 12.6 Parts Screening Control 12.7 Document Flow Through HA 12.8 Manufacturing Support 13.0 Emerging Issues 13.1 Future Radiation Trends in Electronic Technologies 13.2 Single Event Upset, Transients, Latch-up 13.3 Extremely Low Dose Rate Sensitivity 13.4 SOI Future Technology 14.0 Conclusion 15.0 Acknowledgements 16.0 References
I-1
Page I-2 I-2 I-4 I-5 I-6 I-7 I-7 I-7 I-9 I-10 I-11 I-12 I-14 I-15 I-16 I-17 I-18 I-20 I-21 I-23 I-25 I-26 I-27 I-27 I-29 I-29 I-30 I-31 I-34 I-34 I-35 I-35 I-37 I-37 I-38 I-39 I-40 I-41 I-41 I-42 I-44 I-44 I-50 I-51 I-52 I-54 I-54
1.0 Introduction Within a space program system hardness assurance (HA) plays a very important function. Although HA plays a major role at the parts level, HA at higher assembly levels, at the board, subassembly, package and system must be addressed. When radiation requirements are imposed on a system, an engineer must design the board or package to meet those requirements. Hardening designs must be traded against cost, shielding analysis performed against weight, parts tested for radiation hardness and then results documented. However, the job in the survivability community is only partially complete. Transitioning a developed product to manufacturing is the next step and that is not a simple task. One has to determine how to transition the design and work that has been accomplished to the manufacturing floor. Along the way if the circuit design, procurement of components or fabrication of a board were not well controlled or the translation was poorly interpreted by the assembler, margins achieved in the original design may not be ever manufactured. Hardness assurance provide the guidelines or rules that are established to assure that the final product will be what was designed and still maintain its original radiation design margins. Within a survivability program there has to be a hardness assurance plan. An overview will be presented of what this plan covers. Because HA involves hardening a design to a set of space environments, we will briefly review the types of radiation environments that various space systems have to encounter and understand the kinds of effects that can affect a system and what hardening techniques a designer can use. Radiation codes for analyses and radiation test facilities for testing will be described. An important section of system design is also the selection of parts. We will show how parts are selected and based on their hardness, how they must be maintained and monitored during the life of a program. Finally, some future issues will be presented. In light of the fact that we are designing systems better, faster and cheaper, we need to continue to find alternative ways of making sure that we can maintain the hardness of our systems. These are challenges that we have to face in the future. The first twelve references are for general reading on the subject of radiation [1-12] followed by references on the discussed topics. References [141-195] are military specifications, handbooks, test methods, and standards. 1.1 Perspective As one can imagine, building systems according to Fig. 1 to fly in space successfully is very expensive. Imagine the amount of time and dollars that goes into a design, assembly and qualification of systems, such as the Hubble Space Telescope, the Space Transportation System (Space Shuttle), the International Space Station or a launch system. Tens to hundreds of millions of dollars are spent and 5 to 15 years or more are spent developing, assembling, testing, launching and maintaining these space systems. If a system were to be lost during launch, the payload lose control during flight, the payload lose communication due to a misinterpretation of information due to poor documentation by the engineer to the assembler or manufacturer or poor quality control of the parts that go into the system, the prime contractor could lose incentives and profit from the contract. Setbacks could occur.
I-2
If the systems were similar to the disasters of the Columbia or Challenger space shuttle incidents, setbacks in schedules would occur. If the problem were similar to the optical flaw in the Hubble Space Telescope, valuable scientific data would be lost because time had to be spent developing a solution to the problem. Until the cause is identified and a solution is found, large costs are incurred. • Development of a system is expensive • Challenges - Advanced electronics, Plastic Encapsulated Microelectronics (PEMs), COTS • Failure of a system in space results in loss of data, profit, and large program setbacks • Loss in 1998 was $1.6B during the advocation of faster, better and cheaper systems • Loss totaled $597M in 1999
International Space Station
Space Shuttle Hubble Space Telescope
• System Hardness Assurance is the methodology/discipline of assuring that the production processes do not adversely reduce the margins designed into our Space Systems during development. Figure 1. Perspective - Importance of Radiation System Hardness Assurance.
Radiation hardness assurance is defined as a piece of reliability for which the radiation experts are responsible. Their responsibility is to assure that the system they design has not lost its margins when going into manufacturing. This degradation may be the result of not monitoring a process change in a part or the result of not providing adequate design margins to account for radiation degradation. This implies that the role of radiation experts in the design and manufacture of a space system is to demonstrate that the methodology or discipline inherent in the parts used, the boards and packages one tested, analyzed and assembled or the system they integrate will still maintain the design margins during manufacturing and throughout the entire life of a program. The methodology or discipline will be described in this short course. In 1998 the space industry lost $1.6B, when there was a strong avocation to develop space systems faster, better and cheaper. The Mexican satellite, Solidaridad 1 failed on orbit. The loss per vehicle was around $50M or higher which is too many dollars. In 1999 about $611M was lost in the satellite industry due to various problems. A number of these problems occurred during launch. Whether or not this was the complete list of failures during that year, changes occurred to minimize many of the failures that were observed during 1998, the previous year. The message here is that even though failures may not be all associated with radiation hardness assurance issues, these could easily apply to us if we do not adhere to strict hardness assurance disciplines [16].
I-3
2.0 System Hardness Program A system hardness program has two important phases as shown in Fig. 2. They are (a) Design and Development and (b) Manufacturing and Production. During the Design and Development phase, a program contract is defined, describing the type of space program, its mission objectives and the contractual requirements imposed by the customer on the contractor. Along with the program requirements, a system survivability program plan is defined. This may contain the radiation specifications that the system must meet, the transported radiation requirements at the parts and a system compliance document describing what types of radiation testing or analysis must be performed in order to satisfy compliance at the package or subassembly levels. At the system level compliance may not be a requirement because testing may not be achievable or may be cost prohibitive. Generally, testing or analyses performed at the lower levels, such as parts, boards and packages are applied to assess the hardness of the system. Manufacturing / Design / Development Phase Info / Production Phase Technology Mission Objectives
System Survivability Program
Transfer
Radiation Environments (free field) Hardness Compliance Plan Defines test / analysis approaches for compliance of packages, system
Hardness Assurance
Hardness Surveillance
Radiation Requirements (transported)
Hardness Characterization and Qualification
Hardened Design
Figure 2. System Hardness Program.
During the Design and Development phase various tradeoff concepts are performed. These concepts are important for optimizing cost and performance. The selection of candidate process and design technologies, such as CMOS versus bipolar or FPGAs (Field Programmable Gate Array) versus ASICs (Application Specific Integrated Circuit), are evaluated. Sample products from various suppliers are radiation tested. Cost tradeoff is made between procuring hardened versus radiation tolerant parts. Schedules are laid out to determine when parts should arrive versus when boards and packages can be assembled for evaluation. An approach may be to invest the time to design by applying a foundry’s layout rules to harden a design [132], called “Hardened by Design,” versus procuring radiation hardened parts. Radiation testing and documentation must be performed to qualify such a part. Parts engineering must determine eventually how to establish monitoring controls with the manufacturer in order for parts to be procured for the program. Going into Manufacturing and Production the circuit designs are fairly established. Parts procured during the development phase may have consisted of a small sample size and I-4
originated from a particular lot. Now if one had to procure larger quantities for manufacturing or production, it is possible that the parts to be ordered going into manufacturing may not be the same pedigree as those tested during development. Therefore lot sampling testing or life of buys from the same lot that the smaller samples originated must be considered. Based on lots tested during manufacturing, statistical analysis must be performed to determine how margins have changed as compared to those measured during development. Surveillance testing may also be required for parts with small design margins or critical to the system performance. More will be described below about what hardness assurance practices are needed during manufacturing. 2.1 Hardness Assurance Management Plan A typical Hardness Assurance (HA) management plan is described below. In Fig. 3 HA has a role in every facet of a program lifecycle as described above: (a) Program Initiation sometimes also referred to as Demonstration / validation (Demo/val), (b) Full Scale Development (FSD), and (c) Production [125, 126]. The activities that occur during the Program Initiation include the establishment of the radiation requirements, cost tradeoffs, evaluation of candidate process technologies and the establishment of hardening techniques and design concepts. Typically, this period spans 2-5 years. Program Life Cycle Program Initiation Phase (2-5 yrs)
Full Scale Development Phase (4-7 yrs)
Production Phase (5-10 yrs)
Engineering Radiation Requirements Cost/design trade offs Hardening techniques Implement techniques Board design Package design Parts procurement/qualification Testing & Evaluation - Radiation testing and analyses Test facility usage Rad code usage - Parts boards package compliance Hardness Assurance Plan • Procurement procedures • Lot sampling • Hardening procedures • Qualification documents • Lot sampling testing
• • • •
Configuration Mgt. Production Quality Assurance Program Control
Figure 3. Hardness Assurance Management Plan.
During FSD, designers focus on their selected architecture design and begin their concept demonstrations. Typically, this period spans about 4-7 years. Non-hardened form-fit and functional parts may be used in certain small scale circuits to proof out the design. FPGAs may be used to rapidly design a complex device for electrical performance evaluation. The parts procurement process starts. During the latter half of the FSD program, some of the radiation hardened parts are procured. Designs on FPGAs are transferred to a hardened process. Typically, the integrated circuit fabrication, radiation evaluation and board assembly will take 2-3 years. This assumes a minimum of two fabrication runs. When prototype hardened parts arrive during the middle of FSD, new boards are built. Board test I-5
plans are created. Once several boards or packages are completed with the first prototype radiation hardened parts, engineering will evaluate the electrical performance and radiation hardness. In the final years prior to production, package or subassembly testing may be conducted. Here refinements are made in the designs. After compliance has been completed, designs are pretty frozen, results are documented and guidelines are being established for the manufacturing of the packages or systems. The results of the analyses and testing are documented and guidelines are established as to how components will be procured and qualified. HA needs to manage the guidelines for manufacturing of the final system. HA is so important here. HA must determine that the radiation design margins (RDMs) have not degraded or changed from the development period until the start of manufacturing. A flight experiment may also be part of the Development phase for a satellite or launcher program to demonstrate design concepts suitable for space application. Generally, this flight experiment is used as a risk mitigation approach. If considered, it should be as early as possible in the program in order to minimize cost in design changes. Certainly, the parts selected must be hardened to the space environment. During the Production phase, HA has inputs to provide in establishing the procedures for lot procurement, lot sampling and surveillance testing. Hardness procedures and guidelines have to be established for transferring to manufacturing. Any changes or deviations in the manufacturing cycle must be carefully monitored and reviewed closely through a control board process [ 127, 129, 130, 131]. This control board process is what we call configuration management or managing the configuration. Changes must require many organizations of various responsibilities to now approve the modification. They may include disciplines from reliability, safety, survivability, procurement, subcontracts, system engineering, packaging, manufacturing, and support equipment. 3.0 Requirements Generation Radiation requirements for satellite programs are created in the following way as shown in Fig. 4. A mission objective is defined in a document that may be contractual, that is, if a problem occurs during the development phase, the prime contractor may be obligated to fix it. This may be a scientific or telecommunication mission. From the mission objective, orbital parameters are obtained, such as the launch time, location, altitude, inclination and mission duration. These parameters are inputs to a space environment code that will determine the proton, electron, and heavy ion flux or fluence spectra for the trapped radiation belt, the solar event activities and the galactic cosmic environment over the entire mission. The frequency and duration of these environments are also defined. This may also include meteoroid or orbital debris occurrences or the outgassing from surrounding system materials. A requirement that addresses spacecraft charging may require certain selections of insulating materials or a method for bleeding charge from the system. [12] Calculations are performed to transport the free field environments through a typical package shield to determine the radiation levels at the electronics. Ray-tracing analyses are important for determining the dose upon a part. For many low cost designs, 1D transport modeling is performed. However, this can be an overestimate of the actual dose, so 2D or 3D ray-tracing analyses are performed to refine the dose requirements. These latter dose analyses account for board and package materials, and other adjacent packages
I-6
on the satellite. When these radiation levels are generated, they may be traded against several parameters: (a) cost of a system, (b) testability, (c) weight, (d) technologies for the system and (e) performance of the system. Technologies include hardened versus radiation tolerant or non hardened electronics. These latter parameters are important in driving the cost of a program and the complexity of a system design. The fewer the number of components, the higher reliability is the system. A super hardened system can be made, but cost or affordability may be so high that it can break the bank. Therefore, tradeoff studies are performed between cost and the radiation requirements until an agreeable median is found.
SPACECRAFT MISSION PARAMETERS • Orbital trajectory • Mission duration • Launch period
• Technology capability • Testability • Cost
• • • •
• • • •
SPACE RADIATION ENVIRONMENTS Trapped protons Trapped electrons Solar events Galactic cosmic
REQUIREMENTS Proton fluence Heavy ion LET Total ionizing dose Flare frequency, duration
Steps involved ... • Obtain orbital parameters, mission duration, launch time • Develop space environments, requirements • Conduct trade studies Figure 4. Generation of Space Requirements.
4.0 Program Initiation Phase During the program initiation phase the space environment specifications should be defined as one of the key documents going into a contract. It is important to have this document ready because it takes time to develop this material. Let us review some of the key characteristics of this environment. 4.1 Space Radiation Environment 4.1.1 Composition of the Space Environment The natural space environment is described by four important contributions as shown in Fig. 5 when we are dealing with the development of a space system. The top three will be the focus of this short course [13,14]. They are (a) the radiation from the Van Allen or trapped radiation belts, (b) the solar particle events (SPE) and (c) the galactic cosmic particles. The fourth environment consists of atmospheric neutrons that affect systems in the atmosphere or at ground level. The solar particle events can be described in several forms. The system designer considers the two types of solar flare events: the 90% worst case flare and what is sometimes called an Anomalous Large Solar flare are generally in one’s environment specification. Another type of solar event is called a CME or Coronal Mass Ejection. These first three environments are the primary radiation environments I-7
that a satellite system may have to deal with. For low altitude aircraft or ground systems, such systems have to deal with the atmospheric neutrons. These neutrons arise from the interaction of the galactic cosmic radiation with the atmospheric molecules. The population of these neutrons is greatest over the polar caps, while dropping by a factor of 2-3 over the equator.
• Trapped Radiation Belts (Van Allen) • Solar particle events – 90% worst case flare – Anomalous large solar flare – CMEs (Coronal Mass Ejections) • Galactic Cosmic Rays • Atmospheric neutrons Figure 5. Composition of the Natural Space Environment.
The earth has a magnetic core that is dynamic, constantly in motion. Magnetic field lines emanate from the north pole and reenters from the south. The field lines are compressed by the solar wind on the side facing the sun, expanding on the opposite side away from the sun several earth radii out and focusing down around the north and south poles. Some of the protons and electrons from the solar wind will be trapped by the geomagnetic fields of the earth. These particles will spiral along the earth's magnetic field lines entering from the north and south poles. With the right energy and momentum, the protons and electrons will spiral back and forth between the north and south poles. Since the protons and electrons have opposite charges, they also have a lateral velocity, such that protons and electrons drift in eastward and westward directions around the earth. Hence, this is how the radiation belts are formed. The two trapped radiation belts that girdle the earth are the inner and the outer belts. Protons and electrons from the sun constantly replenish the particles in the radiation belt. Without an iron core, there would be no magnetic field and we would constantly be exposed to the radiation from the sun. System designers take advantage of the geomagnetic shielding. For example, the Space Shuttle orbits beneath the geomagnetic field in order to optimize protection from extraneous exposure to the proton radiation. Likewise, commercial satellites orbit beneath the belts to minimize exposure to the radiation. By doing this, they can leverage on Commercial-Off-The-Shelf (COTS) electronics that may be sensitive to radiation. Depending on the type of mission, telecommunication, scientific experiments or surveillance, the orbit will determine the type of environments the satellite will see over I-8
the entire mission. LEO (low earth orbit, MEO (medium earth orbit) and GEO (geosynchronous earth orbit) are affected by the trapped radiation, from either the lower or the upper radiation belts. Solar events that emanate from the sun consist of protons and heavy ions. The protons have energies much less than those from the galactic cosmic environment. The solar wind carries these particles towards the earth and constantly replenishes the protons and electrons in the belt. These particles affect satellites that orbit over the north and south poles and for satellites that are far away from the earth's geomagnetic fields. These satellites could be planetary satellites or ones at the Langragian L1, L2 points in space. The galactic cosmic particles originate from outer space, from other stellar events. The galactic cosmic environment consists of protons and heavy ions of much higher energies than those coming from the sun. The galactic cosmic particles affect all orbits, especially those that fly over the poles or are outside the earth's geomagnetic field. Finally, the galactic cosmic particles interact with the atmospheric atoms, oxygen and nitrogen atoms creating neutrons in the reactions. Neutrons can penetrate into our atmosphere, effecting our aircrafts and ground equipment. One can learn more about the space environment today with monitors on several satellites [17, 18, 19]. If we look at the geomagnetic field lines more closely, the magnetic field is really not aligned with the rotational axis of the earth, but is tilted by 11 degrees with respect to the rotational axis of the earth and is skewed by about 200 km inward above Rio de Janeiro, Brazil. Here the magnetic field is weakest such that protons can penetrate to low altitudes, such as 300-400 km. On the opposite side, one would expect that the belt would be higher. Because the magnetic core is within the earth, the magnetic field lines are stationary with the earth. Hence, the belts rotate with the earth’s rotation. Over Rio de Janeiro, folks refer to this region as the South Atlantic Anomaly (SAA) region. Most satellites orbiting at low altitudes take advantage of the geomagnetic shielding and deal only with the protons in the SAA. We refer to these satellites as LEO or (Low Earth Orbit) satellites. As mentioned earlier many LEO satellites leverage on COTS technologies, giving them high performance, lower power and higher speed. However, because of the high sensitivity to radiation, care must be taken in the system design [20, 21]. Also to minimize the cost of hardening 2D or 3D ray tracing analysis are needed to determine whether COTS technologies may even be implemented with good design margins. If we look at the SAA region more closely, we find that these SAA protons are only located in a very small region of the earth. Hence, much of the LEO orbits that traverse this region may be 3-4 orbits out of 14 revolutions per day. 4.1.2 Trapped Proton Dose Between 800 to 4000 km the geomagnetic fields are strongest. The magnetic field lines are weakest around Brazil. As altitude increases, the proton isocontours begin to increase spreading around the earth, eventually forming the entire belt. The belts span approximately ±30-40° latitude. As one can see, by designing a satellite that skirts beneath the belt, one can avoid much of the space radiation and have to only deal with the radiation during a small fraction of the time. For LEO satellites with orbits within ±30°
I-9
latitude, only the SAA and possibly the galactic cosmic particles are of concern. For LEOs that have an inclination angle above 30°, the SAA, galactic cosmic and solar events over the poles must be considered. Minimizing the amount of radiation allows the designer to leverage on COTS technologies. The Space Transportation Shuttle (STS) and Hubble Space Telescope (HST) have LEO orbits. The space shuttle takes advantage of this belt to help minimize any unnecessary radiation exposure to the crew during a solar event.
Total Dose [rad(Si)/yr]
Dose from trapped protons for circular orbits as a function of altitude and inclination. GEO does not see any trapped protons. 350-8000 km, 30 deg. inclination 2000 km, 0 deg inclination 2000 km, 60 deg inclination 2000 km, 90 deg inclination
1x105 8x104 6x104 4
4x10
2x104
LEO SAA protons
0
MEO 0
2000
4000
6000
8000
Altitude [km] Figure 6. Trapped Protons Dose Versus Altitude and Inclination.
Trapped protons are located at altitudes between 400 km up to around 9000 km as shown in Fig. 6. The SAA is located between 300 km to about 1000 km. Above 1000 km the belt is formed. As one can see the LEOs are between 400 to about 800 km. MEO (Medium Earth Orbit) lie in the heart of the belt where the satellite would be exposed constantly to the ionizing protons and electrons. Hardening is very difficult in this vicinity. At GEO, 35,000 km, the protons are gone. However, in Fig. 7 a GEO (Geosynchronous Orbit) satellite will have to face the outer electrons. 4.1.3 Trapped Electron Dose The location of the electrons is shown in the lower left plot in terms of dose versus altitude. We see that for LEO orbits, the protons really dominate. For the MEO and GEO satellites, the inner and outer electron belts dominate. They are represented by the two humps in the curve of Fig. 7 [22]. The external dose on a spacecraft depends on the location of the satellite and the internal dose depends on the material. For example, at LEO, the total dose for a ten year mission behind 200 mil of aluminum shield is around 20 krad(Si). For MEOs at 1000 to 3000 km, the dose can be as high as 600 krad per 10 yrs behind 200 mil of aluminum. For GEO at 35,000 km the dose is 60 krad(Si) for 10 years with 200 mil of aluminum shield. With thinner aluminum shields, the dose will increase.
I-10
106 10-year electron dose behind 200 mils Al shielding
Protons (AP-8)
Electrons (AE-8)
104
GEO
MEO
105 105 104 103 102
LEO
Dose, rad(Si), Equivalent Al
Orbital altitudes significantly change the dose Van Allen Trapped Belts • LEO (550 – 1000 km) - 20 krad (Si) • MEO (1000 – 3000 km) - 600 krad (Si) • GEO (36,000 km) - 60 krad (Si)
103
102
103
104 105 4 3 2 1 0 1 2 3 4 5 6 7 8 9 10 Altitude (km) L- Shell (1 L-shell = 6370 km = 1 earth radius) Figure 7. Electron Dose Versus Altitude.
4.1.4 Solar Cycles Solar event activities follow a typical solar cycle that lasts for about 11 years as shown in Fig. 8. A system designer typically considers 4 years of low solar activities and about 7 years of active time from the sun. The galactic cosmic environment tracks in the opposite way. During the 4 years of low solar activities, the galactic cosmic environment is high. A satellite designer with a 5 to 7 year mission, might want to launch during this benign period in order to take advantage of the low radiation exposure. Fig. 8 shows a plot of the solar activities taken over three solar cycles. Each solar cycle is quite different with the largest solar event in terms of the proton fluence and energy. As one can see many of the 8 2 events have proton fluences about 10 to 109 protons/cm for protons that have energies greater than 10 and 30 MeV, respectively [23, 24]. 10
2
The largest solar events have protons as high as 1x10 protons/cm . The August 1972 and October 1989 flares were the large ones that were observed. They represent the Anomalous Large Solar (ALS) flares. The purple outline shows the number of sunspots as observed on the sun. There appears to be a correlation between the number of sunspots and the number of flare activities. The number of events and the number of occurrences of these protons versus fluence are plotted on a probability graph, one finds that they will lie fairly well on a linear line [25, 9 2 26, 27]. The 90% worst case events are represented by the 10 protons/cm occurrences. For the ALS flare, this falls in the 99.9% bracket. Such a flare has chance of 1 in 1000 to occur during a mission lifetime. The general design requirements are to account for the 90% worst case flares. That is, there is a 90% chance over the entire mission that the 9 2 satellite will be exposed to a flare of 10 protons/cm in fluence. For a GEO mission that may last for 15 years, one solar cycle needs to be accounted for such that, one ALS flare must be included in the environment specification. I-11
Solar max ~ 7yrs Solar min ~ 4yrs
Sunspot Number
Fluence (No. of protons/cm2)
Solar sunspots and solar activities during solar cycles 20, 21, and 22. August 1972 flare October 1989 flare
• August 1972 and October 1989 solar events (anomalous large solar flares, ALSF) are the largest ever observed over three solar cycles • Solar events of 109 protons/cm2 or less are considered the 90% worst case condition J. Barth, NASA/GSFC, 1997 IEEE Nuclear and Radiation Effects Conference Short Course, pg I-58.
Figure 8. Solar Cycles 20, 21 and 22 (typically 11 years).
Solar flares can last for several days. The largest solar flare, the October 1989 event, lasted for at least 8 days. Other lower intensity events lasted for a day or two. This is important to note when designing any memory refreshing or circumvention techniques into the system. Since the number of refreshes to memory can consume a large amount of computing time, a system engineer must factor this into one’s system maintenance plan in order to minimize the impact to the scientific data collection of a satellite. If an ALS flare were known to occur during a mission, the system engineer may have to scrub system only during the event. Recently studies show that there are two types of solar event activities. A solar flare is one that contains only protons of energies up to 600 - 700 MeV with little heavy ions and CMEs (Coronal Mass Ejections) contain electrons and heavy ions [28, 29, 31]. CMEs are also associated with disturbing the earth's magnetic field in terms of geomagnetic disturbances [30]. There are websites that illustrate these solar event activities. Today’s scientific missions with satellites and the electronics that are taking these pictures provide a marvelous source of scientific data. There are movies that show some of the solar event activities as detected by the SOHO project. Some of these movies were shown by Joe Mazur of Aerospace Corp in his 2002 short course. 4.1.5 Shielding Against Solar and Galactic Environments Fig. 9 illustrates the effect of shielding against the ions from a CME at GEO. This is a plot of the integral flux as a function of LET (Linear Energy Transfer) of the ions. LET is the amount of energy that is deposited per unit length by an ion. This integral curve is a composite of ion flux starting with protons to alphas, oxygen, silicon, nitrogen and ions up to iron. The curve continues with much lower ion intensities above iron to a LET of 2 100 MeV-cm /mg. The lower ion intensities contain heavier ions, such as zirconium,
I-12
barium, platinum and lead. Note that by going from 100 mil to 3500 mil of equivalent aluminum shielding, one can attenuate the intensity of these solar ions by at least two orders of magnitude.
Integral Flux [ions/(m2-sr-s)]
At a LET of about 30 MeV-cm2/mg, the flux drops off dramatically • LET (Linear Energy Transfer) defined as amount of energy an ion can deposit along a unit path length
105 June 91 spectrum 104 GEO, Z = 1 - 92 3 10 2 10 100 mil 101 200 mil 0 10 500 mil protons 10-1 1000 mil Fe -2 10 2000 mil 10-3 3500 mil -4 10 10-5 alphas, O, Si, Ni 10-6 3.5-in. equivalent -7 10 -8 aluminum can 10 10-9 CHIME Zr, Ba, U reduce flux 10-10 -1 intensity by about 10 100 101 102 2 orders of LET [MeV-cm2/mg] magnitude
Figure 9. Combined Flare and CME Flux Spectra Versus LET and Shielding. Cosmic ray particles are much less intense, but extremely more energetic than solar ions. 3.5 inches of equivalent aluminum can reduce flux intensity by only 1 order of magnitude as compared to 2 orders of magnitude for solar particles. Cosmic Ray Integral Spectrum Integral Flux [ions/(m2-sr-s)]
Solar minimum, GEO, Z = 1 - 92
101 100 10-1 protons 10-2 Fe 10-3 10-4 10-5 alphas, O, Si, Ni 10-6 10-7 10-8 10-9 10-10 Zr, Ba, U 10-11 -12 10 Space Radiation 5.0 10-13 -1 10 100 101 102
100 mil 200 mil 500 mil 1000 mil 2000 mil 3500 mil
LET [MeV-cm2/mg] Figure 10. Effect of Shielding on Cosmic Rays: LET Spectrum.
The galactic cosmic particles consist of ions ranging from protons to uranium ions [32]. The most abundant element is hydrogen, followed by helium and on down to iron before the abundance drops dramatically. The ion energies of the solar events are also much lower than the energies from the galactic cosmic environment. The galactic cosmic ions have energies in the GeV range. It is postulated that these ions arrive from distant stars, supernovae or from the “Big Bang.” However, the fluxes of these particles are much lower than the fluxes coming from solar events. The galactic cosmic flux is about 3-4 orders of magnitude lower. For example, the flux for protons from a solar event is about 106
I-13
protons/(m2-sr-s), while the galactic cosmic flux is 101 protons/(m2-sr-s). The galactic cosmic environment is most active during solar minimum or during the 4 years of minimum solar activity. Since the galactic cosmic energies are much higher in the GeV range than those from the sun, stopping these particles is almost impossible. Fig. 10 is similar to Fig. 9 showing the integral flux of the galactic cosmic ions versus their LET or stopping power at GEO. By going from 100 mil to 3500 mil of equivalent aluminum shielding, the intensity of the galactic cosmic flux only diminishes by 1 order of magnitude as compared to the particles from the sun which dropped by at least 2 orders of magnitude. Mitigating single event effects (SEE) from the galactic cosmic environment is generally done by other means other than shielding. Trapped Radiation (Van Allen belt) - protons, electrons • Trapped belts - MEO • South Atlantic Anomaly region - LEO – Geomagnetic axis tilted by 11° from rotational axis – Skewed by 200 km, intensity highest over Rio de Janeiro, Brazil LEO
Solar flares - protons, heavy ions, electrons • LEO, MEO, GEO • 11 yr solar cycle (7 active yrs) – LETs between 1 - 100 MeV-cm2/mg – flux ≤ 106 ions/(m2-sr-s) 90% worst case Aug. 72, Oct. 89 ~ 8 days CMEs (Coronal Mass Ejections) Galactic Cosmic Rays - protons, heavy ions • LEO, MEO, GEO • LETs between 1 - 100 MeV-cm2/mg • flux ≤ 10 ions/(m2-sr-s)
Solar Particle Events
Solar Cycles 20, 21, 22
Figure 11. Space Environment Highlights.
4.1.6 Environment Highlights Fig. 11 summarizes the characteristics of the space environment that LEO, MEO and GEO space systems may have to face as a requirement. In the trapped radiation belts, protons and electrons are the dominant contributors between 300-800 km. The protons are from a region where the magnetic field is weakest. This region, called the South Atlantic Anomaly region is located over Rio de Janeiro, Brazil. The weakness of the magnetic field comes about from a 11° tilt of the geomagnetic axis with the earth's rotational axis. The geomagnetic axis is also skewed such that it is 200 km closer over Brazil. At higher altitudes, the SAA spreads and merges into the belt itself. LEO satellites orbiting within ±30° latitude have the geomagnetic field to protect the spacecraft and only have to tend to the radiation from the SAA. At higher altitudes, the second belt contains predominantly electrons. GEO satellites are affected by total ionizing dose. Solar events occur predominantly during 7 years of a 11 year solar cycle. The particles primarily range from protons up to iron. Elements up to uranium may exist at much lower
I-14
magnitudes, although this is a subject under investigation. The solar events can be described as three kinds. For spacecraft designers, the most familiar requirement is the 9 2 90% worst case flares that have fluences as high as 10 protons/cm for proton energies greater than 10 MeV. These flares are expected to have a 90% chance of happening during a mission lifetime. The second kind of solar event is the Anomalous Large Solar flare, 10 such as the August 72 or Oct 89 flare. These flares have proton fluences as high as 10 2 protons/cm or greater and occur possibly once during a solar cycle. The third kind of solar activity is called CMEs. These solar events have heavy ions and electrons. The LET 2 of the heavy ions can reach as high as 100 MeV-cm /mg, although the majority of the ions reach iron before falling off rapidly. Solar events impact satellites that orbit over the poles, since they orbit outside the earth's geomagnetic field. Galactic cosmic particles have ionized particles similar to those from the flares and CMEs. However, their energies are much higher, but the flux is orders of magnitude down. Shielding is difficult in stopping these particles. Their LETs range as high as 100 MeV2 cm /mg. 4.2 Radiation Requirements Within the framework of the various radiation environments, a set of radiation requirements can be developed. Examples of some of the data in a requirement are shown in Fig. 12. A dose depth curve based on the orbital parameters and the radiation environments the system will see is derived by various radiation environmental codes [33, 34, 35]. Then there are various integral flux and fluence plots for the galactic cosmic radiation, the solar event activities and the trapped belt radiation. The upper left plot is the dose per year versus shielding thickness as derived by radiation transport codes [36, 37, 38, 39]. The dose is the accumulation of all the environments the system will see over the entire mission. On the far right is an example of an integral flux versus LET curve for several shield thicknesses for a 90% worst case solar event. Integral Flux versus LET Curve 90% Worst Case Solar Event Integral Flux [ions/cm2-sr-sec]
Dose per year (rad(Si)/yr)
Dose Depth Curve
102 101 100 10-1 10-2 10-3 10-4 10-5 10-6 10-7 10-8 10-9 10-10 10-11 10-12 10-13 10-14
0.001
500 mil 300 mil 100 mil
0.01
0.1 1 LET[MeV-cm2/mg]
10
100
Others: • micrometeoroids • Dose depth curve • spacecraft charging • Trapped e- integral fluence/flux versus MeV • UV radiation • Trapped p+ integral fluence/flux versus MeV • Atomic oxygen erosion • Galactic cosmic integral flux versus LET • Outgassing • Solar flare event proton and heavy ion spectra • Solar electromagnetism Figure 12. Typical Space Environment Requirements.
I-15
Other plots include integral flux curves versus LET curves of the galactic cosmic and ALS flare. Tabulated data of these plots are also included to allow the engineer to perform one’s own risk analysis. The radiation environments mentioned above penetrate into the electronic packages. There are other space environments that are included in the specification but will not be covered in this short course. They include (a) plasma, (b) solar electromagnetism, (c) UV radiation, (d) micrometeoroid and orbital debris, (e) thermal radiation, (f) atomic oxygen erosion and (g) out gassing [12]. This set of radiation specification provides the requirements within which a space program must comply. Let us understand what radiation effects a system can exhibit based on these requirements. 4.3 System Radiation Effects Taking the radiation specifications that were discussed in the previous chart to the system level, one finds the following results, shown in Fig. 13. For example, in a MEO mission, total ionizing dose contributed by protons and electrons depositing charge in the gate oxides of a MOSFET technology will result in voltage threshold shifts in the n and p channel transistors. Leakages will occur making n channel transistors harder to turn-off. At the package level consumption of supply current will increase possibly on a number of parts, exceeding the power capability of the system. Eventually, the parts will fail to function properly and the result will lead to a system hard failure. For a LEO mission that orbits in the trapped radiation belt, the accumulation of dose can lead to system hard errors as mentioned above. With COTS technologies, one needs to be aware of another effect called Single Event Latch-up (SEL). An unprotected latch-up without current limiting and power recycling can lead to the burn-out of a device with the possibility of severing a bond wire or creating an opening of metallization from electromigration. Transient and single event upsets that disturb the memory state of a register or a memory cache are considered as soft errors in the system. These errors generally do not cause a catastrophic problem. They can be corrected by refreshing the memory back to its original state. In other cases, it can be an issue if a processor were to shut down due to double errors within a word that can not be corrected without a reboot. Reboot of a system may require times longer than a few minutes and may impact the mission of a program. In order to apply the discipline of hardness assurance to our designs, it is necessary to have an understanding of the important radiation phenomena in order to apply techniques properly. Radiation as described in the environment section, comes in the form of protons, electrons and heavy ions. In space the majority of the ionizing particles are protons. When these particles penetrate into electronic technologies, such as MOS (metal oxide silicon) or bipolar transistors, they will ionize along their path depending on the LET they have and create electron-hole pairs. It takes 3.6 eV to create an electron-hole pair in silicon and 17 eV in silicon dioxide [40, 41]. When a particle traverses a semiconductor device, the particle will ionize along its path passing through the passivation layers or oxides before entering into the silicon material. Charges that are created in the oxide may lead to radiation effects called Single Event Gate Rupture (SEGR) in MOSFETs or linear devices, Single Event Dielectric Rupture (SEDR) in Field Programmable Gate Arrays, and
I-16
Total Ionizing Dose (TID) in MOSFETs and bipolar technologies. In MOSFETs TID is in the gate oxide, while in bipolar technologies, the TID is in the field oxide. Charges created in the silicon material results in mobile charges that cause radiation effects called Single Event Upsets (SEU), Single Event Transients (SET), Single Event Burnout (SEB) or Single Event Latch-up (SEL). A few of the more important effects will be discussed below. The following references [43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57] describe various types of single event effects in electronic components. Several of these effects will be discussed in further detail in this short course. • Soft errors can be cleared by refreshing the electronics
• Hard errors are permanent (i.e., latchup, total dose, gate rupture, single event burnout) Natural Environment
Effect on Electronics
System Results
Trapped Protons
SEU, SEL, transients, ionizing & damaging dose
Soft and hard errors
Trapped Electrons
Total dose ionization
Hard errors
Solar Flares
SEU, SEL, transients, ionizing & damaging dose
Soft and hard errors
Galactic Cosmic Rays
SEU, SEL, transients
Soft and hard errors
Atmospheric neutrons
SEU, SEL, transients
Soft and hard errors
Figure 13. Environments and System Level Radiation Effects.
SEL stands for Single Event Latch-up that is the result of a pnpn path that latches into a high current state. SES (Single Event Snapback) is found in NMOS devices. This is a result of the parasitic npn inherent beneath an NMOS transistor. SEB is the result of a SES causing the parasitic npn to draw high current and to burn out the transistor. SEFI stands for Single Event Functional Interrupt that is associated with processor upsets. MBU are multibit upsets from the same single particle event. In a system one is concern with MBUs from a single word. EDAC (error detection and correction) addresses the detection and correction of single bit errors, but can not correct double bit errors within the same word. When an ionizing particle strikes a semiconductor device, the event happens very quickly, on the order of several hundred picoseconds. The electron-hole pairs that are created, drift or diffuse, depending on the presence of an electric field. Electrons drift very rapidly, leaving the holes behind to move slowly. 4.4 System Strategies for Reducing Radiation Effects Depending on the mission objective, a system engineer must balance the performance criteria and minimize exposure of the system by taking advantage of the radiation environment as described in Fig. 14. One approach would be to limit the orbit beneath the trapped radiation belt to minimize exposure to the solar events or the galactic cosmic
I-17
environments, such as the Space Shuttle and the Hubble Space Telescope. If a LEO mission can stay within ±30° latitude, the only radiation requirement would be from the SAA protons and the galactic cosmic particles. Other techniques that a system designer can choose from include localized shielding, software mitigating methods, such as error detection and correction, parity checking or refreshing, watch-dog timing and verifying a write or read command before executing the instruction. Many of these methods are addressed above the parts level. Know characteristics for space environment well – Take advantage of geomagnetic shielding – Orbit between ±30 deg latitude, below 1000 km – Orbit between inner and outer belts – Orbit outside belts – Short orbital missions – Know realistic number of flares Localized shielding Package shadowing for more sensitive packages Electronics mitigating techniques – current limiting, EDAC, redundancy, recycling power or watch-dog timing Parts selection Software mitigating techniques – parity checking, scrubbing or refreshing Hamming codes, verifying write before execution Figure 14. Other Strategies of Reducing Radiation Effects.
4.6 System Hardening Approaches System design encompasses a variety of different hardening techniques as described in Fig. 15. It should be noted that component hardening at the Integrated Circuit (IC) level is just one possible hardening solution. IC hardening is supported by DTRA (Defense Threat Reduction Agency), Sandia National Labs, NASA and various European agencies. The suppliers of hardened parts include Honeywell, BAE Systems, ST Microelectronics, and Intersil. The radiation tolerant devices may include: Xilinx, Actel, Analog Devices, Aeroflex, National Semiconductor Corporation and others. System hardening techniques cover (a) parts selection, (b) shielding, (c) circumvention, (d) current limiting, (e) watch dog-timing, (f) security coding, (g) fault-tolerant software, (h) degraded electrical performance design, (i) environment modeling, (j) IC modeling, and (m) GBS (grounding, bonding and shielding) for EMI /EMC (Electro-Magnetic Interference / Electro-Magnetic coupling) protection. Along with these hardening techniques, there are various associated software codes to support the analysis. The codes are used to help perform radiation transport for shielding analyses, simulate a circuit response to SEE, determine the SEE rate [122, 124], perform
I-18
risk analysis [121] and help determine parameter limits. In addition studies include the interpretation of heavy ion and proton data [123] and how they can be used interchangeably [119, 120]. System design encompasses a variety of different techniques. Parts hardening is just one of those disciplines. System Hardening Techniques
Supporting Codes
Parts selection Fault-tolerant software Shielding - materials SpaceRad, CREME96, NOVICE Circumvention, recycle power PSPICE Current limiting PSPICE Port entry protection PSPICE Watch-dog timer PSPICE • Degraded parameters PSPICE • Environment modeling SpaceRad, CREME96 Radiation physics FLUKA, GEANT4, SRIM • IC modeling PSPICE, WINSPICE
Parts Hardening Silicon process hardening is funded by various organizations, i.e., • DTRA, NASA, JPL, etc. Typical foundries are • STMicroelectronics • BAE Systems • Honeywell • National • Analog Devices • Intersil, etc.
Figure 15. System Hardening Approaches. Radiation Effect
Solutions
Ionizing dose - electron and proton
Shield Replace with hardened or rad-tolerant parts Account for component degradations
Displacement damage
Replace with hardened CMOS Account for bipolar gain degradation Current Transfer Ratio in electro-optics
Single-event latchup
Current limit and cycle power Shield Use thin epitaxial, SOS, SOI technologies
Single-event upset
Refresh, error detection, and correction, redundancy or watch-dog timing Replace with hardened or rad-tolerant parts Shield
Other suggestions: 1. Review parts lists for sensitive parts 2. Verify by irradiation or acquire radiation data
Figure 16. Hardening Approaches.
It's also important that the parts list be reviewed by a radiation analyst in order for a designer to know what steps must be taken to design a robust system that will meet the radiation requirements. Generally, a radiation analyst will search for radiation data that might exist in any of the open literatures, internal databases or websites at NASA/GSFC, ERRIC or JPL. If data does not exist, the analyst would recommend testing the part. I-19
Several hardening solutions have been mentioned along the way. Fig. 16 summarizes a number of these solutions at the parts, board, package and system levels [128]. For ionizing dose effects, shielding can reduce the electron dose for a GEO mission. However, for other orbits, such as LEO, MEO or even a mission to Jupiter, protons are difficult to shield. The best solution is to select hardened parts with extra design margin added to account for degraded parameters in the circuit design. For proton displacement damage, MOS technologies are generally immune to this effect. For bipolar transistor technologies, gain loss is serious and must be accounted for by including degraded DC and AC parameters in the circuit design. Single event latch-up can be mitigated with current limiting and recycling power. However, radiation testing is needed to determine the latch-up sensitivity of the parts and to determine whether current limiting and recycling power will work. Single event upsets can be mitigated at the part and system levels. At the parts level, LETs 2 of greater than 20 MeV-cm /mg will allow the parts to survive through a proton environment such as a large solar flare. However, with upsets in a memory device or in a processor, an EDAC circuit and a watch-dog timer are necessary to status the health of these complex devices. Other software error checking techniques may be needed, such as Reed Solomon or Hamming decoding. 4.7 Space Environment Modeling Tools There are analysis codes that support the radiation analyst [37, 38, 39]. In Fig. 17 a list of space environment codes is available to support satellite or launcher programs. A few can be found on the web [33]. For a satellite program, the codes will support circular and elliptical orbits. For a launcher program Space Radiation [35] has the capability to generate a ballistic trajectory. Many of the codes will support radiation transport calculations of heavy ions and protons. Space Radiation has the capability to model neutron effects at atmospheric levels. In addition some of these codes will calculate the total ionizing dose that passes through a shield and onto an electronic part. Others will also calculate the Single Event Effects rates. These space environment models have improved over the years. One of the major accomplishments has been the revision of the heavy ion spectrum originating from the solar events. The based line code was CREME developed by Naval Research Laboratory. Using the earlier model would have given overly conservative SEE rates that made many satellite programs difficult to leverage on COTS. With the reductions made in the heavy ion content based on the October 89 and March 91 solar events, many satellite programs can now leverage on COTS parts because their SEE rates are more reasonable.
I-20
Codes are used to: • Model vehicle's space environments • Transport radiation through spacecraft wall/package materials • Calculate rate of occurrence of a radiation effect Satellite Launcher Heavy Protons Total Parts ions transport dose prediction transport
Code Space Rad 4.0/5.0
X
CHIME
X
X
X X
X
X
X
CREME96 – web based
X
X
Boeing-MACREE
X
X
X
SPENVIS – web based
X
X
X
X X
X
• Improvement in space environment models helped pave the way for insertion of COTS in space systems • Predicted heavy ion flux from flares reduced by several orders of magnitude Figure 17. Space Environment Codes.
4.8 Shielding Example Typical dose versus shielding thickness curve
At about 100 mils electrons fall off rapidly, while further shielding does not provide any further benefit from energetic protons
Dose per year (rad(Si)/yr)
Curves represent • Trapped electrons • Bremsstrahlung radiation • Solar event protons • Total
650 km, 90° inclination orbit Trapped electron Bremsstrahlung Trapped proton Solar Proton Total Natural Dose
1x106 1x105 1x104 1x103 1x102 1x101 1x100 1x10-1 0
100
200
300
400
500
Shielding Thickness [mils of Al]
Figure 18. Total Ionizing Dose versus Shielding.
Determining the dose at a part is important in determining the level of hardness required for the parts. If the margins are high, surveillance may not be necessary. If the margins are low, lot sampling may be required. As one can see ray tracing or transport analysis is important in determining accurately whether one will have adequate margin without additional testing later. Therefore it is important to generate a dose versus shielding thickness curve. As an example of a shielding calculation, Fig. 18 shows the dose per year versus shielding thickness for a LEO system. The trapped electrons are negligible above 200 mil of aluminum. The trapped protons or solar flare protons are more difficult to
I-21
shield. As one can see in Fig. 18, adding 500 mil of shielding only dropped the dose by a factor of two from 1krad(Si) to 400 rad(Si). Example: 650 km, 90° inclination circular orbit How much shielding is needed for a 3krad (Si) commercial part? Given: Need 3X margin and there's already 150 mil of package shielding available around the part Shield(mil)
Trapped e-
brems. e-
Trapped p+
90% w/c flare
ALSF*
Total Dose
0.1 0.5 1 3 5 7 10 20 50 60 80 100 200 300 400 500
5.02E+05 4.80E+05 3.50E+05 1.53E+05 8.85E+04 5.82E+04 3.57E+04 1.26E+04 3.21E+03 2.39E+03 1.41E+03 8.75E+02 1.02E+02 1.01E+01 1.10E+00 1.17E-01
1.54E+02 1.53E+02 1.51E+02 1.04E+02 7.63E+01 6.11E+01 4.68E+01 2.59E+01 1.26E+01 1.09E+01 8.48E+00 6.96E+00 3.77E+00 2.70E+00 1.73E+00 1.84E+00
8.28E+04 8.60E+03 3.48E+03 1.31E+03 9.69E+02 8.37E+02 7.16E+02 5.22E+02 3.59E+02 3.36E+02 3.03E+02 2.77E+02 2.18E+02 1.90E+02 1.71E+02 1.57E+02
1.78E+03 1.74E+03 1.69E+03 1.55E+03 1.44E+03 1.33E+03 1.20E+03 8.75E+02 3.01E+02 2.37E+02 1.64E+02 1.26E+02 6.26E+01 4.14E+01 2.97E+01 2.22E+01
4.15E+03 4.08E+03 3.99E+03 3.73E+03 3.52E+03 3.33E+03 3.09E+03 2.50E+03 1.45E+03 1.28E+03 1.04E+03 8.68E+02 4.75E+02 3.18E+02 2.34E+02 1.81E+02
5.91E+05 4.95E+05 3.59E+05 1.60E+05 9.45E+04 6.38E+04 4.07E+04 1.65E+04 5.33E+03 4.25E+03 2.93E+03 2.15E+03 8.61E+02 5.62E+02 4.38E+02 3.65E+02
* ALSF - Anomalous large solar flare
Figure 19. Total Dose Calculation.
Solution: Design margin = Failure threshold/Mission lifetime dose on part or Failure threshold / Design margin = Mission lifetime dose on part 3krad (Si) commercial part / 3X margin of safety = 1.0 krad (Si) – According to table, one obtains 1k- 861rad (Si) with 190-200 mil of aluminum – With 150 mil already available, then 200 mil - 150 mil = 50 mil more to achieve 3X margin – If volume space is a problem, use tantalum, ρTa/ρAl = 6:1 effective shielding to aluminum (i.e., 200 mil/6 = 33.3 mil of tantalum) – Location of tantalum shield must be analyzed for dose enhancement
Figure 20. Shielding Calculation for Total Ionizing Dose.
Let us look at a shielding problem. Suppose the designer must choose a part that is three krad(Si) radiation hardened for a commercial space program. The board will be contained in a 150 mil aluminum walled package. A margin of 3X was determined to be
I-22
appropriate for this program. Fig. 19 tabulates dose from the different space environment constituents. By dividing the hardness of the part by the design margin, one obtains the minimum dose at the part over the entire mission. In this example, the dose would be 1 krad(Si). According to Fig. 20 1krad(Si) to 861 rad(Si) requires 190 to 200 mil of aluminum. Therefore, with about 150 mil from the package itself, only an additional 50 mil of aluminum is required. If the package did not allow enough room to accommodate the additional 50 mil of aluminum, one can use a higher atomic number material. For example, the equivalent amount of tantalum to aluminum is the ratio of their mass densities. Therefore, 200 mil of aluminum is equivalent to 33.3 mil of tantalum. More elaborate transport models can be used to refine the dose at a part.[37] Most 1D analysis does not account for unique angles where the walls may be thinner allowing more dose at a particular section of a part. A system analyst has to assess the vulnerability of the part. With possible part to part variation, a commercial part may degrade below the anticipated design margin due to process variations. With a 3X margin surveillance testing is recommended for this part according to Mil. Handbook 814/815. 4.9 Upset Rate / Risk Example If a designer has to select a higher performance memory device, one of the common questions that arises is “what is the risk and upset rate for 200 memory devices in a LEO satellite”. An assumption is given that the critical exposure time is 8 min during a 90% worst case solar event. Also the LET threshold of the memory devices is 3 MeV-cm2/mg with a saturated cross section of 2x10-2 cm2/device. Fig. 21 shows the integral flux spectrum versus LET of such a solar event. At a LET of 3 MeV-cm2/mg, the integral flux is 8x10-6 ions/(cm2-sr-s). LET spectra of ion contribution from 90% worst case solar event through 500 mil Al shielding, 650 km, 90° inclination
Integral Flux [ions/cm2-sr-sec]
90% Worst Case Solar Event, 500 mil Al 102 101 100 10-1 10-2 10-3 10-4 10-5 10-6 10-7 10-8 10-9 10-10 10-11 10-12 10-13 10-14
Integral Flux = 8x10-6 ions/cm2-sr-s
0.001
LET = 3 MeV-cm2/mg
0.01
0.1 1 LET[MeV-cm2/mg]
10
Figure 21. Solar Event LET Heavy Ion Spectrum.
I-23
100
By taking the product of the integral flux at a LET of 3 MeV-cm2/mg with the saturated cross section, an estimate of the upset rate is obtained. It is assumed that the upset cross section of the device has a sharp knee. The upset rate in an 8 minute interval is determined to be 0.19 upsets in a system of 200 memory devices as shown in Fig. 22. The risk is determined by using the Poisson probability distribution and placing the upset rate into the argument of the exponential function as shown in Fig. 23. Including the 8 min duration into the equation, we obtain an 18% chance for an upset to occur during the 8 min interval.
Problem: For spacecraft at 650 km, 90° inclination the system operates for 8 min over the pole. Calculate the upset rate and assess the risk for 200 memory devices with a LET threshold of 3 MeV-cm2/mg and a cross section of 2x10-2 cm2/device behind 500 mil of aluminum. Number of upsets = cross-section of device x ion flux x exposure duration = 2x10-2 cm2/device x 8x10-6 ions/(cm2sr-s) x 8 min x 60s / min x 4πsr x 200 devices = 0.19 upsets/system Figure 22. Upset Rate Analysis Use Poisson's distribution to determine the probability of an upset in the system: Poisson's equation: P(k;t) = e-λt(λt)k/k! Probability of no upsets in the time interval t is P(0;t) = e-λt, where λ is the upset rate and the total number of bits. Probability of success, Psuccess = exp (- no. of bits x upset rate per flight per bit x bits/device x no. of devices x flight time x no. of flights) Flight time = exposure time of 8 min. No. of flights = 1 No. of devices = 200 Psuccess = exp[ - number of upsets/system-flare] = exp[- 0.19 upsets/system-flare] = 0.82 Pfail = 1 - Psuccess = 0.18 or 18% chance that an upset will occur during the 8 min exposure period
Figure 23. Determining the Risk or Probability of Success.
I-24
By performing this risk analysis, one can establish the guideline for procuring parts during the manufacturing phase of the program. For example, if parts tested during the development phase exhibited a LET of 3 MeV-cm2/mg, but was discovered to have a LET lower than 3 MeV-cm2/mg when purchased during production, then the margin is reduced and the system may exhibit a higher probability for upsetting. Likewise, in the above example of calculating the dose at a part, if the parts procured later exhibited a higher sensitivity dose, the 200 mil of aluminum or the 3X would not be adequate. 5.0 Hardening Against Proton Since protons are the predominate specie in the space environment, one needs to understand how they interact with the specific electronics in a semiconductor device and how their effects can be mitigated. Protons interact differently than heavy ions in causing SEEs (Single Event Effects). One uses different units to characterize protons versus heavy ions. For protons MeV is used as the unit of energy, whereas for heavy ions the stopping 2 power or Linear Energy Transfer (LET) in MeV-cm /mg is used as the unit. It should be noted that for protons greater than 200 MeV, the energy lost per unit length, LET, is very small, such that very little charge is deposited in the silicon material. Yet a large number of upsets is observed with protons. Here lies the reason. proton
28Si 14
29P 15
LET, Stopping Power [MeV-cm2/mg]
Primarily one of the following reactions occur
4He 2
24 24Mg 12 12
+
27 27Al 13 13
+
Proton (>50 MeV) + Si nucleus results in by-products • Al+, Mg+ ions (1-5 MeV), alpha, p+ • LET (Al+, Mg+) < 14 MeV- cm2/mg Hydrogen
proton
+ protons
+
102
Helium Magnesium Aluminum
LET = 13.9 MeV-cm2/mg
101
The maximum linear energy transfer (LET) of Al+ or Mg+ establishes the necessary proton hardness, i.e., 20 MeVcm2/mg
100 10-1 10-2 4 10
105
106
107
108
109
Energy [eV]
Figure 24. Proton Physics. 23
When an energetic proton enters a silicon material as shown in Fig. 24, it sees 10 silicon nuclei, i.e., Avogadro’s number of atoms in a mole of material. Based on the nuclear capture cross section between protons and a silicon nucleus, there is a probability for a proton to be captured by a silicon nucleus. Silicon-28 now becomes phosphorus-28 that is unstable and decays in a few microsecs to either an aluminum-27 nuclei plus two protons or a magnesium-24 nuclei plus an alpha and a proton. The energies of the aluminum and magnesium ions range between 5-10 MeV. The chart in Fig. 24 is a plot of the LET of a proton, alpha, aluminum and magnesium ions versus energy. According to this chart, the 2 aluminum or magnesium ions have a maximum LET of 14 MeV-cm /mg between 10-50 I-25
MeV. For this reason, if one wanted to design a circuit that could be immune to proton 2 upsets, the LET hardness of the circuit has to be at least 15 MeV-cm /mg or greater. This would make the circuit very difficult to upset with protons. Similar principles can apply to neutron single event effects because of similar nuclear reaction properties. From a hardness assurance viewpoint, if a new supply of parts were to be procured, and the supplier made improvements to the part by making a faster part, it may result in the lowering of the LET hardness to upsets. Similarly if the supplier eliminates the need for an epitaxial silicon layer, the LET threshold for latch-up may drop from 26 to 10 MeV2 cm /mg. Process controls need to be applied in order to monitor these changes. Heavy ions can exhibit the same reactions as protons in space. However, the ion energies at a test facility, such as Brookhaven National Labs are lower so their predominant reaction in silicon is primarily direct ionization rather than from nuclear interactions. 6.0 Key Integrated Circuit / Semiconductor Degraded Parameters and Effects When one looks at radiation effects at the semiconductor level, one can distinguish two types of effects, those that are transients and those that exhibit long term degradation. Fig. 24 highlights some of these effects, the cause and how they impact the MOS or bipolar electronic technologies [42]. For example, dose ionization is caused by trapped charges in the oxide. For MOS devices, there would be voltage threshold shifts from the gate oxide and from the field oxide. Leakage currents can occur due to an inversion of the silicon beneath the gate oxide between a drain and a source of a transistor. These leakage currents can slow the charging or discharging of a capacitive node, hence change the propagation delay times or rise and fall times of the AC characteristics of a circuit. Similarly, for a bipolar transistor, the increase in base leakage current decreases the Hfe or DC gain performance of a bipolar transistor. The AC and DC circuit parameters are degraded. Effect
Cause
MOS
Bipolar
1. Ionizing Dose
Oxide – trapped charge
Threshold voltage shift Leakage current Increased current Degraded AC/DC parameters
Current gain degradation Increased base leakage Increased current Degraded AC/DC parameters
2. Very low dose rate
Interface traps
Threshold voltage shift
Gain degradation
3. Particle interaction
Lattice damage
minimal
DC gain degradation (increased base leakage) Degraded AC/DC parameters
Other effects 4. Particle ionization
Capacitance discharge
Single-event upset Single-event upset Single-event transient Single-event transient
5. Gate oxide rupture Particle ionization Power MOS – SEGR 6. Parasitic SCR (pnpn) Initiation
Latchup
SEGR in linear caps.
Single-event latchup
Latchup
Single-event induced breakdown
Figure 24. Key Integrated Circuit / Semiconductor Degraded Parameters and Effects.
I-26
Displacement damage arises from elastic or inelastic scattering of protons with a silicon atom recoil creating trapping sites in the base region of a bipolar transistor. Other effects include SEGR [105, 106, 107, 108], SEU, SEB [104] and SEL [58, 59, 61, 63] of which some will be discussed later [109]. Having an awareness of these effects is important in not only choosing the correct hardening technique in the design, but also to know what effect might occur that could impact the design margin when a supplier changes their process. Hardness assurance is to make sure that the hardening techniques still apply during manufacturing and that the design margin has not changed since the beginning of the design. 7.0 Total Ionizing Dose Effects 7.1 MOS Devices Total ionizing dose (TID) in MOS devices is created by the ionization of a particle in the insulating materials, such as SiO2, in a semiconductor device as shown by the semiconductor band diagram of Fig. 26. For an MOS device, the two regions that are sensitive to TID are the gate oxide and the field oxide. The field oxide is the isolation region separating individual transistors. For a bipolar device the sensitive area would only be the field oxide region over the base and isolation regions. With positive gate bias on an electrode over an oxide, electrons are swept rapidly towards the gate electrode leaving holes behind either trapped in the oxide or transported through the oxide by drift or diffusion towards the silicon dioxide/silicon interface. Holes will recombine along the way towards the silicon dioxide/silicon interface. However, some holes will react along its path creating hydrogen that in turn will drift towards the interface. If a hydrogen atom does reach the Si-SiO2 interface, it will create an interface trap or a dangling bond at the Si-SiO2 interface. This dangling bond is defined as an interface state. Band diagram of an MOS capacitor with a positive gate bias. Illustrated are the main processes for radiation-induced charge generation and trapping.
e-h pairs created by ionizing radiation
Nit: interface trap formation (Pb )
SiO2
Si
+ + +_ _ _ Not: deep hole + _ _ + trapping (E’) + near interface +
+ gate bias
proton transport
gate proton release
+
H+ hopping transport of holes through localized states in bulk SiO2
Figure 26. Radiation Effects of a MOS Device.
For the holes trapped in the oxide that we refer to as oxide traps, a negative bias is required to compensate for these extra positive charges. Therefore one observes a shift of the
I-27
transconductance I-V curve as shown in the traces of Fig. 27. When the I-V curve exhibits a parallel shift as indicated by the arrow, only positive charges are observed. However, with a change in the slope of the curve, interface states are then involved. Leakage currents between the drain and source can be observed, created by the inversion of the silicon beneath the field oxide around the edges of a transistor as shown in the traces of Fig. 27. Measuring techniques such as the subthreshold I-V method can be used to extract the number of oxide traps and interface states created by the total ionizing dose. However, if leakage currents are too high, then the dual transistor measurement technique is recommended. As mentioned above charges are generated by the ionizing particles that traverse the device. Charges are generated in the field oxide located in the bulk of the passivation layers and in the gate oxide of a MOS transistor. These extra charges will cause voltage threshold shifts in an n or p channel device. For an n channel device, one observes a negative voltage shift. Similarly, there is a negative threshold shift in the p channel as shown in Fig. 27.
Drain Current [A]
P- Channel Power MOSFET I-V characteristic
1.E+00 1.E-01 1.E-02 1.E-03 1.E-04 1.E-05 1.E-06 1.E-07 1.E-08 1.E-09 1.E-10 1.E-11 -10
Pre-Irrad 1 krad 10 krad 100 krad 250 krad 450 krad 1000 krad -8
pre 1Mrad(Si)
-6
-4
-2
0
Gate Voltage [V] Solutions: • Account for threshold voltage shifts and leakage currents in the n and p channel transistors Figure 27. Radiation Total Ionizing Dose Effects in MOSFETs. [196]
If parts are annealed with time, bias and temperature, as recommended by the Military Standard 1019.4 methodology, this will allow some of the holes to migrate towards the SiSiO2 interface. The creation of interface states is commonly known as the “rebound” effect. The hydrogen that is generated will migrate to the interface, creating a dangling bond or an interface state. For n channel transistors, the interface state is a negative charge that causes the transconductance curve to shift to the right. If the voltage shift passes the original pre-irradiated threshold voltage of the device, this phenomenon is known as the "rebound" effect, known to exist only at low dose rates such as in the space environment. For n channel transistors, the “rebound” effect requires a higher voltage to turn on the I-28
device [81, 84, 85]. The “rebound” effect has been observed in MOS devices. However, this should be distinguished from the Extremely Low Dose Rate Sensitive (ELDRS) effect that has been only associated with particular bipolar devices. Irradiation has been performed at varying rates with leads shorted to various biasing conditions. This topic will be further discussed in the Future Issues section below. Other related areas where total ionizing dose effects are being studied include PEMs (Plastic Encapsulated Microelectronics) and post manufacturing processes. Microelectronics packaged in plastic provide a low cost approach to design. However, the radiation effects are not that well understood [83]. Studies are being conducted to look at high temperature stress steps during packaging and burn-in that may impact the radiation sensitivity of a part [82, 86]. 7.2 Bipolar Devices In a bipolar transistor one observes total ionizing dose effects from the charges that are trapped in the field oxide. Here we observe base leakages that are evident in the drop of the DC gain or Hfe at Icollector currents below 100 mA as shown in Fig. 28. The gain of the transistor is normally above 100. However, after each level of irradiation, the gain of the device diminishes as indicated by the arrow on measured data of a pnp transistor, a 2N2907. The gain drops dramatically with small Icollector currents, i.e., 1 mA or less. This degradation in Hfe must be accounted for in a circuit design. In other words, the designer should use degraded parameters from radiation and should not just use the degradations called out in reliability standards.
Hfe, Collector / Base Current
Gain degradation from gamma irradiation Pre-Irrad 5 krad 10 krad 25 krad 50 krad 100 krad 250 krad 500 krad 1000 krad
pnp Bipolar Transistor
180 160 pre 140 120 100 80 60 40 20 0 1Mrad(Si) -20 1.E-13 1.E-11 1.E-09 1.E-07 1.E-05 1.E-03 1.E-01 1.E+01
Collector Current [A] Solutions: • Account for gain degradation and leakage currents in the npn & pnp transistors Figure 28. Radiation Total Ionizing Dose Effects in Bipolar Transistors. [196]
8.0 Proton Displacement Damage Displacement damage is a known problem in the military world or when parts are near a nuclear reactor. On the space side there are similar problems with protons. From a
I-29
physicist view point, neutrons and protons are considered to have similar nuclear interaction properties, because they both belong to the baryon nucleon family of elementary particles. Protons, as distinguished from neutrons, can ionize along its path and deposit charge besides causing displacement damage. Note again that the ionization by a proton is still less than the ionization contributed by a heavy ion. This is described in the Hardening Against Proton section. It has been shown that the clusters of damage created by a 50 keV recoiling silicon atom from elastic scattering of high energy protons can be on the order of 0.1 x 0.1 micron square in cross section. With future COTS technologies, feature sizes of 0.1 x 0.1 micron square are achievable. This raises the concern of how vulnerable future electronics will be to damage from incoming particles. 8.1 Gain Degradation In bipolar technologies, as mentioned earlier, damage in the base region will create recombination or detrapping sites that will lower the gain of a transistor as shown by the data on the left in Fig. 28. •
Bipolar technologies with base regions with fT < 10 MHz are most susceptible Hardening solution: Account • 1/hfefinal - 1/hfeinitial = kΦprotron for degradation in hfe gain hfe Degradation from with adequate radiation 140 Displacement Damage design margin, RDM 46 mA, hfe = 136 hfe
design level
pre-irradiation hfe 14.0/div
100 46 mA, hfe = 52 50
post-irradiation 0
10-11
RDM = 2X
0 9 1010 1011 1012 1013 10-1 10 proton fluence, Φ (p+/cm2)
IC [A], decade/div.
Figure 29. Total Dose Proton Displacement Damage.
Bipolar transistors are sensitive to displacement damage with protons from the trapped radiation belt or from solar flares, especially those with large base depths with fT less than 10 MHz. (fT is defined as the frequency of a bipolar transistor at which the gain is unity.) If dose ionization can be separated out, the displacement degradation in the DC gain of a bipolar transistor can be characterized by the Messenger-Spratt equation: 1/Hfefinal 1/Hfeirrad = kφ, where k is the proton damage constant and φ is the proton fluence. The DC gain, Hfe, is defined as the ratio of IC over IB.
I-30
Similar to neutron damage, the gain of a bipolar transistor will degrade as a function of IC as shown in the left set of measurements. At 46 mA, the initial DC gain was 136. After a given fluence, the DC gain has degraded to 52. If one were to plot the degradation of Hfe where the gain is highest with IC as a function of proton fluence, one obtains the curve 11 2 shown on the right. Initially, Hfe decreases slowly at fluences less than 10 protons/cm . 13 2 Above this the Hfe drops rapidly. Above 10 protons/cm the Hfe does not appear to decrease rapidly, but approaches values between 1-4. In finding a hardening solution, an engineer needs to account for the degradation in Hfe. To allow for robustness in a circuit design or some radiation design margin (RDM), a designer needs to include additional degradation other than the gain at the operational limit. The margin can be applied to the circuit or to the radiation dose. For the radiation dose, RDM applies to the radiation fluence level, whereas for a circuit performance a design margin (DM) applies to the gain of the device. Also note that at low IC currents, the DC gain degrades very rapidly. Therefore, for robustness in a design, a designer should try to avoid designing at low IC currents. The IC current at the maximum DC gain is generally recommended. From a HA point of view, any degradation in the Hfe gain of a bipolar transistor can impact the original RDM or DM. Likewise, similar degradation in the voltage threshold of a MOS transistor can affect the original DM. In terms of affecting the ionizing dose requirement, this would impact the RDM. 9.0 Displacement Damage in Other Technologies and System Solutions In electro-optical devices it was observed by Johnson et al. that damage mainly in the LED could result in degradation to the current drive that supports the operation of the device. With inadequate current drive capability designed in, again, not using radiation degraded parameters, failure can occur [98, 99, 100, 103] to the system. • Low-frequency opto-couplers are sensitive to CTR (Current Transfer Ratio) degradation • Solution: Design for adequate LED drive (i.e., >1 mA) and design for receiver gain loss
Vcc 3
Output, 2
npn
100k ohm
Input 5
7
GND
300 ohm
Current Transfer Ratio (Ic(5V)/Iforward)
4N49 Opto-coupler CTR Degradation with Proton Energies
193, 45-MeV protons
1 beam normal incidence to chip
10
192.4 MeV protons 45 MeV protons 0
10
-1
10
-2
10
-3
10
-4
10
109
Preirradiation
1010 1011 Proton Fluence [protons/cm2]
Figure 30. Electro-Optical Device Degradation.
I-31
1012
In Fig. 30 we show a typical electro-optical circuit with a 300 Ω load in series with the LED (Light Emitting Diode). The npn transistor is connected with a 100 kΩ load. On the right is a plot of the current transfer ratio (CTR) defined as the npn transistor collector current divided by the LED current in the forward biased mode. The CTR as a function of 11 2 proton fluence is shown on the right in Fig. 30. As we approach 10 protons /cm , the CTR begins to roll off rapidly for a given LED drive current. The photoionization current collected by the transistor is dropping due to the inefficiency of the LED to generate adequate light. Higher drive currents on the LED will provide more light emission, hence hardening the circuit design. Note also that by going from 192 to 45 MeV protons, slightly more displacement damage is observed. Hence, a designer needs to use degraded CTR data at low proton energies. Several LED drive currents should be measured in order to allow a designer to select the best optimal design performance for a mission. In addition to bipolar discrete transistors that are sensitive to displacement damage, other semiconductor devices such as analog or linear devices show similar effects [101] as described in Fig. 31. These include operational amplifiers, linear regulators, voltage references, comparators, Analog to Digital (A/D) converters and Digital to Analog (D/A) converters, and pulse width modulators. These technologies have similar effects in their bipolar transistors as shown in Fig. 29. For example, an operational amplifier will exhibit degradation with the open-loop gain. A voltage reference may degrade in the precision to output the proper voltage. A/D and D/A converters may lose accuracy by starting in the least significant bit (LSB) first due to offset, bias and reference degradation and then extending to higher bits. In addition to linear devices, as already mentioned, there are electro-optical devices that also exhibit displacement damage as described above. These include optocouplers, opto-isolators and light emitting diodes (LEDs) [102]. Other devices that can degrade from displacement damage are diodes. At extremely high fluences, diodes can become ohmic.
Displacement Damage (Non-Ionizing Energy Loss) in linear devices • Op amp, linear regulator, voltage reference, comparators, A/D, D/A, pulse width modulator Displacement Damage (NIEL) in electro-optical devices • Optocouplers, optoisolators, LEDs, CCDs Displacement Damage (NIEL) in discretes • Diodes, bipolar transistors Single event upset (protons & neutrons) • Linears, sequential logic devices, i.e., SRAMs, FFs, registers in CPUs, DRAMs, SDRAMs Figure 31. Proton Effects on Other Electronic Devices.
I-32
A recent issue in bipolar devices is the effect of single event upsets or transients created by 10 the neutron reaction with B, an isotope of boron. Besides proton SEUs, the by-products 10 of the n- B reaction can occur for devices with LETs (Linear Energy Transfer) less than 3 2 MeV-cm /mg. This SEU effect is similar to the proton interaction with a silicon nucleus as described in Section 5.0. Boron is commonly found in the passivation layer over an integrated circuit chip called BPSG (borophosphate silicate glass), in the source-drain junctions of p channel devices, in the active region of n channel transistors, in the base region of an npn and the emitter and collector of a pnp. The nuclear capture cross section is high at low neutron energies. Neutrons are generated at low altitudes by the interaction of the galactic cosmic environment with the oxygen and nitrogen in the atmosphere. Neutron SEUs have been observed by various ground and aircraft systems. Very similar to the p+ - Si reaction, neutrons behave in a similar way, in the sense that the neutron byproducts are the ions that will cause the single event effect. In order to mitigate the neutron SEE problem, parts with a LET greater than 14 MeV-cm2/mg should be selected. Another solution that some semiconductor foundries are doing is to eliminate 10B from their ion implantation or passivation process. There are a number of hardening solutions to protons as listed in Fig. 32. For BJT devices, as mentioned earlier, one needs to account for gain degradation that will occur when a device is exposed to very high fluences, such as from an ALS flare, in a MEO through the proton enriched trapped radiation belt or a mission to the intense trapped radiation belt of Jupiter. To harden against displacement damage, several gain stage design can be considered in the circuitry. Time dependent annealing can be considered. • Commercial manufacturers are working with the submicron feature sizes, i.e., size of a defect cascade • Silicon recoils from proton inelastic scattering can creat a cluster of damages within 0.1 x 0.1 µm2 area
Solutions • Account for gain degradation in BJTs • Use several gain stages in circuitry • Account for time dependent annealing • Account for CTR loss in optical couplers • Filter transient glitches in linears • Circumvent and refresh memory • Use CMOS, power MOSFETs, GaAs, Si-Ge HEMT Figure 32. Proton Hardening Solutions.
In designing electro-optical devices, such as optical couplers, a designer needs to account 12 for the degradation in the CTR (current transfer ratio). Otherwise, after a fluence of 10 2 protons/cm , there may not be enough drive to switch the transistors. For linear devices, I-33
such as op amps, voltage regulators and references, transient glitches can be removed by adding filter capacitors on the signal lines. Software codes can be used to check for upsets or glitches and sequential logic devices should be refreshed to correct for any corrupted data that is inadvertently loaded into memory. Hardened technologies to displacement damage include CMOS, power MOSFETs, GaAs and silicon germanium strained HEMT (high electron mobility transistor) devices. These technologies are extremely hard to proton damage and to total ionizing dose effects. 10.0 Comparison of Commercial versus Hardened Technologies Fig. 33 is a comparison of commercial versus hardened technologies that are available to a designer. Considering the application of the program, one can choose commercial parts, but one has to be aware of their radiation sensitivity. Care must be taken to prevent any catastrophic failures when these are implemented in a system. The SEUs and dose ionization sensitivity levels of these parts are high. For example, the TID of COTS can be as low as a few krad(Si). The bipolar TID hardness is based on Mil Std. 1019.4 radiation testing at 50-300 rad(Si)/s. However, with ELDRS effect, low dose rate testing is needed because some linears can fail at low doses. Some of the hardened technologies include CMOS/epi, CMOS/SOS and CMOS/SOI processes. These parts have total ionizing dose hardness levels as high as a Mrad(Si) and -10 single event upset rates less than 10 upsets/bit-day in a 90% worst case galactic cosmic environment. • Commercial technologies - sensitive to SEUs, low total dose hardness and possible lot-to-lot variation – Commercial bipolar is SEU/SET sensitive and soft to total dose at very low dose rates. It is also sensitive to lattice displacement damage from protons. • Hardened CMOS is immune to SEUs and has a high total dose hardness Technology
Single Event (upsets/bit-day)
CMOS/epi commercial
10-5 – 10-8
Bipolar/bulk commercial
10-6
CMOS/epi, SOS, SOI (hardened)
Ionizing dose* (rad(Si)) 5k – 60k ~100k (>10 rad/s) 30 – 50k (<1 rad/s)
<10-10
>1 Meg
Lattice Damage (particles/cm2) 1010 – 1012 -
* tested at 50-300 rad(Si)/s Figure 33. Comparison between Commercial and Hardened Technologies.
11.0 Radiation Test Facilities Fig. 34 shows the different radiation test facilities that are available to the users for conducting either total ionizing dose irradiation, proton displacement damage or SEE testing [116, 117, 118].
I-34
Cobalt-60 testing radiation resources include the Gamma Cell-220 and the J.L. Shepherd sources. Dose rates can vary from 0.1 rad(Si)/s to as high as 300 rad(Si)/s. Some users have also conducted TID testing using a 10 keV X ray source. For proton testing the facilities include (1) the Crocker Laboratory at the University of California, Davis, operating at a maximum of 60 MeV, (2) Indiana University's Cyclotron Facility at 198 MeV, (3) Massachusetts General Hospital at 210 MeV, and TRIUMF in Vancouver, Canada. For heavy ion testing possible facilities include (1) Brookhaven National Laboratory, Upton, New York, (2) Lawrence Berkeley National Laboratory, Berkeley, CA, (3) Texas A/M University, (4) Michigan State University, (5) GANIL in Caen, France and (6) GSI at Darmstadt, Germany. The machines that can provide GeV ions are Texas A/M University, Michigan State University, GANIL and GSI. Total Dose Part Level ~1 in2
SEE
60
proton cyclotrons
Heavy ion facilities Brookhaven National Labs, Sandia, Berkeley National Labs, GSI Texas A/M Univ., Michigan State, GANIL
Laser JPL, NRL LANL neutrons
60
proton cyclotrons
Heavy ion facilities Brookhaven National Labs, Sandia, Berkeley National Labs, GSI Texas A/M Univ., Michigan State, GANIL
None
Co, Indiana Univ. 10keV Mass. Gen. Xrays U.C. Davis, TRIUMF
Co, Indiana Univ. 10keV Mass. Gen., ~12"x12" Xrays TRIUMF module
Others
Figure 34. Radiation Test Facilities.
There are also other available sources for conducting SEE effects. For very small spot sizes, the micro-ion beam at Sandia National Laboratories or the lasers at Aerospace Corp., Jet Propulsion Laboratory or Naval Research Laboratory are available. For neutron single event upset testing, Los Alamos National Laboratory has a neutron source that matches the atmospheric neutron spectrum. Another alternate approach to proton and electron testing of electronic parts is the use of cobalt-60 and 10 keV irradiation, respectively [133]. 12.0 Production / Deployment Phase 12.1 System Hardness Assurance During the program initiation phase parts technologies and hardening techniques are being evaluated. Radiation characterization data for selected technologies are being performed. After a program has defined their environment specifications, designers defined their design baseline approach, performed their tradeoffs, and a detailed full-scale development and compliance plan defined, the program begins full-scale engineering development (FSED) or full-scale development (FSD) as illustrated in Fig. 35.
I-35
Timeline
Hardness assurance needs to be applied throughout the entire life cycle
PROGRAM FULL-SCALE PRODUCTION INITIATION DEVELOPMENT
DEPLOYMENT
DESIGN HARDENING & TEST DEVELOP HARDNESS ASSURANCE & MAINTENANCE PLAN
LIFE CYCLE SURVIVABILITY HARDNESS ASSURANCE HARDNESS MAINTENANCE & SURVIELLANCE
DEVELOP QUALITY ASSURANCE QA PLAN CONFIGURATION CONTROL
Figure 35. Hardness Assurance Plan.
During this FSD phase boards and packages are developed. A system may also be assembled for evaluation. A detailed schedule plan must be developed to logistically determine when parts can be procured so that engineering can assemble them on boards or modules. Further electrical characterization and radiation testing at the board level will be performed. When packages are beginning to be assembled, parts development efforts begin their qualification stages. Technologies have been generally frozen and documentations are prepared describing the procurement process, circuit designs and radiation performance characteristics. Circuit designers are simulating critical paths in their designs. A qualification plan is being developed, describing the methodologies of preparing or assembling the packages. Design drawings at the part, board, module and subassembly levels are prepared. This includes documentation delivered to manufacturing and production. Engineering packages and an engineering system are being assembled for electrical characterization. This includes performance testing, functional operation, ElectroMagnetic Coupling /ElectroMagnetic Interference (EMC/EMI) testing, thermal and vacuum testing, vibrational characterization and temperature cycling evaluation. During this final phase of FSED, radiation testing of individual packages or a system may be evaluated. Due to cost most systems will be assessed based on radiation testing at the parts or board levels. Going into production and deployment, a hardness assurance maintenance plan needs to be developed. This describes what needs to be performed if parts issues occur. For example, parts may not be procured as life of buys, but as lots. Therefore further qualification testing may have to be performed to requalify the new purchase. Change must be documented and approved through a configuration control committee. This committee includes reliability, safety, survivability, structural, materials, quality assurance, tooling, manufacturing and procurement engineering. The customer of the system may be included. This maintenance plan establishes procedures as to how a system is controlled. Hardness assurance and surveillance procedures are implemented after full-scale
I-36
development. When FSD is completed, generally all qualification documents have been completed. Procedures for manufacturing which includes assembly drawings, step by step manufacturing processes are described. Tooling and support test equipment must also be in place. Any alterations or changes from the processes that were characterized during FSD must now be carefully reviewed for acceptance or rejection. Obviously, cost becomes an issue. When a manufacturer of a part decides to improve their product, such changes must be forewarned and must be evaluated by the configuration control committee. During FSD, surveillance testing is planned. Certain critical parts may require surveillance testing. These may be parts critical to the performance of the system. They may also be parts that are below a Radiation Design Margin. Guidelines describing the acceptance criteria and the test procedures at a radiation test facility are prepared. 12.2 Maintenance / Surveillance Program Some of the physical and management tasks of a maintenance and surveillance program are described in Fig. 36. Some of the physical tasks include electrical functional testing or characterization of Hardness Critical Processes. These include electrical AC / DC performance parameter measurements that characterize a part. Measurements are made and compared with those obtained during FSD. Trends are plotted to determine if degradations have occurred. Surveillance continues with possible lot sampling testing of parts. From the management side, test results are reviewed and analyzed. From the radiation side a survivability analyst has to determine if a degradation will impact the program during its life cycle. If such degradation is of concern, the configuration committee must determine what possible solutions are available. Inspections, Test, Configuration Audits
Physical Assurance Management
Configuration Control, Quality Control, Engineering Data (Drawings), Hardness Baseline, Production Control Repair Procedures, HCP’s, Functional Tests, Identify Degradations
Physical Maintenance Management
Configuration Management, Logistics, Degradation Evaluation, Data, Training Testing & Analysis, Monitoring Repairs, Identify Degradations
Physical Surveillance Management
Predict System Effectiveness Assessment HM, Report Degradations
Figure 36. Composition of Hardness Assurance, Maintenance and Surveillance Program.
12.3 Types of Hardness Degradation There are various types of degradation, some obvious and some not so obvious. These are illustrated in Fig. 37. The general ones include process changes from an electronics manufacturer. This could be a process improvement that includes a die shrink, a
I-37
replacement to an obsolete or discontinued part or a mask change. A process could be discontinued or transferred to another foundry. A new lot may be procured and radiation testing has indicated a lower radiation sensitivity level. Hardness degradations may occur during production or deployment Change must be identified PART Tested / analyzed VARIATION Assess margins PART DESIGN SUBSTITUTION MODIFICATIONS Documented Engineering Control Screen Review Board • Degradation reviewed by Manufacturing Survivability Support equipment Procurement Reliability
NEW VENDOR
STOP
INCORRECT ASSEMBLY
PROCESS CHANGE
INCORRECT PROCEDURE
ERRONEOUS INSPECTION
Figure 37. Types of Hardness Degradation.
The not so obvious degradations include an error in an assembly drawing, such as the mislabeling of a part, wrong part identifier, a new additional manufacturing step, wrong manufacturing procedure, wrong orientation of a part, wrong type of solder, poor ESD (ElectroStatic Discharge) handling procedure or the wrong wiring. These changes or degradations must be reviewed and assessed by the configuration committee. A survivability analyst must assess the degradation and determine whether the system can still meet the mission requirements. If the margin is impacted, possible solutions must be proposed to the customer by the contractor. 12.4 Implementing Hardness Assurance In HA there are procedures and monitoring techniques as described in Fig. 38. One of procedures requires the one to identify the hardness critical items among all the components in a design. This is called HADD or Hardness Assurance Design Documentation. This procedure allows manufacturing to identify whether the component is non-critical or critical and whether electrical testing is required to screen the component, such as parts, boards or packages. If a part is flagged for radiation screening, incoming parts will have to be sampled from each lot. Monitoring of electrical test data of parameters to track a process may be provided by the semiconductor manufacturer for a particular fabrication run that was associated with the parts procured. Inspection and receiving provide proper labeling for handling and storing of stock items for future manufacturing. Safety may provide handling procedures for radioactive materials from neutron or proton irradiation. Procurement may have a procedure for authorizing the final acceptance of a qualified product into the system with proper loggingin and receiving of a product from a manufacturer. I-38
PROCEDURES • Hardness control based on item and margin • Hardness Assurance Design Documentation (HADD) identifies hardness critical items and procedures MONITORING TECHNIQUES • Configuration controls • Process controls • Electrical screens • Lot sample radiation tests • Acceptance tests • Inspection Figure 38. Procedures and monitoring techniques of HA.
12.5 Classifying Parts in a Space Program HCC 1 SPECIAL PART SCREENING CAT 1 CONTROLS REQUIRED MISSION CRITICAL 2 < RDM < 10 PARTS VULNERABLE TO INSPECTION HCC 1 H THE SPACE REQUIRED CATEGORY 1 ALL ENVIRONMENT HCC 2 10< RDM < 100 NORMAL PARTS HCI CATEGORY 2 PART CONTROLS MISSION CRITICAL REQUIRED BUT NOT VULNERABLE TO THE SPACE ENVIRONMENT NOT MISSION CRITICAL
Figure 39. Classifying parts in a space program.
Parts are categorized to various levels of HA as shown in Fig. 39. Radiation design margins (RDM) are determined by taking the ratio between the failure level of a part to the radiation level established at the part inside the package. The RDM is described further in the next section. Depending on what the RDM is, the parts may be classified as not hardness critical or Hardness Critical Items (HCI) and then further categorized as Hardness Critical Category (HCC) Item 2, or HCC Item 1 M or H type depending on its application. Depending on the level of importance, this establishes certain types of handling procedures. They may be electrical, radiation testing, quality inspection or destructive product assessment (DPA). The HCIs are vulnerable to the space environment. They are divided into HCC 2, HCC 1M or 1H. At the bottom of the pyramid are parts that are considered non-mission critical. The next level of parts is considered to be mission critical, but not sensitive to the space environment. HCI Category 2 (HCC 2) parts are considered the third level up which are critical parts sensitive to the space environment but with sufficient RDM. At the fourth
I-39
level HCC 1 M or H Category 1 parts require special inspection which may include radiation testing. These parts are considered critical based on RDM or application. Finally, the HCC 1 Category 1 parts are considered most critical requiring special screening that includes radiation testing. 12.6 Parts Screening Control Parts selection and how their hardness is maintained are important issues in a hardness assurance program. Once the parts have been selected, a plan of how to maintain their hardness is shown in Fig. 40. In the beginning to establish what parts are required for a program such as total ionizing dose, a shielding analysis must be performed to determine the amount of dose a part will see over the entire mission. Elaborate ray tracing analysis may be performed in order to optimize the hardness against cost and weight of a system. Once an estimated design margin has been applied, parts can then be selected. A sample size of a lot is radiation tested to quantify the hardness of the part. Based on the sample size and the statistics of the results, a radiation design margin (RDM) can then be determined by taking the ratio between the failure thresholds of the parts to the dose that the parts will see. Radiation Requirements Radiation Design Margin
failure threshold RDM = design spec
Parts Selection
2< RDM< 10
HCC-1 Lot Acceptance Tests Required
HCC-2 and HNC* Lot Acceptance Tests Not Required
Test Lot
Failed - reject lot
10< RDM< 100
Test Lot
Pass - accept lot
Failed - corrective action Shield, part replacement, redesign circuit, replace part, harden process
* HCC - Hardness Critical Category HNC - Hardness NonCritical
Figure 40. Parts Screening Control Per Mil-Handbook 814 / 815.
If the RDM lies between 10 and 100, then radiation surveillance testing may not be required. If the lot were tested and failed, then corrective action must be implemented. This may include additional shielding of the part, replacing the part, redesigning the circuit or hardening the part. In the case where the RDM is less than 10, but greater than two, sample lots of the part must be tested to monitor the variability in the radiation hardness that might occur. If the
I-40
lot fails, it should be rejected, since the RDM is small. The uncertainty and variability of the lot may place the program at risk. Lot sampling and continuous monitoring during production for HCC-1M parts is the methodology for assuring that the hardness of a part will not change during manufacturing. 12.7 Document Flow Through HA Documents are generated by engineering during FSD. Fig. 41 describes some of these types of documents and what types are transported to manufacturing. The final radiation test reports, design and analysis work are summarized. Design engineers prepare final design drawings. Parts engineers prepare final parts specification drawings and procurement procedures that manufacturing can use. Radiation engineering works together with the parts, design engineers to make sure the proper parts will be procured, and the design complies with the radiation requirements. Manufacturing has to know exactly where to obtain the part types, where to assemble them, how to assemble them and how to keep inventory of the parts. Training and maintenance manuals are generated based on what was developed during the FSD program. PLANS and REPORTS SPACE HARDENING PLAN PLAN
Engineering
HARDNESS ASSURANCE PLAN
Manufacturing DESIGN DRAWINGS DESIGN AND DRAWINGS PRODUCTION AND CONTROLS
TEST PLANS & REPORTS
HARDNESS CRITICAL ITEMS REPORT
TRAINING MANUALS HARDNESS MAINTENANCE & SURVEILLANCE PLANS
MAINTENANCE MANUALS MANUALS SURVEILLANCE & MAINTENANCE ACTIVITIES
Figure 41. Flow of documents through HA.
12.8 Manufacturing Support In manufacturing there are a number of organizations that are required to support a production program. Logistics is just one of the important groups as shown in Fig. 42. Other groups include organizations from (1) support equipment which provides the automated test equipment for testing complex semiconductor devices, (2) parts engineering for providing the test vectors for characterizing incoming parts, (3) personnel training for providing the documents for floor personnel in the guidance in assembling a package or a system, (4) computer resources for automating data processing and logistic tracking, and (5) supply for providing administrative office needs. Logistics must be sure that the drawings or tracking documents are provided in the proper format that an assembler or procurement personnel can use. An example is described in Fig. 43. A designer or a parts
I-41
engineer may be more familiar with a parts list in alphabetical order. An assembler on the manufacturing floor may need an assembly drawing in numerical part order. Therefore a list with the drawing part number is more appropriate on the manufacturing side. Maintenance Planning
Facilities
Support Equipment
Information and Data Management
Personnel and Training
SYSTEM SUPPORT
Packaging Transport Storage
Supply
Design Interface
Computer Resources Support
Logistics Support Management
Figure 42. Engineering Support Organizations.
Engineering: • Summarizes the verification analyses • Indicates which assemblies were analyzed, analysis results, relative location of assembly in system, and location of analysis in HADD • Two lists are generated: 1. Engineering list - arranged by subsystem, assembly, subassembly and piecepart for ease of design reviews 2. Logistic list - arranged by part number in alphanumeric sequence for ease of assembling components on the manufacturing floor Figure 43. Logistic documentation.
13.0 Emerging Issues Some of the arising issues one has to face in future designs are shown in Fig. 44. In the space environment models the heavy ion content extends out to a LET of 80-100 MeV2 cm /mg for both the galactic cosmic and solar event activities [134, 135, 136, 137, 138, 139]. Physicists seem to state that ions heavier than iron do not exist. Ions heavier than iron have not been strongly correlated with solar event activities. This may be important for the use of COTS technologies where one may be able to operate during a solar event.
I-42
• Modeling of heavy ion content - solar events versus galactic cosmic rays • SEE testing of complex ICs in plastic encapsulation • SEE testing of ICs in flipped chip packaging • SEE testing of embedded ICs in oscillators, motors • SEE evaluation of expensive modules or boards • Rad hardening by Hardness by Design • Single event transients in future electronics • Availability/affordability of higher energy accelerators • Hardness assurance of COTS parts • A cost effective HA test for ELDRS • Obsolescence of parts for long term space programs Figure 44. Emerging Issues.
Single event transients may be a growing concern for future complex electronics. Some of these transients when latched as bad data, can cause serious errors to a system if not controlled. As electronics become harder to test, there is the question as to whether existing high energy accelerators can correlate well with the space environment. The energies of particles in space differ greatly from the energies used for testing. Recent experiments of parts have indicated possible differences in effects. During the 1990’s this period opened up the opportunity for space programs to leverage on COTS. However, it was not long that one realized that COTS had their own unique issues. How does one control obsolescence when technologies and processes are changing far more rapidly than any rad hard foundries in the past? How does one monitor the possible lot to lot variation and control the radiation hardness of the parts that are selected? Due to the reduced number of radiation hardened suppliers, many design houses are using layout design rules from commercial foundries to produce radiation hardened parts. Can Hardness By Design (HBD) provide a solution to radiation hardened products [132]? Programs that require many years to develop have to struggle to keep up with the commercial electronics industry. When parts are evaluated and selected during FSD, they may be obsolete by the time manufacturing comes along. How should such programs handle obsolescence? In SEU testing most parts are delidded and the beam exposes a die from the top. However, as the die increases, packaging grows in input /output pad count, or pin count. Manufacturers are considering Ball Grid Arrays (BGA) for higher packaging density. Chips become more difficult to test. Some chips are mounted such that the back side is facing up, hence making it difficult to irradiate unless the backside of the package is grinded off. Decapsulation of plastic encapsulated chips is difficult. Complex devices include packages that house a microprocessor or a FPGA (Field Programmable Gate Array). There are chips in oscillators or in motor drives. When a piece of the assembly is removed, the entire part will fail to function properly. Testing will then require the entire assembly and this may be difficult for heavy ions testing.
I-43
These issues do affect HA. For example, the increased sensitivity of SET in future electronic technologies may indicate that a change in the process to a smaller die size may cause the SETs to increase in the system. This would reduce the original design margin. Parts testing issues will impact how one can conduct surveillance testing in the future. If HBD is successful, this technique could be used to support obsolescence when stock item parts are low on the assembly line. This would be a viable low cost solution. 13.1 Future Radiation Trends in Electronic Technologies Electronic products in the future are becoming more complex, but will have better performance, lower power consumption and more capabilities as shown in Fig 45. As a function of time, the density of bits will increase while the memory cell size will decrease along with its feature size [42].
Potential benefits of non-hardened parts • High speed, high performance, greater capabilities, low cost, low power, higher memory capacity Technology Parameter Trends Parameter/device Units 2001 2003 2005 2007 2010 2013 2016 Feature size, MPU (nm) 65 45 32 25 18 13 9 Equiv. Tox, ASICs (nm) 1.3-1.6 1.1-1.4 0.8-1.3 0.6-1.1 0.5-0.8 0.4-0.6 0.4-0.5 Core voltage (V) 1.8 1.2 1.0 0.9 0.6 0.5-0.6 0.3 6 2 Logic density (10 xtrs/cm ) 38.6 61.2 97.2 154.3 309 617 1235 Product Trend - Complexity and functionality increase Pattern recognition computerized home appliances robotics computerized cars global telecommunication
from 2002 International Technology Roadmap for Semiconductors
Camcorders CDs laptops, PCs mass storage medias cellular phones
Color TVs calculator portable radios microwave ovens
Time
Past (1970) Present (2003) Future (2010)
Figure 45. Future Trends in Electronic Technologies.
MOS gate oxide thicknesses are projected to decrease from 1.3 nm in 2001 to 0.4 nm by 2016, especially for commercial products. Metallurgical junction depths will be shallower based on improved ion implants and the use of RTA (Rapid Thermal Annealing) processes. High k dielectrics for gate oxides are being considered whereas low k dielectrics will be used for the passivation layers with copper in the metallization vias. Supply voltages will be lowered. 13.2 Single Event Upset, Transient and Latch-up Latch-up is a phenomenon that is associated with a semiconductor device that contains a four layer doped structure [58, 59]. In Fig. 46, the picture on the bottom left, is a typical cross section of a CMOS (Complementary Metal Oxide Silicon) transistor. We identify the pnpn four layer doped path starting from the p drain in the nwell to the psubstrate and then to the drain of the n channel. Under normal operating conditions, the p drain to the nwell junction and the psubstrate to the drain of the n channel junction are forward biased. The nwell
I-44
to psubstrate junction is reversed bias and acts as the gate of the pnpn structure. In the picture we depict the pnpn layer by a pnp and an npn bipolar transistor. The resistors represent the substrate and nwell resistances.
Figure 46. What is pnpn Latch-up or SEL (Single Event Latch-up)?
This pnpn layer is also called a silicon controlled rectifier (SCR). It can be used as a switching device for motor control. However, this pnpn layer is inherent in all CMOS bulk processes, typical of many commercial processes. The SCR has two operating conditions as shown in the trace on the bottom right of Fig. 46. It can operate in the low current, high voltage condition or high current, low voltage condition. Typically, SCRs can handle high currents. However, for an inherent parasitic device that is embedded in a CMOS process, this parasitic device may not be current limited. If charge from a particle strikes across the nwell and psub junction, the following actions occur. The base current of the pnp originates from the collector current of the npn. In turn the collector current of the pnp is fed back into the base of the npn. This current loop can be magnified when the product of the npn and pnp gains is greater than one. When this current is exceeded, the pnpn device is turned on, such that all three junctions are now forward biased. If the current is not limited, catastrophic failure can occur affecting the part. The pnpn structure has two modes of operation as mentioned above, (a) low current and high voltage or (b) high current and low voltage. Normally, the device operates in the state of (a). When latch-up occurs, the device goes into the state of (b). In the high current, low voltage condition, the current will continue to rise along with temperature until a metallization, such as the ground or power via, or even a bond wire evaporates, hence severing the connection. In the future electronic technologies and geometries are continuing to shrink. When this occurs, adjacent transistors are closer together, such that the pnpn structure is more I-45
sensitive to latch-up [42]. Although supply voltages are dropping, one hopes that the operating voltage will drop below the turn-on of the SCR where latch-up is sustained. This needs to be assessed in order to determine whether latch-up has been eliminated.
2V/div
Out1
2V/div
Out2
Circuits left in latchup state can induce electromigration damage
500mV/div 20ms/div 2V/div
500mV/div
Out1
Icc
Latch-up of a Samsung 4Mbit SRAM, K6R4016V1C, S/N 3 20µs/div
Solutions: • Screen parts by radiation testing to eliminate parts sensitive to SEL • Use thin epitaxial, SOI, or SOS technologies • Current monitor, current limit the device to ~ 50-100 mA above the operation of the device followed by recycling power Figure 47. Latch-up can be Disruptive to Normal Operations.
Latch-up can be disruptive to normal operation as shown in Fig. 47. It can also be a serious problem if it is catastrophic. The upper two curves are measurements of two output channels of a 4Mbit Samsung SRAM. The third trace is indicated by the trigger (arrow) which is monitoring the supply current, Icc. One of the outputs and the Icc current is shown in a magnified horizontal scale as the bottom two traces. Here, one can see the initiation of latch-up by the sudden rise in the supply current and the drop in the output signal to ground. The current monitor is a current transformer (CT) probe that only monitors the sudden rise in current. However, a DC current meter from a power supply indicated that the chip was drawing a large excess of current, i.e., greater than 200 mA. When latch-up occurred, the outputs failed by dropping to ground. The initial output pattern was an alternating checkerboard pattern. Power had to be turned off and turned back-on in order to remove the part out of its latched-up state. The checkerboard pattern had to be reloaded also in order to return the part to normal functional conditions. Unless satellite systems are autonomous with automated power recycling, ground intervention can take minutes to hours. If a latch-up were to occur, then large current densities could easily pass through the metallization vias for long periods. If power were not recycled quickly, high currents through the metallization vias can result in electromigration [63]. Some of the hardening techniques are as follows. External to the semiconductor device at the board level, one can use a current limiting resistor between 2-20 Ω/V or (limiting the current to 500 to 50 mA), depending on where the holding current is on the latch-up I-V I-46
characteristic curve. The holding current is defined as the minimum amount of current to sustain the high current condition of the pnpn latch-up structure. The resistor is connected in series with the part to be protected. Note however, that when a resistor is applied external to the device rather than internally in an integrated circuit (IC) chip, a voltage drop will be applied to the part. When this occurs, the part may be operating below the minimum specified voltage and needs to be evaluated for operation and design acceptance. When the resistor is added in an IC chip, the voltage drop is accounted for in the design. Parts sensitive to SEL can be tested at heavy ion facilities, such as Brookhaven National Laboratory and Texas A/M University. Other resources to search for SEL include ion micro-beam at Sandia National Laboratory or the micro-beam lasers at Jet Propulsion Laboratory, Naval Research Laboratory or Aerospace Corp. [60, 61, 62]. Other solutions at the semiconductor level in reducing the amount of charge generation are to reduce the volume of the space charge region around the pn junctions. This can be done by using an epitaxial layer, truncating the collection volume to a very thin layer. In the epitaxial layer, the nwell and psubstrate resistances are designed to be very low. By making the resistances low, one shorts the p drain to the nwell junction and the psubstrate to the drain of the n channel junction such that the pnpn device can not be turned on that easily. Another method would be to eliminate the silicon substrate by either using silicon on sapphire (SOS) or silicon on insulator (SOI) processes. A Single Event Upset (SEU) is an electronic circuit effect caused by the passage of an ionizing particle that temporarily changes the logic state of a memory cell in an integrated circuit. This effect impacts sequential logic devices. Examples of these devices are flip flops, latches, cache memory, registers in microprocessors, DRAM, SDRAM (synchronous DRAM), or SRAM memories [71, 73]. All these devices can lose their original memory bits if a particle could upset these devices [66, 68, 69, 70]. The loss is not permanent in the sense that the device can be refreshed or restored with its original information later. SEUs are considered soft errors. Also we distinguish SEUs from single event transients or glitches where the latter will cause a circuit to recover to its original state without the need for a refresh or restoring process. In the case of glitches, they can not upset or disturb the memory bit of a sequential logic circuit. The critical charge, Qc, depends on the square of the critical dimension or feature size, such as a contact opening or channel length of a MOS transistor. As the critical dimension or the feature size is reduced, capacitance at a circuit node will be reduced. Since Q = CV, the critical charge will decrease also as the capacitance is lowered. However, it appears that if the commercial manufacturers were to mitigate alpha particle upsets in their products, Qc will be limited by the maximum alpha particle deposited charge. Although a lower charge limit is predicted, future electronic technologies may still be sensitive above this limit to upsets or even transients. In order to increase the speed of a device we also know that higher performance and lower operating voltages lead to less charge on a circuit node. With less capacitance and voltage, it is anticipated that a circuit node will become more sensitive to radiation. We find that many commercial complex chips, such as microprocessors [75, 77, 78], ASICs and FPGAs are sensitive to upsets.
I-47
SEU (Single-Event Upset) is an electronic circuit effect caused by the passage of an ionizing particle that temporarily changes the logic state in a memory cell in an integrated circuit. • Refreshing the cell corrects the error • Reducing a memory feature size decreases the charge holding the logic bit in a memory cell (i.e., Q = CV) • High performance and radiation hardening are competing attributes • Manufacturers may limit critical charge to alpha upset (10-2 pC) • Future memory devices may still upset above 10-2 pC Vcc BIT • Sensitive circuits include: SEU sensitive – memories - SRAMs Node1 – DRAMs, SDRAMs NA1 – Flip flops Wordline – Registers in a microprocessor – Latches
BITBAR P1
P2 Node2 NA2 Wordline
N1
N2 Vss
SRAM memory bit
Figure 48. What is an SEU (Single Event Upset)?
Soft Error FIT Rate/Mbit
Fig. 48 shows a typical SRAM memory bit cell circuit. The flip flop is composed of six transistors, two are access or pass transistors while the other four transistors form two cross coupled CMOS inverters tied together as shown. The sensitive nodes are at the gates of the CMOS inverters. If a particle were to deposit enough charge on either of those nodes, the flip flop will change its state. The most vulnerable state is where either the n or p channel transistors of the flip flop are in the OFF or HI state. If an ionizing particle were to discharge the capacitor, the charge would leak to ground. This event would bring that leg of the inverter to a LO state and force the other inverter to a HI state, hence the upset [65]. 0.14 µm, Shrunk Cell 0.14 µm 0.16 µm, Shrunk Cell 0.16 µm 0.25 µm
1000
100
Voltage scaling has a first order (exponential!) impact on singleevent upset rate
Terrestrial
0.0
FIT Rate in NYC
1.0
2.0
3.0
4.0
5.0
Power Supply (V) The increasing SEU sensitivity of advanced commercial technologies is making terrestrial upset a primary reliability concern in ICs.
Figure 49. Impact of Silicon IC Scaling on Single Event Upset. [140]
I-48
SEUs may increase as shown in Fig. 49 as the supply voltage is reduced in the future. The FIT (Failure In Time) is predicted to increase with decreasing supply voltage. Feature sizes from 0.25 micron to 0.14 micron are shown for these predictions as calculated by Dodd et al [140]. The FIT appears to scale exponentially with voltage. This is alarming because this indicates that one might expect more SEUs or SETs from commercial products in the future [72, 74, 76]. Besides sequential logic circuits, there are a number of different types of combinatorial logics devices. These devices are comprised of NANDs, NORs, ORs and inverters. Other devices that are not combinatorial parts, but are sensitive to transients are the linear devices, such as voltage references, regulators, comparators, operational amplifiers, oscillators and electro-optical devices [64, 67, 79, 80]. In Fig. 50 a combinatorial logic circuit is shown, comprised of NANDs, ANDs and inverters that finally lead to two sequential logic D flip flops. If the timing were such that the transient signal that is generated propagates to the Data (D) input of both sequential logic devices at a time when data is acquired by the clock pulse, then an erroneous signal can be latched. We have already mentioned the concerns with lower critical charges at circuit nodes for future electronic circuits. Transient upset signals can vary from a few mV to the supply voltage level. Glitches are momentary, and can vary from a few ns to microsecs. A sample of a SET is shown as measured from a voltage regulator chip. The top trace is the current response from a CT (current transformer) probe on the supply line. The second trace is the output response. The AC signal is a few hundred mVs for about 150 µs. • Single-event charge deposition creates a transient noise glitch SET in a combinatorial circuit • Occurs in combinatorial logic, linears, electro-optical devices from Prof. Lloyd Messengill, Vanderbilt Univ. • Effect depends on: S2 G D Q – Combinatorial logic paths D A1 Possible Q – State of sequential logic Path H State 000 E – Timing S1 B1 D Q – If erroneous signal Q I Possible is latched - soft error F Path C1 occurs State 100 Combinational SE Hit
Icc SET in a linear voltage regulator chip
output
CLK
200mV/div 100mV/div
50 µs/div Figure 50. Upsets in Combinatorial Logic Circuitry.
Single Event Transients can also occur in electro-optical devices. Glitches may vary from a few mVs to 3-4V. SETs will occur in high speed optocouplers [110, 111]. To mitigate I-49
SETs [115], one can use capacitors that will filter small glitches. Slowing down the response time of a device is another possible solution. This way the circuit can respond in the millisecond time frame and would not respond to glitches in the microseconds or nanosecond time frame. Other ways of mitigating SETs at the system level include parity checking, constant refreshing or voting logic. Redundancy of circuits, such as triple modular redundancy (TMR) is a possible solution if space is available. A watch-dog timer is used to monitor the status of a processor. Some watch-dog timers may upset so they will require protection from resetting a microprocessor frequently. Besides implementing hardening techniques to SETs, one needs to perform an analysis to assess the risk of a SET in a system. Radiation testing of opto-couplers should also be considered [112, 113, 114]. 13.3 Enhanced Low Dose Rate Sensitivity A radiation effect that applies to a number of bipolar linears is the ELDRS (Enhanced Low Dose Rate Sensitivity) effect. High dose rate at 100 rad(SiO2)/s has been the standard for testing parts with cobalt-60. However, as lower dose rates are used which are more representative of the space environment, certain linear devices have shown to degrade more rapidly. An example is the input offset current of an operational amplifier, OP42 [87]. The input offset current rose as the dose rates was lowered below 1 rad(SiO2)/s. • ELDRS is a radiation effect observed only at very low dose rates similar to the space environment in bipolar technologies • Parts irradiated at very low dose rates (< 1 rad(SiO2)/s) may show higher excess leakages than 100 rad(SiO2)/s • No proven approach to ELDRS testing; best guide - American Standard for Testing and Materials F-1892-98 no Bipolar Device? Standard room test yes unclear Include any preelevated no dose rate ELDRS data? temperature stress in dependence samples Conduct dose yes rate tests Option 3 - 5 mrad/s test at yes, dose rate dependence 25°C, apply a 1.5X design Lot acceptance test Option 1 - Test at application dose rate
Option 2 - Conduct test 0.5-5 rad(SiO2)/s at 100°C, apply 3X design margin
margin (<30krad) or irradiate at 25°C for 2500 hrs at higher dose rate (>30krad), apply a 2X design margin
Figure 51. ELDRS (Enhanced Low Dose Rate Sensitivity).
The discovery of this effect and the understanding of the ELDRS have been presented at past Nuclear and Space Radiation Effects Conferences and have been published in the IEEE Transactions of Nuclear Science [88, 89, 90, 91, 92, 93, 94, 95, 96, 97]. The most recent theory suggests that the effect may be caused by competing mechanisms of how rapid hydrogen or the hole can migrate to the interface. If the holes were to migrate to the
I-50
interface before the hydrogen atoms at high dose rates, a charge barrier is created, preventing the hydrogen from creating interface states. Whereas at low dose rates, some of the hydrogen atoms can reach the interface and at very low dose rates, more hydrogen atoms will get through to create the interface states. Presently there is no proven accepted methodology for a cost effective screen for ELDRS. Recommendations as suggested in the ASTM F 1892-98 are shown in Fig. 51. To determine if a bipolar device exhibited ELDRS, a literature search should be first conducted. If data do not exists, then dose rate testing at two levels should be performed, one at 50 rad(Si)/s, while the other at a lower dose rate, such as 1 rad(Si)/s. A comparison of the two dose rate data should reveal any evidence of ELDRS. An indication of ELDRS would require a lot acceptance test. Three possible approaches are given. The first approach suggests irradiating at a dose rate that the device will be used. A second approach suggests irradiating the devices at 0.5-5 rad(SiO2)/s at 100°C to a specified dose. Then a factor of three design margin is applied to the sensitive parameter. The third approach suggests irradiating the parts at 25°C at 5 mrad(SiO2)/s and apply a factor of 1.5 design margin or irradiate at a higher dose rate for 2500 hrs for (>30 krad requirement) and apply a factor of 2X design margin. 13.4 SOI Future Technology SOI, silicon on insulator may be a very promising technology in the future as described in Fig. 52. It has a number of benefits. A cross section of the technology is shown on the upper right. Each individual transistor is isolated from each other by a silicon dioxide layer called the buried oxide. This eliminates SEL completely. With less active volume, less charge can contribute to an upset or transient effect. With less capacitance the speedperformance goes up with the technology. Although SOI has many performance benefits, inherent hardness must be understood
Source-n+ Field oxide
Poly
Drain n+
P-body ----++++++++
Field oxide
Buried oxide
Silicon substrate
Benefits of SOI - less volume → less charge collection, higher LET - pnpn immune - reduced capacitance → high speed - smaller junctions → less carrier generation, higher LET Issues with SOI - parasitic bipolar transistor • Transistor snapback - secondary breakdown, SEB • Total dose induced breakdown • SEUs possible with poor design, i.e., LET of 4 MeV-cm2/mg • Buried oxide creats trapped charge effects Solutions: body ties, ion implant, cross-coupling poly-resistors From J. Schwank, Sandia National Laboratories
Figure 52. SOI Future Technology – Radiation Effects
I-51
However, SOI has its own issues if the design were not properly done. For example, the buried oxide like any ordinary oxide can build up charge from the passage of an ionizing particle. The charge in the buried oxide will create a back-gate voltage shift in the MOS transistor. For an n channel transistor, the charge from the back-gate oxide could cause an inversion layer and create a leakage path between the drain and source of the transistor. Also with less capacitance, a flip flop circuit can now be prone to SEUs as low as a LET of 2 4 MeV-cm /mg. To mitigate some of the problems, cross-coupling poly resistors are considered for hardening the transistor to SEUs. Ion implantation on the back substrate can reduce the possibility of back-gate threshold shifts. Finally, body ties can eliminate the possibility of the parasitic bipolar transistor within a MOS device from contributing to any SEE enhancement effects. 10.0 Conclusion In conclusion as shown in Figs. 53 and 54, system hardness assurance is about the methodologies or disciplines needed to assure that a system designed during development will maintain the design margins during manufacturing. We began by identifying three phases in the system hardness assurance program. These three phases are the program initiation, the full-scale development and the manufacturing or production. In the program initiation phase, the space environment specification is defined. Also a compliance document is generated describing what tests and analyses are needed to comply to the environment specification at the package level. Parts technologies are being evaluated against the environments and traded off with cost in order to make the system affordable. During the early phase of FSD circuit designs are developed with FPGAs and non hardened parts in order to validate design concepts. Degraded parameters of the selected technologies are obtained from proton and cobalt-60 testing. The degraded parameters are included in the circuit design. Once the designs are approved, the designs are fabricated in the selected hardened technologies. Selections of part types are irradiated to determine if the designs will meet the radiation requirements. Single event effects, protons and total ionizing dose testing are conducted. Once the parts have passed the radiation tests, boards and packages are assembled during the latter years of FSD for electrical, thermal, vibrational, and shock evaluation. Results are documented and lessons-learned procedures are written. Interfacing with manufacturing is required to provide the training manuals, procedures and data necessary for the assembly personnel to use. In manufacturing critical parts are carefully monitored and radiation tested for changes in the design margin. Degradations, such as parts obsolescence, miswiring, mislabeling, dents in a package, or process change on a product from the manufacturer are notified to the configuration management board for review. Discrepancy reports are written explaining any changes that are approved and are filed. Radiation hardening techniques that a designer can use were presented. A shielding and an upset example were described. In each case, the results are related back to how HA can be affected. The hardening techniques included (a) parts selection, (b) shielding, (c)
I-52
circumvention, (d) EDAC, (f) parity checking, (g) analyses and modeling codes, and (h) radiation testing. These are some of the important methodologies that go into the design of a space system to assure that the system will meet the radiation requirements over the lifetime of a program. System Hardness Assurance Methodology/discipline of assuring that the production processes did not adversely reduce the margins designed during development Program Initiation Defined space environment specification Compliance document - test & analysis approach Full Scale Development Apply radiation design margin – Shielding – Upset risk analysis – Account for radiation effects - TID Types of system hardening approaches Figure 53. Conclusion
Finally, design issues were raised. Some of these included possible limitations in testing parts at a heavy ion facility. An example is the case of a die in a flip chip configuration. Still another issue is the absence of a proven HA for bipolar parts with ELDRS. Another issue is the possible increased sensitivity of SETs in future electron technologies. It is important to remember that the space industry lost several hundred millions of dollars during 1998 and 1999 due to various launch problems. It could also recur if we ignored the methodologies or disciplines discussed here in this short course today. By understanding some of the radiation effects and knowing how the hardening techniques are applied, this will help us know what to look for when a change occurs. We will know whether an increase in SETs or test issue will impact the hardness of our system design. Many challenges lie before us in developing our future space systems and many hardening tools and methodologies are available to us today. Manufacturing / Production Composition of Maintenance / Surveillance tasks Types of hardness degradations / changes Configuration management Monitoring techniques Parts categorization Future issues SEE testing limitations Increased sensitivity of future technologies to SETs Cost effective HA test for ELDRS Web location of hardness assurance-related documents http://www.spwghac.org/Radiation%20Info.htm Figure 54. Conclusion
I-53
11.0 Acknowledgements The combination of writing two short courses within a year has been a challenging experience. The first author would like to thank Dr. Jim Schwank and Dr. Joe Srour for allowing me that opportunity. Dr. Srour’s valuable and encouraging suggestions are much appreciated. The first author would like to also thank his wife, Anna for reviewing the manuscript and his three sons, especially the youngest one, Andrew for patiently waiting while his dad was working long hours over the Christmas holiday, Spring break and throughout the year in putting together this manuscript and assembling the charts. The first author would also like to thank Marion and Harvey Eisen for reviewing the manuscript and my colleagues for their wonderful technical discussions/support and again to Harvey for providing a complete list of HA related references. The second author wishes to acknowledge the extensive effort of the first author in preparation of this short course and to thank him for the opportunity to share in this work. 12.0 References General [1] IEEE Trans. Nuclear Science, Dec. issues, selected papers from the Nuclear and Space Radiation Effects Conference (NSREC). [2] IEEE NSREC Data Workshop Proceedings. [3] IEEE Nuclear and Space Radiation Effects Conference Short Course. [4] Proceedings of RADECS (Radiation and its Effects on Components and Systems) - European version of IEEE NSREC. [5] V.A.J. van Lint, T.M. Flanagan, R.E. Leadon, J.A. Naber & V.C. Rogers, Mechanisms of Radiation Effects in Electronic Materials, Wiley. [6] E.J. Daly, “The Radiation Environment: The Interaction of Radiation with Materials,” computer methods ESA, 1989. [7] A. Holmes-Siedle, and L. Adams, Handbook of Radiation Effects, Oxford University Press. 1993. [8] G. Messenger, M. Ash, The Effects of Radiation on Electronics Systems, 2nd edition, Van Nostrand Reinhold, 1992. [9] G.C. Messenger and M.S. Ash, Single Event Phenomena (Kluwer Academic Publishers, New York, 1997). [10] S.M. Sze, Physics of Semiconductor Devices, Wiley. [11] T.P. Ma and P.V. Dressendorfer, Ionizing Radiation Effects in MOS Devices and Circuits (WileyInterscience, New York, 1989). [12] E.M. Silverman, NASA Contractor Report 4661 Part 1, Space Environmental Effects on Spacecraft: LEO Materials selection guide, August 1995.
Space Environment [13] J. Barth, “Modeling Space Radiation Environments,” 1997 IEEE NSREC short course, Snowmass, Jul. 1997. [14] J.E. Mazur, “The Radiation Environment Outside and Inside a Spacecraft,” 2002 IEEE Nuclear and Space Radiation Effects Conference Short Course Notebook, Section 2, pp. 1-69 (Phoenix, June 2002). [15] European Cooperation for Space Standardization (ECSS), “Space Engineering Space Environment,” ECSS-E-10-04A, Jan. 2000. [16] H.C. Koons, J.E. Mazur, R.S. Selsnick, J.B. Blake, J.F. Fennell, J.L. Roeder, and P.C. Anderson, “Impact of the Space Environment on Space Systems,” Aerospace Report TR-99 (1670)-1 (July 20, 1999). [17] National Oceanic and Atmospheric Administration Space Environment Center, http://www.sec.noaa.gov/today.html (June 16, 2003). [18] Canadian Space Weather Forecast Centre at http://www.spaceweather.gc.ca. [19] Naval Research Laboratory Large Angle and Spectrometric Coronagraph Experiment (LASCO), http://lasco-www.nrl.navy.mil/rt-movies.html (June 16, 2003).
I-54
Trapped Radiation [20] S.L. Huston, and K. A. Pfitzer, “A New Model for the Low Altitude Trapped Proton Environment, “IEEE Trans. Nuc. Sci., vol. 45, no. 6, pp. 29972-2978, Dec. 1998. [21] D.M. Sawyer, and J.I. Vett, “AP8 Trapped Proton Environment for Solar Maximum and Solar Minimum,” NSSDC/WDC-A-R&S 76-06, NASA-GSFC, 1976. [22] J.I. Vette, “The AE-8 Trapped Electron Model Environment,” NSSDC/WDC-A-R&S 91-24, NASA-GSFC, 1991.
Solar Events [23] J. Feynman, G. Spitale, J. Wang and S. Gabriel, “Interplanetary Proton Fluence Model: JPL 1991,” J. Geophys. Res. 98, A8, pp. 13281-13294, 1993. [24] J.H. King, “Solar proton fluences for 1977-1983 Space Missions,” J. Spacecrafts and Rockets, 11, 401, 1974. [25] 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. On Nucl. Sci., vol. 46, no. 6, pp. 1481-1485, Dec. 1999. [26] M.A. Xapsos, J.L. Barth, E.G. Stassinopoulos, “Space Environment Effects: Model for Emission of Solar Protons (ESP) – cumulative and worst case event fluences,” NASA report NASA/TP-1999-209763, Dec. 1999. [27] M.A. Xapsos, G.P. Summers, J.L. Barth, E.G. Stassinopoulos, E.A. Burke, “Probability Model For Cumulative Solar Proton Event Fluences,” IEEE Trans. On Nucl. Sci., vol. 47, no. 3, pp. 486-490, Jun. 2000. [28] Carlowicz, M.J., Lopez, R.E., “Storms from the Sun: The Emerging Science of Space Weather: Joseph Henry Press: Washington, DC, 2002, 256 pp. [29] Golub, L., J.M. Pasachoff, Nearest Star: The Surprising Science of Our Sun, Harvard University Press, Cambridge, MA, 2001, pp. 267. [30] J.G. Kappenman, L.J. Zanetti, W.A. Radasky, Geomagnetic Storms Can Threaten Electric Power Grid, Earth in Space 1997, 9, (7), 9. [31] Parker, E.N. Interplanetary Dynamical Processes, Interscience Publishers: New York, 1963, pp. 272.
Galactic Cosmic [32] D. Binder, et al., “Satellite Anomalies from Galactic Cosmic Rays,” IEEE Transactions on Nucl. Sci., vol. 22, pp. 2675-2680 (1975).
Radiation Codes [33] A.J. Tylka, J.H. Adams, P.R. Boberg, B. Brownstein, W.F. Dietrich, E.O. Fleuckiger, E.L. Petersen, M.A. Shea, D.F. Smart, and E.C. Smith, “CREME96: a Revision of the Cosmic Ray Effects on MicroElectronics Code,” IEEE Trans. Nucl. Sci., vol. 44, no. 6, pp. 2150-2160, Dec. 1997. [34] J.H. Adams, “Cosmic Ray Effects on Microelectronics, Part IV,” NRL Memorandum Report 5901, Naval Research Laboratory, Washington DC 20375-5000, USA, 1986. [35] Space Radiation, electronics space radiation effects commercial software available from Severn Communication Corp., 1023 Benfield Blvd., Millersville, MD 21108. [36] L.J. Lorence, “II. Radiation Transport Phenomena and Modeling,” in 1997 IEEE Nuclear and Space Radiation Effects Conference Short Course, Snowmass, Colorado, July 1997, pp. IIA-1 to IIA-29. [37] T. Jordan, “NOVICE a Radiation Transport/Shielding Code,” Experimental and Mathematical Physics Consultants, Report #EMP.L82.001, Jan 1982. [38] P. Truscott, F. Lei, C. Dyer, C. Ferguson, R. Gurriaran, P. Nieminen, E. Daly, J. Apostokalis, S. Giani, M. Grazia Pia, L. Urban, and M. Maire, “GEANT4 – A new Monte Carlo Toolkit for Simulating Space Radiation Shielding and Effects,” IEEE NSREC 2000 Data Workshop Proceeding, pp. 147-152, 2000. [39] S. Seltzer, “SHIELDOSE: A computer Code for Space Shielding Radiation Dose Calculations,” NBS Technical Note 1116, National Bureau of Standards, May 1980.
Radiation Effects of Materials [40] F. McLean, and T. Oldham, “Basic Mechanisms of Radiation Effects on Electronic Materials, Devices and Integrated Circuits,” 1987 IEEE NSREC short course, Snowmass, Jul. 1987.
I-55
[41] R. Mangeret, T. Carriere, and J. Beaucour, “Effects of Material and/or Structure on Shielding of Electronic Devices,” IEEE Trans. Nucl. Sci., vol. 43, no. 6, pp. 2665-2670, Dec. 1996.
Future Electronics [42] A.H. Johnston, “Radiation Effects in Advanced Microelectronics Technologies,” RADECSS 1997 Proceedings, pp. 1-16, 1998.
Single Event Effects [43] J. Barth, “Ionizing Radiation Environment Concerns,” SEECA Single Event Effect Criticality Analysis, http://radhome.gsfc.nasa.gov/radhome/papers/seecal.htm, 1996. [44] S. Buchner, D. McMorrow, J. Melinger, and A.B. Campbell, “Laboratory Tests for Single Event Effects,” IEEE Trans. Nucl. Sci., vol. 43, no 2, pp. 678-686, Apr. 1996. [45] S. Duzellier, R. Ecoffet, “Recent Trends in Single Event Effect Ground Testing,” IEEE Trans. Nucl. Sci., vol. 43, no.2, pp. 671-677, Apr. 1996. [46] M. Gates, “Single Event Effect Criticality Analysis (SEECA) Functional Analysis and Criticality,” http://radhome.gsfc.nasa.gov/radhome/papers/seecai.htm, 1996. [47] A.H. Johnston, “Single Event Effect Criticality Analysis (SEECA) Effects in Electronic Devices and SEE Rates,” http://radhome.gsfc.nasa.gov/radhome/papers/seecai.htm, 1996. [48] K. LaBel, “SEECA Single Event Effect Criticality Analysis, SEU propagation analysis: system level effects,” http://radhome.gsfc.nasa.gov/radhome/papers/seecai.htm, 1996. [49] P. Marshall, “SEECA, SEE Criticality Assessment Case studies,” http://radhome.gsfc.nasa.gov/radhome/papers/seecai.htm, 1996. [50] R. Koga, S.H. Crain, P. Yu, and K.B. Crawford, “SEE Sensitivity Determination of High Density DRAMs with Limited Range Heavy Ions,” IEEE NSREC 2000, dataworkshop proceedings, pp. 45-52, 2000. [51] R. Koga, P.Yu, K.B. Crawford, S.H. Crain, and V.T. Tran, “Permanent Single Event Functional Interrupts (SEFI) in 128 and 256 Mbit Synchronous Dynamic Random Access Memories (SDRAMs),” IEEE NSREC 2001, dataworkshop proceedings, pp. 6-13, 2001. [52] K.A. LaBel, A.K. Moran, E.G. Stassinopoulos, J.M. Barth, C.M. Seidleck, P. Marshall, M. Carts, C. Marshall, J. Kinnison and B. Carkhuff, “Single Event Effect Test Results for Candidate Spacecraft Electronics,” p. 14 – 21, 1997 IEEE Radiation Effects Data Workshop. [53] 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,” p. 39 – 50, 1998 IEEE Radiation Effects Data Workshop. [54] B.G. Henson, P.T. McDonald and W.J. Stapor, “SDRAM Space Radiation Effects Measurements and Analysis,” p. 15 – 23, 1999 IEEE Radiation Effects Data Workshop. [55] M.V. O’Bryan, C.M. Seidleck, K.A. LaBel, R.A. Reed, J.L. Barth, C.J. Marshall, D.K. Hawkins, J.W. Howard, H.S. Kim, M.A. Carts, P.W. Marshall and K.E. Forslund, “Recent Radiation Damage and Single Event Effect Results for Microelectronics,” p. 1 – 14, 1999 IEEE Radiation Effects Data Workshop. [56] W. Hajdas, J. Bialkowski, M. Fivian, R. Henneck, A. Mchedlishvili, E. Sturcke, K. Thomsen and A. Zehnder, “Components Testing for HESSI Satellite Aspects Modules,” p. 92 – 95, 1999 IEEE Radiation Effects Data Workshop. [57] J.W. Howard, M.A. Carts, R. Stattel, C.E. Rogers, T.L. Irwin, C. Dunsmore, J.A. Sciarini, and K.A. LaBel, “Total Dose and Single Event Effects Testing of the Intel Pentium III (P3) and AMD K7 Microprocessors,” IEEE NSREC 2001, dataworkshop proceedings, pp. 38-47, 2001.
Single Event Latchup [58] W.A. Kolasinski et al., “Simulation of Cosmic-Ray Induced Soft Errors and Latchup in IntegratedCircuit Computer Memories,” IEEE Trans. Nucl. Sci., vol. 26, pp. 5087-5091, (1979). [59] " D.K. Nichols, J.R. Coss, R.K. Watson, H. R. Schwartz and R.L. Pease, An Observation of ProtonInduced Latchup," p. 1654-1656, IEEE Trans. Nucl Sci., vol. 39, Dec. 1992. [60] S.C. Moss et. al., “Correlation of Picosecond Laser-Induced Latchup and Energetic Particle-Induced Latchup in CMOS Test Structures,” IEEE Trans. Nucl. Sci. vol. 42, pp. 1948-1956 (1995). [61] A.H. Johnston, G.M. Swift and L.D. Edmonds, “Latchup in Integrated Circuit from Energetic Protons,” p. 2367 – 2377, IEEE Trans. Nucl. Sci., vol. 44, Dec. 1997.
I-56
[62] J.V. Osborn, D.C. Mayer, R.C. Lacoe, S.C. Moss, and S.D. LaLumondiere, “Single Event Latchup Characteristics of Three Commercial CMOS Processes,” Proceedings, 7th NASA Symposium on VLSI Design (1998). [63] T.F. Miyahira, A.H. Johnston, H.N. Becker, S.D. LaLumondiere, and S.C. Moss, “Catastrophic Latchup in CMOS Analog-to-Digital Converters,” IEEE Trans. On Nucl. Sci., vol. 48. pp. 1833-1840 (2001).
Single Event Upset [64] R. Koga, S.D. Pinkerton, S.C. Moss, D.C. Mayer, S. LaLumondiere, S.J. Hansel, K.B. Crawford, and W.R. Crain, “Observation of Single event Upsets in Analog Microcircuits,” IEEE Trans. Nucl. Sci., vol. 40, pp. 1838-1844 (1993). [65] T. Calin, M. Nicolaidis, and R. Valazco, “Upset Hardened Memory Design for Submicron CMOS Technology,” IEEE Trans. Nucl. Sci., vol. 43, pp. 2874-2878 (1996). [66] B. Doucin, T. Carriere, C. Povey, P. Garnier, J. Beaucour, “Model of Single Event Upsets Induced by Space Protons in Electronic Devices,” RADECS 1995 proceedings, pp. 402-408, 1996. [67] R. Koga, S.H. Penzin, K.B. Crawford, W.R. Crain, S.C. Moss, S.D. Pinkerton, S.D. LaLumondiere, and M.C. Maher, “Single event Upset Sensitivity Dependence of Linear Integrated Circuits on Bias Conditions,” IEEE Trans. Nucl. Sci., vol. 44, pp. 2325-2332 (1997). [68] R. Koga and W. A. Kolasinski, “Heavy-Ion Induced Single Event Upsets of Microcircuits: A Summary of the Aerospace Corporation Test Data,” IEEE Trans. Nucl. Sci. vol. 31, pp. 1190-1195). [69] R.A. Reed, M.A. Carts, P.W. Marshall, C.J. Marshall, O. Musseau, PJ. McNulty, D.R. Roth, S. Buchner, J. Melinger and T. Corbière, “Heavy Ion and Proton-Induced Single Event Multiple Upset,” p. 2224 – 2229, IEEE Trans. Nucl. Sci., vol. 44, Dec. 1997. [70] K.A. LaBel, P.W. Marshall, J.L. Barth, R.B. Katz, R.A. Reed, H.W. Leidecker, H.S. Kim and C.J. Marshall, “Anatomy of an In-Flight Anomaly: Investigation of Proton-Induced SEE Test Results for Stacked IBM DRAMs,” p 2898 – 2903, IEEE Trans. Nucl. Sci., vol. 45, Dec. 1998. [71] S. Duzellier, R. Ecoffet, D. Falguère, T. Nuns, L. Guibert, W. Hajdas and M.C. Calvet, “Low Energy Proton Induced SEE in Memories,” p. 2306 – 2310, IEEE Trans. Nucl. Sci., vol. 44, Dec. 1997. [72] V.V. Miroshkin, M.G. Tverskoy, “A Simple Approach to SEU Cross Section Evaluation,” IEEE Trans. Nucl. Sci., vol, 45, no. 6, pp. 2884-2890, Dec. 1998. [73] C. Poivey, B. Doucin and R. Harboe-Sorensen, “Radiation Characterization of Commercially Available 1Mbit/4Mbit SRAMs for Space Applications,” p. 68 – 73, 1998 IEEE Radiation Effects Data Workshop. [74] J. Barak, J. Levinson, A. Akkerman, E. Adler, A. Zentner, D. David, Y. Lifshitz, M. Hass, B.E. Fischer, M. Schlögl, M. Victoria and W. Hajdas, “Scaling of SEU Mapping and Cross Section and Proton Induced SEU and Reduced Supply Voltage,” p. 1342 – 1353, IEEE Trans. Nucl. Sci., vol. 46, Dec. 1999. [75] D.M. Hiemstra and A. Baril, “Single Event Upset Characterization of the Pentium MMX and Pentium II Microprocessors Using Proton Irradiation,” p. 1453 – 1460, IEEE Trans. Nucl. Sci., vol. 46, Dec. 1999. [76] S.P. Buchner, T.J. Meehan, A.B. Campbell, K.A. Clark and D. McMorrow, “Characterization of SingleEvent Upsets in a Flash Analog-to-Digital Converter (AD9058),” p. 2358 – 2364, IEEE Trans. Nucl. Sci., vol. 47, Dec. 2000. [77] D.M. Hiemstra, “Single Event Upset Characterization of the Pentium® MMX and Celeron® Microprocessors using Proton Irradiation,” p. 39 – 44, 2000 IEEE Radiation Effects Data Workshop. [78] G.M. Swift, S.M. Guertin, F. Farmanesh, F. Irom and D.G. Millward, “Single Event Upset in the Power PC750,” presented at the 2001 IEEE NSREC.
Single Event Transients [79] G. Soelkner, P. Voss, W. Kaindl, G. Wachutka, K.H. Maier and H.W. Becker, “Charge Carrier Avalanche Multiplication in High-Voltage Diodes Triggered by Ionizing Radiation,” p. 2365 - 2372, IEEE Trans. Nucl. Sci., vol. 47, Dec. 2000. [80] R.L. Pease, A. Sternberg, L. Massengill, R. Schrimpf, S. Buchner, M. Savage, J. Titus, and T. Turflinger, “Critical Charge for Single Event Transients (SET) in Bipolar Linear Circuits,” IEEE Trans. Nucl. Sci., vol. 48, no. 6, pp. 1966-1972, Dec. 2001.
Dose Ionization [81] S. Dowling, “Comparative effect of Gamma Total Dose on Surface Mount and Non Surface Mount Bipolar Transistors,” RADECS 1993 Proceedings, pp. 338-343, 1994.
I-57
[82] M. Shaneyfelt, D.M. Fleetwood, J.R. Schwank, T.L. Meisenheimer, and P.S. Winokur, “Effects of burnin on Radiation Hardness,” IEEE Trans. Nucl. Sci., vol. 41, no. 6, pp. 2550-2559, 1994. [83] S. Clark, J.P. Bings, M.C. Maher, M.K. Williams, D.R. Alexander, and R.L. Pease, “Plastic Packaging and Burn-in Effects on Ionizing Dose Response in CMOS Microcircuits, “IEEE Trans. Nuc. Sci., vol. 42, no. 6, pp. 1607-1614, Dec. 1995. [84] J.V. Osborn, R.C. Lacoe, D.C. Mayer, and G. Yabikul, “Total-Dose Hardness of Three Commercial CMOS Microelectronics Foundries,” IEEE Trans. Nucl. Sci., vol. 45, pp. 1458-1463 (1998). [85] G.Anelli, M. Campbell, M. Delmastro, F. Faccio, S. Florian, A. Giraldo, E. Heijne, P. Jarron, K. Kloukinas, A. Marchioro, P. Moreira, and W. Snoeys, “Radiation Tolerant VLSI Circuits in Standard Deep Submicron CMOS Technologies for the LHC Experiments: Practical Design Aspects,” IEEE Trans. Nucl. Sci. Vol. 46, pp. 1690-1696 (1999). [86] J.J. Wall, “The Effects of Space Radiation and Burn-In on Plastic Encapsulated Semiconductors,” p. 96 – 101, 1999 IEEE Radiation Effects Data Workshop. ELDRS Effect [87] A.H. Johnston, G.M. Swift, and B.G. Rax, “Total Dose Effects in Conventional Bipolar Transistors and Linear Integrated Circuits,” IEEE Trans. Nuc. Sci., vol., 41, no. 6, pp. 2427-2436, Dec. 1994. [88] S. McClure, R.L. Pease, W. Will, and G. Perry, “Dependence of Total Dose Response of Bipolar Linear Microcircuits on Applied Dose Rate,” IEEE Trans. Nuc. Sci., vol. 41, no. 6, pp. 2544-2549, Dec. 1994. [89] J. Beaucour, T. Carriere, A. Gach, D. Laxague, and P. Poirot, “Total Dose Effects on Negative Voltage Regulator,” IEEE Trans. Nuc. Sci., vol. 41, no y, pp 2420-2426, Dec. 1994. [90] A.H. Johnston, B.G. Rax, and C.I. Lee, “Enhanced Damage in Linear Bipolar Integrated Circuits at Low Dose Rates,” IEEE Trans. Nuc. Sci., vol. 42, no. 6, pp. 1650-1659, Dec. 1995. [91] T. Carriere, J. Beaucour, A. Gach, B. Johlander, and L. Adams, “Dose Rate and Annealing Effects on Total Dose Response of MOS and Bipolar Circuits,” IEEE Trans. Nucl. Sci., vol. 42, no 6, pp. 2567-2574, Dec. 1995. [92] R.L. Pease, W.E. Combs, A. Johnston, T. Carriere, C. Poivey, A. Gach, and S. McClure, “A Compendium of Recent Total Dose Data on Bipolar Linear Circuits,” 1996 IEEE Radiation Effects Data Workshop, pp. 28-37, 1996. [93] R.L. Pease, M. Gehlhausen, “Elevated Temperature Irradiation of Bipolar Linear Microcircuits,” IEEE Trans. Nucl. Sci., vol. S43, no. 6, p. 3161-3166, Dec. 1996. [94] R.L. Pease, “Total Dose Issues for Microelectronics in Space Systems,” IEEE Trans. Nucl. Sci., vol. 43, pp. 442-452, Apr. 1996. [95] L. Bonora, J.P. David, “An attempt to define conservative conditions for total dose evaluation of bipolar ICs,” IEEE Trans. Nucl. Sci., vol. 44, no 6, pp. 1974-1980, Dec. 1997. [96] R. Pease, M. Gehlhausen, J. Krieg, J. Titus, T. Turflinger, D. Emily, and L. Cohn, “Evaluation of Proposed Hardness Assurance Method for Bipolar Linear Circuits with Enhanced Low Dose Rae sensitivity (ELDRS),” IEEE Trans. Nucl. Sci., vol. 45, no. 6, pp. 2665-2672, Dec. 1998. [97] T. Carriere, R. Ecoffet, and P. Poirot, “Evaluation of Accelerated Total Dose Testing of Linear Bipolar Circuits,” IEEE Trans. Nucl. Sci., vol. 47, no 6, pp. 2350-2357, Dec. 2000.
Displacement Damage [98] G.R. Hopkinson, ClJ. Dale, and P.W. Marshall, “Proton Effects in Charge Coupled Devices,” IEEE Trans. Nucl. Sci., vol. 43, no. 2, pp. 614-627, Apr. 1996. [99] J.P. Spratt, B.C. Passenheim and R.E. Leadon, “The Effect of Nuclear Radiation on P-Channel CCD Imagers,” p. 116 – 121, 1997 IEEE Radiation Effects Data Workshop. [100] M.D’Ordine, “Proton Displacement Damage in Optocouplers,” p. 122 – 124, 1997 IEEE Radiation Effects Data Workshop. [101] B.G. Rax, A.H. Johnston and C.I. Lee, “Proton Damage Effects in Linear Integrated Circuits,” IEEE Trans. Nucl. Sci., vol. 45, Dec. 1998. [102] A.H. Johnston, B.G. Rax, L.E. Selva and C.E. Barnes, “Proton Degradation of Light-Emitting Diodes,” p. 1781 – 1789, IEEE Trans. Nucl. Sci., Dec., vol. 46, Dec. 1999. [103] G.R. Hopkinson, “Proton Damage Effects on P-Channel CCDs,” p. 1790 – 1796, IEEE Trans. Nucl. Sci., vol. 46, Dec. 1999.
I-58
Induced Single Event Destructive Effects [104] E.G. Stassinopoulos, G.J. Brucker, P. Calvel, A. Baiget, C. Peyrotte, R. Gaillard, “Charge Generation by Heavy Ions in Power MOSFETs Burnout Space Predictions, and Dynamic SEB Sensitivity,” IEEE Trans. Nucl. Sci., vol. 39, no. 6, pp. 1704-1711, Dec. 1992. [105] I. Mouret, M. Allenspach, R.D. Schrimpf, J.R. Brew, K.F. Galloway, and P. Calvel, “Temperature and Angular Dependence of Substrate Response in SEGR,” IEEE Trans. Nucl. Sci., vol. 41, no. 6, p. 2216-2221, Dec. 1994. [106] J.L. Titus, C.F. Wheatley, I. Mouret, M. Allenspach, J. Brews, R. Schrimpf, K. Galloway, and R.L. Pease, “Impact of Oxide Thickness on SEGR Failure in Vertical Power MOSFETs, Development of a Semiempirical Expression,” IEEE Trans. Nucl. Sci., vol. 42, no. 6, pp. 1928-1934, Dec. 1995. [107] D.K. Nichols, J.R. Coss, T. Miyahira, J. Titus, D. Oberg, J. Wert, P. Majewski, J. Lintz, “Update of single Event Failures in Power MOSFETs,” 1996 IEEE radiation effects dataworkshop, pp. 67-72, 1996. [108] J.L. Titus and C.F. Wheatley, “Proton-Induced Dielectric Breakdown of Power MOSFETs,” p. 2891 – 2897, IEEE Trans. Nucl. Sci., vol. 45, Dec. 1998. [109] C.I. Underwood and M.K. Oldfield, “Observed Radiation-Induced Degradation of Commercial-OffThe-Shelf (COTS) Devices Operating in Low-Earth Orbit,” IEEE Trans. Nucl. Sci., vol. 45, Dec. 1998.
Radiation Effects of Photonics [110] 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 and E.G. Stassinopoulos, “Proton-Induced Transients in Optocouplers: InFlight Anomalies, Ground Irradiation Test, Mitigation and Implications,” p. 1885 – 1892, IEEE Trans. Nucl. Sci., vol. 44, Dec. 1997. [111] R.A. Reed, P.W. Marshall, A.H. Johnston, J.L. Barth, C.J. Marshall, K.A. LaBel, M. D’Ordine, H.S. Kim, and M.A. Carts, “Emerging Optocoupler Issues with Energetic Particle-Induced Transients and Permanent Radiation Degradation,” IEEE Trans. Nucl. Sci., vol. 45, no. 6, pp. 2833-2841, Dec. 1998. [112] A.H. Johnston, T. Miyahira, G.M. Swift, S.M. Guertin and L.D. Edmonds, “Angular and Energy Dependence of Proton Upset in Optocouplers,” p. 1335 – 1341, IEEE Trans. Nucl. Sci., vol. 46, Dec. 1999. [113] R.A. Reed, C. Poivey, P.W. Marshall, K.A. LaBel, C.J. Marshall, S. Kniffin, J.L. Barth, and C. Seidleck, “Assessing the Impact of the Space Radiation Environment on Parametric Degradation and SingleEvent Transients in Optocouplers,” IEEE Trans. Nucl. Sci., vol. 48, no. 6, pp. 2202-2209, Dec. 2001. [114] P. Marshall, “Electronics Radiation Characterization Project Task on Photonic Technology, Rate Prediction Tool Assessment for Single Event Transient errors,” NASA-GSFC report, Jan 2002. [115] R.A. Reed, “Guideline for Ground Radiation Testing and Using Optocouplers in the Space Radiation Environment,” NASA-GSFC report, Mar. 2002.
Testing and Analysis [116] D.H. Habing, “The Use of Lasers to Simulate Radiation Induced Transients in Semiconductor Devices and Circuits,” IEEE Trans. Nucl. Sci. NS. 12, pp. 91-100 (1965). [117] J.S. Melinger, et al., “Pulsed Laser-Induced Single Event Upset and Charge Collection Measurements as a Function of Optical Penetration Depth,” Journal of Applied Physics, vol. 84, pp. 690-703 (1998). [118] D. McMorrow et al., “Application of a Pulsed Laser for Evaluation and Optimization of SEU-Hard Designs,” IEEE Trans. Nucl. Sci., vol. 47, pp. 559-563 (2000). [119] J.G. Rollins, “Estimation of Proton Upset Rate from Heavy Ion Test Data,” IEEE Trans. Nucl. Sci., vol. 37, no. 6, pp. 1961-1965, Dec. 1990. [120] E.L. Petersen, "The Relationship of Proton and Heavy Ion Upset Thresholds," p. 1600-1604, IEEE Trans. Nucl. Sci., vol. 39, December. 1992. [121] J. Pickel, “Single Event Effects Rate Prediction,” IEEE Trans. Nucl. Sci., vol. 43, no. 2, pp. 483-495, Apr. 1996. [122] P. Calvel, C. Barillot, P. Lamothe, R. Ecoffet, S. Duzellier, and D. Falguere, “An Emipirical Model for Predicting Proton Induced Upset,” IEEE Trans. Nucl. Sci., vol. 43, no. 6, pp 2827-2832, Dec. 1996. [123] P.M. O’Neill, G.D. Badhwar and W.X. Culpepper, “Internuclear Cascade-Evaporation Model for LET Spectra of 200 MeV Protons Used for Parts Testing,” p. 2467 - 2474, IEEE Trans. Nucl. Sci., vol. 45, Dec. 1998. [124] J. Barak, “Empirical Modeling of Proton Induced SEE Rates,” IEEE Trans. Nucl. Sci., vol. 47, no 3, pp 545-550, Jun 2000.
I-59
Hardness Assurance [125] R.L. Pease, and D.R. Alexander, “Hardness Assurance for Space Systems Microelectronics,” Rad. Phys. Chem., vol. 43, pp. 191-204, 1994. [126] R.L. Pease, “Radiation Effects Short Course Hardness Assurance,” Vanderbilt University, August 2001.[109] P.S. Winokur, M.R. Shaneyfelt, T.L. Meisenheimer, and D.M. Fleetwood, “Advanced Qualification Techniques,” RADECS 1993 proceedings, pp. 289-299, 1994. [127] K. LaBel, and M. gates, “Single-Event-Effect Mitigation From a System Perspective,” IEEE Trans. Nucl., Sci., vol. 43, no. 2, pp. 654-660, Apr. 1996. [128] G.K. Lum, N.R. Bennett and J.M. Lockhart, “System Hardening Approaches for a LEO Satellite With Radiation Tolerant Parts,” p. 2026 – 2033, IEEE Trans. Nucl. Sci., vol. 44, Dec. 1997. [129] S.C. Witczak, R.D. Schrimpf, D.M. Fleetwood, K.F. Galloway, R.C. Lacoe, D.C. Mayer, J.M. Puhl, R.L. Pease, and J. Suihle, “Hardness Assurance Testing of Bipolar Junction Transistors and Elevated Irradiation Temperatures,” IEEE Trans. Nucl. Sci., vol. 44, no. 6, pp. 1989-2000, Dec. 1997. [130] K.A. LaBel, A.H. Johnston, J.L. Barth, R.A. Reed and C.E. Barnes, “Emerging Radiation Hardness Assurance (RHA) Issues: A NASA Approach for Spaceflight Programs,” p. 2727 – 2736, IEEE Trans. Nucl. Sci., vol. 45, Dec. 1998. [131] K.A. LaBel, A.H. Johnston, J.L. Barth, R.A. Reed, and C.E. Barnes, “Emerging Radiation Hardness Assurance (RHA) issues: A NASA Approach for Space Flight Programs,” IEEE Trans. Nucl. Sci., vol. 45, no. 6, pp. 2727-2736, Dec. 1998. [132] R.C. Lacoe, J.V. Osborn, R. Koga, S. Brown, and D.C. Mayer, “Application of Hardness-by-Design Methodology to Radiation-Tolerant ASIC Technologies,” IEEE Trans. Nucl. Sci. vol. 47, pp. 2334-2341 (2000). [133] J.R. Schwank, M.R. Shaneyfelt, P. Paillet, D.E. Beutler, V. Ferlet-Cavrois, B.L. Draper, R.A. Loemker, P.E. Dodd, and F.W. Sexton, “Optimum Laboratory Radiation Source for Hardness Assurance Testing,” IEEE Trans. Nucl. Sci. vol. 48, pp. 2152-2157, Dec. 2001.
Issues [134] D.R. Alexander, D.G. Mavis, C.P. Brothers, and J.R. Chavez, “Design Issues for Radiation Tolerant Microcircuits in Space,” 1996 IEEE Nuclear and Space radiation Effects Conference, Short Course, V-1 (1996). [135] R. Koga, “Single Event Effect Ground Test Issues,” IEEE Trans. Nucl. Sci., vol. 43, no. 2, pp. 661670, Apr. 1996. [136] E.J. Daly, J. Lemarie, D. Heynderickx, and D.J. Rodgers, “Problems with Models of the Radiation Belts,” IEEE Trans. Nucl. Sci., vol. 43, no. 2, pp. 403-415, Apr. 1996. [137] P.M. O’Neill, G.D. Badhwar and W.X. Culpepper, “Risk Assessment for Heavy Ions of Part Tested with Protons,” p. 2311 – 2314, IEEE Trans. Nucl. Sci., vol. 44, Dec. 1997. [138] O. Musseau, V. Ferlet-Cavrois, A.B. Campbell, W.J. Stapor and P.T. McDonald, “Comparison of Single Event Phenomena For Front/Back Irradiations,” p. 2250 – 2255, IEEE Trans. Nucl. Sci., vol. 44, Dec. 1997. [139] C. Marshall, and P. Marshall, “Proton Effects and Test Issues for Satellite Designers, part B: Displacement Effect,” 1999 IEEE NSREC short course, Norfolk, Jul 1999. [140] P. E. Dodd, M. R. Shaneyfelt, J. R. Schwank, and G. L. Hash, "Neutron-Induced Soft Errors, Latchup, and Comparison of SER Test Methods 140]for SRAM Technologies," IEDM Tech. Digest, pp. 333-336, Dec. 2002.
Additional List of Radiation Hardness Assurance Related Documents Military Documents Military standards, specifications, and handbooks can be viewed and ordered from the Web site of the DoD Document Automation and Production Services (DAPS), Bldg. 4/D (DPM-DODSSP), 700 Robbins Ave., Philadelphia, PA 19111-5094. Their URL is http://www.dodssp.daps.mil. For assistance, one can phone their help line at 215-697-2179 or Fax –1462. Most can also be viewed and downloaded at DSCC’s web site, http://www.dscccols.com/Programs/MilSpec/default.asp.
Military Performance Specifications [141] 19500, General Specification for Semiconductor Devices.
I-60
[142] 38510, General Specification for Microcircuits. [143] 38534, Performance Specification for Hybrid Microcircuits. [144] 38535, General Specification for Integrated Circuits (Microcircuits) Manufacturing.
Military Handbooks [145] 814, Ionizing Dose and Neutron HA Guidelines for Microcircuits and Semiconductor Devices. [146] 815, Dose Rate HA Guidelines. [147] 816, Guidelines for Developing Radiation Hardness Assured Device Specifications. [148] 817, System Development Radiation Hardness Assurance. [149] 339, Custom Large Scale Integrated Circuits Development and Acquisition for Space Vehicles. [150] 1547, Electronic Parts, Materials, and Processes for Space and Launch Vehicles.
Military Test Methods IN MIL-STD-750 (Test Methods for Semiconductor Devices): [151] 1017, Neutron Irradiation Procedure [152] 1019, Ionizing Radiation (Total Dose) Test Procedure [153] 1032, Package Induce Soft Error Test Procedure (Due To Alpha Particles) [154] 1080, Single Event Burnout and Single Event Gate Rupture [155] 3478, Power MOSFET Electrical Dose Rate Test Method [156] 5001, Wafer Lot Acceptance Testing
IN MIL-STD-883 (Test Methods and Procedures for Microelectronics): [157] 1017, Neutron Irradiation Procedure [158] 1019, Ionizing Radiation (Total Dose) Test Procedure [159] 5004, Screening Procedures [160] 5005, Qualification and Quality Conformance Procedures [161] 5010, Test Procedures for Complex Monolithic Microcircuits
DTRA DOCUMENTS DTRA documents can be obtained from the Defense Technical Information Center (DTIC). Phone 800-2253842. [162] DNA-H-93-140, Military Handbook for Hardness Assurance, Maintenance and Surveillance (HAMS).
ASTM Standards ASTM standards can be purchased from ASTM, 100 Barr Harbor Dr., West Conshohocken, PA 19428-2959. Phone 610-832-9585. Portions of each standard can be viewed at http://www.astm.org. The following are test and measurement standards. They are under the oversight of ASTM Subcommittee F1.11, Nuclear and Space Radiation Effects; Chairman, William Alfonte, 410-326-6044. [163] F528, Test Method of Measurement of Common-Emitter D-C Current Gain of Junction Transistors. [164] F615, Practice for Determining Safe Current Pulse Operating Regions for Metallization on Semiconductor Components. [165] F616, Test Method for Measuring MOSFET Drain Leakage Current. [166] F617, Test Method for Measuring MOSFET Linear Threshold Voltage. [167] F676, Test Method for Measuring Unsaturated TTL Sink Current. [168] F769, Test Method for Measuring Transistor and Diode Leakage Currents. [169] F996, Test Method for Separating an Ionizing Radiation-Induced MOSFET Threshold Voltage Shift into Components Due to Oxide Trapped Holes and Interface States Using the Subthreshold Current-Voltage Characteristics. [170] F1190, Practice for the Neutron Irradiation of Unbiased Electronic Components. [171] F1192, Guide for the Measurement of Single Event Phenomena (SEP) Induced by Heavy Ion Irradiation of Semiconductor Devices. [172] F1263, Test Method for Analysis of Overtest Data in Radiation Testing of Electronic Parts.
I-61
[173] F1467, Guide for Use of an X-Ray Tester (~10 keV Photons) in Ionizing Radiation Effects Testing of Semiconductor Devices and Microcircuits. [174] F1892, Guide for Ionizing Radiation (Total Dose) Effects Testing of Semiconductor Devices.
Radiation Dosimetry Standards The following are radiation dosimetry standards. They are under the oversight of ASTM Subcommittee E10.07; Chairman, Dr. Dave Vehar, 505-845-3414. [175] E265, Test Method for Measuring Reaction Rates and Fast-Neutron Fluences by Radioactivation of Sulfur-32. [176] E496, Test Method for Measuring Neutron Fluence Rate and Average Energy from 3H(d,n)4He Neutron Generators by Radioactivation Techniques. [177] E666, Practice for Calculating Absorbed Dose from Gamma or X Radiation. [178] E668, Practice for Application of Thermoluminescence Dosimetry Systems for Determining Absorbed Dose in Radiation-Hardness Testing of Electronics. [179] E720, Guide for Selection of a Set of Neutron-Activation Foils for Determining Neutron Spectra Used in Radiation-Hardness Testing of Electronics. [180] E721, Method for Determining Neutron Energy Spectra with Neutron-Activation Foils for Radiation Hardness Testing of Electronics. [181] E722, Practice for Characterizing Neutron Energy Fluence Spectra in Terms of an Equivalent Monoenergetic Neutron Fluence for Radiation-Hardness Testing of Electronics. [182] E1026, Methods for Using the Fricke Dosimeter to Measure Absorbed Dose in Water. [183] E1249, Practice for Minimizing Dosimetry Errors in Radiation Hardness Testing of Silicon Electronic Devices Using Co-60 Sources. [184] E1250, Test Method for Application of Ionization Chambers to Assess the Low Energy Gamma Component of Cobalt-60 Irradiators Used in Radiation-Hardness Testing of Silicon Electronic Devices. [185] E1854, Practice for Assuring Test Consistency in Neutron-Induced Displacement Damage of Electronic Parts. [186] E1855, Method for Use of 2N2222 Silicon Bipolar Transistors as Neutron Spectrum Sensors and Displacement Damage Monitors.
Fiber Optic Test Standards The following are fiber optic test standards. They are under the oversight of ASTM Subcommittee E13.09, Optical Fibers for Molecular Spectroscopy; Chairman, Dr. Tuan Bo-Dinh, 423-574-6249. [187] E1614, Guide for Procedure for Measuring Ionizing Radiation-Induced Attenuation in Silica-Based Optical Fibers and Cables for Use in Remote Fiber-Optic Spectroscopy and Broadband Systems. [188] E1654, Guide for Measuring Ionizing Radiation-Induced Spectral Changes in Optical Fibers and Cables for Use in Remote Raman Fiber-Optic Spectroscopy.
EIA Test Methods and Guides Some EIA/JEDEC standards can be obtained from Global Engineering Documents, 15 Inverness Way East, Englewood CO 80112-5704. Phone 800-854-7179. Otherwise see http://www.jedec.org or http://www.eia.org. [189] EIA/JESD-57, Test Procedures for the Measurement of Single-Event Effects in Semiconductor Devices from Heavy Ion Irradiation. [190] JESD-89, Measurement and Reporting of Alpha Particles and Terrestrial Cosmic Ray-Induced Soft Errors in Semiconductor Devices. [191] EIA/JEP-133, Guideline for the Production and Acquisition of Radiation-Hardness Assured Multichip Modules and Hybrid Microcircuits. [192] EIA/TIA-455-64 (FOTP-64), Procedure for Measuring Radiation-Induced Attenuation in Optical Fibers and Cables.
I-62
ESA Test Methods And Guides ESA documents can be obtained from ESA/SCC Secretariat (TOS-QCS), ESTEC P.O.Box 299, 2200 AG Noordwijk, The Netherlands. [193] ESA/SCC Basic Specification No. 22900, Total Dose Steady-State Irradiation Test Method. [194] ESA/SCC Basic Specification No. 25100, Single Event Effects Test Method and Guidelines. [195] ESA PSS-01-609, The Radiation Design Handbook. [196] Private communication with J.M. Baker, Lockheed Martin Space Systems Co., Sunnyvale, CA.
I-63
2004 IEEE NSREC Short Course
Section II
Microelectronic Piece Part Radiation Hardness Assurance for Space Systems
Ronald L. Pease RLP Research
Microelectronic Piece Part Radiation Hardness Assurance for Space Systems Ronald Pease RLP Research Los Lunas, NM 87031 505-565-0548
[email protected]
Table of Contents 1.0
Introduction............................................................................................................... II-2
2.0
Background ............................................................................................................... II-4
3.0
Overview of the “traditional” piece-part hardness assurance (P2HA) approach modified for today’s systems .................................................................................... II-4 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9
4.0
Radiation environment specifications ............................................................ II-7 Failure definitions......................................................................................... II-11 Probability of survival and confidence......................................................... II-14 Piece-part radiation characterization data. ................................................... II-15 Calculation of RDM ..................................................................................... II-20 Categorization .............................................................................................. II-23 Parts categorized as unacceptable ................................................................ II-25 Parts categorized as HCC-1 and HNC ......................................................... II-26 Radiation lot acceptance testing. .................................................................. II-29 3.9.1 Attributes testing................................................................................ II-30 3.9.2 Variables testing................................................................................. II-33
Challenges for piece-part hardness assurance for space systems ........................... II-42 4.1 4.2 4.3 4.4
Knowing the relevant details of the part when it’s commercial................... II-43 Knowing the part response to the radiation environment in the application.II-45 Knowing the radiation environment............................................................. II-46 Knowing how to quantify the results of radiation testing ............................ II-48
5.0
Piece-part hardness assurance management ........................................................... II-48
6.0
Conclusions............................................................................................................. II-49
7.0
Acknowledgements................................................................................................. II-50
8.0
References............................................................................................................... II-50 Appendix: LIST OF RADIATION HARDNESS ASSURANCE RELATED DOCUMENTS........................................................................................................ II-53
II-1
1.0
Introduction
In this section of the short course we will provide a methodology for microelectronic piece-part radiation hardness assurance for space systems and discuss many of the issues involved with the increasing use of commercial parts in space. What do we mean by piecepart radiation hardness assurance? In this context we will define piece-part hardness assurance as “the methods used to assure that microelectronic piece-parts meet specified requirements for system operation at specified radiation levels for a given probability of survival (Ps) and level of confidence (C)”. Using this definition will allow us to quantify the process. The requirement for system operation allows for a failure definition that is determined by the application of the part in the system. The requirement to meet a specified radiation level allows us to test parts as a function of a radiation environment and compare the radiation failure level of the part to the specification level. And finally the specification of the Ps and C for the part will allow us to develop statistical approaches for sample testing of the parts. In the previous section Dr. Gary Lum discussed system level hardness assurance. At the system level there are many approaches one may take to mitigate the effects of radiation on the piece-part. There is, of course, part selection, which is probably one of the most important. In recent years the demand has been for higher performance, lower cost and longer system life (or as some say, better-faster-cheaper), while maintaining high reliability. With these constraints the choices have been narrowing toward state-of-the-art commercial off-the-shelf (COTS) parts. With the growing use of COTS parts in space systems the approach to hardness assurance is somewhat limited compared to that used for what I will call “traditional” methods. However, as I will try to demonstrate, the foundation for a piecepart hardness assurance (HA) methodology, developed by the US military and space agencies and their contractors in the mid 1970s, for mature radiation hardened and radiation tolerant technologies, is still viable for COTS, with some modifications. There are many different approaches to piece-part hardness assurance depending on how much control one has over the design, layout and process of the piece-part. For radiation hardened or radiation tolerant parts, where there is some commitment on the part of the vendor to the hardened system market, one can identify the circuit design and layout factors that affect radiation hardness and apply design and layout rules to assure that the parts comply with these restrictions. Also one may identify the processing parameters that affect hardness and place control limits on the critical processing nodes. Another piece-part hardness assurance technique is to use surrogate parts to assure the hardness of the end product. Examples are the use of test chips containing test structures, test transistors or small circuits that represent critical radiation sensitive elements of the end product. These test chips can be irradiated at the wafer level using such in-line radiation sources as x-ray sources for total dose testing and lasers for single particle response. Another use of surrogates is to design a standard evaluation test circuit, SEC, that is of equivalent complexity to end line products of the same technology and perform lot sample testing on the SEC. All of these techniques are used for parts manufactured under the radiation hardness assured quality manufacture’s list (RHAQML) but cannot be easily applied to COTS. A hardness assurance technique that would work for any microelectronic piece-part, whether RHAQML or COTS,
II-2
is to identify pre-irradiation electrical parameters that correlate to the radiation response of the part and place control limits on these parameters to assure that the radiation response is within specified limits. Many attempts have been made over the years to identify such critical parameters for various radiation environments. However, while some general correlations have been established, no parameters have been shown to provide one to one correlation within a process lot to allow for lot acceptance. Examples of pre-irradiation electrical parameters that show general correlation to radiation response are unity gain cutoff frequency and neutron degradation for discrete bipolar transistors, 1/f noise and total dose induced trapped positive charge for MOSFETs, channel resistance and radiation induced interface traps for MOSFETs and electrical induced latchup and prompt dose rate induced latchup. Another example of a hardness assurance technique for piece-parts is the use of a radiation screen. In this approach all of the samples in the lot are subjected to a radiation test and the parts that pass are used in the system and the ones that do not pass are rejected. This approach has been investigated in two ways for the total dose environment. In the first approach, called irradiate and anneal (IRAN), the parts are exposed to the specification dose and the parts that pass are annealed at elevated temperature to remove the damage. The approach is based on the assumption that upon re-irradiation they will response in the same manner as with the first irradiation and that the anneal will have removed all damage so the reliability and performance will not be affected. The problem with this approach is that the temperature of the anneal cannot be great enough to remove the damage without exceeding the maximum allowable temperature for the part and many parts do not respond to the radiation the same way upon re-irradiation. Another radiation screen for total dose is to irradiate to a level well below the specification dose to show a small amount of parameter degradation. The change in the parameters is then extrapolated to the specification dose to see how much the part would degrade. The problem with this approach is that for most parts the parameter shifts are not linear with dose and hence cannot be extrapolated with confidence. Other 100% radiation screens include prompt dose rate testing for upset or latchup. These techniques have been used in the past for some military systems. However, while the approach does work, it is very expensive and hence seldom cost effective compared to lot sample testing. Our recommendations for piece-part hardness assurance it to utilize radiation hard or tolerant parts where possible and insure that they are designed into the system such that they are hardness non critical and require to radiation lot sample testing by the user. For those cases where rad hard or tolerant parts are not available the approach would be to follow the “traditional” or generic approach described below that is based on lot sample testing. This section of the short course will present the “traditional” piece-part HA methodology using lot sample radiation testing, show step-by-step how it is applied, and provide examples of the statistical approach to lot acceptance testing for various radiation environments. Along with this discussion we will show how the “traditional” method is modified to apply to COTS parts in space. The methodology provides a quantitative approach to hardness assurance. However, it must be considered a tool to be used by the HA engineer and not a set of hard and fast rules. The HA engineer will always have to exercise engineering judgment in assessing the risk involved with using COTS parts in hardened system applications.
II-3
2.0
Background
Piece-part HA is something that every electronic system exposed to a radiation environment has had to address. Many methodologies have been developed over the years by system program offices and system and sub-system contractors. Approaches such as limited parts lists with controlled production lines have been used for some strategic systems on the one extreme, and use of radiation data from data banks, with no additional testing, has been used in the other extreme. In the mid 1970s the US Defense Nuclear Agency (now known as the Defense Threat Reduction Agency, DTRA) began a program to develop the documentation necessary to support piece-part hardness assurance. The group responsible for carrying out this effort was (and still is) the Space Parts Working Group (SPWG) Hardness Assurance Committee (HAC). This group is comprised of representatives from the US Department of Defense, DoD, the Department of Energy, DoE, the National Aeronautics and Space Agency, NASA, and their contractors. Electrical test methods and dosimetry standards have been developed through the American Society for Testing and Materials, ASTM, and most radiation environment test methods and hardness assurance guidelines have been developed as military standards and handbooks. The first 10 years of this effort is described in a summary review paper [Woli-85]. The following definition was adopted by the working group: “Piece-part hardness assurance consists of the procedures, controls, and tests used to insure that a purchased piece-part has a response to nuclear-induced stresses that is within known and acceptable limits” [Woli-85]. If we eliminate the word “controls” and substitute the word radiation for “nuclear”, the definition applies to commercial microelectronics today. While the focus of the DNA funded piece-part HA program has been on mature hardened technologies, the methodology developed is generic and applies to any microelectronic piecepart, whether commercial, radiation tolerant, or radiation hard, as we will attempt to demonstrate. 3.0
Overview of the “traditional” piece-part hardness assurance (P2HA) approach modified for today’s systems
One of the first applications of the P2HA approach was for “permanent” degradation in steady state (long term) neutron and ionizing radiation environments. The methodology developed for piece parts in these environments is summarized in the flow diagram of Figure 1 [HDBK- 94]. This general approach is applicable to all piece-parts with appropriate modifications. MIL-HDBK-814 and the other handbooks that describe piece-part hardness assurance are available in pdf format on-line at http://www.dscc.dla.mil/programs/milspec. For any system piece-part hardness assurance program one must start with the application of the part in the system. The application will determine how one defines failure, and the system mission, as well as the location of the part in the system, will determine the radiation environments. The probabilities of survival, Ps, and confidence levels, C, for the piece-parts are specified by the program office based on the overall system Ps and C. For the purpose of the P2HA program Ps and C are considered givens. In the methodology described in Figure 1, all of the microelectronic parts are categorized for each system application and each radiation environment based on a radiation design margin. The radiation design margin, RDM, is defined as the mean (often geometric mean) of the radiation failure level of the part, Rmf, divided by the radiation specification level, Rspec. The failure levels of the individual samples in the radiation data are determined from the degradation of performance of the part
II-4
as a function of the radiation environment and the failure criteria determined by the application of the part.
Device type
System Requirements Survival probability And confidence requirements
Radiation specification requirements
YES Circuit requirements Worst case Circuit analysis
HCC-2 and HNC Lot acceptance Testing not required Acceptable device
PASS FAIL
Re-evaluate device category
Radiation data NO
Part categorization
Characterization test
Hardness Critical Category- 1M lot acceptance test required Sample test
Device not Acceptable Corrective Action required
Additional Localized shielding Part substitution
FAIL
Test new lot
PASS
Lot rejected
Lot Acceptance Tests Lot control
FAIL
PASS
Lot acceptable
Re-evaluate Device category Circuit redesign
HCC-2 not applicable to most space systems
Figure 1. The piece-part hardness assurance methodology reproduced from Figure 2 in MIL-HDBK-814 [HDBK-94]. The dashed box is added for noted comments.
There are basically three categories in which a part may be placed: (1) unacceptable, (2) hardness non-critical and (3) hardness critical. If a part is unacceptable, then either a different part must be used in the application or one of the RDM factors must be changed in order to make the part hardness critical or hardness non-critical. The factors that may be changed include the radiation specification level, e.g. use of shielding, the failure definition, e.g. application circuit re-design, or the mean radiation failure level, e.g. by part substitution, hardening of the part or specific lot selection. In the initial part categorization procedure, the data that are used to determine the mean failure level of the part type must be based on a statistically significant sample randomly chosen from a population that includes all of the variations that might be expected over the lifetime of purchases of the part for the system. Typically the sample should include, as a minimum, units from three wafer lots manufactured over a period of at least 1 year from each manufacturer on the approved parts list for that part type. In the “traditional” method, it is assumed that the parts are mature, having been in production for a long period of time and with nothing changing in the circuit design, layout or process. This is one of the key differences that distinguish the “traditional” method from a modified method that would
II-5
be applicable to COTS in space systems. For many COTS parts, both the design and process can change frequently without customer notification. For parts that are categorized in the hardness non-critical (HNC) category no further testing or analysis is required. This is obviously the most desirable category, and for many systems with moderate radiation requirements, a majority of the microelectronics parts will fall in this category. However, with the reduction in the number of radiation hardened parts on the market and the increased use of COTS in military and space systems, the HNC category often does not have a high percentage of parts, unless the radiation levels are very modest. Parts in the Hardness Critical Category, HCC-1M, must be lot sample radiation tested for every purchase lot. There are other HCC categories such as HCC-1S, HCC-1H and HCC-2, but they are special cases that are not important for an understanding of the methodology. For example 1S is a category that does not require lot sample testing based on RDM but must be purchased from a single source (manufacturer). Category 1H is a hardness-dedicated category for parts that must be placed in the system in a specific location for the system to be hard. Examples are terminal protection devices for electrical overstress hardening and resistors or inductors for latchup prevention. Category HCC-2 is a category for parts that have RDMs in between HNC and HCC-1 and, while they do not require radiation lot sample testing for every lot, they do require “periodic” radiation lot sample testing as shown in Figure 1. This category is usually not meaningful for COTS parts since few COTS parts can be considered mature with design, layout and process fixed. Hence in Figure 1 we have highlighted this category and indicated that it is not applicable to space systems. A more detailed discussion of the categories is given in a later section. The criterion that is used to determine in what category a part is placed is either based on historical radiation effects data for the class of parts and the environment or it is based on the statistics of the radiation data for the part and the environment. The criterion based on historical data (and some engineering judgment) is called the design margin breakpoint, DMBP, method and is often used for systems with moderate radiation requirements. The statistical approach is called the part categorization criteria, PCC, method and is based on an analysis of the distribution of radiation failure levels for all of the samples used to determine the RDM. Each of these methods will be discussed in detail. Figure 1 shows the general framework for the P2HA methodology. The concept is straightforward. We look at how a part is used in a system. We analyze its application to determine how much the electrical parameters or the functional performance would have to degrade to constitute a failure. This becomes the failure criterion. We then either determine the various radiation environments that the part will see during the life of the system to establish a radiation requirement or we use a radiation specification that is imposed on us by the program office. We then look at radiation data on the part type and from the failure criteria determine a mean radiation failure level for the sample in each environment. We take the ratio of the mean failure level to the specification level and establish an RDM. Applying our categorization criterion, we place the part in one of three categories. If the part falls in category HCC-1 we perform a lot sample radiation test on every purchase lot to determine whether the lot is acceptable. While this sounds straightforward, there are many complications that must be addressed, as we shall discuss in the following sections.
II-6
3.1
Radiation environment specifications
The piece-part radiation specifications are derived specifications based on the freefield radiation environment, whether natural or combined natural and nuclear weapon, and the mission requirements. The categorization of all parts and the lot acceptance tests for HCC-1 parts requires a specification of the radiation environment. There are many different ways to specify the various radiation environments. For a typical space mission a complete description would involve all of the particle types, the particle energies, fluxes and fluences that the microelectronic parts would encounter for the duration of the mission. This would, of course, be impossible since the space environment is very dynamic, and even if it were possible, the information would require a considerable amount of analysis to reduce it to a form that could be used in conducting a radiation test in the laboratory. Also, trying to duplicate the actual radiation environments in the laboratory in terms of particle type, energy, flux and fluence would be not only daunting but cost prohibitive (and in some cases impossible). While information about the actual radiation environment is often very useful when it comes to understanding the radiation response of a given microcircuit for characterization testing of a part, for hardness assurance programs the radiation is usually specified in terms of a parameter that is correlated to a degradation mechanism. In other words we specify the radiation environment by the radiation effects. For example we know that for most types of ionizing radiation the response of a microcircuit is a function of the amount of absorbed energy in the sensitive material. For most microelectronic parts the sensitive material is silicon dioxide and the measure of absorbed energy is the rad which is defined as 100 ergs per gram material. Hence we usually specify a long term ionizing radiation environment for all types of ionizing radiation as so many rad(SiO2). Specifying the ionizing radiation environment in terms of rad allows us to use many different forms of ionizing radiation for the purpose of radiation testing. By specifying the total dose environment in terms of rad we do not have to provide the actual fluence and energy spectrum of the proton, electron and photon environments. While this approach seemed to work, it was discovered in the 70’s that CMOS parts did not just degrade a certain amount for a given dose but, instead, exhibited responses that were a function of time (including both the irradiation time, i.e. dose rate, and the post irradiation time, i.e. annealing). These phenomena were labeled time dependent effects, TDE. These TDE include the buildup of trapped positive charge, the annealing of trapped positive charge, the buildup of interface traps, and the eventual annealing of interface traps, each with different time constants. It was shown that failures related to both types of defects could be bound with a two part test that included an irradiation at a reasonable laboratory dose rate (50-300 rad/s as specified in the US military test standard) followed by an elevated temperature anneal. To further complicate things, it was shown in the early 90’s that many bipolar linear parts were also sensitive to the dose rate but their response at low dose rate could not be bound with a combination high dose rate plus elevated temperature anneal test. Hence the dose rate sensitivity for this phenomenon was referred to as a true dose rate effect (TDRE), as opposed to a time dependent effect. The time dependent effect similar to that in CMOS, was also present in many of these bipolar linear parts, further complicating the development of a standard test procedure. Even though all of these issues have arisen because of the use of rad, the current practice is still to specify the ionizing radiation environment in terms of rad. For space environments, the average dose rate (rad/s) is also either specified, or at least known. The range of dose rates that are a) specified in test standards, b) available in radiation test
II-7
facilities and c) representative of system radiation environments are shown in Figure 2 taken from reference [Holm-02]. This figure does not include the recent change in MIL-STD883/Test Method 1019.6 to add low dose rate testing at 0.01 rad/s for bipolar linear circuits.
Figure 2. The dose rates for various radiation environments compared to the range of dose rates used for testing and the dose rates available from various radiation facilities [Holm-02]. TM1019.6 now includes testing at 0.01 rad/s for bipolar linear circuits.
Another radiation environment important for many microelectronic parts is the high energy particle environment that is capable of causing displacement damage. This includes electrons, protons and neutrons as well as heavy ions. In the 1980’s, it was demonstrated that displacement damage was proportional to the non-ionization energy loss, NIEL [Summ-87], in the materials, even though the nature of defects is different for the different particle types. This allowed for the specification of radiation environments in terms of (1) displacement damage dose, DDD or Dd, (2) NIEL or (3) 1 MeV equivalent silicon damage neutron fluence. While this environment has always been important for nuclear weapons effects and is usually specified in terms of 1 MeV equivalent neutron fluence (n/cm2) it is also important for many microelectronic parts in the natural space environment such as bipolar linear circuits and transistors, solar cells and some opto-electronic parts. Although this approach works for many part types, there are exceptions. Recent work on opto-electronic devices shows that some parameter degradation depends on total NIEL while other parameters depend only on the columbic NIEL for high energy protons [Walt-03]. Another radiation effect that is important for microelectronics in space environments is known collectively as single event effects, SEE. These effects include single event upset, latchup, functional interrupt, transients, burnout, gate rupture and probably others [see, for example, the special issue on single-event effects and the space radiation environment, April 1996 Trans. on Nucl. Sci.]. The radiation environments that are primarily responsible for these effects are the proton trapped radiation, solar particles (including protons and heavy ions), atmospheric neutrons and galactic cosmic rays, GCR (including protons and heavy
II-8
ions). The principal measure of these environments that correlates to the observed effects for the heavy ions is the linear energy transfer or LET. Hence the heavy ion environment is often specified as integral particle fluence vs. LET. While LET is certainly very important it is not the only parameter of interest for heavy ion SEE. Many laboratory sources of heavy ions are capable of producing the required range of LETs but only a few can produce particle energies sufficient to get the penetration depth in microcircuit materials that is required for good simulation of the space environment. This is especially important in devices that either have very thick overlayers or very thick sensitive device regions. Another factor in trying to duplicate the radiation environment in the laboratory is with the different LET values. The space environment consists of a near continuum of LETs, especially for values up to about 40 MeV-cm2/mg. In laboratory testing a large number of LET values is achieved by using a small number of different ions and varying the LET by hitting the part at an angle. This produces an effective LET that is equal to the normal incident LET times the secant of the incident angle. Again, there are limitations to this technique that must be considered. For protons and neutrons it is not the LET that is of primary interest, since there are few microelectronic device failures caused by the direct energy deposition of the protons, and even less for neutrons. The important parameter for protons and neutrons is the particle energy. Hence the space proton and atmospheric neutron environments are usually specified in terms of integral fluence vs. energy. While the effectiveness of protons and neutrons in causing SEE are roughly the same for energies above 30 MeV, they are very different for the lower energies. For protons the failure cross section decreases rapidly below 30 MeV whereas for neutrons the failure cross sections in some circuits are still appreciable at 14 MeV and may increase significantly for thermal or epithermal neutrons [Dyer-03]. While the high energy particle and heavy ion environments are normally specified by a fluence vs. particle energy or fluence vs. LET there are other ways the radiation requirements may be specified. For example the system program office may decide to handle the piece part latchup issue by specifying that no parts may be allowed in the system that have a certain threshold LET for latchup, for example 40 MeV-cm2/mg. A more conservative approach that is often used specifies no latchup may occur for an LET > 120 with the part biased at 120% of its highest operating voltage and at the upper end of its operating temperature range. The heavy ion induced upset specification may be handled by specifying that no parts will be allowed in the system that have a bit error rate above a specified value. The bit error rate is then determined from the experimental data on the part (given in terms of the bit error cross section vs. LET) and the radiation environment for the system mission parameters. This method of specifying the radiation requirements does not address the actual radiation environments that are expected to be encountered in the system mission but rather puts the radiation requirement in terms that are often used to specify the radiation performance of a piece part. Another class of radiation effects that occur as a result of uniform transient ionizing radiation in silicon is referred to as transient dose or prompt dose rate (to contrast it with dose rate effects that occur at very low dose rates such as ELDRS). These effects are associated with nuclear weapon environments. The prompt dose rate environment is specified in terms of peak dose rate in rad/s along with a full width, half maximum pulse width. Effects associated with this environment include dose rate upset, latchup and burnout. While the
II-9
actual peak dose rate may be specified, for many weapon systems where power down techniques are employed, the prompt dose rate environment may be specified by two dose rates, 1) a level somewhat above the power down dose rate level that is considered a minimum threshold operate-through dose rate for which no upsets or latchups are allowed, and 2) a peak dose rate that is the maximum dose rate that the system may see for which no catastrophic failures, such as burnout, are allowed. Additional radiation environments associated with nuclear weapons include electromagnetic pulse radiation and thermo-mechanical shock damage (from highly nonuniform transient ionizing radiation energy deposition). While these environments are interesting and can cause significant expenditures in hardening, they are generally addressed at the system level and not with piece-part hardness assurance programs. Therefore, we will not address these radiation environment effects. In summary what the piece-part hardness assurance engineer is usually given to work with in the way of radiation specifications for a space environment is total dose (rad), a displacement damage specification that may be in terms of a 1 MeV equivalent neutron fluence (n/cm2) or displacement damage dose (MeV-cm2/g), a heavy ion integral fluence (particles/cm2) vs. LET (MeV-cm2/mg), a proton integral fluence (p/cm2) vs. proton energy (MeV) and for atmospheric neutrons a neutron integral fluence (n/cm2) vs. neutron energy (MeV). There is often a specification for threshold LET (MeV-cm2/mg) for no latchup as well. For the weapon environment one is usually given a total dose (rad), a dose rate (rad/s) (hopefully with pulse width) and 1 MeV equivalent neutron fluence (n/cm2). When a system radiation environment is specified it is sometimes given as the radiation incident on the system, also referred to as the free field case. To determine the RDM for a part one must know the radiation environment that will be received at the piecepart level. For space systems the major environments, specified in terms of ionizing dose, displacement damage dose, LET and proton fluence vs. energy are often given as a function of shielding thickness in mils aluminum. A top level specification is then determined by estimating the material between the piece-part and the free field in equivalent mil Al. The simple approach, which is often taken in a P2HA program, is to use the top level number. There are two potential problems with this approach. First the radiation environment at the part level may be lower than the top level number because of the potential errors in estimating the effective shielding. What often happens is that an initial categorization is performed using the top level number, and if the RDM is not sufficient (e.g. the part is categorized as unacceptable) then a calculation is performed to determine a more realistic environment at the part location. There are several 3-D codes available for transporting the free field environment through various layers of materials to determine a more realistic radiation specification for the part. The NSREC short courses given in 2002 (Section II-5.0, Mazur and Section V-3.2, Poivey), 1997 (Section IIA-Lorence) and 1993 (Section II-Garrett) include discussions of codes that address radiation transport through spacecraft materials. One of the more popular codes for calculating dose is SHIELDOSE2 which has been incorporated in space radiation environment codes such as SPACERAD and SPENVIS. Monte Carlo codes such as the GEANT4 simulation toolkit [Lei-02] [Trus-00] can be used for more accurate simulation of radiation environments inside spacecraft. Additional codes include NOVICE and MEVDP. Many spacecraft manufacturers start with a 3-D code
II-10
calculation of the internal radiation environment and, hence, do not have to refine the calculation for a less conservative estimate of the piece-part environment. The second problem that may occur with the use of the top level radiation specifications is an under estimation of the actual environment at the piece-part level. The reason for the possible under estimate is dose enhancement that may occur with low energy particles incident on high Z materials adjacent to the piece-part die. The issue of dose enhancement in spacecraft has been addressed in publications by Solin [Soli-98] [Soli-03]. Special attention should be paid to the details of the packaging materials and configuration used for the piece-part to determine whether dose enhancement may be an issue. Also, the first approach for estimating effective shielding as a homogeneous barrier of uniform thickness without consideration of actual shielding details may prove to be non-conservative. 3.2
Failure definitions
To determine the radiation failure levels of the samples used in the radiation characterization testing of the piece-part, a failure definition is required. The radiation failure definition is established for each part type and radiation environment based on a circuit analysis of each application of the part in the system. While this sounds like a difficult task, it is usually addressed by looking for the most critical or worst case application of the part. The conservative approach is to find the worse case failure definition of the part for all of its applications and use that definition to establish the RDM for that environment. However, there may be exceptions to this rule in cases of complex, expensive parts, such as ASICs, microprocessors and memory. The failure definition for some non-critical applications of the part may be quite different from the worst case failure definition. In this case, one may choose to have multiple failure definitions depending on application, which will allow the use of a more susceptible production lot for the application with the more relaxed failure definition. For such a case, the logistics are very important. A system would have to be in place to provide different markings for the parts that would identify the specific application for which the part qualified. In general, this would not be considered good practice and a single failure definition representing worst case should be used. Piece-part failure may be defined in terms of parametric failure or functional failure. Both types of failure must be considered since it is possible that functional failure can occur in some parts before certain critical parameters that degrade with radiation exceed their failure limits. A good example of this would be the case of maximum stand-by power supply current in digital CMOS. A part may fail to function properly at a dose well below where the stand-by current exceeds the post radiation specification or vice versa. Functional failure should always be a part of the failure definition and a functional test should always be included as part of the electrical characterization of the part to determine radiation failure levels. Determining the parametric failure limits of a part will require knowledge of the part radiation response, to identify those parameters that are most sensitive to the radiation environment, as well as an analysis of the circuit application to identify the most critical parameters for system operation and performance. In a given space system there may be parts of the system that were designed with radiation as a consideration and other parts that were designed with no consideration for radiation, e.g. sub-systems purchased as commercial units. The task of setting failure limits on “critical” parameters will depend on which approach was taken in the application circuit for the part. If the application circuit was
II-11
designed with radiation as a design constraint then the “critical” parameters and their allowable tolerances should be known and documented. On the other hand if radiation was not a consideration in the design, an analysis must be performed to identify the “critical” parameters and their allowable limits. The allowable limits will be determined by how much system performance degradation can be allowed without compromising the systems operation. Hence a considerable amount of engineering judgment must be used in setting the failure limits on parameters. Although it is possible to establish the parameter failure limits through circuit simulation, in practice it is usually done by consulting the system designer since he is the one most familiar with the part application and requirements. While it is important to investigate each application of a part and establish the critical functional and parametric failure limits for each of the radiation environments, there are some failure definitions that are generic. For example, having a part latch up and stay latched would not be acceptable for any part in any application. Some systems allow the use of parts that are latchup sensitive by providing system designs that either prevent the latchup from occurring (inductors or current limiting resistors) or cycle the power when a dose rate event occurs or a latch occurs. If a latchable part is susceptible to latchup-induced burnout the burnout may be prevented for the prompt dose rate environment by detecting the dose rate event and removing power before burnout can occur. However, in the case of single event latchup, the latchup can only be detected after the part has latched and draws sufficiently more current in order to detect the latchup. This means that only latchup susceptible parts that do not burn out from latchup should be used in space applications. Another issue concerning latchup is the impact of the momentary current surge on device reliability where the metallization may be damaged but not destroyed [Beck-02]. Preventing latchup from occurring with the use of inductors and series resistors does not work for microlatches that may occur in a section of a chip with only minor increases in power supply current. Hence the best policy is to use parts that do not latch up for the specified system radiation environments, i.e. a generic failure definition of latchup as failure. Other generic failure definitions can be applied to all forms of catastrophic failure. Catastrophic failure does not occur with long term ionization or displacement damage dose but may occur with the transient ionization environments such as single particle ionization or prompt dose rate. Single particles can cause single event burnout (SEB) and single event gate rupture (SEGR) in power MOSFETs, SEB in bipolar transistors, and SEGR in non-volatile memory devices such as EEPROMS that require the application of "high" voltage to implement write and erase functions. Although some failure definitions are generic in all cases, there are some failure definitions that can be treated generically in many cases, to avoid detailed application circuit analysis. For example in most cases one can define functional failure without regard to application. However, there are two types of functional failure, transient and “permanent”. Transient functional failures do not affect the operation of the part but can cause system failure, whereas “permanent” functional failure is much like catastrophic failure in the sense that the part is essentially no longer usable. Bit flips in memories are failures regardless of application. However, depending on how much of the memory is being used at any given time and whether the information in the memory is critical to system operation, many of the bit flips are a “don’t care”. In this case, using the definition of failure as a single bit flip may be too severe. Single event transients can be a problem for both digital and analog circuits. In the case of digital circuits, they are only considered failures if they propagate to a latch II-12
and change the stored information. The single event transients (SETs) in analog circuits are those that occur on the output and may be of sufficient amplitude and duration to cause a failure in external circuitry. The failure definition for analog SETs will depend on the application and hence analog SETs cannot be given a generic failure definition. Generic failure definitions can also be used for parametric degradation. While it is usually a very conservative definition, failure can be defined as exceeding the pre-irradiation electrical specification limits for the part. Using a generic definition for parametric failure does avoid performing the application circuit analysis, but the risk is that the failure definitions are far too conservative. However, this approach is often used if the radiation design margin is sufficient to place the part in the hardness non-critical category without performing a detailed application circuit analysis. For parts that are purchased with “guaranteed” radiation specifications, either by manufacturer claims or by a government approved qualification testing program, the generic failure definition can be post irradiation parameter limits (PIPLs) established for the part for the specified environment. The manufacturer “guarantees” that the parameters will not exceed the specified PIPLs at the specified radiation level. If parts are purchased with PIPLs then the system designer should use these degraded parameters as the specification limits for the part in the system design. There are cases where a generic failure definition is not feasible. For analog single event transients, SETs, one must determine the amplitude and pulse width of an SET that would cause failure in the circuit application since this will differ widely for different applications. The following is an example of a failure criterion established for two operational amplifiers using a combination of input from the system designer and a SPICE circuit analysis. The circuit of interest is shown in Figure 3 [Boul 03].
Figure 3. System application circuit to monitor power distribution in a satellite.
II-13
This circuit is used to monitor the power distribution in a satellite. The LM124 and OP27-1 are used for current limiting and OP27-2 is used as a current sensor. The environment of interest is heavy ions and the failure criterion is an SET at the output of the op amp causing a failure of the application circuit. The system design engineer has determined that this application circuit will cause a system failure if a transient at output 3 exceeds 1.8V for more than 6 µs. This requirement is then flowed down to the three linear circuits in the application circuit. Since the output of OP27-2 is the output of the application circuit, the failure criterion for OP27-2 is the same as the failure criterion of the application circuit. The analysis for the LM124 and the OP27-1, using macro-models for the linear circuits not struck by the heavy ion pulse, and a validated SPICE micro-model to generate the SET, determined that the failure criterion for these linear circuits is an SET that exceeds +/1.25V for 6µs. Hence the worst case failure definition for both the LM124 and OP27 for this application is an SET of +/-1.25V for 6µs. Another example would be the peak allowable photocurrent generated by a component at the specification prompt dose rate. Such radiation responses do not easily fall into the category of functional or electrical parametric failure but rather are radiation responses of the component in the environment. In summary, radiation induced failure of a microelectronic part depends on the application of the part in a system. A rigorous failure definition would require an application circuit analysis for each application of the part for each radiation environment. However, in practice such a rigorous definition is only established when there is a question about whether the part can meet the radiation requirements, e.g. when the part is categorized as unacceptable or if the part is borderline between hardness critical and hardness non-critical, or when the radiation response limits must be defined. Whenever possible, a generic failure definition will be used to avoid the detailed application circuit analysis. Generic failure definitions should always include those failures where the part is essentially no longer useable, such as catastrophic failure and permanent functional failure. In the case of parametric failure generic failure definitions are usually conservative and are only used if the outcome is positive. If the part is border line between categories then a more careful analysis is required to establish a more realistic failure definition. 3.3
Probability of survival and confidence
The probability of survival, Ps, and the confidence, C, are necessary to determine the acceptance or rejection of a lot for hardness critical parts. Ps and C are also necessary when the part categorization criterion (PCC) method is used, as will be discussed later. The probability of survival for a given piece-part is usually derived from the overall Ps assigned to the system, often with the conservative assumption that the failure of a critical piece-part will mean the failure of the system. All microelectronic piece-parts are usually considered critical. For systems like spacecraft that are usually not repairable, the requirements for system Ps can be quite high, whereas for many military systems that may have thousands of units in the field, the requirements are often much lower. For example, if one requires that a given spacecraft have a Ps of 0.9 (90% probability of survival) and there are 200 critical microelectronic piece-parts in the system, then the Ps for each one would be 0.9995, if each were considered equally capable of causing system failure. Hence, for this system we would require that each part type have a Ps (at a given C) of 0.9995. This analysis, of course
II-14
assumes that the Ps for each part type will be 0.9995 and not higher. In reality, parts that are hardness non-critical may have much higher value of Ps than 0.9995. For example, if we require that a hardness non-critical part have a minimum RDM of 20 (some systems require as little as 10 and as much as 100), and we assume a lognormal distribution of the radiation failure levels with a standard deviation of 10% of the mean, then the mean will be 9.5 standard deviations above the specification level. The actual Ps for having the mean 9.5 standard deviations above the specification will depend on the type of failure distribution assumed, the sample size and the assumed confidence level. However, it will be well above 0.9995 in most cases. The actual system Ps is usually determined by the number of parts that are hardness critical and have a Ps near the minimum requirement. This minimum piece-part Ps with specified confidence level is normally set by the system contractor or the program office and not by the piece-part hardness assurance engineers. For the purposes of this presentation, it will be assumed that the Ps and C are determined at the program office and are constraints imposed on the piece-part HA program. The value of C is usually set at 0.9 but may be set at 0.95 for some systems. The value of Ps required for piece-parts can have a broad range, depending on the type of system. For space systems and strategic military systems the Ps is usually set between 0.99 and 0.9999. 3.4
Piece-part radiation characterization data.
The final set of information that is required to establish the category of a piece-part for a given system application is the radiation data on the part. In order to minimize the time and cost of the categorization effort, the optimum solution is to use data that were taken by someone else. There are many sources of radiation data on piece-parts. As discussed in the 2002 short course, there are a number of radiation effects data bases available on the internet. A list of these websites was presented by Christian Poivey and is reproduced in Table 1. Table 1. Radiation effects databases on the web
Agency
Web Site
DTRA
http://erric.dasiac.com
NASA-GSFC
http://radhome.gsfc.nasa.gov
NASA-JPL
http://radnet.jpl.nasa.gov
ESA
http://escies.org
NRL
http://redex.nrl.navy.mil
The information on specific manufacturer’s part types (usually including date codes) varies from averaged summary data to raw data on individual samples. In some cases, only a reference is given to a report. Irradiation details are not always given. There are many other sources of radiation effects data including the open literature and government and contractor technical reports. The best open literature source is the IEEE NSREC Radiation Effects Data Workshop Record, published each year since 1992. This workshop record includes all of the papers presented each year at the Workshop, held as part of this conference. The workshop record includes many radiation effects data compendia, e.g. SEU and SEL response for heavy ions, protons and neutrons, SET response from heavy ions, ELDRS in bipolar linear devices and response of optoelectronic devices. Additional open literature sources include the II-15
December issue of the IEEE Transactions on Nuclear Science, TNS, (papers from this conference) and the Proceedings of the European Conference on RADiation and its Effects on Components and Systems, RADECS, (a subset of papers is published in the June TNS the year following the conference). Government technical reports are usually available but are often restricted in their distribution and contractor technical reports are often considered proprietary. Several for-profit radiation effects test organizations have data available for purchase. While many microelectronic part types, often from multiple manufacturers, have been characterized for their radiation response in several environments, the relevance of these data for categorizing a specific part type in a specific system application is seldom clear. There are many potential problems with the use of data from outside sources. The major problem is whether the samples used for a given data set are representative of the parts that will be used in your system. The chances that the test samples are representative are much greater if the part type is controlled in its circuit design, layout and process. This is usually only the case for standard military parts controlled by a specification or source control drawing (SCD) or standard microcircuit drawing (SMD). For commercial parts there are no controls on the many design, layout and process parameters that potentially influence the radiation response. Therefore, unless the radiation data are taken on samples from the same wafer lots as will be used in the system, the data may not be applicable if the technology has undergone any significant process and/or design changes as often occurs with commercial technologies. Unfortunately with commercial parts there is often no way of knowing the wafer lot from which the samples were taken. The only information that is usually available is the date code, which indicates the week or period of time that parts from various lots were packaged. A date code lot often represents more than one wafer lot. For some parts that are sold by a company that is not the manufacturer, and may have multiple sources for the dice, there is no guarantee that the same date code lot is comprised of dice from the same manufacturer, let alone the same wafer lot. This is one of the major problems confronting users of commercial microelectronic parts in systems with radiation requirements. In short, any data that exit may not representative of the parts that will be used in the system and should be verified by performance of radiation lot acceptance testing to support a system’s requirements. There are other potential problems with the use of existing radiation effect data such as the irradiation conditions. Are the radiation sources used for the test specified and proper? For example if the tests are heavy ion tests, were the ion energies sufficient to provide uniform LET in the sensitive volume, especially for angular irradiation? If the tests are for total dose was the dose rate specified and is it appropriate? Another consideration is the irradiation bias conditions. Was the irradiation bias the worst case, or at least appropriate for the operational conditions of the system application? If not, the test results may be nonconservative. Are there any cases where the use of existing data may be appropriate? While the use of existing data is generally not advised there are cases where it may be applied with engineering judgment. For example if there are cases where, based on an understanding of the radiation response mechanisms, backed by modeling and generic data, the part would be considered hardness non-critical, HNC, then the use of specific data on a part type may be used to verify that assessment. There are many examples of this including the 1) use of neutron data to show that the RDMs for displacement damage in space from protons and
II-16
electrons would place the part in the HNC category, 2) total dose data on digital logic bipolar part families using the same process technology, and 3) prompt dose rate burnout data on most CMOS VLSI parts. In addition to using existing data to place parts in the HNC category, when that assessment is backed by analysis and modeling, the data may be used to decide that a part is HCC-1. However, there are risks in that decision. If one decides to use the existing data to place a part in HCC-1 and the part should actually have been placed in the unacceptable category then every lot that is tested for use in the system may end up failing. This would be an expensive proposition, not only in the cost of the production lots but in the time and cost of the testing. A more conservative approach would be to use the existing data to decide that the part is unacceptable and substitute a different part, if that is possible. In summary, while there many sources of existing radiation effects data and some of the databases are rather extensive, the existing data are often not useful for a specific system P2HA program. The best use of existing data is to place parts in the HNC category when there are already sound engineering reasons to decide that the part is HNC, or to place the part in the unacceptable category and substitute a harder part. While placing a part in the HCC-1 category using existing data may be considered safe, one does run the risk that all of the production lots will fail, resulting in lost time and money. Piece-parts that are purchased with “guaranteed” radiation hardness are a special case. There are two categories of radiation hard parts, those sold under a government program and guaranteed by a specification to meet certain radiation levels, and those sold by manufacturers that claim the part meets certain radiation levels. Parts manufactured under a government program such as the US Radiation Hardness Assured Qualified Manufacturers List, RHAQML, can be considered to meet the specified radiation level. The program is structured to assure that none of the population will fail at the specified levels. If the ratio of the part specification level to the system specified radiation level is sufficient (as will be discussed later) the part can be considered to be HNC. If the part is purchased with a manufacturer’s “guaranteed” hardness level then the specific wafer lot radiation data should be requested and verified that it is appropriate for determining the failure level for the application of the part in the system. If the manufacturer can only offer generic data, data taken by other agencies on the part or data on selected wafer lots then the parts should be subjected to additional characterization testing. If there are no existing data for the part in the radiation environment or there are doubts about the appropriateness of the data, then a radiation characterization test must be conducted. The radiation characterization testing to establish the part category must be taken on a sample that is representative of the variations that will be encountered in the total population of samples that will be used in the fielded system. If the system is a one of a kind satellite that uses a small number of each part type then a single buy should be sufficient for the “life” of the system build. In this case, the characterization test and the lot acceptance test for parts that are HCC-1 can be one and the same. On the other hand if the total number of copies of the system is large and/or the number of parts of a given type used in each copy of the system is very large then there may be a requirement for multiple lots purchased at one time or multiple lots purchased over a long period of time. In either case the total population will consist of multiple lots and the characterization test sample should reflect this. Hence there is a requirement that the sample include multiple wafer lots representing production II-17
over a minimum time period of 6 months to a year. While this does increase the probability that the sample will be representative it certainly does not guarantee it for commercial parts. There is a risk that the several lots selected for the characterization will be selected from a time where the design, layout and process were fixed and the part will be categorized as HNC. In the future, when the design or process is changed, the radiation response may change significantly, causing the part to be HCC-1 or unacceptable. However, once the part is categorized as HNC no further testing will be done. For this reason it is recommended that commercial parts be placed in HCC-1 unless there are sound reasons, based on analysis and modeling, that would warrant putting the part in the HNC category. Once an appropriate population is assembled from which to randomly select the radiation test samples, the radiation characterization testing may be performed for each of the relevant radiation environments. The first thing to do before beginning the radiation characterization testing is to generate a radiation test plan. A test plan is an essential part of any RHA program and must be done with care to ensure completeness, accuracy and traceability. The plan is required to document what is to be done and how, and to force the test engineers (or at least someone in the organization) to think through all aspects of the testing. A test plan should include the following as a minimum: 1) All of the pertinent information about the part, including part type, manufacturer, package, pinouts, electrical specifications, date code (or other legacy information if available, e.g. wafer lot, production lot), and process technology, if known. 2) Sample size and sample preparation. This item will be discussed in more detail when examples are given for variables test data. The sample size should be a minimum of 11 packages for each environment with one part being used as an un-irradiated control device, unless there are extenuating circumstances for using a smaller number, such as cost or availability. [An example of the availability argument would be that only 100 parts were available from the single production lot, and at least 50 are needed for system production with 25 spares, and tests must be conducted for three radiation environments. In this case only 25 parts would be available for radiation testing or any other type of destructive testing.] The method for marking the individual samples should be given and any preconditioning should be specified, such as burn-in or other pre-irradiation elevated temperature stress, if required. 3) Reference should be given to a standard test procedure, if one exists, and the standard procedure should be followed. There are many standard procedures for radiation testing including those from ESA, the US military (MIL-STD-750 and MIL-STD-883), ASTM and JEDEC. A list of standards and guidelines is included in the appendix. 4) List of electrical parameters and functional tests that must be measured as a function of the radiation environment for all step stress tests. There may be tests that are critical to the operation of the part in the system that are not included in the electrical specifications for the part. 5) Description of all in-situ or in-flux tests that must be made for radiation testing that is performed in these modes. Examples are the tests to monitor a) prompt dose rate upset, latchup or burnout, and b) heavy ion or proton induced upset, latchup, burnout, gate rupture or functional interrupt.
II-18
6) Description of radiation sources and the source parameters to be used. This would include the radiation type, energy, flux or dose rate, and dose or fluence levels, as well as the irradiation temperature. 7) The irradiation bias conditions and a circuit diagram of the irradiation test circuit or bias circuit as well as a diagram of the proposed test setup for complex testing such as SEE and dose-rate upset/survivability. 8) A description of the dosimetry system to be used to measure the radiation environment including calibration and frequency of use. 9) A step by step test procedure to include time intervals between irradiation and electrical measurements. 10) A description of the data recording, file formatting and data analysis procedures to produce the data required to determine the individual part failure levels, if required. In most radiation characterization testing the objective will be to determine the radiation failure levels of the individual samples. However, in some cases this will not be possible, as will be discussed later. In those cases where the failure levels cannot be determined then either parameter degradation will be recorded at specific radiation levels or a pass/fail on each sample will be determined at specific overtest levels. 11) An analysis, as required, concerning failure modes, rates and levels for tests such as SEE and dose-rate upset/survivability, due to the complex nature of the tests and failure modes. 12) DUT/test facility analysis to determine critical volumes, sensitive depths, beam penetration and other device and facility interactions. 13) Software required for implementing the testing, especially for microprocessor or I/O bus testing where the DUT’s response is often a function of the software programs and data being exercised. The details of how to conduct the radiation characterization testing in each of the radiation environments has been covered in previous short courses [see e.g. Poivey 2002] and do not need to be repeated here. However, there are issues that need to be emphasized. For example, the part ought to be tested as it will be used. If there is anything that is unique about the application of the part that requires non-standard parameter measurements or nonstandard operating conditions, it should be noted. For example recent data on the low dose rate response of bipolar linear voltage regulators and references has shown that the worst case bias condition is with all leads grounded. Hence, if some of the parts are used in the system as unbiased spares they would represent worst case bias. This is contrary to what has been considered the worst case bias for most microelectronic devices in a total dose environment, which is maximum supply voltage. Research has also shown that the worst case electrical parameter for many positive voltage regulators is the maximum output drive current, a parameter that is not usually included in the electrical specifications [Peas-98]. A recent paper on proton induced degradation of a light emitting diode/phototransistor (LED/PT) pair [Knif 03] showed that the failure in a space system application depended on the size of the aperture between the pair. With a 20 mil aperture, as occurred in the system, the degradation of a critical parameter was much greater than with no aperture (as is normally done in characterization testing). Testing parts the way they are used in the system
II-19
is becoming even more of an issue with the increased use of application specific integrated circuits, ASICS, and configurable logic devices, such as PLDs and FPGAs. When testing configurable devices, the engineer must not only test them with the appropriate bias conditions, but also configure them as they will be in the system. 3.5
Calculation of RDM
In paragraphs 3.1 to 3.4 we have discussed the inputs to the calculation of RDM. RDM is the ratio of the mean failure level of the part type sample to the system radiation level for the part. The system “radiation level” may be given as a minimum or threshold level for which no failures are allowed, as in the case of threshold LET or threshold dose rate level, or it may be given as a maximum allowable level, such as the maximum bit error rate for heavy ion induced upset. As discussed in 3.1 the initial value of the system radiation level for the part should be the top level requirement. The top level number is determined by transporting the free field environment through an effective shielding thickness using a simple transport calculation. The radiation failure values for each sample in the test population are found by analyzing the data on the parts to determine the environment level where failure occurs according to the definition of failure determined from the worst case application of the part in the system. This process can be used for each of the radiation environments based on the effect that is being evaluated and the way the radiation requirement is specified. Table 2 is a list of the different radiation environments and effects that may require evaluation. Table 2. Radiation failure levels for various radiation environments.
Radiation environment
Radiation effect
Radiation failure level
Total ionizing radiation
Functional failure
Maximum dose level for full functional performance
Total ionizing radiation
Parametric failure
Dose level where parameter exceeds allowable limit
Total displacement damage dose (DDD)
Functional failure
Maximum DDD level for full functional performance
Total displacement damage dose (DDD)
Parametric failure
DDD level where parameter exceeds allowable limit
Minimum LET
Latchup
Minimum LET
Single event upset
Maximum LET where no upset occurs
Bit error rate or device error rate for specified environment
Single event upset
Calculated error rate based on cross section vs. LET
Analog single event transients
Calculated error rate based on cross section vs. LET where failure is defined by amplitude and pulse width
Device error rate for specified environment
Maximum LET where no latchup occurs
Threshold prompt dose rate
Upset
Threshold prompt dose rate
Latchup
Maximum dose rate for no latchup
Maximum prompt dose rate
Burnout
Maximum dose rate for no upset
II-20
Maximum dose rate for no upset
For step stress testing, failure has been defined by the maximum test level where failure does not occur. This is an overly conservative definition in many cases, especially when the steps in the radiation levels are large. Getting a more reasonable estimate of the failure level for step stress testing where the failures are abrupt, as in upset, latchup, burnout and functional failure, can be determined using a method of maximum likelihood estimation as described by Namenson [Name-84]. However, if all of the failures fall in one bin, this method cannot be applied and no estimation of the mean and standard deviation is possible. If any of the parts fail at the first radiation test level, then the test should be repeated since there is no way to bound the lowest failure level. If the data are adequate and all of the parts have experienced failure in the test environment, then a radiation failure level should be determined for all of the samples in the test. However, there are a number of cases where this may not occur. For example, if the parts are taken to the highest radiation level for which the facility is capable, e.g. prompt dose rate on a flash x-ray machine or LINAC or highest LET for the available ions, and no failures occur, then one can determine only a lower limit for the response, and the mean failure level cannot be ascertained. In this case, one must assume that the mean failure level is just the highest test level. In SEU testing, for example, if the total particle fluence for each ion is limited to 107 ions/cm2 (to avoid total dose damage), then the device cross section at that LET is just 10-7 cm2 for purposes of calculating a device error rate. The mean radiation failure level for the sample is calculated based on the assumed distribution. For many radiation environments the distribution of failure levels is lognormal. One can perform a goodness of fit calculation for different parametric distributions (normal, lognormal, Weibull, etc) to see which one shows the best fit, but it is usually safe to assume that the distribution is lognormal and calculate the geometric mean. The failure level data should be plotted as the cumulative probability vs. log of the failure level just as a check to see if the distribution is reasonably close to a straight line. As an example consider that we have a sample of 30 parts of a given manufacturer’s part type, 10 each from three different wafer lots fabricated over a period of two years. The total dose failure levels are determined by interpolation between successive radiation steps for the most sensitive parameter that has a failure level determined by the application circuit analysis. Table 3 is a list of the radiation failure levels for each of the samples in arbitrary units (column 2). In Table 3 we have taken the failure levels for the 30 samples and ordered them in increasing value in column 3. In column 4 we have taken the cumulative percent as n/(N+1), where N is the total number of samples, and in column 5 we show the natural log (ln) of the failure level. The last column is the inverse standard normal cumulative probability distribution (NORMSINV of column 4) which has a mean of zero and a standard deviation of 1. Note that the function NORMSINV is available in Microsoft EXCEL. A cumulative probability plot of the ln failure values is shown in Figure 4. This is the same plot one would get on normal probability paper. If the ln failure values lie on a straight line then the distribution is lognormal. For the example we have chosen here the lognormal is a reasonable fit. The failure values vary by a factor of 4 (15 to 60) and the mean of the ln failure values is 3.38 giving a geometric mean of the failure levels of 29.4. The standard mean is 31.3, as shown in the table. If we assume that the specification level for the radiation environment is 10 then the RDM would be 2.94.
II-21
Table 3. Failure levels for 30 samples.
Sample #
Failure level
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
20 55 25 28 22 22 42 15 19.5 17.5 24 28 38 25 33 18 50 33 32.5 35 60 34 16 30 26 46 40 29 31.5 44 mean
Ordered failure level 15 16 17.5 18 19.5 20 22 22 24 25 25 26 28 28 29 30 31.5 32.5 33 33 34 35 38 40 42 44 46 50 55 60 31.3
n/(N+1)
log failure level
NORMSINV
0.032 0.065 0.097 0.129 0.161 0.194 0.226 0.258 0.290 0.323 0.355 0.387 0.419 0.452 0.484 0.516 0.548 0.581 0.613 0.645 0.677 0.710 0.742 0.774 0.806 0.839 0.871 0.903 0.935 0.968 mean stdev exp(mean)
2.71 2.77 2.86 2.89 2.97 3.00 3.09 3.09 3.18 3.22 3.22 3.26 3.33 3.33 3.37 3.40 3.45 3.48 3.50 3.50 3.53 3.56 3.64 3.69 3.74 3.78 3.83 3.91 4.01 4.09 3.38 0.365 29.36
-1.85 -1.52 -1.30 -1.13 -0.99 -0.86 -0.75 -0.65 -0.55 -0.46 -0.37 -0.29 -0.20 -0.12 -0.04 0.04 0.12 0.20 0.29 0.37 0.46 0.55 0.65 0.75 0.86 0.99 1.13 1.30 1.52 1.85
II-22
Inverse normal cumulative probability distribution of log failure levels 2.5 2.0 mean
1.5 1.0 0.5 Stdev
0.0 -0.5 -1.0 -1.5 -2.0 -2.5 2.5
3.0
3.5
4.0
4.5
Ln of failure levels Figure 4. Inverse normal cumulative probability distribution of ln failure levels for 30 samples.
3.6
Categorization
Once we have determined the RDMs for the part types in each of the specified environments, we can proceed to categorization. The are two methods for placing the parts in one of the three categories, the design margin breakpoint method, DMBP, and the part categorization criteria, PCC, method. The DMBP method is an overly conservative method that is often used for systems with moderate requirements, such as Army manned ground combat systems, aircraft or satellites in low earth orbit. The breakpoints between the categories are somewhat arbitrary but are supposedly based on historical data for large categories of part types for each of the radiation environments. If we ignore the category HCC-2, that is shown in Figure 1 as requiring periodic testing but not lot acceptance, then the breakpoints between the categories are as shown in Table 4. Table 4. DMBP breakpoints.
RDM < 2
2 ≤ RDM < 10-100
10-100 ≤ RDM
Unacceptable
HCC-1
HNC
II-23
In Table 4 we have shown the breakpoint between HCC-1 and HNC to be 10-100 since it is not a fixed number but is usually specified by the system program office. Ten is normally the lowest number that is used for most environments but it tends to be overly conservative in some cases and not conservative enough in other cases. In MIL-HDBK-814 (also in MIL-STD-1766 for missiles) the category HCC-2 is specified as having an RDM between 10 and 100 and the HNC category is for parts with RDMs above 100. In reality the assignment of breakpoints is an engineering judgment call, although it should be based on historical data when there are sufficient data to establish a statistical base. The breakpoints between HCC-1 and HNC can be different for different environments. The selection of 2 as the breakpoint between HCC-1 and unacceptable may be too conservative in some cases, since many systems automatically put a factor of 2 design margin on the radiation specification. However if the extra safety margin is not included and the RDM is 1 then by definition one half of the samples will fail, so setting the breakpoint at 2 does give a reasonable chance of having lots pass if the standard deviations are small. The DMBP method of categorization is not only used when the radiation requirements are moderate but must also be used when the failure distribution cannot be determined, such as discussed above where parts cannot be taken to failure. For our example in Table 3, RDM is 2.94 so the part would be categorized as HCC-1. The PCC method is a statistical approach based on the actual failure distribution. It can be applied when there are sufficient data, such as the example given in Table 3. However, it must be emphasized that when categorizing the parts the data base must be sufficiently universal to be representative of the full range of expected failure levels for all of the population of a given part type that will be used over the life of the system. Hence, one cannot use this approach if there are only data on one date code lot of a device and multiple lots will be used in the system. In the case where only one lot will be used for the entire system build and spares, and there is no database to establish a device category, then the part will of necessity be considered HCC-1 and a lot sample test will be performed, making categorization irrelevant. This issue will be addressed in more detail later. Since it is often difficult to establish the proper database for the PCC method, especially if one attempts to used archive data, this method, although more quantitative, is not frequently used. The breakpoints for the categories using the PCC method are shown in Table 5. Table 5. Categorization using the PCC method.
RDM < 2
2 ≤ RDM ≤ PCC
PCC < RDM
Unacceptable
HCC-1
HNC
The value of PCC depends on the assumed distribution. If one can assume that the distribution of radiation failure levels is lognormal then the definition of PCC is the following PCC = exp[KTL*sln(RFAIL)] Where KTL is the one sided tolerance limit that is a function of sample size, Ps and C and sln(RFAIL) is the sample standard deviation of the natural logarithms of the failure levels. This formula is derived from the relationship that the mean failure level minus KTL standard deviations must be greater than the radiation specification level, where RDM is defined by
II-24
RFAIL(mean)/RSPEC. If we apply the PCC method to our example in Table 3 then we must determine a value of KTL for a sample size, n, of 30 and a specified value of Ps and C. Tables of KTL values for typical values of n, Ps and C are given in MIL-HDBK-814. A plot of KTL vs. n for C = 0.9 and values of Ps of 0.9, 0.95, 0.99 and 0.999 is shown in Figure 5.
10 9
Ps=0.9 Ps=0.95 Ps=0.99 Ps=0.999
8
KTL
7
Confidence level = 0.90
6 5 4 3 2 1 0
5
10
15
20 25 30 Sample size
35
40
45
50
Figure 5. The value of the one sided tolerance factor, KTL, vs. sample size for a confidence of 0.9 and various values of Ps.
If we choose a value of Ps of 0.999 for a C of 0.9 then for our example with an n of 30 KTL is 3.79. Hence PCC is exp(3.79*0.365) = 3.99 and RDM is less than PCC, so the part is categorized as HCC-1. This is the same result we got with the DMBP method. Looking at it another way if we take the mean minus KTL standard deviations (3.38-3.79*0.365 = 1.99) and convert it to a radiation level we get 7.36, which is well below the specification level of 10. 3.7
Parts categorized as unacceptable
There are several alternatives for parts with unacceptable RDMs. If the part is manufactured by more than one company, then one or more of the different manufacturer’s parts may be found to have an acceptable RDM. This is certainly the case with the total dose response of bipolar linear circuits [Krie-01]. If a pin for pin replacement from another manufacturer that meets the RDM cannot be found, then a functional equivalent part using a different process technology may work if the parameter limits are acceptable. For example in the case of logic gates and some linear circuits there are often bipolar, CMOS, BiCMOS, and high speed CMOS versions of the same circuit function. If the part has only a single source and part substitution does not work because of system performance or cost, or some
II-25
other factor, then the RDM of the part must be increased. This may be done by changing the radiation specification or the failure definition. The radiation specification should be reevaluated to assure that a realistic value is being used. This may mean doing a more rigorous calculation of the environment at the part location in the system, in case the top level value was too conservative. It may also mean that some type of additional shielding may have to be used, such as spot shielding, or having the part packaged in a radiation shielded package. Another method for increasing the RDM is to re-evaluate the failure definition to assure that it is realistic. If a generic failure definition is used that is too conservative, the application circuit for the part should be analyzed to provide a more realistic failure definition. If this approach does not sufficiently increase the RDM, another option is to redesign the application circuit to relax the post radiation performance requirements for the part. Of course, if the part is failing functionally there is not much that can be done to change the failure definition. Another option is to review the irradiation test conditions to see if they are representative of the normal operating conditions for the part. If worst case bias were used rather than nominal operating bias then a retest may be in order to see if the failure levels increase. For environments where the failure levels are sensitive to processing parameters that may vary from lot to lot, such as with total dose response, it may be possible to meet the RDM requirement by simply testing several date code lots and picking the best one. If none of these approaches result in an acceptable RDM, a final option is to harden the part. This may be done with a number of approaches, all of which will be expensive, but hardening may be the most cost effective approach over the life of the system. If a hardening effort is undertaken it should be sufficient to put the part in the HNC category so that the lot acceptance testing can be eliminated. There are several options for hardening a microcircuit that are currently available. The most cost effective approach depends on a variety of items that include e.g., device complexity, integration density, operating speed, and radiation specifications. The options to achieve a more robust device include 1) radiation hardened by design using a commercial foundry, 2) using the current design on a hard foundry, 3) combining hardened by design with a hardened foundry and finally 4) a full custom redesign in a hard foundry. The options that are available for parts whose RDM places them in the unacceptable category are summarized in Table 6. 3.8
Parts categorized as HCC-1 and HNC
One of the objectives of the piece-part hardness assurance program should be to get as many parts as possible into the HNC category since they do not require lot acceptance testing. Radiation lot acceptance testing is very expensive for all environments but especially so for heavy ion or proton testing. Because of this, if there are parts that have an RDM close to the breakpoint between HCC-1 and HNC then it may be cost effective to re-evaluate these parts in a manner similar to that described in the previous section in order to see if they will fall into the HNC category upon re-evaluation.
II-26
Table 6. Options for increasing the RDM of an unacceptable part.
Factor to be changed
Options for increasing RDM
Radiation specification
Perform a more detailed calculation of environment Re-locate the part in the system to increase shielding Use spot shielding Have the part packaged in a shielded package
Failure definition
Re-evaluate the application circuit for failure Re-design the circuit where the part is used to relax the post radiation requirements
Part response
Select another manufacturer of the same part Select the same functional part in another technology Select a different lot from the same vendor Substitute a different part type Re-evaluate the irradiation bias conditions Harden the part
Many parts can be placed in the HNC category on the basis of analysis, modeling or engineering judgment. For example, if there is a DDD requirement and the total 1 MeV equivalent silicon neutron damage fluence is below 1013 n/cm2, all MOS devices would fall in the HNC category. If the parts use an SOI technology, e.g. SOS, bonded wafer, SIMOX, or the older dielectrically isolated (DI) technology, and the layout rules preclude multiple components in the same isolation region (which is normally the case), the parts can be considered HNC for SEL and prompt dose rate induced latchup. However, one may still have to be concerned about snap-back in SOI that can occur with no body ties or with improper body ties. While it is not a universal rule, in most cases bipolar technologies are immune to single particle induced latchup, even though they may be susceptible to prompt dose rate latchup. There is only one case reported in the literature where single particle latchup has been observed in a bipolar part [Shog-93] and it was later shown to be a form of latch called a “high current anomaly” [Koga-94]. It is not the conventional pnpn latchup, but rather a second breakdown in a pn junction. To our knowledge this type of phenomenon has only been observed in one circuit, an AD9048 8-bit ADC. Another radiation induced failure that is extremely rare, though not universally excluded, is prompt dose rate burnout at the typical peak dose rate specification level of 1011 to 1012 rad/s. Based on analysis and engineering judgment, many parts can be placed in the HNC category for this environment using data on similar part types in the same technology. Another method for getting parts into the HNC category is to buy parts which have guaranteed radiation tolerance. As previously mentioned, the US the Department of Defense has an RHAQML program that provides the framework for assuring that the approved parts meet or exceed the stated radiation specification levels. This framework includes review and
II-27
approval of an RHA Quality Management (QM) plan that describes the management structure, policies and procedures and documentation controlling the RHA process. For each part that is qualified under the program the test methods, sampling plans, radiation facilities and data analysis procedures must be reviewed and approved. The customer has access to the data on each lot. The testing is performed in accordance with radiation test standards and guidelines and the radiation levels and post irradiation parameter limits are provided in a standard microcircuit drawing (SMD). For microcircuits the RHA requirements are specified in Appendix C of MIL-PRF-38535 and for parts qualified to space (class V) the requirements are described in Appendix B. For discrete devices the requirements are described in MIL-S19500 Appendix C. Parts that are purchased under the RHAQML program should be considered HNC as long as the ratio of the SMD specified radiation level to the part system application radiation specified level meets the RDM requirement for HNC using the DMBP method. Because the SMD specified radiation level is a guaranteed minimum and not the mean failure level, the RDM determined in this manner will be overly conservative. Hence the breakpoint for HNC may need to be lowered for RHAQML parts. However, this would be a decision by the program office. One other factor that must be considered when using RHAQML parts is the post irradiation parameter limits, PIPLs, given in the SMD. These limits must be compared to the failure definition for the part to assure that they are acceptable for the part application. As an example of the use of RHAQML parts, if we have a part with a total dose requirement of 20 krad and the breakpoint for HNC is set at 5 for RHAQML parts, then we should buy a part that is specified as a 100 krad part (level R). Table 7 is a list of the RHA level designators for RHAQML parts showing the corresponding total dose levels. Table 7. MIL-PRF-38535 RHA level designators.
RHA level designator
Total dose level (krad)
M
3
D
10
P
30
L
50
R
100
F
300
G
500
H
1000
While the total number of suppliers that are participating in the RHAQML program is growing and a number of RHA part types are currently available, there are a substantial number of manufacturer’s part types that are advertised as radiation tolerant or radiation hard that are not part of the RHAQML program. The only recommendation that can be given for these parts is “caveat emptor”. While many of these products are radiation tested on a lot by lot basis using sound testing techniques, others are advertised as meeting certain radiation II-28
levels based on anecdotal evidence or data taken on single lots of the product by outside testing agencies. If these parts are to be used, it is best to treat them as HCC-1 and perform lot sample testing as a backup. For those part types that cannot be placed in the HNC category (hopefully a small number) radiation lot acceptance testing must be performed. 3.9
Radiation lot acceptance testing.
For all parts in the HCC-1 category radiation lot acceptance testing (RLAT) must be performed. The first issue that must be addressed in RLAT is the definition of a lot. In a strict sense a lot should, as a minimum, refer to a wafer lot, i.e. a set of wafers that are batched processed at the same time using the same equipment. While not all steps in the process are performed in wafer batches, the designation wafer lot (or diffusion lot) is used by most manufacturers, and is well defined and controlled, being documented by a run sheet. In the narrowest sense, a lot is considered to be a single wafer. Many manufacturers do their RLAT on a single wafer, since for some environments (e.g. total dose) there can be significant variations in the radiation response from wafer to wafer. In fact, one manufacturer not only performs lot sample testing on a wafer by wafer basis but only sells product as RHA that comes from the center 2/3 of the wafer. This is based on the assumption that product from the center of the wafer will have a tighter distribution of radiation failure levels. While the preferred lot definition is that of a single wafer or wafer lot, with commercial parts it is often not possible to buy product from a known wafer lot. The exception is when the product is offered in die form. If the commercial parts are purchased in die form then the wafer lot can usually be determined. However, in most cases the only indication of a lot is the date code. The date code, normally a 4 digit number that specifies the week and year of final assembly, is printed on the package. If the “date code” is given with numbers and letters, the time of assembly is not obvious and the manufacturer would have to be contacted to determine this information. Parts with the same date code were assembled at the same time. This does not mean that they come from a single wafer lot. It is possible to have a single date code with multiple lots as well as a single wafer lot with multiple date codes. Finally, a lot can be considered as a purchase lot that may consist of several date codes. This definition of a lot should not be used for selecting parts for RLAT testing, unless it can be assured that all of the parts are from the same mask set and use the same process. The only way to verify that the circuit revision is the same is by examining a die from each data code. Verifying the consistency of the process would require extensive physical analysis such as spreading resistance profiling (SRP). In Table 8 we list the various lot definitions in order of inclusion.
II-29
Table 8. Possible lot definitions for lot acceptance testing.
Lot definition
Description
Single wafer
A single wafer is lot sample tested
Wafer lot
Batch of wafers processed at same time
Date code lot
All parts assembled at same time possibly from several wafer lots
Purchase lot (not recommended)
Parts purchased in one buy that may contain several date codes
In MIL-PRF-38535 Appendix J, Table J-1, the lot for RHA testing is defined as a wafer lot. In addition, Table C-1 allows for the sample selection to be on the basis of a single wafer, wafer lot, or inspection lot. The definition of inspection lot in MIL-PRF-38535 is a quantity of ICs “manufactured on the same production line through final seal by the same production techniques”. For the system P2HA program the RLAT lot should be considered a wafer lot if that can be determined but, as a minimum, a date code lot if the wafer lot cannot be identified. Every part type lot must be RLAT tested in each environment for which it is categorized as HCC-1. There are two types of test methods that can be used for RLAT testing, attributes testing and variables testing. 3.9.1
Attributes testing
Attributes testing is a go/no-go test where each sample is determined to either pass or fail at each radiation level tested. If the test is performed at the system radiation specification level, then there is but one test and each part either passes or fails. This approach is widely used for electrical testing and non-radiation environmental testing. It is also the approach that is used for RHA testing in MIL-PRF-38535. The sampling plans for attribute testing are based on the assumption of a normal or Gaussian distribution and tables are given based on either an AQL (acceptance quality level) or LTPD (lot tolerance percent defective). For the LTPD method, tables have been developed for sample size requirements for a confidence level of 0.9 as a function of LTPD and accept number. The LTPD corresponds to Ps and the accept number is the maximum number of samples in the population that can fail the test. Table 9 shows an example of the LTPD table for accept numbers of 0 and 1. Table 9. Sample size LTPD table for C=0.9 and accept numbers of 0 and 1.
Accept #
LTPD = 20
10
5
2
1
0.5
0.2
0.1
Ps = 0.8
0.9
0.95
0.98
0.99
0.995
0.998
0.999
0
11
22
45
116
231
461
1156
2303
1
18
38
77
195
390
778
1946
3891
II-30
It is clear from Table 9 that if the Ps requirement for the piece-parts in the system is much greater than 0.9 (which is usually the case for space and strategic systems), then the use of testing by attributes is not practical because of the large sample size. While this technique is used for technology conformance inspection (TCI) testing in MIL-PRF-38535 and the samples sizes are such as to only guarantee a Ps of 0.9, the total RHS program is more extensive. A variety of techniques are used to control the hardness, such as identifying and controlling the radiation sensitive process and design parameters and performing radiation testing on test structures and standard evaluation circuits. These controls are lacking in a commercial product and hence the RLAT testing is the only means of determining the radiation performance. In addition RHAQML parts are required to include an overtest factor of 2. Attributes testing is the preferred test technique by both manufacturers and system contractors because it is more production oriented and does not require extensive analysis or engineering judgment. The result of the test for each sample is a simple pass/fail according the failure criteria programmed for the test. Recognizing that attributes testing is preferred and that for many systems a high value of Ps is required, Namenson [Name-82] developed the mathematical framework for a test method he called overtesting. The overtest method allows one to significantly reduce the sample size while meeting the Ps and C specified for the piece-parts by testing the parts at a single level that is much higher than the specification level. The only problem with this method is that a maximum standard deviation for the radiation failure levels for the part must be known or assumed. For many radiation part types and environments this can be a major issue since there is often an insufficient database to make an educated guess for the maximum standard deviation of the failure levels. In the development of the overtest method the assumption is made that the failure levels can be described by a lognormal distribution. The maximum standard deviation is then the standard deviation of the natural logarithms of the failure levels, σln(max). Estimated values of this parameter have been determined for some environments and some part types, but in general good engineering judgment must be used to come up with a reasonable number for σln(max) that will not be overly conservative but reflect the expected variation in failure levels. In the overtest method an overtest factor RT/RS (RT is the radiation test level and RS is the radiation specification level) is calculated for the environment that will provide the required Ps and C for a given sample size, n, and accept number, c. For a lognormal distribution of failure levels the probability of survival Ps (at the system specification level) for confidence C is given by the following relation: Ps = F[ invF(PT) + ln(RT/RS) ] σln(max) where F is the cumulative normal standard probability distribution and PT is the probability of survival of the test samples at test level RT. From this relation the ratio RT/RS can be determined as follows [Name-82]: RT/RS = exp{σln(max) [invF(Ps) – invF(PT)]}
II-31
For a given confidence, C, the value of PT is determined from the following equation PT = exp[ln(1-C)/n]. Using these relations Table 10 shows some typical values for the overtest factors vs. sample size to achieve a Ps of 0.999 at a C of 0.9 specified for a system assuming σln(max) = 0.5 and as a function of σln(max) for sample sizes of 5 and 10. Table 10. Overtest factors for selected values of Ps, C, n, and σln(max)
Overtest Factors for C=0.9 and Ps=0.999 Sample size
PT
4 5 6 7 8 9 10 Sample size 5 5 5 5 5 5 5
0.562 0.631 0.681 0.720 0.750 0.774 0.794 Sigma ln(max) 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Inv F(PT) Inv F(PS) 0.157 0.334 0.471 0.582 0.674 0.753 0.822
3.090 3.090 3.090 3.090 3.090 3.090 3.090
RT/RS 1.317 1.735 2.286 3.011 3.967 5.225 6.883
Inv F(PS)Inv F(PT)
Sigma ln(max)
2.933 2.756 2.619 2.508 2.416 2.337 2.269 Sample size 10 10 10 10 10 10 10 10
0.500 0.500 0.500 0.500 0.500 0.500 0.500 Sigma ln(max) 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.75
RT/RS 4.335 3.967 3.704 3.505 3.347 3.218 3.109 RT/RS 1.255 1.574 1.975 2.478 3.109 3.901 4.894 5.482
The value for σln(max) of 0.5 is a rather conservative number that was calculated for a large data set of neutron failure levels for discrete bipolar transistors and is not relevant for many part types and radiation environments [Name-82]. However, it serves to illustrate how the sample size can be significantly reduced and give the same Ps using a simple attributes test at a prescribed overtest level. To illustrate the value of σln(max) for a sample of ln radiation failure levels refer to Table 3. In this example we showed a distribution of failure levels for 30 samples that varied by a factor of 4. For this example, the value of σln(max) was 0.365. This number would remain the same if the failure levels were scaled since we are calculating the stdev of the logarithms of the failure levels. To illustrate an extreme failure distribution let us assume that we have total dose failure levels that vary by a factor of 10 for a sample of only 10 parts, e.g. 20, 30, 45, 60, 75, 90, 115, 130, 170, and 200 krad. In this case the value of σln(max) is 0.75 and for such a large stdev, for a sample of 10 and an accept number of zero, we would have to use an overtest factor of 5.5 to get a Ps of 0.999 at a C of 0.9. For RLAT testing that is restricted to a worst case lot definition of a single date code, it is highly unlikely that one would see failure levels that varied by a factor of 10.
II-32
There are a number of restrictions for overtest. To use the overtest method the radiation response of the part should be based on a single failure mechanism and the degradation of critical parameters should be monotonic with the radiation environment. An example of where the method should not be used is in rebound testing per TM1019.5. The rebound test requires that the part be exposed to the specification dose, measured to see if it passes, then given an additional dose of 50% of the specification and annealed at 100°C for 168 hours and measured again. The problem with using overtest for rebound testing is that there is a greater possibility of having the part fail functionally when it is given the additional dose prior to the anneal. If the part fails functionally the test is invalid since the bias will be improper during the anneal. If overtest is to be used on CMOS for low dose rate applications, it would be better to perform the total dose test at low dose rate rather than do the two part rebound test. Another example is with non monotonic response, where a parameter degrades with increasing radiation up to a maximum degradation and begins to recover with further irradiation. This behavior has been seen with the so called “latchup window” effect from prompt dose rate in CMOS as well as with certain leakage currents from total dose in digital CMOS. It has also been observed with input bias current in bipolar op amps and comparators irradiated at elevated temperature, such as allowed in TM1019.6. The complex response from the total dose exposures is a result of the different time responses of the buildup and annealing of the trapped positive charge and interface traps. In these cases testing at the higher overtest radiation level will show less degradation than at lower radiation levels, hence the overtest may be non conservative. The overtest method should not be used when the radiation response is not monotonic. 3.9.2
Variables testing
For RLAT testing the preferred method of test is the variables method where data are either taken at a single radiation level and each sample has all of the critical electrical parameters recorded and analyzed, or the data are taken at several levels to record the radiation induced changes in the critical parameters or determine the level where functional failure occurs. The variables data method is more expensive since the results of each test parameter must be recorded and analyzed. However, with automated data recording and software programs for data analysis, the task is much less demanding than it once was. Variables data involves a statistical analysis of the data to determine a mean failure level and standard deviation for the assumed parametric distribution. There are two approaches to variables data testing. If the radiation response results in the gradual degradation of electrical parameters, such as with total dose and DDD, then one may perform the analysis on the changes in the critical parameters at a fixed radiation level, often the specification level. This technique will work only if the changes in the parameters with radiation are either linear or sub-linear. If the change in the parameter is super-linear or abrupt, as it is for many threshold effects, the parts must be taken to failure or at least to where the slope of the response increases abruptly. A good example of an abrupt response is the change in offset voltage on an operational amplifier with total dose. This is illustrated in Figure 6 for an OP42 [John95]. Here we see a linear plot of the average change in the offset voltage vs. total dose at several dose rates. At a dose rate of 22 mrad/s we see that Vos shows very little change out to a total dose of around 27 krad. Above this value the Vos degrades rapidly reaching what would probably be considered a failure (∆Vos of ~13 mV) at 33 krad. When data such as these are encountered, the parts must be tested at several radiation levels to determine a
II-33
radiation failure level distribution. However, if the change in the parameter is known to be linear or sub-linear, the statistics can be performed on the change in parameter.
Figure 6. Change in Vos for OP42 op amp vs. total dose at different dose rates [John-95].
The accept/reject criteria for radiation sensitive parameters at the radiation specification level is the following, assuming a normal distribution: For parameters that increase with radiation the lot is accepted if Para(mean) +KTL*s(para) < Para(lim), and for parameters that decrease with radiation, the lot is accepted if Para(mean) – KTL*s(para) > Para(lim), where Para(mean) is the average value of the post radiation parameter, s(para) is the standard deviation of the post rad parameter values and Para(lim) is the failure level for the parameter. If the distribution of the post radiation parameter values is log normally distributed, the statistical analysis is performed on the logarithms of the post radiation parameter values before being converted. As an example, we will illustrate the parameter variables data test using actual data taken on a sample of 6 quad op amps from a single wafer. In this case the RLAT lot is a single wafer. We will assume that the radiation specification is 100 krad, the most critical parameter is the input bias current and the post radiation allowable limit is +/- 100 nA. We will further assume that the Ps is 0.999 and the C is 0.9. Since there are 4 op amps in each package the total sample size is 24. With the specified n, C and Ps the KTL is 3.9. The results of the analysis are shown in Table 11.
II-34
Table 11. Pre and post irradiation data on Ib+ for 6 quad op amps from a single wafer.
Pre rad Ib+
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
(nA) -7.08 -6.97 -6.71 -6.95 -7 -6.83 -6.7 -6.96 -7.13 -6.88 -6.81 -7 -7 -6.86 -6.71 -6.99 -7.15 -6.93 -6.87 -7.13 -7.12 -7.05 -6.94 -7.1
Post rad Ib+ at 100 krad (nA) -61.2 -55.1 -53.6 -59.8 -61.9 -56.1 -54.5 -60.1 -65.8 -58.6 -56.6 -64.1 -61.4 -55.1 -53.9 -60.2 -62.6 -56.6 -55.0 -61.4 -64.6 -59.5 -57.7 -62.1
n/(N+1)
0.04 0.08 0.12 0.16 0.2 0.24 0.28 0.32 0.36 0.4 0.44 0.48 0.52 0.56 0.6 0.64 0.68 0.72 0.76 0.8 0.84 0.88 0.92 0.96 mean sdev
Ordered NORMSINV abs Ib+ (nA) 53.6 -1.75 53.9 -1.41 54.5 -1.17 55.0 -0.99 55.1 -0.84 55.1 -0.71 56.1 -0.58 56.6 -0.47 56.6 -0.36 57.7 -0.25 58.6 -0.15 59.5 -0.05 59.8 0.05 60.1 0.15 60.2 0.25 61.2 0.36 61.4 0.47 61.4 0.58 61.9 0.71 62.1 0.84 62.6 0.99 64.1 1.17 64.6 1.41 65.8 1.75 59.06 3.61
A cumulative probability plot of the absolute value of Ib+ is given in Figure 7 plotted with a linear trendline fit.
II-35
Inverse normal cumulative distribution of post rad Ib 2.0
inverse cum prob
1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 52
57
62
67
Post rad Ib (nA)
Figure 7. Cumulative probability plot of 100 krad Ib+ for 24 op amps from a single wafer.
From Figure 7 it is seen that the data provide a reasonable fit to a normal distribution. Using the acceptance criterion given above for parameters that increase with radiation we get a value of Para(mean) + KTL*s(para) of 73.1 nA, which is well below the specification limit of 100 nA, so the lot passes. The use of the variables data approach for RLAT is normally quite limited since it only applies to those environments where there is gradual parametric degradation that is well behaved. For many cases, the variables data approach will be used to determine a distribution of failure levels for the test samples. This will be the case for functional failure and for all parameters that show abrupt changes (threshold effects) with radiation. Examples of threshold effects are almost all transient ionization effects (prompt dose rate and single particle) where a minimum amount of charge or junction photocurrent must be reached before the circuit responds. For those environments the radiation level, dose rate or LET or proton energy must be increased until failures are observed. Hence for these environments the variable parameters are with the radiation environment and not the device parameters. Examples are given here for prompt dose rate upset in an inverter and single event upset in an SRAM. Let us assume that we have a system that will have to operate in the presence of a nuclear weapons threat and that the system will be powered down during the transient environment. The power down dose rate level will need to be set at a level high enough so that the system cannot easily be disabled repeatedly but low enough so that all of the electronics will be able to operate through the prompt dose rate pulse. For purposes of illustration, let’s say that the level is set at 106 rad/s. This means that the specification for no upset is 106 rad/s. Again assume the Ps is 0.999 and C is 0.9. We have a simple inverter that is in category HCC-1. The failure criteria for the inverter are that the Vout high cannot drop below Vcc/2 and the output Vout low cannot go above Vcc/2. A sample of 10 hex inverters from one date code lot is tested on a linac for the threshold dose rate upset with ½ of the inverter outputs in each package high and ½ low. The minimum dose rate upset level is
II-36
determined for each package according to the stated failure criteria. The dose rate is adjusted for each inverter so that the minimum threshold upset level is determined within 50%. The results are shown in Table 12 for the highest dose rate for no upset. These values will be used for the threshold failure levels. If a better value of threshold dose rate upset level is desired then the method of maximum likelihood can be used [Name-84]. Figure 8 is a plot of the inverse cumulative probability distribution of the ln of the threshold dose rate upset levels. Table 12. Highest dose rate for no upset for a sample of 10 hex inverters. sample # 1 2 3 4 5 6 7 8 9 10
Thres.upset (rad/s) 6.50E+06 7.00E+06 7.00E+06 7.50E+06 8.50E+06 9.00E+06 1.00E+07 1.00E+07 1.20E+07 1.30E+07
n/(N+1) 0.09 0.18 0.27 0.36 0.45 0.55 0.64 0.73 0.82 0.91 mean stdev KTL
ln of fail NORMSINV dose rate n/(n+1) 15.69 -1.34 15.76 -0.91 15.76 -0.60 15.83 -0.35 15.96 -0.11 16.01 0.11 16.12 0.35 16.12 0.60 16.30 0.91 16.38 1.34 15.99 0.24 4.63
2.0
inverse cum prob.
1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 15.5
15.7
15.9
16.1
16.3
16.5
ln (thes. upset) Figure 8. Inverse cumulative probability distribution for the ln of the highest dose rate for no failure for 10 hex inverters.
If we take the mean minus KTL* sln(fail) we get 2.9 × 106 rad/s so, again, the lot passes. II-37
In the next example we will use the OP27 that was shown in Figure 3, where we determined that the failure definition is any SET that exceeds +/-1.25V for 6 µs. We will consider a geostationary orbit at solar minimum as our worst case environment. We will use a Ps of 0.99 at a C = 0.9. Our sample size will be 5 giving us a KTL of 4.67. The allowable error rate for this system application will be given as 1 failure per year or 2.74 × 10-3 errors/day. In order to determine an experimental error rate for the samples we will have to test them with heavy ions to get an error cross section vs. LET and input this data into a code to calculate the error rate. We will use CREME96 for this calculation. The parts are tested with heavy ions and all SETs are recorded. From the data we determine a peak amplitude and pulse width for each SET. In Figure 9 on the left we show a plot of the SET amplitude vs. pulse width, where we have highlighted in solid black all of the pulses that exceed +/1.25V for 6 µs. In Figure 9 on the right, we show the SET failure cross section vs. LET with a Weibull fit to the data. 0
1.E+05
-2
cross section (um^2)
Pulse heigth (Del V) (V).
-1 -3 -4 -5 OP27 INV G = 10 Vin = -0.4V
-6 -7 -8
Weibull A 3.5E4 um2 X0 1 MeV-mg/cm2 W 5 MeV-mg/cm2 s 5 MeV-mg/cm2
1.E+04
1.E+03
Weibull OP27(-0.4V INV) OP27(-0.06V INV)
-9 -10 0.E+00
5.E-06
1.E-05
1.E+02 0
20
Pulse width (PW) (s)
40
60
80
LET
Figure 9. OP27 SET pulse amplitude vs. pulse width on left and Weibull fit to SET failure cross section vs. LET on the right.
The calculated values of error rate are shown in Table 13, along with the ln of the error rate and the average and stdev of the ln error rates. Figure 10 shows the inverse cumulative probability of the ln of the error rates to demonstrate that they fit a log normal distribution. Table 13. SET error rates for 5 samples of OP27 for SET fail of 1.25V at 6 µs.
Sample # 1 2 3 4 5
ER (e/d) 4.50E-03 5.50E-03 6.00E-03 4.00E-03 7.00E-03
ordered n/(N+1) ER 4.00E-03 0.17 4.50E-03 0.33 5.50E-03 0.50 6.00E-03 0.67 7.00E-03 0.83 mean stdev KTL
II-38
ln(ER)
NORMSINV
-5.52 -5.40 -5.20 -5.12 -4.96 -5.24 0.22 4.67
-0.97 -0.43 0.00 0.43 0.97
Inverse cumulative prob. dist.
1.50 1.00 0.50 0.00 -0.50 -1.00 -1.50 -5.60
-5.40
-5.20
-5.00
-4.80
Ln error rate [ln(e/d)] Figure 10. Inverse cumulative distribution of ln SET error rates for OP27.
The value of exp[ER(mean) + KTL*sln] is 1.5 × 10-2 error per day so the lot fails. For our last example we consider an RLAT for a 4 Mbit SRAM for heavy ion induced upset. The specification for the device is that the bit error rate will not exceed 10-8 errors per bit day for the specified orbit and mission duration with a Ps of 0.99 and a C of 0.9. We have purchased the 4 Mbit SRAMs in die form from a single wafer lot. Because of the expense of the test and the parts, we will only test a sample of 5 which will mean that the KTL is 4.67. The choice of sample size is a risk factor since doubling the sample size to 10 will reduce the KTL to 3.53 and potentially reduce the risk of failing the lot. However, we are willing to take the risk because of the cost of parts and the facility time and our confidence that the parts will pass. The parts are subjected to a full Group A electrical test. Five samples are randomly drawn from the population of parts passing electrical tests. Our five samples are then irradiated at a heavy ion accelerator using approved test methods to obtain data of bit error cross section for increasing values of effective LET until a saturation cross section is obtained. The data are then analyzed to generate a plot of bit error rate vs. LET(eff) for each sample and the data are fit to a Weibull curve. The threshold LET and saturated cross section are determined for each sample and the data input to a software program such as CREME96 or SPACERAD to calculate the bit error rate for the specified environment. The results for the five samples are shown in Table 14 for the bit error rate given in error/bit/day (e/b-d). We have assumed a lognormal distribution.
II-39
Table 14. Bit error rates for 4 Mbit SRAM for specified environment
Sample #
e/b-d
ordered n/(N+1) ln(e/b-d) NORMSINV e/b-d
1
6.0E-09 5.6E-09
0.17
-19.00
-0.97
2
5.6E-09 6.0E-09
0.33
-18.93
-0.43
3
7.6E-09 6.7E-09
0.50
-18.82
0.00
4
7.2E-09 7.2E-09
0.67
-18.75
0.43
5
6.7E-09 7.6E-09
0.83
-18.70
0.97
mean
-18.84
stdev
0.13
Figure 11 is a plot of the inverse normal cumulative distribution of the logarithms of the bit error rates. From this figure we see that the lognormal assumption is reasonable.
Inverse cum. prob. dist.
1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -19.1
-19.0
-18.9
-18.8
-18.7
-18.6
Ln (e/b-d) Figure 11. Inverse normal cumulative distribution for ln of bit error rate for the 5 samples of the 4 Mbit SRAM.
From Table 14 the standard deviation for this sample is 0.13 and the value of exp[(mean) + KTL*s] is 1.2 × 10-8 e/b-d, hence the lot fails. For this example the assumption that we could get by with a sample of 5 was not a good one. If we had taken a sample of 10 and gotten the results shown in Table 15 the lot would have passed, even though the range of values of bit error rate is greater (2.8-5.4 × 10-9 e/b-d) than for our sample of 5 (3-5 × 10-9 e/b-d). This is because the standard deviation and KTL are smaller than for the sample of 5. Again we have assumed a lognormal distribution which is a reasonable fit as shown in Figure 12.
II-40
Table 15. Same as Table 14 but for a sample of 10.
Sample #
e/b-d
ordered e/b-d
n/(N+1) ln(e/b-d) NORMSINV
1
6.30E-09 5.40E-09
0.09
-19.04
-1.34
2
6.80E-09 5.80E-09
0.18
-18.97
-0.91
3
7.90E-09 6.00E-09
0.27
-18.93
-0.60
4
5.40E-09 6.30E-09
0.36
-18.88
-0.35
5
6.70E-09 6.50E-09
0.45
-18.85
-0.11
6
7.40E-09 6.70E-09
0.55
-18.82
0.11
7
5.80E-09 6.80E-09
0.64
-18.81
0.35
8
7.10E-09 7.10E-09
0.73
-18.76
0.60
9
6.50E-09 7.40E-09
0.82
-18.72
0.91
10
6.00E-09 7.90E-09
0.91
-18.66
1.34
mean
-18.84
stdev
0.12
Inverse cum. prob. dist.
1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 -19.1
-19.0
-18.9
-18.8
-18.7
-18.6
Ln (e/b-d) Figure 12. Inverse normal cumulative distribution for ln of bit error rate for 10 samples of the 4 Mbit SRAM.
For the sample of 10, the value of exp[(mean) + KTL*s] is 9.8 × 10-9 e/b-d which just manages to pass the acceptance criterion.
II-41
4.0
Challenges for piece-part hardness assurance for space systems
We have described a framework for piece-part hardness assurance that is applicable to any microelectronic system that has a radiation requirement. While the methodology was developed for military systems using established parts with stable production it is generally applicable, with some modifications, for space systems using a combination of radiation tolerant and commercial parts, as we have tried to demonstrate. The modifications, as we have discussed consist of 1) eliminating the HCC-2 category, 2) restricting the definition of a lot, 3) recognizing that when a single lot buy is made for the life of a system the characterization test serves as the RLAT, and 4) changing the definition of RDM for the SEE environment. An example of how this methodology can be modified for commercial parts is illustrated in Figure 13 which shows the methodology adapted for analog SETs in linear circuits. Device type
System Requirements WC test conditions Risk assessment ER(allow) Ps and C
Mission parameters
Application analysis SPICE/macros
∆V, PW fail
Archive data sufficient?
Laser data
no
yes Error rate cal. e.g.CREME96 ERexp(mean)
Get Weibull parameters
Calculate RDM= ER(allow) / ERexp(mean)
Heavy ion tests
∆Vmax, PWmax< fail?
yes
no
Part categorization Unacceptable action required
Circuit redesign
Re-evaluate
shielding
Part substi.
Hardness critical HCC
HNC no further action
Lot acceptance tests
Figure 13. Piece-part HA methodology for ASETs in linear circuits.
In the flow diagram of Figure 13, an example of which has been presented for the OP27 in a circuit to monitor current distribution in a satellite, the system input is given in terms of an allowable error rate, Ps, C and mission parameters (used to calculate a worst case environment). The application circuit is analyzed to come up with failure criteria in terms of a maximum allowable pulse amplitude and width. If we have no heavy ion data on the part under the worst case operating conditions, we can use a laser to see if the selected part type is capable of generating a failing SET pulse. If not the part is HNC. If the SET pulses with the laser exceed the failure criteria then we must take heavy ion data to get an SET failure cross section vs. LET to calculate the error rate. This error rate is compared to the allowable error
II-42
rate to see if the part is HNC, HCC or unacceptable. If unacceptable then we must lower the error rate. If HCC we must perform lot acceptance testing. The DMBP value that separates HCC from HNC will be set by the program office, and for this environment may be less than 10. This methodology is quantitative, based on known statistical relations and sampling techniques. It is in contrast to the qualitative risk assessment described in [LaBe-98] that ranks parts in three loosely defined risk categories of low, medium and high, based on the availability of radiation effects data and the engineering judgment of the lead radiation effects engineer. The NASA approach to hardness assurance for space flight programs [LaBe-98] is described in terms of risk management rather than risk avoidance since “… one is unlikely to have enough knowledge about radiation effects in commercial devices to deal with risk in a precise, mathematical way. There are simply too many unknowns.” While we make no claims about the precision of the methodology described herein, it does present a mathematical approach that allows for a quantitative assessment of risk, rather than a loosely defined qualitative assessment. It should be emphasized that this methodology is to be used as a tool and not a set of inflexible rules. In this section we will discuss some of the challenges in using this quantitative methodology, especially with commercial parts. The use of radiation tolerant parts is straightforward if the parts are purchased through a system such as the US RHAQML program since the parts will have a well known, documented, and verified radiation performance. The situation with commercial parts is quite different and requires further discussion. However we do believe that the problem can be quantified in order to meet the required probability of survival of the system in the radiation environment. 4.1
Knowing the relevant details of the part when it’s commercial
One of the first challenges with commercial parts is in knowing what you have. The information on the circuit design and architecture as well as the process technology is often incomplete or not available. With the competitive nature of the commercial market many microcircuits are frequently being redesigned, or scaled down, and the processes are continually being modified to improve yield and performance. Radiation response is, of course, not a consideration for any of these changes. Hence the radiation response of the part can change dramatically whenever one of these changes takes place. An example of this is the bipolar linear circuit line at National Semiconductor. I use this example because it has been the subject of numerous open literature publications. Many of the common circuit types of operational amplifiers and comparators (that are frequently used in space systems) have been in production at National since the 1980’s. In the late 1980’s, these parts were fabricated in Santa Clara, CA and many were qualified for space radiation levels. In the early 1990’s, when the enhanced low dose rate sensitivity (ELDRS) phenomenon was discovered in linear bipolar circuits [McCl-94] [John-94], the production of the National bipolar linear parts had been moved to the UK. When the process was moved several changes were made, including the final passivation layer. The parts fabricated in the UK on a 4 inch wafer line were found to exhibit ELDRS. Although very little testing was ever performed at low dose rate on the parts fabricated in Santa Clara, one group of LM124s that came from a wafer bank was flown on the Microelectronics and Photonics Test Bed (MPTB) space experiment [Titu-98] Titu-99]. Both the ground and space tests on these parts showed
II-43
that they did not have enhanced degradation at low dose rate, whereas the parts made in the UK did. In the late 1990’s, the decision was made to change the fabrication line from 4 inch wafers to 6 inch wafers because of the obsolescence of the 4 inch equipment and the 4 inch silicon wafers. When the line was changed to a 6 inch line, some of the higher production parts were sent to Arlington, TX and some remained in the UK. In addition, some of the commercial parts were also fabricated in China. In the change from 4 inch to 6 inch wafers several of the anneal steps were changed, as well as the passivation equipment. All of the parts fabricated on the 6 inch line started showing severe degradation at high dose rate and low dose rate [Peas-02]. On the LM139 quad comparator, in an effort to reduce the size of the chip, the circuit was re-laid out, causing one of the comparators to degrade at a much lower dose than the other three comparators on the same die [Peas-02]. For this example, several circuits that have been in production for over 20 years with only minor, if any, circuit re-designs, have gone from relatively hard to total dose at high and low dose rate (4 inch line in Santa Clara) to hard at high dose rate but soft at low dose rate (4 inch line in UK) to soft at high and low dose rate (6 inch line in three locations). This issue has been addressed, and it has been demonstrated that the hardness at high and low dose rate can be restored [Mahe-04]. However, this example illustrates how much the radiation response can change with what were considered to be minor changes in layout and processing. It also illustrates that the radiation response of the same circuit type from the same manufacturer can vary widely from date code to date code. Because of the possibility that the part may undergo changes from one production run to the next, it is always a good idea to examine the die. This is very easy to do if the parts are purchased in die form, but requires the assistance of a failure analysis group if the parts are in plastic, since the plastic above the die must be etched away. For parts in hermetic packages the lids can usually be removed with judiciously applied pressure. The die should be examined to identify all markings. There is usually an indicator of the manufacturer, where the die was fabricated and the circuit revision. A photomicrograph(s) with sufficient resolution to identify all of the critical indicators should be kept with radiation test report for future reference. While an examination of the die surface will reveal critical information about the chip vendor, wafer foundry, circuit revision and layout, it will not indicate critical processing details. A TEM cross section and a spreading resistance doping profile can provide useful information about the process, but there is almost no way to determine if minor process changes have been made from one lot to the next. Even though all of the surface features are the same, critical processing changes can result in different radiation response that would invalidate using data on one lot to accept another lot, when dealing with commercial parts. With today’s changing world in microcircuits (due in large part to the cost of state-ofthe-art fabrication facilities and mask sets), there is an increasing number of microcircuit suppliers that do not fabricate their chips. There are fabless circuit design houses that use commercial foundries for processing, and there are packaging companies that buy dies from several sources and package and screen the parts. There are also “radiation hardened” sources of supply that manage the integration of the several tasks of circuit design, fabrication, packaging and post assembly testing. There are also “radiation hardened” sources of supply that merely buy commercial parts and perform a simple LTPD lot sample radiation test as the basis of their claim that the parts meet a specified radiation level. Other sources of supply take commercial parts with moderate radiation performance and package II-44
the dies in shielded packages to increase their radiation tolerance. Whenever parts are purchased from a non-vertically integrated vendor (one that does not perform all of the functions in-house), one should make sure what type of die is in the package. For example, one may purchase a given circuit type from a packaging vendor and receive a single date code lot that is comprised of dies from multiple sources. The only way you would have of verifying which dies are from which manufacturers would be to open all the packages and check the dies, unless you could have some documentation from the seller that all of the dies were from a single source. It is not recommended that parts purchased from a packaging firm that uses multiple die sources be used in a system with radiation requirements unless one can be certain that all of the parts in a given purchase lot come from a single wafer run. It would be better to purchase the parts in die form from a single wafer lot and have them packaged by a qualified packaging house. 4.2
Knowing the part response to the radiation environment in the application.
Once the part has been uniquely identified with respect to circuit design, layout, processing and packaging, one must consider what radiation data may exist on that part that would be useful in categorizing the parts for its specific application in a system. This has become an increasingly difficult task as the complexity of the parts increase, as the variety of the way parts are “manufactured” increases and as designs and processes are continually changed to increase yield and performance. With commercial parts, almost every wafer lot is in some ways unique. This is not the case, however, with parts that are fabricated in a wafer foundry, where the design rules and process are held constant for a given technology. The use of archival data also depends on the system application of the part and whether the part was tested with the proper bias and operating conditions and whether the critical parameters for system application were monitored. Because of these, and other reasons, the opportunity to use archival data to categorize parts is diminishing, while the desire to use these data is increasing, because of the cost of testing. Hence, for the most part, whenever a commercial part is being used in a space system, the part will need to be characterized for its radiation response in each environment in order to confidently place it in one of the three categories. The exceptions to this have been addressed and consist of those cases where there is historical data based on technology, or analysis and modeling that can demonstrate that the part has sufficient margin to be placed in the HNC category. For many part types in space radiation environments the test methods and procedures are known and have become standardized. The Appendix lists many of the radiation test methods that are available in the US and Europe that have been developed either as military standards, European Space Agency (ESA) standards, ASTM standards and guidelines or Electronic Industries Association (EIA)/Joint Electronic Devices Engineering Council (JEDEC) standards and guidelines (also listed in reference [Flee-03]). While these test standards and guidelines provide all of the necessary information about irradiation test facilities, dosimetry, irradiation test fixtures and test procedures, they only cover known radiation response mechanisms in standard devices. The list of effects in state-of the-art microcircuits is growing with the new developments in technology and the testing guidelines cannot always keep up with the new information. Therefore, the radiation test engineers must keep abreast of the most recent information in the literature to adequately design a radiation test on some of the newer microcircuit technologies that are emerging in today’s market. Such issues as single event transients in bipolar linear circuits and sub-half-micron II-45
digital CMOS technologies, enhanced low dose rate sensitivity (ELDRS) of bipolar linear circuits, displacement damage in optoelectronic devices, single event functional interrupt in CMOS VLSI circuits, single event gate rupture in thin gate oxides and effects of direct ionization from protons at steep angles in optoelectronic devices are all recently identified phenomena. While the ELDRS phenomenon was identified in 1991 [Enlo-91] and was reported in microcircuits in 1994 [McCl-94], [Beau-94], and [John-94], it has taken 12 years before the phenomenon was specifically addressed in a test method (MIL-STD-883 test method 1019.6 dated March 2003). An updated guideline on ELDRS (ASTM F1982) was just approved in September 2003. Many of these other recently discovered phenomena have not been addressed in the test standards and guidelines. When using complex or new technology commercial parts in space systems, it is important to fully understand the functional and parametric characteristics of the part and how the part will be used in the system application. This will in most cases necessitate working closely with the design engineer using the part in order to assess the operating conditions as well as the critical electrical parameters and functional requirements of the part in order to adequately design the radiation test and determine failure levels. Sometimes this will have to be an iterative process as more is learned about the radiation response of the part through characteristic testing. It is not the purpose of this short course section to address test methods for emerging technologies since testing techniques have been covered in other short courses [NSRE-00]. However, it is necessary to point out that how a part is tested (irradiation operating and bias conditions) as well as what characteristics are monitored to determine failure, are extremely important in establishing the failure distribution of the test sample. Another value of the characterization testing is for establishing the test procedure that will be used for any lot acceptance testing that may follow. As we have mentioned before, if the characterization testing is performed on a single lot buy for the life of the system, then the RLAT has essentially been performed by default. However, if subsequent lots of the same circuit type are purchased, RLAT testing will be necessary, unless, of course, the part is categorized as HNC. 4.3
Knowing the radiation environment
There are two types of radiation threats in space, the natural environment and the potential nuclear weapon environment. In the case of the nuclear weapon environment, there is the immediate threat of the high altitude or space weapon detonation and the delayed threat of the increase in the radiation belts or “pumped” belts which may last for weeks or months. Most non-military space systems are only hardened against the natural environment without regard for the possibility of “pumped” belts. On the other hand military space systems must consider both the natural environment and the potential weapon environment. In dealing with the natural radiation environment, there are generally three near earth environments that are considered: trapped radiation, galactic cosmic rays (GRC) and solar activity. As discussed in numerous short courses, there are computer models based on historical data for estimating the environments for a given mission, based on the orbital parameters and the mission duration. These radiation environments can then be converted into familiar radiation specifications such as total dose in rad(Si), fluence vs. LET, nonionization energy loss (or displacement damage dose) or particle fluence vs. energy (such as for protons). As a first approximation, the radiation external to the spacecraft is determined
II-46
from these models and then the environment may be transported through an equivalent spherical aluminum shield using a one dimensional transport code. From this radiation environment as a function of shield thickness, one can get a “top level” estimate of the radiation environment at the part level by estimating an equivalent aluminum shield thickness for the part location in the system. While this is often conservative, it has usually worked well in the past when multiple layers of design margins (margins on radiation specifications, margins on failure definitions and margins on part response) were imposed to assure conservative designs. With today’s emphasis on better, faster, cheaper the designs are becoming more aggressive, the margins are being reduced and the radiation response of the parts is becoming less known with the increased use of commercial parts to increase performance and reduce cost. Hence another challenge for today’s space systems is that the radiation environment at the part level must be known more precisely. This means that better methods of modeling the environment and transporting it through complex multi-layered systems are required. As we have mentioned in a previous section, if the part is borderline between either HCC-1 and acceptable or between HCC-1 and HNC, then we may have to do a better job of determining the radiation environment. An example of how the total dose environment can change dramatically with shielding thickness is shown in Figure 14 for a five year mission with a sun-synchronous, circular orbit at different elevations.
Total Ionizing Dose [rad(Si)]
1E+08 1000 km 800 km 600 km
1E+07 1E+06 1E+05 1E+04 1E+03 0
50
100
150
200
Aluminum Shielding Depth (mils) Figure 14. Total dose vs. aluminum shield depth for a 5 year mission (sun-synchronous, circular orbit).
This has been approached on two fronts, modeling the natural radiation environment and developing codes to improve the accuracy and speed of transporting the environment through a complex three dimensional system. In recent years the radiation models of the environment have been improved with new data, e.g. CRÈME 96, and methods of estimating
II-47
solar activity using extreme value theory have been developed [Xaps-99] [Xaps-00]. Transport code development has been bolstered by increasing computing power and Monte Carlo codes for three dimensional transport calculations are readily available [NSREC Short Course-1997 and 2002]. For parts that are determined to be very marginal in RDM a more accurate estimation of the radiation environment is often necessary. For the threat of nuclear weapon generated environments in space, the radiation environments definitions are usually not the responsibility of the piece-part hardness assurance engineer but rather the system program office or prime system contractor or their consultants and contractors. For the piece-part HA engineer the potential weapon environment is normally a given. However, the piece-part HA engineer, should be an active participant in the discussion about how the environment is defined since he will be the one to translate this into a radiation test to determine failure levels. Often key information about the radiation environment that is necessary to properly design a radiation test is missing, such as timing or particle energy spectra. One example is prompt dose rate pulse width. If a peak prompt ionizing radiation dose rate specification is given, it should always be specified with a full width half maximum (FWHM) pulse width. In most cases, the weapon environment is specified as a free field environment that is incident on the spacecraft. This free field environment is often considered the “top level” environment and transporting the environment to the part location is not done unless the part has a borderline RDM. Similar tools are used for the transport of a weapon environment as for a natural environment. 4.4
Knowing how to quantify the results of radiation testing
Another major challenge of space system piece-part HA is meeting a high level for probability of survival at specified confidence. The Ps that is commonly used for military systems that are built in large numbers, some of which may be expendable in a battlefield scenario, are not acceptable for strategic systems or for spacecraft that are extremely expensive and difficult to replace or repair. The probability of survival requirements for these systems must of necessity be very high, forcing the Ps for piece-parts to very high levels. Where a piece-part Ps for a relatively inexpensive cheap high volume system may be 0.8 or 0.9 the requirements for strategic systems and space systems may be 0.99 or higher. The challenge for the HA engineer is to meet this requirement with a cost effective test. The techniques that we have described as the “traditional” piece-part HA method provide approaches for meeting these stringent requirements based on overtest and variables testing. 5.0
Piece-part hardness assurance management
For P2HA to be effective the system program office must be committed to some form of the methodology as described herein. This means that the P2HA engineers must be given the support of the parts, materials and processes control board (PMPCB), the procurement department, the design engineers and all levels of management. Without the commitment to the P2HA program and the support of management team, implementing such a methodology will have very little chance of success. The P2HA engineers should be represented on the PMPCB, which must approve all electronics parts to be used in the system and any changes made to the parts list. The design engineers must be willing to work with the P2HA engineers to establish radiation failure criteria for the parts. The procurement personnel must
II-48
be willing to work with the P2HA engineers to establish the methods of assuring that all HCC-1 parts are purchased according to the established lot definitions. Another important aspect of the P2HA program is documentation. A radiation effects/hardness assurance file should be kept on all microelectronic parts in the system. The file, whether physical or electronic or both, should contain all information pertinent to the hardness assurance for the part type, as follows; 1.
The location of every application of the part in the system,
2.
The electrical and radiation specifications on the part,
3.
Any archival radiation effect data on the part,
4.
Worst case and special case failure definitions for each radiation environment,
5.
Part specific radiation environment requirements, if they exist,
6.
Results of the categorization analysis including the calculations,
7.
Test plans for any characterization testing performed,
8.
A compete test report for each characterization test performed,
9.
Test plans for RLAT testing, and
10. A test report for each RLAT test with the resulting acceptance or failure of the lot stated. In addition a radiation effect database should be established which contains the results of all of the characterization and RLAT testing on all of the microelectronic parts for the system. This can be used for new system designs, re-designs and upgrades. 6.0
Conclusions
In this section of the short course we have presented a piece-part hardness assurance methodology based on the methodology developed in the mid 1980s for US military and space systems and presented in MIL-HDBK-814. We have shown how this methodology can be adapted for COTS is space systems. This methodology is quantitative, but is flexible and should be used as a tool and not a set of hard rules with fixed numbers. The method is based on categorizing parts in three categories, unacceptable, hardness critical and hardness noncritical. The categorization is based on a radiation design margin that is determined from the radiation environment at the part level, the failure criteria based on the system application of the part and the part radiation response. If parts are purchased as radiation hardened or radiation tolerant they should be designed in the system so that they are categorized as hardness non-critical. There are many variables that lead to uncertainties in the environment, the failure criteria, and the part response, but it is better to have a means of quantifying the risk of having a part fail in a radiation application than to just hope for the best. We have discussed methods for reducing some of the uncertainties, but recognize that uncertainties will always exist. For the piece-part hardness assurance methodology to work it must have the backing of management and the assistance of the parts control board, the system designers and the procurement department.
II-49
7.0
Acknowledgements
This effort was funded by DTRA under contract DTRA01-02-D-0008, Delivery Order 4. I would like to thank the DTRA radiation Hardened Microelectronics Program for support. I would also like to thank Lew Cohn, Dave Alexander and Joe Srour for technical discussions, Christian Poivey for information on European HA documents and Harvey Eisen for the list of HA documents in the Appendix. 8.0
References
[Beau-94]
J. T. Beaucour, T. Carriere, A. Gach, D. Laxague and P. Poirot, “Total Dose Effects on Negative Voltage Regulator”, IEEE Trans. Nuc. Sci. NS-41, No. 6, 2420-2426, December 1994.
[Beck-02]
H. N. Becker, T. F. Miyahira and A. H. Johnston, “Latent damage in CMOS Devices from single-event latchup”, IEEE Trans. Nucl. Sci. Vol. 49, No. 6, pp 3009-3015, Dec 2002.
[Boul-03]
Y. Boulghassoul, P. Adell, J. D. Rowe, L.W. Massengill, R.D. Schrimpf and A. L. Sternberg, "System-Level Design Hardening Based on Worst-Case ASET Simulations," Proceedings of the 7th RADECS, Noordwijk, The Netherlands, September 2003.
[Dyer-03]
C. Dyer, S. Clucas, F. Lei, P. Truscott, R. Nartello, and C. Comber, “Comparative simulations of single event upsets induced by protons and neutrons in commercial SRAMS ”, Proceedings of the 7th RADECS, Noordwijk, The Netherlands, September 2003.
[Enlo-91]
E. W. Enlow, R. L. Pease, W. E. Combs, R. D. Schrimpf and R. N. Nowlin, "Response of Advanced Bipolar Processes to Ionizing Radiation", IEEE Trans. Nuc. Sci. NS-38, No. 6, 13421351, December 1991.
[Flee-03]
D. M. Fleetwood and H. A. Eisen, "Total Dose Hardness Assurance", IEEE Trans. Nucl. Sci. NS-50, No. 3, 552-564, June 2003.
[HDBK-94]
MIL-HDBK-814, Military Handbook: Ionizing dose and neutron hardness assurance guidelines for microcircuits and semiconductor devices, Feb. 8, 1994.
[Holm-02]
Handbook of Radiation Effects, 2nd Edition, Andrew Holmes-Siedle and Len Adams, Oxford University Press, 2002.
[John-94]
A. H. Johnson, G. M. Swift, and B. G. Rax, “Total dose effects in conventional bipolar transistors and integrated circuits”, IEEE Trans. Nucl. Sci. Vol. 41, No. 6, pp 2427-2436, Dec 1994.
[John-95]
A. H. Johnson, B. G. Rax and C. I. Lee, “Enhanced damage in linear bipolar circuits at low dose rate”, IEEE Trans. Nucl. Sci. Vol. 42, No. 6, pp 1650-1659, Dec 1995.
[Knif 03]
S. D. Kniffin, et al “The impact of system configuration on device radiation damage testing for optical components”, Proceedings of the 7th RADECS, Noordwijk, The Netherlands, September 2003.
[Koga-94]
R. Koga, et al, “Ion-induced sustained high current condition in a bipolar device”, IEEE Trans. Nucl. Sci. Vol. 41, No. 6, pp 2172-2178, Dec 1994.
[Krie-01]
J. Krieg, T, L, Turflinger and R. Pease “Manufacturer variability of enhanced low dose rate sensitivity (ELDRS)” IEEE NSREC Radiation Effects Data Workshop Record, p. 167-171, 2001.
[LaBe-98]
K. A. LaBel, A. H. Johnston, J. L. Barth, R. A. Reed and C. E. Barnes “Emerging Radiation Hardness Assurance (RHA) issues: A NASA Approach for Space Flight Programs”, IEEE Trans. Nucl. Sci. NS-45, No. 6, 2727-2736 Dec 1998.
[Lei-02]
F. Lei, et al, “MULASSIS: A Geant4-Based Multilayered Shielding Simulation Toolkit”, IEEE Trans. Nucl. Sci. Vol. 49, No. 6, pp 2788-2793, Dec 2002.
II-50
[McCl-94]
S. McClure, R. L. Pease, W. Will and G. Perry, “Dependence of total dose response of bipolar linear microcircuits on applied dose rate”, IEEE Trans. Nucl. Sci. Vol. 41, No. 6, pp 2544-2549, Dec 1994.
[Mahe-04]
M. Maher, M. Shaneyfelt, and R.L. Pease, “Development of a passivation to eliminate ELDRS in bipolar linear circuits”, Presented at the HEART Conference, Monterey, CA, March 2004.
[Name-82]
A. I. Namenson, “Hardness assurance and overtesting”, IEEE Trans. Nucl. Sci. Vol. 29, No. 6, pp 1821-1826, Dec 1982.
[Name-84]
A. I. Namenson, “Statistical analysis of step stress measurements in hardness assurance”, IEEE Trans. Nucl. Sci. Vol. 31, No. 6, pp 1398-1401, Dec 1984.
[Name-85]
A. I. Namenson and I. Arimura, “Estimating electronic parameter end points for devices which suffer abrupt functional failure during radiation testing”, IEEE Trans. Nucl. Sci. Vol. 32, No. 6, pp 4250-4253, Dec 1985.
[Name-87]
A. I. Namenson and I. Arimura, “A logical methodology for determining electrical end points for multi-lot and multi-parameter data”, IEEE Trans. Nucl. Sci. Vol. 34, No. 6, pp 1726-1729, Dec 1987.
[Peas-94]
Pease, R. L. and D. R. Alexander, "Hardness Assurance for Space System Microelectronics", Radiat. Phys. Chem., Vol. 43, No. 191-204, 1994.
[Peas-96a]
R. L. Pease, W. E. Combs, A. Johnston, T. Carriere, C. Poivey, A. Gach, and S. McClure, "A Compendium of Recent Total Dose Data on Bipolar Linear Microcircuits", IEEE Radiation Effects Data Workshop Record, 28-37, 1996.
[Peas-96b]
Pease, R. L., "Total Dose Issues for Microelectronics in Space Systems", IEEE Trans. Nuc. Sci. NS-43, No. 2, 442-452, April 1996.
[Peas-98]
Pease, R. L., S. McClure, J. Gorelick and S. C. Witczak, “Enhanced Low-Dose-Rate Sensitivity of a Low-Dropout Voltage Regulator”, IEEE Trans. Nucl. Sci. NS-45, No. 6, 2571-2576, December 1998.
[Peas-02]
R. L. Pease, M. C. Maher, M. R. Shaneyfelt, M. Savage, P. Baker, J. Krieg, and T. Turflinger "Total Dose Hardening of a Bipolar Voltage Comparator", IEEE Trans. Nucl. Sci. NS-49, No. 6, 3180-3184, Dec 2002.
[Peas-03]
Pease, R. L., “Hardness Assurance for Commercial Microelectronics”, International Journal of High Speed Electronics and Systems, 2003.
[Shin-98]
J. L. Shinn, et al, “Validation of a Comprehensive Space Radiation Transport Code”, IEEE Trans. Nucl. Sci. Vol. 45, No. 6, pp 2711-2719, Dec 1998.
[Shog-93]
M. Shoga, et al “Observation of single event latchup in bipolar devices” IEEE NSREC Radiation Effects Data Workshop Record, p. 118-120, 1993.
[Soli-98]
J. R. Solin, “The GEO Total Ionizing Dose”, IEEE Trans. Nucl. Sci. Vol. 45, No. 6, pp 29642971, Dec 1998.
[Soli-03]
J. R. Solin, “Bremsstrahlung Dose Enhancement Factors for Satellites in 12-, 14.4- and 24- Hr Circular Earth Orbits”, Proceedings of the 7th RADECS, Noordwijk, The Netherlands, September 2003.
[Summ-87]
G. P. Summers, C. J. Dale, E. A. Burke, E. A. Wolicki, P. Marshall, M. A. Gehlhausen and M. A. Xapsos, “Correlation of Particle-Induced Displacement Damage in Silicon”, IEEE Trans. Nucl. Sci. Vol. 34, No. 6, pp 1134-1139, Dec 1987.
[Titu-98]
J. L. Titus, W. E. Combs, T. L. Turflinger, J. F. Krieg, H. J. Tausch, D. B. Brown, R. L. Pease and A. B. Campbell, “First Observations of Enhanced Low Dose Rate Sensitivity (ELDRS) in Space: One Part of the MPTB Experiment”, IEEE Trans. Nucl. Sci. NS-45, No. 6, 2673-2680 Dec 1998.
II-51
[Titu-99]
J. L. Titus, D. Emily, J. F. Krieg, T. L. Turflinger, R. L. Pease and A. B. Campbell, “Enhanced Low Dose Rate Sensitivity (ELDRS) of Linear Circuits in a Space Environment”, IEEE Trans. Nucl. Sci. NS-46, No. 6, 1608-1615, Dec 1999.
[Trus-00]
P. Truscott, et al, “GEANT4- A new Monte Carlo Toolkit for Simulating Space Radiation Shielding and Effects”, IEEE NSREC Radiation Effects Data Workshop Record, p. 147-152, 2000.
[Walt-03]
Walters, Rob, et al. “High energy proton irradiation effects in GaAs devices”, Proceedings of the 7th RADECS, Noordwijk, The Netherlands, September 2003.
[Woli-85]
E. A. Wolicki, I. Arimura, A. J. Carlan, H. A. Eisen and J. J. Halpin, “Radiation hardness assurance for electronic parts: Accomplishments and Plans”, IEEE Trans. Nucl. Sci. Vol. 32, No. 6, pp 4230-4236, Dec 1985.
[Xaps-99]
M. A. Zapsos, et al, “Probability model for worst case solar proton event fluences”, IEEE Trans. Nuc. Sci. NS-46, No. 6, 1481-1485, December 1999.
[Xaps-00]
M. A. Zapsos, et al, “Characterizing solar proton energy spectra for radiation effects applications”, IEEE Trans. Nuc. Sci. NS-47, No. 6, 2218-2223, December 2000.
II-52
Appendix: LIST OF RADIATION HARDNESS ASSURANCE RELATED DOCUMENTS MILITARY PERFORMANCE SPECIFICATIONS 19500, General Specification for Semiconductor Devices. 38510, General Specification for Microcircuits. 38534, Performance Specification for Hybrid Microcircuits. 38535, General Specification for Integrated Circuits (Microcircuits) Manufacturing. MILITARY HANDBOOKS 814, Ionizing Dose and Neutron HA Guidelines for Microcircuits and Semiconductor Devices. 815, Dose Rate HA Guidelines. 816, Guidelines for Developing Radiation Hardness Assured Device Specifications. 817, System Development Radiation Hardness Assurance. 339, Custom Large Scale Integrated Circuits Development and Acquisition for Space Vehicles. 1547, Electronic Parts, Materials, and Processes for Space and Launch Vehicles. 1766, Nuclear Hardness and Survivability Program Guidelines for ICBM Weapon Systems and Space Systems. MILITARY TEST METHODS IN MIL-STD-750 (Test Methods for Semiconductor Devices): 1017, Neutron Irradiation Procedure 1019, Ionizing Radiation (Total Dose) Test Procedure 1032, Package Induce Soft Error Test Procedure (Due To Alpha Particles) 1080, Single Event Burnout and Single Event Gate Rupture 3478, Power MOSFET Electrical Dose Rate Test Method 5001, Wafer Lot Acceptance Testing IN MIL-STD-883 (Test Methods and Procedures for Microelectronics): 1017, Neutron Irradiation Procedure 1019, Ionizing Radiation (Total Dose) Test Procedure 1020, Dose Rate Induced Latchup Test Procedure 1021, Dose Rate Upset Testing of Digital Microcircuits 1023, Dose Rate Response of Linear Microcircuits 5004, Screening Procedures 5005, Qualification and Quality Conformance Procedures 5010, Test Procedures for Complex Monolithic Microcircuits Military standards, specifications, and handbooks can be viewed and ordered from the Web site of the DoD Document Automation and Production Services (DAPS), Bldg. 4/D (DPMDODSSP), 700 Robbins Ave., Philadelphia, PA 19111-5094. Their URL is
II-53
http://www.dodssp.daps.mil. For assistance, one can phone their help line at 215-697-2179 or Fax –1462. Most can also be viewed and downloaded at DSCC’s web site, http://www.dscc.dla.mil/Programs/MilSpec/default.asp. DTRA DOCUMENTS DNA-H-93-52, Program Management Handbook on Nuclear Survivability. DNA-H-95-61, Transient Radiation Effects on Electronics (TREE) Handbook. DNA-H-93-140, Military Handbook for Hardness Assurance, Maintenance and Surveillance (HAMS). DTRA documents can be obtained from the Defense Technical Information Center (DTIC). Phone 800-225-3842. ASTM STANDARDS The following are test and measurement standards. They are under the oversight of ASTM Subcommittee F1.11, Quality and Hardness Assurance; Chairman, Allan Johnston, 818-3546425. F448, Test Method for Measuring Steady-State Primary Photocurrent. F526, Test Method for Measuring Dose for Use in Linear Acceleration Pulsed Radiation Effects Tests. F528, Test Method of Measurement of Common-Emitter D-C Current Gain of Junction Transistors. F615, Practice for Determining Safe Current Pulse Operating Regions for Metallization on Semiconductor Components. F616, Test Method for Measuring MOSFET Drain Leakage Current. F617, Test Method for Measuring MOSFET Linear Threshold Voltage. F676, Test Method for Measuring Unsaturated TTL Sink Current. F744, Test Method for Measuring Dose Rate Threshold for Upset of Digital Integrated Circuits. F769, Test Method for Measuring Transistor and Diode Leakage Currents. F773, Practice for Measuring Dose Rate Response of Linear Integrated Circuit. F980, Guide for the Measurement of Rapid Annealing of Neutron-Induced Displacement Damage in Silicon Semiconductor Devices. F996, Test Method for Separating an Ionizing Radiation-Induced MOSFET Threshold Voltage Shift into Components Due to Oxide Trapped Holes and Interface States Using the Subthreshold Current-Voltage Characteristics. F1190, Practice for the Neutron Irradiation of Unbiased Electronic Components. F1192, Guide for the Measurement of Single Event Phenomena (SEP) Induced by Heavy Ion Irradiation of Semiconductor Devices. F1262, Guide for Transient Radiation Upset Threshold Testing of Digital Integrated Circuits. F1263, Test Method for Analysis of Overtest Data in Radiation Testing of Electronic Parts. F1467, Guide for Use of an X-Ray Tester (~10 keV Photons) in Ionizing Radiation Effects Testing of Semiconductor Devices and Microcircuits. F1892, Guide for Ionizing Radiation (Total Dose) Effects Testing of Semiconductor Devices. F1893, Guide for the Measurement of Ionizing Dose-Rate Burnout of Semiconductor Devices.
II-54
The following are radiation dosimetry standards. They are under the oversight of ASTM Subcommittee E10.07; Chairman, Dr. Dave Vehar, 505-845-3414. E265, Test Method for Measuring Reaction Rates and Fast-Neutron Fluences by Radioactivation of Sulfur-32. E496, Test Method for Measuring Neutron Fluence Rate and Average Energy from 3H(d,n)4He Neutron Generators by Radioactivation Techniques. E665, Practice for Determining Absorbed Dose Versus Depth in Materials Exposed to the X-Ray Output of Flash X-Ray Machines. E666, Practice for Calculating Absorbed Dose from Gamma or X Radiation. E668, Practice for Application of Thermoluminescence Dosimetry Systems for Determining Absorbed Dose in Radiation-Hardness Testing of Electronics. E720, Guide for Selection of a Set of Neutron-Activation Foils for Determining Neutron Spectra Used in Radiation-Hardness Testing of Electronics. E721, Method for Determining Neutron Energy Spectra with Neutron-Activation Foils for Radiation Hardness Testing of Electronics. E722, Practice for Characterizing Neutron Energy Fluence Spectra in Terms of an Equivalent Monoenergetic Neutron Fluence for Radiation-Hardness Testing of Electronics. E1026, Methods for Using the Fricke Dosimeter to Measure Absorbed Dose in Water. E1249, Practice for Minimizing Dosimetry Errors in Radiation Hardness Testing of Silicon Electronic Devices Using Co-60 Sources. E1250, Test Method for Application of Ionization Chambers to Assess the Low Energy Gamma Component of Cobalt-60 Irradiators Used in Radiation-Hardness Testing of Silicon Electronic Devices. E1854, Practice for Assuring Test Consistency in Neutron-Induced Displacement Damage of Electronic Parts. E1855, Method for Use of 2N2222 Silicon Bipolar Transistors as Neutron Spectrum Sensors and Displacement Damage Monitors. E1894, Guide for Selecting Dosimetry Systems for Application in Pulsed X-Ray Sources. The following are fiber optic test standards. They are under the oversight of ASTM Subcommittee E13.09, Optical Fibers for Molecular Spectroscopy; Chairman, Dr. Tuan BoDinh, 423-574-6249. E1614, Guide for Procedure for Measuring Ionizing Radiation-Induced Attenuation in SilicaBased Optical Fibers and Cables for Use in Remote Fiber-Optic Spectroscopy and Broadband Systems. E1654, Guide for Measuring Ionizing Radiation-Induced Spectral Changes in Optical Fibers and Cables for Use in Remote Raman Fiber-Optic Spectroscopy. ASTM standards can be purchased from ASTM, 100 Barr Harbor Dr., West Conshohocken, PA 19428-2959. Phone 610-832-9585. Portions of each standard can be viewed at http://www.astm.org. EIA TEST METHODS AND GUIDES EIA/JESD-57, Test Procedures for the Measurement of Single-Event Effects in Semiconductor Devices from Heavy Ion Irradiation.
II-55
JESD-89, Measurement and Reporting of Alpha Particles and Terrestrial Cosmic Ray-Induced Soft Errors in Semiconductor Devices. EIA/JEP-133, Guideline for the Production and Acquisition of Radiation-Hardness Assured Multichip Modules and Hybrid Microcircuits. EIA/TIA-455-64 (FOTP-64), Procedure for Measuring Radiation-Induced Attenuation in Optical Fibers and Cables. Some EIA/JEDEC standards can be obtained from Global Engineering Documents, 15 Inverness Way East, Englewood CO 80112-5704. Phone 800-854-7179. Otherwise see http://www.jedec.org or http://www.eia.org. ESA TEST METHODS AND GUIDES ESA/SCC Basic Specification No. 22900, Total Dose Steady-State Irradiation Test Method. ESA/SCC Basic Specification No. 25100, Single Event Effects Test Method and Guidelines. ESA PSS-01-609, The Radiation Design Handbook. ESA documents can be obtained from ESA/SCC Secretariat (TOS-QCS), ESTEC P.O. Box 299, 2200 AG Noordwijk, The Netherlands. Compiled by Harvey Eisen, chairman of the AF/NASA/DTRA Space Parts Working Group, Hardness Assurance Committee and consultant to ITT Advanced Engineering and Sciences Corp.
II-56
2004 IEEE NSREC Short Course
Section III
Optical Sources, Fibers, and Photonic Subsystems
Allan H. Johnston Jet Propulsion Laboratory
Optical Sources, Fibers, and Photonic Subsystems A. H. Johnston Jet Propulsion Laboratory California Institute of Technology
I
Introduction and Background A. Some Basic Principles of Optics B. Some Important Semiconductor Properties
II
Optical Emitters A. Light-Emitting Diodes B. Laser Diodes
III Radiation Environments and Radiation Damage A. Space Environments B. Special Environments: Nuclear Reactors and Particle Accelerators C. Fundamental Interactions D. Energy Dependence of Displacement Damage IV Radiation Damage in Optical Emitters A. Light Emitting Diodes B. Laser Diodes. V
Detectors A. Silicon Detectors B. Radiation Damage
V
Optical Fibers A. Propagation B. Radiation Damage
VII Optical Subsystems A. Digital Optocouplers 1. Operating Principles 2. Radiation Degradation 3. Transients in Optocouplers B. Optical Receivers 1. Operating Principles 2. Radiation Effects in Optical Receivers VIII. Summary
III-1
Optical Sources, Fibers and Photonic Subsystems A. H. Johnston Jet Propulsion Laboratory California Institute of Technology
I. Introduction and Background This section of the Short Course discusses radiation effects in photonic components, along with some examples of the radiation response of photonic subsystems. It begins with a review of some optical principles that are important in the discussions that follow. A. Some Basic Principles of Optics Wavelength and Energy Light can behave either as a wave (classical optics) or particle. The energy of the photon associated with light of wavelength λ is given by the relationship below:
E = hc/λ
(1)
where E is the photon energy, h is Planck’s constant (6.62 x10-27 erg-s), c is the velocity of light, and λ is the wavelength. A practical way to apply this is to remember the simplified relationship for common units
E = 1.24 / λ
(2)
where E is in electron volts, and the wavelength λ is in µm. Absorption Absorption in optical materials is described by Beer’s law, -αx
I = Ioe
(3)
where I is the light intensity after the light has traveled a distance x in the material, Io is the light intensity at the surface (ignoring surface reflection), and α is the absorption coefficient in The research in this chapter was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration (NASA).
III-2
reciprocal cm. A penetration depth can be defined by 1/α, corresponding to a drop in intensity by the factor 1/e. Equation 3 applies to a wide range of materials, including amorphous materials such as glass and many semiconductors. It will be used later in discussions of optical detectors, laser diodes, and optical fibers. Absorption in amorphous materials decreases with increasing wavelength because of Rayleigh scattering. The functional dependence is quite strong, obeying a λ-4 power. Fig. 1 show how absorption in silica (silicon dioxide) depends on wavelength. This example is for typical core materials in optical fibers, so the units – dB/km – relate to long-distance optical fiber applications. The Rayleigh scattering region is apparent at shorter wavelengths. The presence of small amounts of water in the silica increase the absorption in the regions shown in the figure. The three “windows” at 850, 1300 and 1500 nm correspond to regions that are of practical interest for fiber communication because fiber losses are acceptably low at those wavelengths. Many optoelectronic components have been developed for those specific wavelengths because of the strong interest and demand for optical communication applications. 7
Attenuation, dB/km
6 5 4 Water peaks
3 2 "First window"
"Third"
1 "Second"
700
800
900
1000 1100
1200 1300 1400 1500 1600
Wavelength, nm Fig. 1. Absorption dependence on wavelength for highly purified amorphous silicon dioxide with small amounts of water present.
Absorption in crystals depends on the bandgap and the nature of the material, decreasing near the band edge when the photon energy falls below the energy required to raise carriers from the valence to the conduction band. Fig. 2 shows how the absorption coefficient of three different semiconductors depends on wavelength. The bandgaps of the three materials are 1.08 µm (silicon), 0.86 µm (GaAs) and 1.65 µm (InGaAs). Because silicon has an indirect bandgap (direct and indirect semiconductors are discussed in more detail in the next subsection), the absorption coefficient of silicon changes gradually with increasing wavelength, compared to the more abrupt wavelength dependence of the other two materials that have direct bandgaps. Although not shown in the figure, germanium also has an indirect bandgap, with an absorption edge at about 1.8 µm. Germanium detectors are useful for longer wavelengths, but detectors with III-V semiconductors (such as InGaAs) have largely supplanted germanium in fiber optic receiver applications because they have lower dark current as well as a relatively constant absorption coefficient over a wide range of wavelengths.
III-3
Fig. 2. Dependence of the absorption coefficient on wavelength for three different semiconductor materials.
Reflection and Refraction Index of Refraction Light travels more slowly in an optical medium compared to its velocity free space, by a factor n, where n is the index of refraction. The index of refraction is approximately equal to the square root of the dielectric constant of the material. The table below shows the index of refraction for several materials of interest. The index of refraction depends on wavelength, decreasing at shorter wavelengths. The values in Table 1 are approximate. For AlGaAs (a solid solution) the index of refraction depends on the mole fraction of aluminum, decreasing with increasing Al concentration. Table 1 Index of Refraction for Several Materials of Interest
Material amorphous silica (glass) silicon GaAs Al0.3Ga0.7As InP
Index of Refraction 1.5 3.45 3.62 3.39 3.54
III-4
Snell’s Law Refraction of light at an interface between two different materials obeys Snell’s law,
n1 sin(θ1) = n2 sin(θ2)
(4)
where n1 and n2 are the refractive indices of the two materials, and θ1 and θ2 are the angle of incidence and the angle of refraction, as shown in the simple diagram of Fig. 3, below. In this case the ray approaches the interface from the medium with lower refractive index, and Eq. 4 is valid for all angles because the angle of refraction is always less than the angle of incidence.
θi n1 n2 θr
Fig. 3. Refraction of a light wave at the interface between two materials with indices of refraction n1 and n2.
For the case where the light ray approaches the interface from the region with higher refractive index, the situation is more complex because the angle of refraction is greater than the angle of incidence. Equation 4 has no real solution when the angle of refraction exceeds 90 º. For that angle (called the critical angle or Brewster angle) and all angles exceeding the Brewster angle, light is no longer transmitted from the first medium to the second, but is reflected. This is called total internal reflection. It is fundamental to an understanding of optical fibers, discussed in Section III, as well as in light extraction from LEDs.
θr
n1 n2 θi θi
>
θc
Total reflection for
θi > θc
θc = sin-1
( nn
2 1
n2 > n1
Fig. 4. Refraction when the light approaches the interface from the region with higher index of refraction, showing the critical or Brewster angle.
III-5
Even though the simplified diagram of Fig. 4 shows complete reflection at the interface when the angle of incidence exceeds the Brewster angle, the incident wave actually extends slightly beyond the interface region. This evanescent wave extends into the surface of the dielectric. The penetration distance of the evanescent wave into the material with lower dielectric constant is ≈ 0.1 λ. This will be important when we consider optical confinement in very thin layers of materials used in laser diodes, where the dimensions of different materials are much less than one wavelength. Antireflection Coatings It is possible to increase light transmission at an interface by depositing a thin film with a lower refractive index. The film properties are chosen so that the effective film thickness is equal to ¼ of the wavelength, resulting in constructive interference that eliminates the reflected component at the uncoated interface. The condition for normal incidence is given by
d =
λ 4n
(5)
where d is the thickness of the coating, λ is the wavelength, and n is the refractive index of the coating material. Most optical components incorporate some form of antireflective film to increase light transmission because of the large Fresnel reflection losses that occur for materials with large refractive index. In practice, multiple films are used for more effective antireflection properties when the light strikes the interface at more oblique angles. B. Some Important Semiconductor Properties Bandgap The bandgap of a semiconductor corresponds to the energy difference between bound electrons in the valence band and mobile electrons in the conduction band. Recall from the band theory of solids that the motion of carriers in a crystal is described by a wave function, ψk = u(r)eik·r, where u(r) is a modulating function, and k and r are momentum (wavenumber) and distance variables within the lattice. The energy E of the carriers is expressed by the equation
E = ћ2k2/2m
(6)
where ћ is Plancks’ constant divided by 2π, k is the wave number, and m is the effective mass of the carriers within the semiconductor lattice. This is illustrated in the simple diagram of Fig. 5a for an ideal material with quadratic dependence of the wavenumber k on energy. Real materials are much more complex, with energy bands that depend on the crystal orientation. Direct and indirect semiconductors The selection rules of quantum mechanics apply to transitions between the valence and electron bands of a solid. For the case where the maximum energy in the valence band corresponds to the same value of k as the lowest energy in the conduction band, it is possible for a carrier to move between the two regions with a direct transition, corresponding to absorption or emission of a photon. Semiconductors with this alignment of the minimum and maximum of the two band regions are direct bandgap semiconductors, and they are the class of semiconductors III-6
that are the most important for optoelectronic emitters because they can absorb or emit photons with high efficiency. E
E
Conduction band
Eg Eg
Valence band
0
k changes with energy transition (requires phonon for energy transition)
k does not change with energy transition k
0 <000>
a) Direct Semiconductor
<111>
kxyz
b) Indirect Semiconductor
Fig. 5. Simplified diagram showing the alignment of valence and conduction bands in direct and indirect semiconductors.
For many semiconductors (particularly silicon), the minimum in the conduction band does not correspond to the same wavenumber as the maximum in the valence band. In this case, transitions between the two bands require absorption (or emission) of a phonon in addition to a phonon in order to change the wavenumber, k. Consequently, photon absorption and emission have relatively low probabilities, and are usually not the dominant process in these classes of materials. Although indirect semiconductors are useful as detectors, they are not capable of efficient light emission because of the requirement for phonons as part of the transition process. Interband recombination is proportional to the product of the n- and p-carrier densities, R = Bnp
(7)
Values of the proportionality constant, B, are shown in Table 2 for several different semiconductors. For semiconductors with direct bandgap the band-to-band recombination probability is 5 to 6 order of magnitude greater than for semiconductors with indirect bandgap. Table 2. Rate Constant for Band-to-Band Recombination for Various Semiconductors
Semiconductor
Bandgap
Rate Constant, B (cm3/s)
silicon
indirect
1.79 x 10-27
germanium
indirect
5.25 x 10-26
GaP
indirect
5.37 x 10-26
GaAs
direct
7.21 x 10-22
InP
direct
1.26 x 10-21
III-7
Band Tailing Real semiconductors have many possible states, distributed with slightly different energies along the conduction and valence band boundaries. Transitions are possible between different states, leading to a distribution of photon energies for optical transitions. The bandgap energy also depends on doping concentration. For doping levels > 1017 cm-3 the presence of a large number of easily ionized dopant atoms distorts the band structure, causing states with high occupation to spill into the normally forbidden region that exists when the semiconductor is lightly doped. This effect is called band tailing [Pank76], as shown in Fig. 6. It is important for laser diodes as well as for light emitting diodes with compensated doping. There are several important consequences. First, the photon energy is reduced somewhat compared to lightly doped material because electrons in the conduction band can make a transition to the highly populated “tail” region in the valence band, which extends into the forbidden region of the bandgap. Second, the wavelength is slightly longer (corresponding to lower energy transitions), reducing absorption losses in the material. That characteristic is particularly important for compensated semiconductors.
With high doping, Impurity states extend into forbidden region of bandgap
E
N(E) Fig. 6. Distortion of band edges at high doping levels. The extension of impurity bands into the forbidden region of the bandgap allows transitions between lightly populated states in the valence band and the heavily populated region in the valence band, reducing the energy of photons that are emitted during the transition.
Absorption and Recombination Mechanisms For direct bandgap semiconductors, there are several processes that involve absorption or emission of photons: • Spontaneous emission of a photon, moving an electron from the conduction band to the valence band • Absorption of a photon, elevating an electron to the conduction band and simultaneously creating a hole in the valence band, and • Stimulated emission, where the presence of photons within the semiconductor “triggers” emission of additional photons. An important property of stimulated emission is that photons produced by this process have the same direction and frequency of the initiating photon. III-8
Those three processes are illustrated in Fig. 7. The first process, spontaneous emission, is responsible for light emission in light-emitting diodes. The last process, stimulated emission, is the mechanism involved in semiconductor lasers. The probability of stimulated emission is very low under low injection condition, but it becomes the dominant recombination process when the density of carriers is sufficiently high. The fact that the photons that are emitted by that process have the same direction and frequency as the initiating photon is critically important for semiconductor laser operation.
Ec
Ev Spontaneous emission
Stimulated emission
Absorption
Fig. 7. Absorption and emission processes that are accompanied by photon emission or absorption.
It is also possible for carriers to recombine through non-radiative processes (loss mechanisms). Two important processes are illustrated in Fig. 8. Non-radiative recombination through traps is shown at the left. We will see later that one of the important effects of radiation is to increase non-radiative recombination losses through deep level recombination centers, which are introduced by displacement damage. Another loss mechanism is Auger recombination [Vurg97, Fehs02], shown at the right. In this example, two electrons interact, releasing energy to one of the electrons involved in the initial interaction. The electron with the higher energy is released from the lattice, eventually releasing the excess energy in the form of heat. The other electron loses energy in the initial interaction, falling to the valence band without releasing a photon (the lost energy was gained by the “hot” electron in the original interaction). Auger recombination can be considered the inverse of the impact ionization process. High carrier densities are required; the probability for Auger recombination depends on the third power of the carrier density, as well as on basic material properties. A third loss mechanism, surface recombination, must also be considered. However, surface recombination from radiation damage is less significant for most III-V semiconductors compared to silicon, partly because it is not possible to form insulators with the same high quality as silicon dioxide in III-V semiconductors.
Ec
Ev Recombination through trap
Auger recombination
Fig. 8. Two important non-radiative recombination mechanisms for direct-bandgap semiconductors.
III-9
Quantum Efficiency The term quantum efficiency is often used to describe the effectiveness of radiative processes. The internal quantum efficiency, which ignores loss mechanisms in light extraction, is related to the lifetime of non-radiative and radiative processes by the equation τr ⎞ ⎛ Q.E. = ⎜1 + ⎟ τnr ⎠ ⎝
−1
(8)
where Q.E. is the internal quantum efficiency, τr is the radiative recombination lifetime, and τnr is the non-radiative lifetime. Conventional p-n Junctions (Homojunctions) Highly purified semiconductors have very low conductivity at room temperature. Their properties are highly sensitive to certain types of impurities that are deliberately added as dopants during fabrication. The dopant atoms replace silicon atoms in the lattice structure. For a material such as silicon with a valence of 4, dopant atoms with a valence of five (As or P) will act as donors, giving up their extra electrons in the silicon lattice to the conduction band. Materials with a valence of three act as acceptors, taking an electron from the (filled) valence band, producing a hole. The dopant impurity atoms have energy levels near the edge of the bandgap, and are readily ionized (except at very low temperature). The effect of the impurity atoms is to shift the equilibrium potential within the semiconductor towards the conduction band (donors) or the valence band (acceptors). A p-n junction is formed by placing p- and n-type semiconductors in close proximity (in practice, this is usually done by deliberately adding impurities to modify the doping). The difference in equilibrium potential between the two doped regions creates a retarding potential, as shown in the diagram of Fig. 9. The potential changes near the boundary between the n- and p-regions, creating narrow regions of “uncovered” charge from the dopant atoms. Donor atoms are left with a positive charge, having given up one electron, while acceptor atoms have a negative charge. This boundary extends over regions of approximately 1 to 100 µm, depending on the doping density (it is larger for lightly doped material). The width of the boundary also depends on external bias conditions. Although the example shown is for silicon, the same principle applies to GaAs or other III-V materials. The important point is that unless the doping level is extremely high, the transition region will extend over distances of several microns or more. Such dimensions correspond to more than an optical wavelength for the wavelengths that we are considering in this section of the short course. This establishes the lower dimensional limit for semiconductors fabricated with homojunctions.
III-10
Metallurgical junction
Retarding potential created by “exposed” arsenic and boron atoms
B B B B B
As As As B
Ionized boron and arsenic atoms near boundary
p-type semiconductor
n-type semiconductor
Fig. 9. Diagram showing the retarding potential created when two regions of the same semiconductor type (silicon in this example) with different doping levels are placed in close proximity.
Heterostructures It is also possible to create a p-n junction by placing two different types of semiconductors together, forming a heterojunction. Note the distinction from homojunctions where the doping levels differ but the same basic type of semiconductor is present on each side of the junction. For a heterojunction the barrier for holes and electrons is affected by the difference in bandgap between the two semiconductor types as well as the concentration of dopant atoms. The bandgap discontinuity adds to the potential barrier in the junction, increasing carrier injection efficiency [Pani76]. In order to form a heterostructure that is useful for optoelectronic applications, the lattice spacing of the two semiconductor materials must be closely matched in order to maintain the structure of the lattice. Otherwise defects in the crystalline structure will occur that reduce the mobility and affect the reliability of the device. Figure 10 shows an example of the potential discontinuity formed by a GaAs/AlGaAs heterojunction with no external potential applied [Chua95]. The bandgaps of the two materials differ by 0.40 eV. Note that the band offset is greater in the conduction band compared to the valence band. The band offset ratio is typically about 2:1, and is an important heterostructure property. For the example in Fig. 10 the lateral dimension required by the heterostructure for equilibrium is about 0.1 µm, about a factor of 30 less than the lateral dimension of a GaAs homojunction with comparable doping levels. This example shows a p-N heterostructure (the upper case symbol is used to designate the material with higher gandgap). It is analogous to a conventional p-n junction, except for the increased potential difference that is introduced by the bandgap discontinuity. Note that the energy difference is about twice as large in the valence band compared to the conduction band for this heterostructure. If an external potential is applied it will function in the same way as a conventional p-n junction, but carrier injection will be aided by the bandgap discontinuity, increasing injection efficiency. It is also possible to form p-P or n-N heterostructures (isoheterostructures) because of the energy difference between the two materials. Such heterostructures are frequently used in the III-11
multiple layers used to form modern optoelectronic components. The bandgap discontinuity can be sufficient to confine carriers even though no junction exists. N-AlGaAs
p-GaAs p - GaAs (Na = 1 x 1018 cm-3) N - Al0.3Ga0.7As (ND = 2 x 1017cm-3) Eg for GaAs = 1.42 eV Eg for Al0.3Ga0.7As = 1.82 eV
Electron Energy (eV)
1.8 1.2
Ecp
0.6
∆Ec = 250.6 meV
56.4 meV ECN
Fp 0 Evp
∆Ev = 123.4 meV
-0.6
FN 32.2 meV
-1.2 Xp -1.8 -0.06
-0.03
0.00
EVN
XN 0.03
0.06
0.09
0.12
Distance from Interface (µm) Figure 10. Electron energy vs. distance from the interface for a step junction formed by GaAs and AlGaAs. The bandgap of the two materials differs by 0.4eV.
Ternary and Quaternary Materials for Optical Heterostructures
Requirements Several III-V elements can be combined in various proportions to form solid solutions, either as ternary or quaternary materials. This is called bandgap engineering because it allows the bandgap to be adjusted over a wide range by suitable selection of materials. This allows devices to be tailored over a wide range of wavelengths. The important requirements for the use of III-V solid solutions in heterostructures are as follows: • The lattice spacing must be closely matched to that of the underlying host material in order to maintain orderly crystal structure and avoid introducing defects. Typically this requires a lattice spacing that is matched within approximately 0.2%. • The solid solution must also have a direct bandgap. • The index of refraction of the solid solution is an important property in optoelectronic devices because of the need to confine photons as well as charge carriers in the overall device structure. For these systems it is important to distinguish between the substrate material and the material combinations used for other layers. This can be very confusing, particularly for quaternary materials. We will use parentheses and italics for the substrate material in the discussions that follow. Ternary Systems Using AlGaAs (GaAs) The most widely studied material system is aluminum gallium arsenide, used for the pioneering work in developing LEDs and laser diodes in the 1970s [Raze00]. Unlike many other material combinations, AlGaAs remains closely lattice matched to GaAs over its entire range of composition, a major advantage. Fig. 11 shows how the bandgap of solid solutions of AlGaAs III-12
change as the mole fraction, X, of aluminum is increased. For X = 0.45, the material changes from direct to indirect bandgap, limiting the usefulness for most optoelectronic devices to mole fractions below 0.45. Nevertheless, this material system allows the bandgap to be varied over a range than provides wavelengths from approximately 630 to 870 nm. 3.2
Lattice mismatch < 0.1% Index of refraction increases as X decreases Wavelength range 630 - 870 nm
3.0
2.6 500 2.4 2.2
X is mole fraction of AIAs e.g., AIxGa1-xAs
2.0 1.8
600
Indirect bandgap for X > 0.45
1.6 1.4
700
Wavelength (nm)
Energy Gap, Eg (eV)
2.8
400
800 900 1000
1.2 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Mole Fraction of AIAs, (X) Fig. 11. Variation of the energy gap of solid solutions of AlGaAs as the mole fraction of aluminum is increased from 0 (pure GaAs) to 0.45, where the material no longer has a direct bandgap. This corresponds to a wavelength range of 630 to 870 nm.
Quaternary Systems Using InGaAsP (InP) The fortuitous close lattice matching of AlGaAs is one of the reasons that initial work on heterostructures for optoelectronics concentrated on that material system. Fig. 12 shows lattice matching for AlGaAs, as well as more complex quaternary systems using various combinations of InAs, InP, and GaAs. Note the very poor lattice matching for the GaAs/InAs system. A quaternary system based on InGaAsP provides lattice matching over an extended range of energy gaps from about 0.8 to 1.45 eV, as shown in the figure. That material system has been used to develop LEDs and lasers with wavelengths between 1100 and 1600 nm. Lattice matching considerations restrict the fractional makeup of the constituents. Quaternary compounds have many degrees of freedom. For InGaAsP, the mole fractions of In and Ga are usually interrelated [by (1-X) and X, respectively], with a similar relation between the mole fractions of P and As [(1-Y) and Y] because those combinations maintain close lattice matching.
III-13
2.5
AlAs 2.0
Energy Gap (eV)
AlxGa1-xAs 630 - 850 nm
1.5
InP
GaAs
In1-xGaxAsyP1-y
1.0
1100 - 1600 nm
0.5
InAs 0
5.5
5.6
5.7
5.8
5.9
6.0
6.1
Lattice Constant (Angstroms) Fig. 12. Lattice constant for various material combinations used in optoelectronic devices.
Strained Lattices Although we have stressed the importance of close lattice matching for heterojunctions, it is possible to grow regions with lattice mismatch up to 2% with low defect density provided that the thickness of the transition region between the two materials is less than about 100 angstroms. One advantage of strained lattices is that it allows new material combinations to be used, such as the use of strained InGaAs for lasers and LEDs with wavelengths between 900 and 1100 nm, where there are no suitable materials with close lattice matching. Another advantage of strained layer technology is that it changes the hole mobility. This can be used to improve the performance of optical emitters, deliberately introducing strain into the lattice. If done properly, this can be used to reduce the threshold current density of laser diodes [Cole00]. Many advanced lasers take advantage of this property. Despite the lattice mismatch, it is possible to fabricated highly reliable devices with strained layers as long as the thickness of the strained region is sufficiently low [Selm01]. II. Optical Emitters A. Light-Emitting Diodes For direct bandgap semiconductors, the dominant recombination mechanism under moderate forward bias involves band-to-band emission of a photon. Light-emitting diodes use this property to generate light, with an intensity that is approximately proportional to the forward current through the p-n junction. There is a very important difference between conventional diodes and LEDs. For conventional diodes the purpose is to produce a p-n junction with low internal loss mechanisms and efficient rectification. For an LED, the purpose is to maximize internal losses that result in efficient photon emission, taking advantage of the high radiative loss efficiency. One way to accomplish this is to use a material with very low doping in the active III-14
region. LEDs are analogous to p-i-n diodes, using direct bandgap materials with high probability of band-to-band recombination in the i-region [Pani76]. Fig. 13 shows the light output of an LED along with the forward current through the LED as a function of forward voltage. At low current, non-radiative recombination processes dominate, with negligible light output. When the forward voltage reaches about 1.05 V the injected current is sufficient to overcome non-radiative recombination losses, and there is an increase in the slope of the I-V characteristic (the slope approximately doubles) that is accompanied by an increase in optical power output. Above that value the optical power is approximately proportional to forward current, eventually flattening out because of internal resistance. The region between the change in slope and the saturation point defines the useful range of operating currents for an LED.
Forward Current or Diode Current (A)
10-1 10-2
Slope changes when light output threshold is reached
10-3
LED Current
10-4 Detector Current
10-5 10-6 10-7 10-8 10-9
Detector leakage current
10-10 0.4
0.6
0.8
1
1.2
1.4
1.6
Forward Voltage (V) Fig. 13. Dependence of optical power and forward current on forward voltage for an LED showing the increase in slope when the forward current is sufficient to overcome non-radiative recombination losses.
Although the conversion process from current to light is very efficient, extraction of light is limited by Snell’s law because of the high index of refraction of typical III-V materials, nominally 3.5. As a result, the Brewster angle is about 17 degrees (assuming an air interface), and any photons striking the interface at angles greater than 17 º will be reflected back, as shown in the simple diagram of Fig. 14. The photons that are formed by spontaneous recombination have arbitrary directions. As a result, the extracted optical power is about 2% of the electrical power through the LED unless special designs are used. Methods of improving light extraction include the use of an integral lens, as well as special cavity designs that reduce the number of photon striking the surface at large angles [Vand97].
III-15
Fig. 14. Photons from an LED striking the interface between the III-V material and an interface with n =1. Spontaneous recombination produces photons with arbitrary direction, most of which strike the interface at oblique angles where they are reflected back into the bulk region and eventually recombine.
Homojunction LEDs An older process that is still used for some types of LEDs relies on doping GaAs with an amphoteric impurity; i.e. an impurity than can function either as an acceptor or donor, depending on conditions within the semiconductor. Early LEDs used zinc as an amphoteric impurity, but since the mid-1970’s silicon (as an amphoteric dopant) has been preferred because it provides higher optical efficiency compared to Zn. The growth process relies on a special property of silicon in GaAs (or AlGaAs). When a GaAs crystal is grown in a sample that is heavily doped with silicon, the silicon is a donor at high temperatures. However, it becomes an acceptor if the growth is done at lower temperature [Rupp66]. Thus, if we gradually reduce the temperature during the growth process the initial part of the GaAs will be p-type, while GaAs formed in later stages at lower temperatures will be n-type. The transition temperature is approximately 850 ºC. With this technique, a junction can be formed without introducing special dopants simply by changing the temperature during the growth process. Fig. 15 shows a diagram of an amphoterically doped AlGaAs LED. Note the relatively broad transition region between the n- and p-regions. One reason that this type of LED has high efficiency is that the silicon doping level is so high that the material is compensated. This causes the energy gap to “soften” (band-tailing), reducing non-radiative absorption within the structure. The energy of the photons is slightly below the bandgap energy because of a complex associated with acceptor states. This eliminates band-to-band absorption losses due to shallow impurity states.
III-16
Temperature change during growth creates junction
Anode
p-n junction p -GaAs (Si doped)
Light emission n (regrowth) Amphoteric dopants can be n- or p- type impurities (depends on temperature)
n - GaAs (Si doped)
Cathode
Fig. 15 Diagram of a homojunction LED produced by altering the growth temperature with an amphoteric impurity.
The transition region from p- to n-type material occurs over a relatively wide region because of the somewhat cumbersome mechanical configuration used to grow the crystal, along with the requirement to make this process slow enough to allow the growth temperature to gradually change by about 40 º C. As a consequence, this type of LED requires a high lifetime – several microseconds or more –to function because if the lifetime is too short, the diffusion length will be reduced, and non-radiative recombination will occur before the carriers reach the other side of the extended width of the p-n junction. Heterostructure LEDs Heterostructure LEDs have a far more complicated structure. LEDs with single heterojunctions can be formed by growing AlGaAs on GaAs, using an additional surface diffusion to drive the active junction below the surface. This results in an LED with a much shallower structure and faster response time compared to the amphoterically LED described earlier. That structure has been used for surface-emitting LEDs, matching the circular pattern of the diffusion to the diameter of an optical fiber. More advanced LEDs use double heterojunctions, with light emission from the edge. Figure 16 shows a structure that is typical of the LED technologies used for general purpose and telecommunication applications. The structure has several different layers with different purposes. The active layer is undoped GaAs, with a typical thickness of 0.2 to 0.3 µm. Light produced within that region is confined by the two AlGaAs guiding regions that have lower refractive index than GaAs. The p- and n-doped AlGaAs layers form heterojunctions for carrier injection.
III-17
n-electrode
n-GaAs
AlGaAs guiding layer
Active layer (GaAs)
n-AlGaAs Light emission
p-AlGaAs guiding layer p-electrode
p-GaAs
Fig. 16. Diagram of a double-heterojunction edge-emitting LED.
This type of LED can be fabricated with many different material systems, with wavelengths from 650 nm (using AlGaAs) to 1550 nm (InGaAsP). Longer wavelength LEDs intended for fiber optic applications are often designed for very fast response times, requiring thin active layers. Many recent advances have been made in LED technology, including special designs with transparent substrate to eliminate absorption within the substrate [Vand97], and the use of special microcavity LED designs to improve overall coupling efficiency [Choi04]. Those techniques are particularly important for LEDs in the visible region, increasing light extraction efficiency. In the near infrared, there is more interest in using laser diodes because of the ease of efficient light extraction, as discussed in the next subsection, as well as the requirement for narrow exit beam patterns for fiber optic applications. B. Laser Diodes Laser diodes are similar in concept to light-emitting diodes, but rely on the principle of stimulated emission to provide far higher light generating efficiency than is possible with LEDs. Stimulated emission was shown earlier at the right of Fig. 6. In order for stimulated emission to occur, the carrier density within the semiconductor must be high enough so that photons traveling in the preferred direction have a high probability of creating additional photons within the guided region by spontaneous emission. In an edge-emitting laser the preferred direction is established by cleaving two surfaces to form a Fabry-Perot resonant optical cavity, coating one facet with metal. In order for the device to function as a laser, carrier injection must be high enough to provide an overall optical gain that is greater than one. Before discussing the details of laser operation, it is useful to examine the history and evolution of semiconductor lasers. The first lasers, produced in 1962, were homojunction lasers with threshold current densities above 105 A/cm2 [Pani76]. They could only operate at low temperature (77 K), and were only capable of operating for a few hours. Since that time a great deal of effort has been spent in improving laser efficiency and developing new material technologies to expand the range of available wavelengths. Fig. 17 shows how various changes and breakthroughs in device technology have reduced the threshold current [Lede00]. The first major change was the introduction of heterostructures, which reduced the threshold current by several orders of magnitude, expanded the range of wavelengths and materials, and allowed reliable operation at room temperature. Quantum-well lasers, made possible by advanced III-18
Threshold Current Density (A/cm2)
fabrication techniques that allowed reproducible fabrication of very thin layers of III-V materials, allowed further reductions in threshold current. During the last 8 years, quantum dot lasers have been developed, reducing the threshold current density to about 20 A/cm2. 100,000
Room Temperature
10,000 Quantum Well
Quantum Dot
1,000 Double Heterojunction 100
10
After Ledenstov, et al., J Sel Top Q. Elect, 2000
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 Year Fig. 17. Trends in threshold current since the first semiconductor lasers were developed in the early 1960s.
Gain is established by the properties of the cavity and the probability of stimulated emission. Figure 18 shows a simplied diagram of a laser diode. Although it appears to be very similar to the previous figure of a heterojunction LED, there are important differences. For a laser, the front and back facets must be parallel in order for photons that are reflected from either facet to have the same direction as other photons that are produced within the active region. Although nearly all of the light within the active region is parallel to the main axis of the laser, some of the light will “spill over” beyond the thin active region, but is confined between the two guiding regions. Light emitted from the active region is diffracted, causing the beam to diverge once it leaves the cavity. Key properties of the laser include the reflectivity from the two parallel facets, non-radiative loss within the cavity (including absorption of some of the photons), and the gain of the semiconductor material.
III-19
Current path
Guiding regions
Active region
Reflective back facet
Partially reflective front facet
Key properties: loss in cavity gain properties of the active region reflectance of the two parallel facets
Fig. 18. Basic diagram of a semiconductor laser diode.
The gain depends on current density, as shown in Fig. 19. Three different materials are shown, bulk GaAs; a quantum-well structure in the AlGaAs/GaAs system, and a strained quantum well system using InGaAs. The current density is about three times lower for the InGaAs system compared to bulk GaAs, resulting in lower threshold current. Current densities on the order of 102 to 103 A/cm2 are required for laser operation at room temperature. This corresponds to carrier densities ~ 1018 cm-3. 8 7
Jv /g (mA/µm 2)
6
Bulk GaAs
5 4 3
GaAs/AlGaAs 80A QW
2 1
InGaAs/GaAs 80A QW
0 0
1000
2000
Material Gain
3000
4000
(cm−1)
Fig. 19. Material gain vs. current density for three different systems used for semiconductor lasers.
III-20
5000
For laser operation we have to consider the balance between various recombination and generation mechanisms. From Fig. 19, gain depends on the cavity length. In order to achieve positive gain, the effect of internal gain from spontaneous emission must balance loss mechanisms from optical absorption, recombination, and reflection. Equating these terms is equivalent to the condition
ℜ e [(G th − α )L] = 1
(9)
where ℜ is the reflectivity at the back reflecting surface, Gth is the threshold gain, α is the absorption coefficient (including recombination terms, which for simplicity are lumped into α), and L is the cavity length. We can consider these terms more explicitly as follows:
η I th = (R non − rad + R spont + R stim ) qV
(10)
The recombination terms can be related to carrier density, N. Non-radiative recombination from bulk recombination centers is proportional to N, while Auger recombination is proportional to N3. The spontaneous recombination term is proportional to N2. When the carrier density is increased to the point that the spontaneous recombination term overcomes losses due to recombination and optical absorption, then the probability of stimulated emission increases. This defines the threshold carrier density for the laser. As shown in Fig.19, this typically requires a carrier density of 2-4 x 1018 cm-3. A cross section of an edge-emitting laser with five layers is shown in Fig. 20. Current flows vertically from the top contact to the substrate. The current is confined horizontally to a narrow stripe by a combination of oxide isolation under the top metal contact and the use of an nAlGaAs region at the edge of the two p-regions. That AlGaAs layer has a different bandgap compared to the AlGaAs region immediately below the active GaAs region. This type of construction is referred to as etched mesa, because of the way that the active region is formed. Other fabrication methods include ridge lasers, where the active region is formed within a physical ridge structure on the top surface, and buried crescent lasers, which contain a buried crescent-shaped active region. As discussed earlier, the typical thickness of the active region is 0.2 to 0.3 µm in these types of broad-area lasers.
III-21
Metal
GaAs (p)
Oxide
(AlGa)As (n) (AlGa)As (p)
(AlGa)As (n)
(AlGa)As (n) GaAs (n)
Active Region Fig. 20. Diagram of a GaAs laser diode formed with five separate layers.
A more advanced edge-emitting laser is shown in Fig. 21, using AlGaAsP. This laser contains four quantum wells that provide more efficient carrier confinement compared to lasers with larger active layers. Strained-layer technology is used to reduce the threshold current. The quantum wells have dimensions of about 10 nm. Carriers are confined within the quantum well region, and this is the region where photons are produced by stimulated emission. However, the quantum regions are too thin for optical confinement; the evanescent wave will extend beyond them. Cladding layers are placed above and below the active layer for photon confinement. The n-GaAs layers at the top confine current to the central part of the structure. Additional layers are used to increase efficiency and provide transition regions between materials with different lattice constants (buffer layers). This illustrates the extreme complexity of state-of-theart semiconductor lasers. Modern fabrication techniques allow the sequence of layers with different composition to be grown with high accuracy. It is relatively easy to get high power extraction from lasers because the photon direction is essentially normal to the partially reflecting surface. However, diffraction will cause the beam to diverge after it leaves the laser cavity. Lenses can be incorporated into the laser package to reduce beam dispersion from diffraction. Antireflection coatings are nearly always used to increase light extraction efficiency. Key properties of laser diodes are (a) threshold current, and (b) the linearity of the light output for current above the threshold current (slope efficiency). The threshold current and light output of a laser depend on temperature. There are two ways to deal with this: place the laser in a precisely controlled thermal environment; or monitor the light output from the laser, using feedback to maintain a constant light output when the temperature changes. Many lasers use internal photodiodes – usually discrete photodiodes that are placed within the package- for monitoring purposes, measuring a small amount of light from the back facet. The photodiode signal is then used as part of a feedback circuit to stabilize light output from the laser. However, in some cases degradation of the internal monitor diodes from radiation can be more severe than degradation of the laser.
III-22
p - GaAs p - GaInP p - (AlGa)InP cladding layer n - GaAs current - blocking layer electrode p - GaInP
Strained Quantum Well Active Layer
p - (AlGa)InP cladding layer
undoped strained QW active layer
10 nm
n - (AlGa)InP cladding layer
5 nm
n - GaAs 0
0.6
x
n - GaAs substrate electrode
AI content in (AlxGa1-x) yIn1-yP Fig. 21. Diagram of an advanced quantum-well laser using InGaAsP.
Vertical Cavity Surface-Emitting Lasers Although edge-emitting lasers are very efficient, the geometry is not ideal for coupling to optical fibers, particularly single-mode fibers that have dimensions below 10 µm. Vertical cavity surface-emitting lasers (VCSELs) are lasers with a vertical rather than horizontal structure, using a distributed Bragg reflector at the lower region of the laser for reflection. They were first developed in 1977 using GaInAsP (InP) at 1300 nm [Iga00]. Further details on VCSEL technology are provided in the review paper by Choquette, et al. [Choq97].
A physical diagram of a VCSEL structure is shown in Fig. 22 The optical cavity is vertical, using Bragg reflectors at the top and bottom of the cavity with as many as 70 layers. The layers are designed to reflect a specific wavelength. The top layer is partially reflecting to allow a fraction of the light within the cavity to escape from the surface. The active layer is located at the center. It is very thin, on the order of 0.1 to 0.5 µm. Light produced in that thin region is reflected between the Bragg reflectors at the top and bottom of the optical cavity. Selective oxidation is used to limit current flow to a circular region. Although the diagram shows an overall rectangular geometry, some VCSELs are fabricated with cylindrical geometry. The narrow active region in a VCSEL is buried under many different layers. This limits the ability to extract heat from the laser. Self-heating affects the characteristics of VCSELs, reducing efficiency at higher forward current conditions. Thus, VCSELs are not the best choice for applications involving high power output or linear operation over an extended range of operating currents. However, they can be fabricated with small geometries and very low threshold current, and are particularly useful for fiber optic applications. For example, VCSELs have been produced with threshold current below 1 mA [Iga00]. For VCSELs with small volumes (a necessary condition for very low threshold current), reflection from the cavity causes highly nonlinear behavior with peaks and valleys in the optical power output. This is shown in Fig. 23for 1520 nm VCSELs with two different apertures, 16 and 36 square microns [Sun03]. The oscillatory nature of the L-I characteristics is less evident in
III-23
VCSELs with larger cavities, but all VCSELs exhibit this sort of “structure” in the output characteristics because of the presence of so many layers in the Bragg reflector. Laser output
Selective oxidation limits active region to small cylinder
Mirrors (distributed Bragg reflectors)
p-n junction cavity
Fig. 22 Diagram of a vertical cavity surface-emitting laser. A circular geometry can be used, providing a small emission area that is easily coupled to small-diameter optical fibers.
100 A = 36 µm2
P (µW)
80
60
40
20
A = 16 µm2
0 0
2
4
6
8
10
12
I (mA) Fig. 23. Optical power vs. forward current for very small VCSELs with threshold current below 1 mA. The nonlinearities in the optical power are caused by the properties of the Bragg reflector [Sun03].
Quantum-Dot Lasers Quantum dots are fabricated with dimensions that are small enough to cause quantummechanical confinement of carriers in both the x- and y-directions, compared to quantum III-24
confinement in a single direction that is done for typical edge-emitting lasers. They have much higher carrier density near the band edges, increasing the optical gain in the material, and reducing the effect of temperature on operating characteristics [Lede00]. Although quantum dot lasers are still in their infancy, the predicted threshold current density is ~ 5 A/cm2. Arrays of quantum dots can be placed in a laminar region, resulting in higher efficiency and lower threshold current [Park00] compared to an equivalent continuous region with a conventional laser. Glass-Based Lasers Conventional glass-based lasers use an amorphous host material doped with an impurity. Because of the amorphous structure, each atom can be considered independently compared to a semiconductor laser that is actually a complex system with many different quantum states. The independent property of glass-based lasers results in much narrower spectral width. Glass-based lasers are pumped with an external light source, typically a flashlamp or an array of semiconductor lasers. They are typically used in special applications requiring high power or narrow wavelengths. The typical cavity length is much greater than the cavity length of a semiconductor laser. We will not discuss radiation damage in these types of lasers. However, the results of one study show the effects of radiation-induced absorption in the host material is an important degradation mechanism [Rose95]. III. Radiation Environments and Radiation Damage
This section of the course provides a very brief review of radiation environments and the interactions of the particles in space with semiconductors. More thorough treatments are given in Section II of the 1997 NSREC Short Course by J. L. Barth [Bart97], and Section II of the 2002 NSREC Short Course by J. E. Mazur [Mazu02], as well as the recent review article by Barth, Dyer and Stassinopoulos [Bart03]. A. Space Environments The natural space environment consists of particles in trapped radiation belts, as well as highenergy protons and heavy ions from galactic cosmic rays and solar flares. The interaction of heavy charged particles is usually described by means of linear energy transfer (LET), which is a measure of how much energy is lost when an energetic beam travels through a thin slab of material. The units of LET are MeV-cm2/mg. An equivalent (and more intuitive) way to describe this is in the charge deposited in a material within a unit path length, (LET*). The units of LET* are pC/µm. Although more intuitive for semiconductor device analysis, LET* depends on the density of the target material, while LET does not. For silicon, LET* ~ 0.01 LET; i.e. an LET of 10 MeV-cm2/mg is equivalent to depositing 0.1 pC/µm. For GaAs, LET* is ~ 0.018 LET, and a particle with LET = 10 MeV-cm2/mg deposits 0.18 pC/µm. The distribution of galactic cosmic rays is shown in Fig. 24. Particles emitted by the sun during periods of high solar activity lower the galactic cosmic ray flux by about a factor of 4 compared to the value shown in Fig. 24, which represent the flux during solar minimum conditions (or at extreme distances from the sun). This is referred to as solar modulation. Once the energetic particles reach the earth, the earth’s magnetic field deflects many of the energetic particles, modifying the GCR flux for orbits around the earth. For orbits with high inclination, the GCR flux is relatively unaffected by geomagnetic shielding near the poles, and during the period when the spacecraft is near the poles it will be exposed to nearly the same GCR flux as in free space. For a 705 km high-inclination orbit, the net effect is to reduce the average flux by a factor of about 5 compared to free space. For lower inclination orbits, only particles III-25
with lower atomic number can get through the geomagnetic shielding, as shown in the lower curve in the figure. Note the very strong decrease in particle flux for LETs above approximately 30 MeVcm2/mg. This is called the “iron threshold” because it corresponds to the LET of energetic iron nuclei in the GCR environment. There are relatively few particles with LETs above that value, which is often used as a benchmark for characterizing components. Note however that a particle that traverses a material at an angle will deposit more charge in a given thickness simply because the path length increases by 1/cos(θ), where θ is the angle of incidence. Thus, the effective LET for particles with LET = 30 MeV-cm2/mg at normal incidence is 60-75 MeV- cm2/mg, depending on assumptions about how charge deposited laterally is collected by the active node. 104 103
Avg. Daily Flux (cm -2 )
102 101 Deep Space
1 10−1
Earth Orbit, 600 km, 28 deg
10−2
Earth Orbit, 600 km, 98 deg
10−3 10−4 10−5 10−6 0.1
1
10
100
LET (MeV-cm2/mg) Fig. 24. Distribution of galactic cosmic rays vs. LET for deep space as well as for two different earth orbits during solar minimum. Particles emitted from the sun during solar maximum reduce the average daily flux by about a factor of four near the earth (solar modulation is less significant near the outer planets).
Earth Radiation Belts The earth’s trapped radiation belts consist of an inner proton radiation belt as well as inner and outer radiation belts of energetic electrons (we will ignore the inner electron belt in this abbreviated treatment). The proton belts begin at an altitude of about 700 km, extending to approximately 20,000 km (the altitude also depends on longitude). Proton energies in the trapped belts extend to several hundred MeV. See Barth, et al. for more details. There is a distortion in the belt structure near South America due to slight misalignment between the earth’s axis of rotation and the central axis of the magnetic field. This causes the proton belt to “dip”. Low-inclination spacecraft will be exposed to that distorted region of the proton belt when their orbits pass through the South Atlantic anomaly. The outer electron belt starts at about 15,000 km, extending to much higher altitudes. The distance of the outer electron belt to the earth’s surface depends on latitude. Electron energies extend to about 20 MeV, with a peak in the energy distribution of approximately 3 MeV. III-26
Radiation Belts Near Jupiter Jupiter has a strong magnetic fields, which can trap highly energetic particles. Information on the Jovian radiation belts was obtained from interplanetary missions, but is obviously less complete than information on trapped belts near the earth. The trapped belts near Jupiter contain electrons, protons, and other charged particles [Devi83]. Proton intensities in the Jovian belts are several thousand times greater than the intensities in the earth’s radiation belts, but have lower energies compared to trapped protons near the earth. On the other hand, the electron energies in the Jovian trapped belts extend to several hundred MeV, much higher than electron energies in the earth’s radiation belts. In order to deal with these intense radiation fields, spacecraft that operate near Jupiter have to be designed to meet higher radiation levels. Their orbits are carefully chosen to avoid the peak regions of the radiation belts. Representative radiation levels for typical spacecraft are shown in Table 3, assuming that a 100-mil aluminum shield surrounds the spacecraft electronics. The amount of shielding is very important, particularly for shielding thicknesses that are less than 100 mils (equivalent) of aluminum. Even thin shields remove most of the low energy particles from the environment. Nominal radiation requirements are usually established for a spherical shield with a thickness that approximates the thickness of enclosures around electronics. In most applications far more shielding is actually present on the spacecraft. Most systems take advantage of the inherent shielding provided in the structure and electronic boxes to further reduce the total dose. The values in Table 3 are only approximate, and usually assume that one or more severe solar flares will be encountered during the mission. Table 3 Total Dose Requirements for Representative Space Missions Orbit
Operating Time (years)
Total Dose rad (SiO2)
Space Station
500 km 54 degree
10
5 x 103
Highinclination earth orbiter
705 km 98 degree
5
2 x 104
Geostationary
36,000 km
5
5 x 104
Mars Surface Exploration
NA
3
5 x 103
Mission near Jupiter
NA
9
1.5 x 105 – 2 x 106
Description
Requirements for single-event effects are much more difficult to define because of distribution of ion types and energies in space, as well as changes in solar activity. Proton single-event upset is also influenced by the South Atlantic anomaly in the earth’s trapped radiation belts (see the 1997 and 2002 Short Course references for more details). The error rate for memory cells or registers is often used as a benchmark for single-event upset, although there are other effects (including latchup) that can be even more important. A more complete discussion of single-event upset effects can be found in the review articles by Dodd and Massengill [Dodd03] and Sexton [Sext03]. Commercial silicon-based devices have III-27
error rates on the order of 10-6 to 10-8 errors per bit day in a deep space or geostationary environment. It is often possible to use error-detection-and-correction or other system-level approaches to deal with these types of upset effects. The alternative is to use special hardened circuits. As discussed in Section VII, logic circuits that use compound semiconductors are somewhat more sensitive to single-event upset than comparable silicon technologies [Weat04]. However, compound semiconductor structures are not sensitive to single-event latchup. Basic Effects in Semiconductors from Protons and Electrons Ionization effects from heavy particles were discussed earlier in this section. Because there are so few particles, permanent damage effects from galactic cosmic rays can usually be ignored. The brief discussion below compares ionization and displacement damage effects from highenergy electrons and protons. Space environments can generally be divided into (a) permanent damage effects from highenergy electrons and protons that produce uniform damage within a semiconductor device, and (b) highly localized ionization or displacement effects from the single interaction of a cosmic ray or proton. Although total dose is the dominant damage mechanism for many types of semiconductors, displacement damage is usually more important than ionization damage for optoelectronic devices, and nearly always has to be considered when evaluating optoelectronics for space applications. B. Special Environments: Nuclear Reactors and Particle Accelerators For nuclear reactors, the primary concern is displacement damage from neutrons and decay products from activated material, and ionization damage. Very high radiation levels – in the multi-Megarad region – may be required, particular for worst-case operational scenarios. Many compound semiconductor devices are highly resistant to ionization and displacement damage, making them good candidates for use at nuclear reactor facilities. We will not discuss these environments, but it is important to realize that there are important applications of photonic components that have much higher equivalent radiation levels than typical space environments. High radiation requirements may also occur for electronics associated with particle accelerators. For example, the Large Hadron Collider at CERN has very high radiation requirements, involving particles with far higher energies than those encountered in space. C. Fundamental Interactions Five types of particles are usually considered in radiation environments: gamma rays, electrons, protons, neutrons, and energetic nuclei from space or from nuclear fission. Gamma rays and electrons are often lumped together because the primary way in which they lose energy is through ionization. Ionization produces electron-hole pairs within semiconductors and insulators. This can cause charge to be trapped at interfaces between semiconductors and insulators. Although that mechanism is extremely important for silicon devices, it is much less important for compound semiconductors because insulating materials do not exist in compound devices with the high quality and low interface state density that is present in silicon dioxide. Consequently, most compound semiconductors are relatively immune to ionization radiation damage at levels below 1 Mrad(GaAs) [Srou88]. Protons also produce ionization, and proton ionization is a significant part of the total dose for most space missions. In addition, protons interact with lattice atoms, where they can impart sufficient energy to move atoms from their normal lattice position to a resting point that can be quite distant from the original lattice site. These displacement effects introduce a great deal of III-28
damage in the lattice, and are a major concern for most optoelectronic components. Neutrons produce similar displacement effects. Displacement damage has three important effects: Minority carrier lifetime is reduced; carrier mobility is reduced; and the effective doping level can be altered because of carrier removal. High-energy nuclear particles also produce displacement damage and ionization. However, for those particles the most important consideration is usually single-event upset, not permanent damage effects. Single-event upset is a circuit effect that occurs because the interaction of a single charged particle produces a small but significant amount of charge that can be collected at sensitive circuit nodes, causing stored information in a memory or flip-flop to be altered. Integrated circuits manufactured with compound semiconductors can be very sensitive to singleevent upset effects. D. Energy Dependence of Displacement Damage The energy dependence of displacement damage effects is very important because the space environment consists of a spectrum of electrons and protons with a wide range of energies. We need to understand the dependence of fundamental interactions on energy in order to relate experimental results at one or more energies to the actual space environment. Early work in the 1980’s developed the concept of non-ionizing energy loss (NIEL) for this purpose. The initial work started with theoretical calculations that partitioned energy loss into ionizing and nonionizing components [Summ87]. For silicon, experimental work was in reasonable agreement with theory, as summarized in Fig. 25 [after Summ93]. Electron damage increases with energy, with a minimum energy of approximately 150 keV to transfer enough energy to displace a silicon atom. Protons have a much stronger energy dependence, which increases at low energy because the interaction is electromagnetic. Slow protons spend more time in the vicinity of lattice atoms when they travel at lower velocity, increasing the amount of energy that is transferred to the nucleus. A recent review of displacement damage effects in silicon can be found in the paper by Srour, C. Marshall and J. Marshall [Srou03]. 1000
NIEL (keV-cm2/g)
100
10
1-MeV neutrons
protons
1
0.1
electrons
0.01 After Summers, et al., Trans. Nucl. Sci., 40, 1372 (1993)
0.001 0.1
1
10
100
1000
Particle Energy (MeV) Fig.25. Dependence of non-ionizing energy loss in silicon for electrons, protons and 1-MeV neutrons on the energy of the incident particle.
III-29
The energy dependence of proton damage in GaAs is shown in Fig. 26, taken from the work of Barry, et al. [Barr95]. Unlike silicon, for GaAs there is considerable disagreement between the theoretical calculations and experimental results, particularly for protons with energy above 50 MeV. Differences as high as a factor of 6 have been reported between theoretical calculations of NIEL (with secondary corrections to improve accuracy) and experimental results for lightemitting diodes [Barr95], as shown by the open circles in Fig. 26. The underlying reasons for these differences are still not understood, and are the subject of current research [Reed00, Walt01]. 100
0.1 Experimental Data
Burke et al., 1987 Summers et al., 1988 Unrestricted Energy Loss Restricted Energy Loss
0.01
NIEL (MeV
cm 2/g)
Non-ionizing Energy Disposition 10
Normalized Damage Factor
Burke et al., 1987 Summers et al., 1988 Barry, et al., 1995
1
0.001
1
10
100
1000
Proton Energy (MeV) Fig. 26. Calculations of the energy dependence of NIEL for protons in GaAs, showing discrepancy between theory and measurements of energy dependence in GaAs LEDs for energies above approximately 30 MeV [Barr95].
The uncertainty in energy dependence has important consequences, particularly if radiation tests are done at high energies, for example 200 MeV protons, which are readily available at some facilities. The difficulty is that the actual proton environment of many space systems consists of a spectrum of energies, with mean energies between approximately 25 and 50 MeV. For example, Fig. 27 shows the distribution of proton energies for a high-inclination 705 km earth orbit. Although adding shielding shifts the peak in the energy spectrum to higher energies, the peak energy for this spectrum remains below 50 MeV. Consequently, most of the damage will be caused by protons with energies below 50 MeV, taking into account the increase in NIEL at low proton energies.
III-30
Particles cm -2 Per Steradian
109
108
107
200 mils
400 mils
100 mils 106
105 0.1
60 mils
1.0
705 km, 98 orbit 10 Energy (MeV)
100
1000
Fig. 27. Proton energy spectrum for a 705 km, 98 º orbit with various aluminum spherical shield thicknesses.
If the NIEL energy dependence continues to decrease at high energy, tests at 200 MeV will cause less damage, leading to serious under-estimation of the effect of the actual proton spectrum on the device. For this reason, tests at lower energies – 50 MeV - are recommended for III-V devices. That energy is near the peak energy in typical proton spectra, and is in a region where there is good agreement between NIEL calculations and experimental results. The 50-MeV protons have a range of more than 400 mils in aluminum, which provides sufficient range to penetrate the “dead” regions and packaging of most devices without significantly degrading the proton energy when it reaches the active region of the structure. We will base most of the discussion of device responses in this chapter of the short course on equivalent damage with 50MeV protons. D. Radiation Testing with Protons There are some unique problems that have to be dealt with when radiation tests are done with protons. The strong dependence of NIEL on proton energy for energies below 50-MeV can introduce errors in experimental interpretation because part of the proton energy will be lost when the proton beam goes through packaging or optical windows, reducing the energy of protons that strike the active region. In many cases the active region of an optical device is located beneath several layers of semiconductor, or beneath clear optical compounds that are used for index matching to improve optical coupling efficiency. The range of protons (in aluminum) with various energies is listed in Table 4. Corrections must be made for different types of materials, multiplying by the density of the material. For energies above 50 MeV the range is usually large enough so that corrections for energy loss in the structure are unnecessary, provided the irradiation is done with the beam at normal incidence. However, the range is very limited for lower energies. III-31
Another potential interference effect is darkening of optical windows, lenses, or index matching materials. These effects are not always important, but there are cases where the ionization damage associated with proton tests can produce substantial absorption within such materials, altering the test results. One way to check for this is to irradiate some samples with gamma rays, comparing the results with proton tests at equivalent total dose levels. It is often possible to directly test such materials by disassembling one or more devices, and evaluating absorption on the intervening material separately. Table 4. Range in Aluminum for Protons with Various Energies Proton Energy
Range in Aluminum
Range in Aluminum
(MeV)
(mils)
(µm)
10
5.89
150
15
22.4
569
20
81.6
2,040
50
420
10,670
100
950
24,200
IV. Radiation Damage in Optical Emitters
Optical emitters are relatively unaffected by ionization damage because compound semiconductors do not have high quality insulators, and already have high densities of surface states. In many cases special buffer layers are used that isolate the buried active region from the surface, further reducing the importance of additional recombination at surfaces. The dominant radiation damage mechanism in most optical devices is displacement damage. The relatively slight degradation that occurs in LEDs and laser diodes from irradiation with cobalt-60 sources is actually caused by displacement damage from the Compton electrons produced by the gamma rays, not ionization damage. Despite this general sensitivity, many optical emitters are relatively immune to displacement damage effects because they use thin active regions, with relatively high doping levels. A. Light-Emitting Diodes The sensitivity of light-emitting diodes to displacement damage effects varies by more than four orders of magnitude, depending on the specific design of the LED. Amphoterically doped LEDs are among the most sensitive components, degrading significantly at a nominal fluence of 1010 p/cm2 (50 MeV). This corresponds to about 1.6 krad in a space environment dominated by protons, a very low radiation level. Optical couplers using this type of LED have failed in space applications [Swif03]. The most straightforward way to evaluate LED degradation is to compare the light output after irradiation with pre-irradiation light output. This is shown in Fig. 28 for two types of LEDs made by one manufacturer for high-reliability space applications. The first type of LED is amphoterically doped; the second uses a more complex process with double-heterojunctions. It is clear from this figure that the amphoterically doped LED is far more sensitive compared to the double-heterojunction counterpart. However, this extreme sensitivity is partially offset by the much higher efficiency of the amphoterically doped LED. III-32
Light Output (Normalized)
1.0
50 MeV protons
0.8 OD880
Amphoterically doped
0.6 OP233 0.4
0.2
OD800
Double-heterojunction
0
5x1010
0
1x1011
Proton Fluence (p/cm2) Fig. 28. Fractional light output after proton irradiation for two types of LEDs, normalized to the higher optical output power of the amphoterically doped LED.
When LED damage is evaluated at constant injection conditions (fixed current) the damage is actually superlinear with fluence provided that the recombination centers introduced by the radiation are uniformly distributed within the bandgap. A more sophisticated way to deal with the degradation is to fit the damage to a power law, using the equation below [Rose82]
n
[ (Io/I)
- 1] =
( K τ) Φ
(11)
where Io is the pre-irradiation intensity, I is the light intensity after irradiation, n is an exponent between 0.3 and 1, (Kτ) is a damage constant that includes the minority carrier lifetime, and Φ is the particle fluence. For LEDs that are dominated by lifetime damage, the quantity at the left side of the equation is linearly related to fluence when n = 2/3. Fig. 29 compares degradation data for n = 1 and n =2/3; note the nearly linear behavior for n = 2/3.
III-33
10 I F = 1 mA
[(I o / I)n -1]
n= 1
I F 10 mA
1 I F = 1 mA I F 10 mA 0.1
n = 2/ 3 Optodiode OD880
50 MeV protons 0.01
1010
1011 1012 Proton Fluence (p/ cm2)
1013
Fig. 29. Superlinear damage in an amphoterically doped LED that can be linearized by using an exponent of 2/3 in the expression at the left side of Eq. 11.
Although this is a valid way to examine LED damage, the 2/3 power relationship usually does not linearize damage in heterojunction LEDs. For those types of LEDs, the damage (as described by Eq. 11) is linear for values of n that are very close to unity. Thus, the main advantage of Eq.11 is to linearize the superlinear damage amphoterically doped LEDs. However, the equation can still be used for heterojunction devices with n = 1, providing a linear metric for degradation. A more straightforward way to examine LED damage is to show the fractional remaining optical power output as a function of fluence. This is shown in Fig. 30 for double-heterojunction LEDs with several different wavelengths. The 660 nm LED is fabricated with GaAsP, while the others use AlGaAs. The dashed line shows the degradation of an amphoterically doped LED for comparison. This figure does not take unit-to-unit variability of damage into account, which is typically a factor of two for most types of LEDs. LEDs are also available at longer wavelength that are optimized for high-speed operation in fiber-optic data buses. Those LEDs degrade even less than the heterojunction LEDs in Fig.30, but operate above the cutoff wavelength for silicon detectors.
III-34
Light Output (Normalized)
1 850 660
825 0.1
Amphoterically doped
50 MeV protons IF = 5 mA 0.01 109
880 1010
1011
1012
Proton Fluence (p/cm2) Fig. 30. Fractional change in light output after irradiation with 50-MeV protons for several different types of LEDs.
Optical power degradation in Fig. 30 was evaluated at a forward current of 5 mA, about 5% of the maximum forward current. Less damage occurs when LEDs are measured at high forward current, and it is usually advisable to characterize LED degradation at several different forward current conditions that overlap various use conditions. Note however that most applications use forward current well below the maximum rated value because of reliability concerns. LEDs are subject to gradual degradation (wearout) during operation, and the rate of degradation is much greater at high forward currents [Witt96]. Radiation damage in LEDs that have been operated over extended periods (with up to 20% wearout degradation) is very similar to degradation in LEDs that have not been subjected to wearout [John00], and thus reliability and radiation degradation can be considered independently. Annealing Displacement damage in LEDs anneals after irradiation. Although annealing in non-operating LEDs can be instigated by heating, temperatures above 200 ºC are required [Loo81], which is well above the maximum allowable operating temperature of typical LEDs. The most important factor in LED annealing is forward current injection, which causes damage to anneal even at room temperature. Irradiated LEDs can be stored without forward current injection for six months or more without appreciable change in the degraded characteristics. However, as soon as a forward current is applied, the damage begins to recover. It is possible to take advantage of annealing to reduce the effects of radiation damage. This was done recently in the Galileo space system after proton damage caused circuit failure in an LED application [Swif03]. Although annealing can be used to advantage, the sensitivity of some types of LEDs to injection-enhanced annealing introduces possible errors and inconsistencies in evaluation of radiation degradation. For example, if irradiations are done using samples that are under forward bias, the forward injection will significantly change the amount of degradation that is observed. If the devices are left in this condition after irradiation (as well as before each irradiation if a series of stepped irradiations are done), this will also affect the results. Thus, III-35
injection-enhanced annealing can be a serious interference effect during testing. It is far more straightforward to test devices in an unbiased condition, and evaluate the effects of injection on damage recovery afterwards. Measurements have to be carefully planned to avoid interference from annealing, using pulsed measurements and limiting the current that flows through the device when devices are evaluated after irradiation. Although this is relatively easy for optical power measurements, it may be impossible to measure spectral characteristics without inadvertently annealing the damage because most spectrometers take several minutes to sweep the grating within the light source. An example of injection-enhanced annealing in an amphoterically doped LED is shown in Fig. 31. Two curves are shown, corresponding to two different proton energies. The optical power immediately after irradiation was reduced to about 10% of the initial value, and the damage is calculated using the 2/3 power relationship shown in Eq. 11. The forward current during the extensive annealing time was 10 mA. Initially there is little effect. The damage begins to recover after 10 seconds, and continues to recover for several decades. Saturation in the recovery characteristics is evident at about 3 x 105 seconds (4 days). More of the damage recovers for the device irradiated with 25 MeV protons compared to the device irradiated at higher energy. The maximum amount of damage that can be recovered through annealing is 30 – 40%. Annealing proceeds faster when higher forward currents are used. Fig. 32 compares results for three different currents – 5, 10 and 50 mA – for an amphoterically doped LED with a maximum rated current of 100 mA [John00]. The abscissa shows the total charge that has passed through the LED during annealing for the three different currents. The amount of annealing that has occurred is approximately the same for the three conditions. This provides a way to apply annealing results for one set of conditions to a different end-use condition.
0.5
Fractional Damage Recovery
0.4
Optek OP130 (930 nm)
25 MeV
IF = 10 mA during annealing 0.3
200 MeV
0.2
0.1
0.0 -0.1 0.1
1
10
100 1000 Time (s)
10,000 100,000 1,000,000
Fig. 31. Annealing of an amphoterically doped GaAs LED when a forward current of 10 mA is applied after irradiation. The recovery is faster for devices that are irradiated with lower energy.
III-36
Light Output (Normalized to Initial Value)
0.3 0.25
OD880 LED Irradiated to 8 x 1010 p/cm2 50 MeV Protons
10 mA
0.2
50 mA 0.15 0.1
5 mA
0.05
Results shown for three different current conditions
0 0.001
0.01
0.1
1
10
100
1000
Total Charge After Irradiation (C) Fig. 32. Comparison of annealing in a AlGaAs LED for three different forward currents. The annealing is about the same for all three conditions when total charge is used to normalize the results.
Amphoterically doped LEDs, which require long minority carrier lifetime for efficient photon emission, are highly sensitive to annealing. In contrast, most double-heterojunction LEDs exhbit very little annealing, 10% or less. This is likely related to the shorter lifetime and the influence of mechanisms other than minority carrier lifetime in degradation of DH LEDs. However, we will see in the next section that damage in laser diodes made with double-heterojunctions can be highly sensitive to injection-enhanced annealing. B. Laser Diodes For laser diodes the parameter that is most affected by displacement damage is threshold current. Fig. 33 shows the results of an older study of proton damage in a strained-layer laser diode using InGaAs [Evan93]. They used 5.5 MeV protons in the study, which are about 8.5 times more damaging compared to 50-MeV protons. The first-order effect of the radiation damage is to increase the threshold current; the change in threshold current is proportional to fluence. Note that the slope of the optical power output – slope efficiency – is essentially unchanged for this particular device, except after the highest fluence where it decreases slightly. Measurements were made at constant temperature, using a heat sink. The maximum current was limited to about 70 mA to avoid excessive heating.
III-37
Optical Power (mW)
20 7.8x1012
Strained quantum-well laser λ = 980 nm Active layer: 60A InGaAs
15
Initial 4.6x1011 10
2.1x1012 3.9x1012
5.5 MeV protons 5
1.6x1013 After Evans, et al., TNS, 1991
0 0
10
20
30
40
50
60
70
Current (mA) Fig. 33. Degradation of strained quantum-well laser after irradiation with 5.5 MeV protons [Evan93]
A similar set of results is shown in Fig. 34 for a 650 nm laser. The results are generally similar to the results for the 980-nm laser above, but in this case the slope efficiency changes at moderate fluences. The laser temperature was controlled with a thermoelectric cooler, using pulsed measurements to eliminate self heating at higher currents. 0.5
Detector Current (mA)
0.4 50 MeV protons
Pretest 1012
0.3
3x1012
1013
0.2
2x1013
3x1013
0.1
0
0
10
20 Forward Current (mA)
30
40
Fig. 34. Degradation of a 650-nm laser after irradiation with 50-MeV protons. Note the decrease in slope efficiency compared to the previous figure.
III-38
Although threshold current and slope efficiency are important parameters for laser applications, it is possible to learn more about the internal degradation mechanisms by extending the measurements to low light levels, and plotting the results on a semi-logarthmic plot. The data in Fig. 34 is plotted in this way in Fig. 35. Far more degradation occurs at low currents, below the laser threshold region. The degradation in that region is a better current measure of non-radiative recombination centers than changes in threshold current. 10-3 Pretest 1x1012
Detector Current (A)
10-4
3x1012 1x1013
10-5
2x1013 3x1013
10-6 Transition region from LED to laser operation broadens after irradiation
10-7
10-8
0
10
20
30
40
Forward Current (mA) Fig. 35. Results of Fig. 34 plotted semi-logarithmically to show the optical power at low current, where the device functions as an LED.
Another way to examine laser degradation is to plot the derivative of the optical power vs. forward current. That measurement technique provides more detailed information about the behavior of the device in the lasing mode. Fig. 36 shows the derivative of the detector current (optical power) with respect to forward current as a function of forward current. A large increase in the derivative occurs at the threshold current, and the slope of the derivative near the threshold region is only slightly affected by radiation damage. As the current increases above the threshold current, the slope decreases slightly because of internal losses and efficiencies. The slope in that region changes slightly after irradiation because of increased losses within the cavity that cause the transparency density to change. High-quality lasers have slopes that are nearly horizontal over an extended range of current. Although this technique is useful, it requires accurate, stable measurements. The value of the derivative is highly sensitive to noise or measurement instability.
III-39
Derivative of Detector Current wrt Forward Current
0.05
0.04
0.03
0.02
Fermionics #159 neutron irradiation
0.01
Pretest 8E14 neutrons 0.0 0
10
20
30
40
50
Forward Current (mA)
Fig.36. Derivative of optical power for a 1300 nm laser diode before and after irradiation.
VCSELs Although the operating principles of VCSELs are essentially the same as that of edge-emitting lasers, slight differences in the layer thickness of the Bragg reflector and self-heating are very evident during radiation degradation studies of VCESLs. Self-heating restricts the range of currents over which the device can be operated. Unlike conventional laser diodes, the optical power output has “bumps and wiggles” that correspond to different modes within the complex Bragg reflector. Some VCSELs have significant discontinuities in the optical power curve that can change after irradiation. Fig. 37 shows how optical power in a VCSEL is affected by radiation damage. VCSELs typically have highly nonlinear output characteristics, and nearly always show a decrease in slope efficiency after irradiation. Many VCSELs have a more gradual transition at the threshold current than shown in the example of Fig. 37 due to the complex nature of the Bragg reflector [Scho97]. This causes some ambiguity in defining threshold current. One way to deal with this is to fit the optical power to a linear relationship near the threshold current, extrapolating the slope to define the threshold condition. It is also possible to use the derivative analysis discussed in the previous subsection. However, the discontinuities in VCSEL output characteristics make this technique less useful compared to edge-emitting lasers.
III-40
0.7 50 MeV protons
0.6 Detector Current (mA)
λ = 850 nm
Pretest
0.5
1x1012 3x1012
0.4
1x1013 2x1013
0.3
3x1013 0.2 0.1 0
0
2
4
6 8 10 Forward Current (mA)
12
14
16
Fig. 37. Degradation of a VCSEL after irradiation with 50-MeV protons. Note the substructure and the decrease in slope efficiency at high operating currents.
Degradation of Various Types of Lasers From the earlier discussion of lasers in Section III, one would expect that different types of lasers would respond quite differently to radiation damage because different material systems are used in their fabrication. However, that is not the case for lasers that have been evaluated in recent years. The effect of proton irradiation on threshold current is shown in Fig. 38 for several different types of lasers, showing the percent increase in threshold current. All of the tests were done using 50-MeV protons. Even though these lasers have different wavelengths, using different materials, the results are remarkably similar. The slopes of the threshold current vs. fluence are nearly identical for several of the lasers, and the fluence at which the threshold current first begins to change is remarkably close, except for the VCSEL. We can explain this by re-examining the conditions for lasing in a semiconductor, along with loss mechanisms. In order for lasing to occur, the carrier density must be large enough to “saturate” the bimolecular recombination coefficient, B and obtain a carrier density that is sufficiently high to increase the probability of stimulated emission. However, it turns out that B is nearly the same for the three material systems that we are considering, within about 50%. The other important factor is the gain of the material and laser cavity. The gain curves for the three materials require carrier densities on the order of 1.5-4 x 1018 cm-3. The first-order effect of radiation-induced defects is to increase the number of non-radiative recombination centers. As the number of defects increase, we have to compensate non-radiative losses by increasing the carrier density (i.e., current) to the point where the material gain is higher, which causes the threshold current to increase. Approximations for the gain of these three materials have been developed by Coldren and Corzine, showing that the logarithmic dependence of gain on carrier density can be approximated by a linear relationship for small changes [Cold95]. This explains the observed linear dependence of threshold current on proton fluence over the range of changes shown in Fig. 38. Departures from linearity will occur at higher fluences. III-41
100
% Change in Threshold Current
50 MeV protons InP laser (1330 nm) Strained QW (InGaAs, 905 nm) 10 Distributed feedback laser (InP,1550 nm) VCSEL (AlGaAs,850 nm) 1 1011
1012
1013
1014
Proton Fluence (p/cm2) Fig. 38. Threshold current degradation of several different lasers. Note the remarkable similarity in results despite the different materials and wavelengths.
Annealing of Displacement Damage in Lasers
As discussed previously, LEDs that are fabricated with narrow heterojunctions are relatively insensitive to current-enhanced annealing. Thus it is somewhat surprising that lasers, fabricated with similar structures, are highly sensitive to annealing. Annealing was observed in the first radiation studies of laser diodes, but these earlier devices had extremely high threshold current and would only work reliably at liquid nitrogen temperature [Comp67]. Fig. 39 shows annealing results for a contemporary 1300-nm quantum-well laser. The threshold current is approximately 7 mA. First, note that the device that was not operated under forward bias changes very little, even 3 days after irradiation. The device that continually operated at 5 mA, below the lasing threshold, gradually recovers, with a threshold current shift of about 2/3 as great after 3 days of continuous operation. Two of the devices were operated at currents above the lasing threshold, and their recovery is clearly much faster than the device that operated below threshold, in the LED mode. The current densities are not very different for the device operated below the laser threshold, but the annealing clearly proceeds at a lower rate. This implies that the much higher photon density when the device enters the lasing mode also affects the annealing process. Similar results have been obtained for VCSELs, with even more pronounced recovery for devices where annealing is done above the lasing threshold [John01].
III-42
Change in Threshold Current (%)
35 30 Unbiased anneal 25 5 mA 20 15 mA 15
Lumex laser diode λ = 1300 nm
18 mA
10 5 0 100
All devices irradiated to 3 x 1013 p/cm2
101
102
103 104 Annealing Time (s)
105
106
107
Fig. 39. Annealing of 1300 nm quantum-well lasers under different bias conditions. All four devices were irradiated to a fluence of 3 x 1013 p/cm2.
Annealing in lasers is still an active research area. Annealing progresses more rapidly after irradiation with protons of moderate energy, e.g. ~ 25 MeV, compared to the annealing rate under similar conditions for lasers irradiated with protons of higher energy [John02]. This is probably related to the microscopic nature of the defects produced by the radiation. Low-energy protons produce large numbers of vacancy-interstitial pairs (Frenkel defects) with relatively few cascade damage regions. The Frenkel pairs are in relatively close proximity compared to cascade damage regions, and can recombine more easily. Compared to protons, neutrons produce large numbers of cascade damage regions. Fig. 40 compares annealing results for 1300 nm laser diodes that were irradiated with 50-MeV protons and fission neutrons (normalized to 1MeV equivalent fluence). Proton damage not only proceeds more rapidly, but a larger fraction of the damage recovers. Data for extremely energetic protons- 24-GeV – are also shown in the figure (from [Gill00]), but are for laser diodes from a different manufacturer. Although proton damage is usually stable for devices that are not operated after irradiation, results for neutron irradiation show significant recovery even in unbiased devices [Gill00]. The reason for this difference in stability without applied bias has yet to be explained. It is evident that more work needs to be done on annealing in laser diodes in order to determine how injection, optical power, particle type and energy influence annealing. The device structure and specific semiconductor material may also play important roles in annealing.
III-43
1.0
Unannealed Damage Fraction
Lumex laser (neutrons)
NEC laser (24-GeV protons)
0.8
0.6 Lumex laser (50-MeV protons) Ithresh = 9mA 0.4
0.2
0.0 0.01
0.1
1
10
100
Time after Irradiation (hours)
Fig.40. Annealing of laser diodes after irradiation with neutrons and protons. The annealing proceeds more rapidly after irradiation with 50-MeVprotons. Results for a different laser of the same basic type irradiated with 24-GeV protons are also shown in the figure.
V. An Introduction to Optical Detectors A. Detector Principles
In this section of the short course we will consider detectors formed by conventional p-n or p-i-n regions that respond to light by absorbing a photon, creating electron-hole pairs. Other types of detectors, including highly doped extrinsic detectors, are discussed in another section of the NSREC-04 short course. Silicon is widely used as a detector for wavelengths between 0.4 and 1 µm. The responsivity of a silicon detector is shown in Fig. 41. At short wavelengths the responsivity is low because nearly all of the light is absorbed near the surface, and the energy of the photons is much greater than the bandgap energy. The excess energy when each photon is absorbed is dissipated as heat, not photocurrent. As the wavelength increases, the photon energy is closer to the bandgap energy, increasing the overall efficiency for electron-hole generation. The responsivity at long wavelengths falls to zero when the photon energy falls below the bandgap energy. The thickness of the detector must be greater than the absorption depth. IR-enhanced detectors have increased thickness to maintain responsivity at longer wavelengths, as shown in the figure.
III-44
0.6 IR enhanced
Responsivity (A/W)
0.5 0.4 Standard 0.3 0.2
0.1 0.0 400
500
600
700
800
900
1000
1100
Wavelength (nm) Fig. 41. Responsivity of a conventional silicon detector, optimized for responsivity at approximately 850 nm, and a detector with extended thickness that has a higher responsivity at longer wavelengths.
Conventional detectors consist of a p-n junction that can detect the excess carriers produced by optical absorption. Detectors can be designed to operate in various ways. The two most common are photovoltaic and photoconductive. Photovoltaic. When the photodetector is unbiased, the excess carriers produce a voltage across the region. This self-generated voltage can provide current to a low-resistance load. This is the process that takes place in solar cells. Because the structure is unbiased, much of the current is collected by diffusion, resulting in relatively long collection times. Photoconductive. The photoconductive mode is widely used, biasing the detector to provide an extended depletion width. Photo-generated carriers that are created within the depletion width are rapidly collected. Carriers that are generated beyond the depletion width boundary are collected by diffusion, a much slower process. The diffusion length, L, is related to minority carrier lifetime by the equation
L = Dτ
(12)
where D is the diffusion constant, and τ is the minority carrier lifetime. A diagram of a p-n photodiode is shown in Fig. 42. It is similar to a conventional diode, but is designed so that that the depth for charge collection is consistent with absorption within the material over the wavelength range. This is typically about twice the “1/e” absorption depth (see Fig. 2). Thus, a photodiode that is designed to absorb light up to wavelengths as long as 900 nm will require an overall depth of about 60 µm. It is possible to extend the wavelength to about 1000 nm (near the silicon band edge) by designing a photodiode with a very deep collection depth, 200 µm or more. Guard rings (p+) surround the top surface of the photodiode in order to avoid surface leakage. Although not shown on the figure, most photodiodes have an antireflection surface coating that reduces reflection losses at a specified wavelength. III-45
Incident Light Guard ring Oxide
Oxide
p+
Absorption depth
p+
p+
n Depletion region
n+ Fig. 42. Diagram of a typical photodiode.
p-i-n Photodiodes By reducing the doping level, it is possible to fabricate a photodiode with much the same structure as a basic photodiode operating in the photoconductive mode, but with much faster response time. The lightly doped region allows the depletion width to extend completely through the lightly doped region to the underlying n+ contact, provided that the applied reverse voltage is sufficiently high. This extends the depletion region, allowing all of the carriers to be collected by drift, eliminating the slow diffusion component from the carrier collection process. This type of detector can be made from compound semiconductors, as well as silicon. A diagram of an InGaAs p-i-n detector is shown Fig. 43. It is sensitive to wavelengths between 0.9 and 1.6 µm because of the narrower bandgap. In this example, light is collected from the back instead of the top surface. The detector is fabricated on an InP substrate, which is transparent to light in the wavelength range of interest. Light is absorbed in the InGaAs layer, which can be made quite thin [Liu92] – on the order of 3 µm – because the absorption coefficient is much higher and relatively flat over this range of wavelengths compared to an indirect semiconductor (see Fig. 2). A p-i-n detector fabricated in silicon would typically have a much thicker light-absorbing layer because the absorption coefficient is smaller, and varies much more with wavelength (and would only be useful out to a maximum wavelength of 1100 nm). The construction of a silicon p-i-n detector is similar to that of the conventional silicon detector in Fig.42, but with lower doping in the depletion region.
III-46
Electrode polyimide
p+
n-
InGaAs
Light-absorbing layer
buffer layer
InP substrate
Light Fig. 43. Diagram of an InGaAs p-i-n photodiode intended for wavelengths between 0.9 and 1.6 µm.
One drawback of p-i-n diodes is that they have much higher dark current compared to conventional photodiodes. However, InGaAs devices have relatively thin layers, reducing the dark current compared to silicon or germanium detectors. The dark current has two components, one from surface recombination, and the other from generation-recombination centers in the bulk region where the device is fully depleted. Dark current depends on temperature, doubling approximately every 10 ºC. Avalanche Photodiodes Avalanche photodiodes (APDs) rely on avalanche multiplication to increase the number of electron hole pairs produced by the interaction of a single photon within the structure. Fig. 44 shows a diagram of a silicon avalanche photodiode that is intended for applications with wavelengths up to 900 nm (the absorption depth is about 8 µm at that wavelength). The APD has two distinct regions: a lightly doped drift region which collects carriers from the incident photons where the electric field is relatively low, and a high-field avalanche region below the surface. The basic concept is that minority carriers (in this case electrons, because the multiplication factor for electrons is much higher for electrons) produced by photons in the drift region will be collected in the avalanche region where the high field will cause additional carriers to be produced. The device is biased so that the lightly doped region is depleted, causing the depletion region to extend to the boundary of the avalanche region. Avalanche photodiodes typically have very lightly doped drift regions, as shown in the figure. The avalanche process is temperature sensitive, usually requiring local temperature control of the APD. The APD in Fig. 40 is only intended for wavelengths below 900 nm because the drift region only extends 25 µm below the top surface. Special “IR-enhanced” ADPs are available where the drift region extends to 200 µm. However, those devices have larger dark current because the total volume of the diode is so much higher.
III-47
High-field region below surface
n p E 25 µm
p- (~ 4E13) 0
p+
25 µm Distance from Surface
Fig.44. Diagram of a silicon avalanche photodetector.
B. Radiation Damage Radiation affects detectors in several ways, as listed below.
Ionizing radiation may increase surface recombination because of traps that occur at the interface between the detector surface and the passivation region. Guard rings are usually used at the edge of the detector, which decrease sensitivity to surface recombination • Minority carrier lifetime will decrease, reducing the diffusion length for photogenerated carriers within the detector. • Bulk damage will increase dark current. This mechanism is particularly important for avalanche photodiodes because the dark current from bulk recombination centers will be multiplied by the avalanche multiplication factor of the APD. • Bulk and surface damage will increase noise in the detector. Although noise is only important for detectors used to detect low light levels, it is an important degradation mechanism. •
Fig. 45 shows how the photoresponse of a conventional silicon detector is affected by highenergy protons. Light at shorter wavelengths almost entirely absorbed within the drift region, where it is only slightly affected by lifetime damage. This is the reason that only slight changes occur in the responsivity at 650 nm. However, most of the light at longer wavelengths is absorbed beyond the drift region and relies on diffusion in order to be collected. This causes the photoresponse to degrade more severely at longer wavelengths. A p-i n detector would show nearly the same degradation at all wavelengths, because all of the charge is collected through drift, not diffusion. However, dark current in a p-i-n detector would increase by several orders of magnitude at the maximum fluence levels shown in the figure.
III-48
Photoresponse (normalized)
1.0
0.9
650 nm 0.8
0.7
0.6
930 nm
50 MeV protons
0.5
0.4
0
1 x 1011
2 x 1011
3 x 1011
4 x 1011
5 x 1011
Proton Fluence (p/cm 2 ) Fig. 45. Degradation of a conventional photodiode at two different wavelengths.
Radiation damage in avalanche photodiodes is more complicated because the avalanche gain amplifies radiation-induced dark current from the bulk region, but has no effect on surface current [Swan87]. Figure 46 shows how dark current in a silicon avalanche photodiode is affected by gamma and proton irradiation [Beck03]. Although we often assume that ionization damage is the dominant mechanism for this type of component, displacement damage is actually the dominant process (except at high radiation levels). In this case damage at total dose levels below 20 krad(Si) when the device is irradiated with gamma rays is actually due to displacement damage. The Compton electrons that are produced by the cobalt-60 gamma rays have an average energy of about 500 keV, which are about 250 times less effective in producing displacement damage than the 51-MeV protons, but nevertheless still introduce displacement effects. At higher total dose levels some of the APDs became sensitive to ionization damage, resulting in strongly nonlinear behavior for some of the samples. The total dose where this occurred was about the same for both protons and gamma rays. Thus, damage in these detectors appears to be a superposition of displacement damage and ionization damage.
III-49
104 103
51 MeV protons
102 10
Co-60 gamma rays
1 10−1 10−2 0.1
1.0
10
100
1000
Equivalent Total Dose [krad(Si)] Fig. 46. Degradation of a silicon avalanche photodiode after irradation with protons or cobalt-60 gamma rays (equivalent total dose values are shown). Some samples exhibited large increases at higher levels, consistent with surface damage.
Degradation of Internal Monitor Diodes Used in Laser Diodes Another important class of diodes are the monitor diodes that are frequently incorporated within laser diode assemblies. These are usually discrete diodes, but it is also possible to fabricate integrated monitor diodes [Diec01]. Fig. 47 shows a typical diode assembly where the monitor photodiode measures light irradiated from the back facet.
Laser diode
Heat sink Monitor photodiode Fig. 47. Mechanical configuration of a monitor diode that measures light from the back facet of an edge-emitting laser.
III-50
Fig. 48 compares degradation in the photoresponse of monitor diodes from two different types of laser diodes with the photoresponse degradation of a silicon detector [John01]. In both cases there is much less degradation in the monitor diodes compared to silicon. However, the monitor diodes degrade more rapidly than the threshold current of the laser diodes that they are measuring, and may be the limiting factor in applying these laser diodes in system applications. Thus, it is essential to evaluate monitor diodes when radiation tests of laser diodes are done. Note that little or no annealing will take place in the photodiodes, unlike the threshold current of laser diodes, which anneals rapidly during extended operation, increasing the importance of monitor diode degradation in most applications.
Current (normalized)
1.5
Internal monitor in Lumex laser diode (650 nm)
1.0
0.5
0.0 1011
Silicon detector at 930 nm
Internal monitor in Lumex laser diode (1300 nm)
1013
1012 Fluence
1014
(p/cm2)
Fig. 48. Degradation of internal monitor diodes within laser diode assemblies. They are compared with the degradation of a silicon detector.
III-51
VI. Optical Fibers A. Propagation Optical fibers are designed to confine light within a central region of a fiber. A basic optical fiber consists of a core (which may be uniform or doped with impurities) surrounded by a cladding layer with lower refractive index than the core. The difference in the refractive indices of the core and cladding are key parameters, along with the diameter of the core. Figure 49 shows a diagram of a basic optical fiber. Cladding (n3)
Light is launched into the core from an external light source
Fiber core (n2)
Guiding condition: n3 must be less than n2 Fig. 49. Basic diagram of a step-index optical fiber with index of refraction n2 in the fiber core, n3 in the cladding, and n1 in the medium between the light source and the fiber front surface.
Light that is incident on the fiber will enter the core. Light that enters the core at an angle will be refracted, eventually reaching the cladding. Light that strikes the cladding at angles greater than the Brewster angle will undergo total internal reflection, with very low loss. However, if the angle between the light ray and the cladding is less than the Brewster angle it will enter the cladding, and cannot propagate down the fiber. Thus, light that enters the fiber beyond a critical acceptance angle will be lost in the cladding. The numerical aperture (N.A.) determines the maximum acceptance angle of the fiber. It depends on the refractive index of the fiber as well as on the refractive index of the medium between the optical source and fiber, as described by the equation: N.A. = sin θi =
n 2 2 − n 32
(13)
The N.A. defines the maximum acceptance angle, θi, for light incident at the core of the fiber, assuming that the medium between the fiber core and the light source has an index of refraction of one. Although it is possible to use simple ray diagrams to understand the principles of optical fibers, light actually propagates through the fiber as an electromagnetic wave with various modes. Fibers with a relatively large core diameter (50 to 100 µm) will allow many different III-52
electromagnetic modes to propagate down the fiber. By reducing the diameter, it is possible to fabricate fibers that can only support a single internal mode. A cutoff wavelength can be defined for single-mode fibers, defined by the equation λc =
2π ( N.A.) r J1 (0)
(14)
where λc is the cutoff wavelength, r is the radius of the fiber, and J1(0) is the value of the Bessel function J1 with argument equal to zero (numerically ≈ 2.405). If the wavelength λ is > λc then the fiber will be in single mode, whereas if λ is < λc then the fiber will be capable of multimode propagation. We can estimate the fiber diameter for an example where the refractive indices are 1.46 and 1.45. The numerical aperture is then 0.171. Using the previous equation, the diameter of the fiber must be less than 2.405 λ/ (2π 0.17) or 4.5 µm for a wavelength of 1.3 µm, a very small diameter compared to the diameter of multimode fibers. It is possible to increase the fiber diameter for single-mode operation by reducing the difference between the refractive indices of the core and cladding, but the difference must be high enough to avoid difficulties with nonuniformity of either the cladding or core. Single-mode fibers typically have core diameters between 4 and 10 µm. Dispersion Two different mechanisms cause pulse-width broadening when an optical pulse is transmitted through an optical fiber. The first mechanism is caused by chromatic dispersion within the fiber material because the index of refraction depends on wavelength. It is important because of the finite spectral width of typical light sources (~ 7% for LEDs, and 0.5 to 2% for semiconductor lasers). Chromatic dispersion causes light at different wavelengths to travel at slightly different angles within the fiber, increasing the path length for light at longer wavelength. This will cause the pulse with to broaden. Chromatic dispersion is more important when an LED source is used because of the relatively broad bandwidth. It is possible to eliminate that contribution by using a wavelength near 1.3 nm, where chromatic dispersion in the fiber is nearly zero (a fortuitous property of silica). Fig. 50 shows how chromatic dispersion depends on wavelength for pure silica. Minimum dispersion occurs at about 1300 nm, the “second” optical fiber window.
III-53
Dispersion [ps/(nm-km)]
200
150
100
50 Zero dispersion 0
−50 0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
1.6
Wavelength (µm) Fig. 50. Chromatic dispersion in pure silica. The dispersion is nearly zero for wavelengths near 1300 nm.
The other mechanism is modal dispersion. Fig. 51 shows why this occurs in a step-index fiber that supports several different modes. The optical path length is greater for light that enters the fiber at more extreme angles, delaying the arrival compared to light that enters at steeper incident angles. This causes the optical pulse to be stretched out, as shown in the diagram.
n=1
n=3
Fig. 51. Modal dispersion in a step-index optical fiber. More time is required for higher-order modes to be transmitted through the fiber.
A brief summary of several different types of optical fibers is given below: 1. Step-index fibers, which have a large diameter, uniformly doped core region, surrounded by a cladding with lower refractive index than the core. The core is usually pure silica. Those fibers can transmit several different modes, and have higher dispersion than other types of fibers. Typical fiber core diameters are 50 to 125 µm. 2. Graded-index fibers, which use graded doping to confine photons to the central region of the fiber. The graded doping decreases the index of refraction away from the center, decreasing modal dispersion because the propagation velocity depends on 1/n (n is the index of refraction). Typical core diameter is 50 µm. Germanium is commonly used to dope silica fiber cores in this type of fiber. III-54
3. Single-mode fibers, which use very small fiber diameters. The small fiber diameter limits propagation to a single internal mode, eliminating modal dispersion. Single-mode fibers usually have core diameters between 5 and 10 µm (see Eq. 13). This increases the difficulty of coupling the fiber to the optical power source and detector. 4. Special-purpose fibers. One example is a polarization-preserving fiber. This type of fiber uses an asymmetric doping profile (for example an elliptical core) which has the property of retaining the polarization states of the initial light pattern, even over very long distances. Special fibers can also be made with special core materials for wavelengths > 1600 nm, where intrinsic losses from the bonds within silicon dioxide prevent the use of silica as a core material. Other important properties of optical fibers include (a) losses from bending, which can alter the propagation properties of fibers that are wound in tight spirals; (b) the possibility of launching unstable modes either within the fiber core or within the cladding unless special mode scramblers are used that provide a launch pattern that is equivalent to the optical modes in a long fiber; (c) alteration of fiber properties through stress, which may require careful treatment and winding of fibers on spools; and (d) the practical difficulties of providing stable, known optical power levels, taking Fresnel reflection and mechanical stability into account. Mechanical instability may affect light coupling at either the entrance or exit of a length of optical fiber during radiation testing. All of those factors are important when optical fibers are subjected to radiation testing, as discussed in the next section. B. Radiation Damage in Optical Fibers The primary mechanism involved in radiation damage in optical fibers is formation of color centers within the core (typically silica, although other core materials can be used, including plastic). The presence of impurities increases the rate of color center formation. For pure silica cores, the induced absorption is relatively low, and is strongly temperature dependent. Fibers doped with germanium have relatively low absorption when they are irradiated at low dose rate, but exhibit high absorption losses when they are exposed at high dose rate. The induced loss anneals relatively quickly. Fibers with undoped silica cores do not exhibit this effect. Radiation testing of optical fibers is typically done by exposing a coil of fiber within a radiation source, which can be steady-state (typically a cobalt-60 isotope source), or pulsed (a flash X-ray). There are a number of important details. Stress in the fiber or microbending losses (from short radius coils) can affect the results. A mode scrambler must be used at the input of the fiber. This eliminates “cladding modes” which allow some of the light to travel in the cladding region if the mode scrambler is not used. The optical power launched into the fiber must be accurately known, and low enough to avoid distorting the results because of photobleaching. The diagram in Fig. 52 shows a typical example of an optical fiber test. A lockin amplifier, synchronized to the pulse generator that drives the LED, is used to measure the optical output from the detector and low-noise amplifier. It is usually assumed that the attenuation depends on fiber length. This is generally valid, but the induced absorption will begin to saturate at high total dose levels. Many experimental details are important, particularly the method used to launch light to the fiber and optical power level. Fiber losses are usually reported in units of dB/km. Losses in most fibers are relatively low, requiring fiber lengths of 50 m or more in order to characterize radiation-induced loss with sufficient accuracy.
III-55
Pulse Generator
LED (mounted on thermoelectric cooler)
Detector Mode scrambler
Coupling from LED to test fiber
Lock-in Amplifier
Coupling from fiber to detector
Cobalt-60 irradiation cell
Fig. 52. Diagram of a radiation test of an optical fiber. A spool of the test fiber is placed within the radiation source and monitored with external equipment.
An example of the results of a radiation test of an optical fiber is shown in Fig. 53. In this example the tests were done using an LED with a wavelength of 850 nm. The LED was mounted on a temperature-controlled plate to maintain optical power stability (recall that the output of an LED is very sensitive to temperature). It was driven by a pulse generator, which was synchronized with a lock-in amplifier at the detector. The fiber, a multimode step index fiber with a pure silica core, was cooled within the irradiation cell to a temperature of –55 º C because of concern about fiber degradation at low temperature. Note that the induced absorption would be much lower if the tests were done at room temperature. Several different curves are shown. In each case the fiber was irradiated to a total dose of 10 krad(Si). However, the optical power coupled to the fiber was different for each case. A new section of fiber was used each time that the experiment was repeated. At high optical power levels the absorption increased to approximately 30 dB/km during the irradiation, saturating at about 3 krad(Si). The fiber gradually recovered after irradiation, with nearly complete recovery several hours after the irradiation was stopped. When the optical power level was reduced to low levels, the induced absorption in the fiber was about one order of magnitude greater than the results for the fiber sections that was tested with high optical power levels. At low optical power, the damage does not show any sign of saturation, unlike the results when the tests were done with high optical power. The reason for the lower absorption loss with high power is that when tests are done at high optical power the high density of photons causes some of the color centers within the fiber to anneal. This is called photobleaching.
III-56
1 sec
1h
1 day
−60 dBm
300
Induced Loss (dB/km)
1 min
Irradiation
−50
Recovery
200
−40
100 −30 −20
101
102
103
104 100
Total Dose (Rad)
101 102 103 104 105 Time after Irradiation (sec)
Fig. 53. Absorption and recovery of an optical fiber, irradiated at –55 º C in a cobalt-60 radiation source.
Substantial work has been done to compare test results from various laboratories under a NATO-sponsored committee [Frieb88, Tay901, Tay902]. Important results from the series of tests done by that committee include an investigation of the different variables that affect fiber radiation response, as well as the development of standard radiation test methods to improve the validity of test results by different laboratories. The study used three types of fibers: (1) silicacore multimode fibers; (2) silica-core single-mode fibers; and (3) germanium-doped gradedindex fibers. Two different wavelengths were used, 850 and 1300 nm. The first attempt at interlaboratory comparison revealed several experimental problems that caused substantial differences in the results. The following steps were taken to improve results in the second experimental study: • • • •
The input optical power was standardized at 1 µW A mode-scrambler was used in tests of the multi-mode fibers Mode-stripping was used to eliminate higher-order modes within the cladding Fibers from a single perform were distributed to the participating laboratories
Table 5 summarizes some results from the second study. Several points should be noted. First, the results were reasonably consistent, even though each laboratory had the freedom to implement the tests with their own equipment and procedures. The coefficient of variation (the coefficient of variation is the standard deviation divided by the mean value) was between 0.12 and 0.21 when we compare results from the different laboratories for all of the fiber types. Second, the induced loss in the single-mode fiber was extremely high at a wavelength of 850 nm, but in line with the results from the other types of fibers when it was measured at 1300 nm. The reason for this difference is transient absorption, a characteristic of germanium-doped fibers that causes far more absorption to occur at short times after a pulse of radiation. Fibers with pure silica cores do not exhibit this effect. III-57
Table 5. Steady-State Results for Optical Fibers from the NATO Study
Fiber Manufacturer
Type
Wavelength [nm]
Number of Participating Laboratories
Loss [dB /km]
Coeff. of Variation of Loss
850
4
4.34
0.12
1300
4
2.76
0.20
850
3
17.5
0.16
1300
4
2.7
0.13
Multimode silica core
A
Ge-doped graded index core Single-mode silica core Single-mode silica core
C G H
An additional series of tests was done to evaluate absorption and recovery at short times for Fiber G, a germanium-doped single-mode fiber. This type of fiber has high transient absorption losses that recover rapidly after irradiation. The results of those tests are shown in Fig. 54 [Tay901]. Four different laboratories compared test results for samples of fiber that were obtained from the same preform. The dose rate used for testing was restricted within a narrow range. There is reasonable agreement, although one laboratory measured significantly higher losses, even after the annealing period following the pulsed irradiation. However, pulsed measurements are far more difficult than the steady-state results shown in the previous figure. 20 Fiber coil
Induced Loss [dB/km]
16
dia.
Lab A Lab B Lab C Lab D
12
15 cm 5 cm 5 cm 5 cm
8
4
0 100
Single-mode fiber λ = 0.85 µm Pin = 1 µW
101
102
103
Time [s] Fig. 54. Comparisons of short-time tests of absorption in a single-mode fiber by four different laboratories.
III-58
104
Radiation-Induced Attenuation at 1000 rad
Recent work on radiation effects in optical fibers has included tests of special types of optical fibers, such as polarization-maintaining optical fibers. The core of those fibers are doped with germanium, along with asymmetrical distributions of co-dopants (phosphorus and boron) that produce internal stress within the fiber. Polarization-preserving fibers exhibit far higher losses compared to more conventional types of fibers, as shown in recent test results in Fig. 55 [Gira03]. This figure compares two different types of polarization-preserving fibers along with single-mode fibers made with the same dopant. The polarization-preserving fibers have a strong asymmetry in the doping concentration. The results show very similar transient absorption and recovery for single-mode and polarization-preserving fibers. Although not shown in the figure, the study also examined the polarization properties of the fibers before and after irradiation. There was little change, showing that these fibers can be used in radiation environments while maintaining their polarization properties. The ordinate shows the optical attenuation per krad, assuming a linear dependence. The damage is linear with prompt dose up to approximately 50 krad(SiO2), but becomes sublinear at higher doses. 100 SM1 SM2 PM1 PM2
10
1
After Girard, et al., 2003 RADECS Conference
0.1 10−6 10−5
10−4
10−3
10−2 10−1 Time (s)
Fig. 55. Induced loss in a polarization-preserving fiber.
III-59
100
101
102
103
VII. Optical Subsystems A. Digital Optocouplers
1. Operating Principles Optocouplers are an example of a fundamental type of optical system where we do not have the sophistication of an advanced encoding scheme and special low-noise receiver. An example of a basic single-transistor optocoupler is shown in Fig. 56. An external signal is applied to the LED, and light from the LED is transmitted to the collector of the phototransistor. Light from the LED is absorbed in the collector region, producing a photocurrent in the transistor that can then be used to generate active signals in a separate path that is electrically isolated from the electrical path used to activate the LED.
850 nm
Fig. 56. Diagram of a basic optocoupler with a single phototransistor.
The energy efficiency for this process of first converting electrical energy to light, and then back to electrical energy is typically less than 1% mainly because of the difficulties of extracting light from the LED. (The quantum efficiency for conversion within the LED is approximately 60%, but typically only a small fraction of the LED output can be coupled to the phototransistor). Optocouplers with single transistors like that shown in Fig. 57 have limited bandwidth – typically ≈ 1 MHz - because the optical collection process within the phototransistor involves diffusion, along with the practical difficulty of using a single transistor in a circuit with high bandwidth. The fundamental parameter used to characterize optocouplers is the current transfer ratio (CTR), which is the ratio of the collector current of the phototransistor to the forward current of the LED. It is analogous to transistor gain, but is much lower - typically 1 to 10 – because of the low energy transfer efficiency. There is a nearly linear relationship between LED current and output current for this type of optocoupler that depends on the transistor gain as well as the LED efficiency. Higher bandwidth can be achieved by using a high-gain amplifier circuit in place of the simple phototransistor, with a digital output stage. The designs of these circuits are proprietary, and often include a Schmidt trigger to provide hysteresis. Optocouplers with internal amplifiers may have bandwidths of 50 MHz or more. A double-heterojunction LED is required in order to acheive such high speeds because of the long response time of amphoterically doped LEDs. Unlike basic optocouplers, optocouplers with internal amplifiers have highly nonlinear transfer characteristics, as shown in Fig. 57. These types of optocouplers generally do not specify current transfer ratio because of the nonlinear relationship between LED current and output current. The electrical specifications usually define a guaranteed saturation value at the III-60
output (VOL) that is defined for a specific value of forward current in the LED. However, it is possible to make special measurements of the input current near the transition point for the amplifier, providing better information about the internal operating margin of this type of part. Those measurements are evident in the transfer characteristics shown in Fig. 57. 6
Output Voltage (V)
5
10 mA load current
4
VOL is guaranteed for IF = 10 mA
3
2.5 mA load current
2 1 0 0
1
2
3
4
5
9
10
LED Forward Current (mA) Fig. 57. Transfer characteristics of an optocoupler with an internal high-gain amplifier and digital output stage.
Table 6 lists some common types of optocouplers, including the input current, response time, and LED wavelength. In addition to permanent degradation of the LED and phototransistor or photodiode, we also have to be concerned about transients from heavy ions and protons for transistors with fast response times. However, transients are unimportant for simple phototransistor optocouplers such as the 4N49. The type of LED used within an optocoupler has a large impact on its sensitivity to displacement damage. Many optocoupler manufacturers do not specify the type of LED or the operating wavelength, and this is a major concern in selecting and qualifying optocouplers for space applications. If the type of LED is changed by the manufacturer, it may change the susceptibility of the optocoupler to displacement damage (see Fig. 30). Table 6. Properties of Some Common Types of Optocouplers Optocoupler Type
Circuitry
4N49
Single transistor
6N139
Darlington transistor
6N134 5203
LED Wavelength 870 nm
Input Current
Current Transfer Ratio
Response Time (ns)
1 mA
2
700 nm
0.5 mA
50
High-speed amplifier
700 nm
10 mA
---
50 ns
High-speed amplifier
850 nm
3 mA
---
20 ns
III-61
As long as a digital optocoupler is heavily overdriven, the effects of noise are minimized, provided the pulse width is compatible with bandwidth requirements. However, if we reduce the input current of the LED to the point where it is just able to function the output will be unstable, exhibiting noise characteristics that are similar to that of a high-gain optical receiver. Although this is not a mode that is allowed for normal optocouplers, degradation of the LED output within the optocoupler can result in conditions with marginal overdrive where the noise margin is very low. 2. Radiation Degradation Optocouplers with amphoterically doped LEDs, such as the 4N49 optocoupler, are severely degraded by displacement damage. As shown in Fig. 58, the current transfer ratio falls to 0.4 - a factor of 2.5 reduction of the initial value - at a 50-MeV proton fluence of only 1010 p/cm2. That is an extremely low fluence, corresponding to a total dose of about 1.6 krad(Si). Fluences of that magnitude can be produced by a single solar flare in space. Protons with lower energy will cause even higher degradation because NIEL increases with decreasing proton energy. Although we will not discuss electron displacement damage, displacement damage from electrons is also important for optocouplers that use this type of LED. Separate measurements were made of the gain of the internal phototransistor during the proton tests, and those results are also shown in the figure. Generally, gain degradation is insignificant at the radiation levels where this type of LED degrades, showing how LED degradation dominates the radiation response. Although not shown in the figure, tests with cobalt-60 gamma rays cause the CTR to degrade by less than 20% at 50 krad(Si), underscoring the need to consider displacement damage effects for optocouplers. 1
Parameter (normalized)
Gain
4N49 optocoupler (Micropac)
Photoresponse
IC
0.1
CTR
ID
0.01 109
1010 Proton Fluence (p/ cm2)
Fig. 59. Degradation of the current transfer ratio of the 4N49 optocoupler after irradiation with 50-MeV protons.
III-62
1011
It is interesting to contrast the 4N49 with a more modern optocoupler, the 6N139. The minimum forward current of the 6N139 is 0.5 mA, only a factor of two below that of the 4N49. However, the 6N139 is far more resistant to radiation damage, as shown in Fig. 59 [Miya02]. The current transfer ratio is nearly unchanged at radiation levels where the 4N49 degrades severely. The 6N139 is about a factor of 50 more resistant to proton damage that the older 4N49. The main reason is the different LED technology, although more efficient optical coupling also plays a role. Another feature is evident in this figure. When we examine the effect of LED input current on CTR, we see that the peak CTR shifts to higher currents. The reason for this is the reduced light output from the LED, along with the current dependence of phototransistor gain. The CTR degrades more rapidly than one would expect from LED degradation alone because the lower light output of the LED “starves” the phototransistor, changing its operating mode to lower current where the effective gain is much lower. Linear optocouplers are designed to take advantage of the current dependence of CTR on LED drive current, choosing an operating region that is well above the peak in the CTR-drive current relationship. This reduces the dependence of CTR on temperature (LED optical power decreases about 1% per degree Centigrade) and makes it possible to produce optocouplers with a narrow range of CTR. Radiation damage in these types of optocouplers is highly nonlinear, because small fluences shift the operating region to lower currents where the increased efficiency of the phototransistor compensates for the reduced light output of the LED. The degradation increases rapidly once the LED light output moves the operating region to the low-current side of the CTR-drive current relationship. 60 Pretest
Current Transfer Ratio
2 x 1010 6N139 40
4 x 1010 1 x 1011 2.5 x 1011 5 x 1011
20
0 10−6
10−5
10−4
10−3
LED Current (A) Fig. 59. Degradation of a more contemporary optocoupler with a double-heterojunction LED and far more efficient optical power coupling between the LED and phototransistor. The proton energy was 50 MeV [Miya02]
Additional work has been done by Germanicus, et al., on degradation of optocouplers at several different proton energies [Germ02], assuming that damage in the LED is the dominant mechanism. They showed good agreement with NIEL values for energies from 21 to 100 MeV.
III-63
3. Transients from Protons and Heavy Ions Low-speed optocouplers such as the 4N49 are essentially immune to transients from heavy ions or protons because of the slow response of the phototransistor. However, high-speed optocouplers such as the 6N134 are readily upset when they are exposed to heavy ions or protons. The input photodiode used in these devices has a circular geometry with a diameter of about 500 µm. This is far larger than the area of typical logic or analog devices that are sensitive to single-event upset effects. Consequently, direct ionization from protons – a phenomenon that is usually ruled out because the LET of protons is on the order of 10-2 MeV-cm2/mg - plays a significant role in the response of these devices. The earliest measurements of single-event upset effects considered upset from heavy ions. Those results showed that the threshold LET for upset was about 0.3 MeV-cm2/mg, a low threshold LET, but not sufficiently low to be concerned about direct ionization from protons. Fig. 60 shows the cross section for upset from heavy ions in the 6N134 optocoupler. The cross section is significantly less than the area of the photodiode near threshold, and increases gradually as the LET increases. At first this is difficult to understand, because one would expect the cross section to quickly rise to the area of the photodiode. However, modeling results show that charge from ions striking the photodiode away from the center partially recombines before it can be collected. This is the reason for the gradual increase in cross section. Note that the cross section is greater than the diode area for LET values above 10 MeV-cm2/mg. Two mechanisms are involved that contribute to the larger effective area: (1) charge deposited outside the diode area can be collected by diffusion, and (2) single-event transients in the amplifier. These mechanisms produce different pulse widths (amplifier transients produce transients with extended pulse width), allowing them to be distinguished from the basic nature of the output transient of the optocoupler. 0.004
Cross Section (cm 2 )
6N134 Optocoupler 0.003
0.002
60 µm range
0.001
Diode area
39 µm range
0 0
20
10
LET (MeV-cm 2 /mg) Fig. 60. Upset cross section of the 6N134 optocoupler from heavy ions.
III-64
30
The 6N134 optocoupler is also sensitive to upset from protons, which is not unexpected because of the very low threshold LET. However, when proton tests were done using incident angles above 80º the cross section was observed to increase by about a factor of six [LaBe97]. This result was difficult to explain, because one would normally assume that the cross section would approach the physical area of the diode. The relatively small increase in cross section along with the gradual nature of the change in cross section with angle seemed to rule out direct ionization from protons as the underlying mechanism. Furthermore, the effect of this increase on the upset rate was not very important because it required extreme angles before the cross section increased significantly. A later study using protons with different energies showed that the cross section of the 6N134 increased by as much as three orders of magnitude when lower energy protons were used, as shown in Fig. 61 [John99]. The smooth increase in cross section with increasing angle of incidence was explained by a detailed model showing the effects of straggling on the charge collection, and that the angular dependence was due to the combined effects of direct ionization from the proton when it traveled through the large-diameter detector, and charge from the proton recoil when it interacted with a lattice atom before completely traversing the detector. Larger incident angles allow reactions with smaller amounts of recoil energy to cause the optocoupler to upset. With lower energy protons, the cross section is much larger and begins to increase at lower incident angles. Calculations using the model developed for that effect showed that the expected error rate increased by a factor of about 4 in a typical earth orbit when the angular dependence was taken into account compared to the error rate calculated for a proton cross section that was independent of incident angle.
10-3 HP6N134 Optocoupler Load current 1 mA
10-4
15 MeV
10-5
30 MeV
10-6
50 MeV 10-7 10-8
0
10
20
30
40
50
60
70
80
90 100
Angle wrt Proton Beam (degrees) Fig. 61. Cross section for upsets in a 6N134 optocoupler for various incident angles during proton irradiation.
III-65
This example shows the extreme sensitivity of some types of optocouplers to transients from protons or ions with very low LET. The effect can be mitigated at the circuit level, provided that transients are taken into account during system design. However, system errors have occurred in some space system when the effect was not taken into account [LaBe97]. The large sensitivity to proton upsets increases its importance in earth-orbiting systems, even for low-altitude spacecraft because of the south Atlantic anomaly. B. Optical Receivers
1. Operating Principles The basic concept of an optical receiver is straightforward: an optical pulse is detected by an input amplifier and then used to generate pulses with a specific coding scheme. However, the presence of noise along with the detected pulse means that there is a finite probability for errors to occur during the detection process. The effect of noise is usually considered by means of a bit error rate, which is the fractional number of errors that occur in the receiver during a transmitted sequence of pulses. For acceptable performance, the bit error rate has to remain very low. For highly sensitive receivers, assuming a Gaussian distribution for noise, the noise and signal level are related as shown in the equation:
BER =
1 erfc 2
⎡ (SN ) 1 / 2 ⎤ ⎢ ⎥ ⎢⎣ 2 2 ⎥⎦
(15)
where BER is the bit error rate, SN is the signal-to-noise ratio, and erfc is the complementary error function. Fig. 62 shows how the bit error rate changes as the signal-to-noise ratio is reduced. When the SN drops below 20 dB, the bit error rate increases rapidly. A common specification for bit error rate is 10-9, which requires a signal-to-noise ratio of 21.6 dB. We will use that criterion in discussing the performance of optical receivers.
Signal to Noise Ratio (dB)
25
20 SNR = 21.6 dB for 10−9 BER 15
10
5
0 −15
−10 −5 Log [Bit Error Rate]
Fig. 62. Dependence of bit error rate on signal-to-noise ratio.
III-66
0
Special modulation techniques such as frequency shift keying, phase modulation, and coherent detection improve the theoretical limit for errors, but the improvements are relatively slight. Those methods are used in long-haul fiber optic systems (such as undersea cables), but are usually not required in systems used in space that use short cables, and generally have higher optical power margins. Most optical receivers are designed to reject noise by means of several sophisticated circuit elements, integrating the signal during each clock cycle [Brai85]. Figure 63 shows this type of receiver, along with the two dominant noise sources, the detector and amplifier. A transimpedance amplifier is typically used that responds to the pulse of current through the detector, but keeps the voltage constant. Integrated receivers can be designed that reduce parasitic capacitance to improve performance at high data rates [Naka91].
Clock Optical pulses
noise Detector
noise Amplifier
Gated Integrator
Shift Register
Logic
ADC
Fig. 63. Block diagram of an optical receiver where the optical signal is integrated during each bit period.
When optical receivers are near their detection limits, noise is clearly evident during each bit cycle. A standard way to examine this is by means of an eye diagram, which is the result of a continuous sampling of the signal over many bit periods. Fig. 64 shows an example. For high signal-to-noise ratios the bit pattern is very much like an ideal pulse train. As the signal-to-noise ratio is reduced, noise becomes evident during each pulse transition and the open regions between the upper and lower logic states “close” with a substantial amount of noise between levels (hence the term eye diagram).
III-67
Detection Threshold
Fig.64. Eye diagram of a the pulses detected after the input amplifier of a sensitive optical receiver. The detection threshold is in the center of the mean pulse amplitude.
In order for an optical receiver to operate properly, the margin between the input optical power and the power detected by the receiver must be large enough to meet the bit rate requirement as well as to account for aging, fiber coupling losses, temperature, and other factors that may cause the internal margin to change over time. For example, the optical power output of LEDs and laser diodes decreases with time, particularly for components that are continually operated, gradually decreasing the BER. We have seen that the signal-to-noise ratio must be on the order of 20 dB or more in order to meet typical bit error requirements. Fig. 65 shows how the link margin is related to the bit rate, the type of optical power source, and the type of detector. The link margin decreases with increasing bit rate because the integration period decreases as bit rate increases (there are fewer photons available during each pulse cycle). Thus, it is clearly more challenging to design receivers at high bit rates. Several tradeoffs must be considered when designing an optical link. More optical power is available from lasers compared to LEDs, increasing the margin when a laser source is used, even though it is more difficult to use lasers because of the need for thermal control and closed-loop current drive circuits. Using an avalanche photodiode increases the margin in the receiver by about 6 dB, but adds more complexity. The overall system design must consider those factors, along with aging and wearout in these more advanced components. Recent work has demonstrated the feasibility of operating fiber optic links above 1-Gb/s using 890 nm LEDs [Akbu01], showing that it is possible to extend the bit-rate to very high frequencies without using laser diodes.
III-68
10 Laser diode source
0
LED source
Power Level (dBm)
−10 −20 −30 −40
Link Margin
n p−i− APD
−50
Coh
−60
ere
ete nt d
n ctio
−70 −80 −90 106
107
108
109
1010
Bit Rate (s−1) Fig. 65. Link margin for optical receivers. Note the decrease in link margin as the bit rate increases.
Optical power and receiver power in these types of receivers are usually described using the unit dBm. This is the power delivered to a specified load resistance, nearly always 50 ohms. With that definition, 0 dBm corresponds to 1 mW of power into 50 ohms; that corresponds to a voltage of 0.2236 volts (rms). The power changes by 10 dB for each factor of ten reduction, but is always referenced to the assumed load of 50 ohms. Radiation Effects in Optical Receivers The earlier example of transient upset in optocouplers demonstrates the importance of charge collected in the input photodiode in generating transients in optical amplifiers. Optocouplers operate with very high signal-to-noise ratios, but nevertheless are extremely sensitive to upset effects. The large area and extended charge collection depth of the silicon detectors used in those devices are the main factors in their high sensitivity. Receivers that are designed for radiation applications usually use direct-bandgap detectors with reduced charge collection depth in order to reduce the collection volume for radiation-induced charge. Fig. 66 shows results from a test of a hardened optical receiver operating at a bit rate of 80Mb/s during irradiation with 59-MeV protons [Facc02]. This receiver used a 1.3 µm laser source with an InGaAs p-i-n detector. The active depth of the detector was approximately 2 µm. The thin active layer reduces the collection volume for transients from the proton beam.
III-69
Bit-Error Cross-Section (cm 2 )
10−5
90o
10−6
80o
10−7
20o
45o 10o
10−8 10−9 10−10
59 MeV protons
10−11 −30
−25
−20
−15
−10
Signal Amplitude (dBm) Fig. 66. Bit error cross section for an 80-Mb/s optical receiver irradiated with protons.
The proton beam flux rate was between 1 and 4 x 108 protons/(cm2-s) during these tests, which is considerably higher than typical flux rates in a space environment. This receiver will be used as part of the Large Hadron Collider at CERN. When high optical power levels were used, the bit error rate remained below 10-9 even with extreme angles of incidence. The cross section for bit errors increased to nearly 10-5 cm2 when the input signal was reduce to –30 dBm, the minimum level required for a BER of 10-9 when there was no radiation present. Different angles of incidence were used during the tests. For angles of 45 º and higher the error rate increased sharply with low optical power levels. This was attributed to direct ionization from protons. This is in general agreement with our expectation that receivers are relatively immune to upset from radiation-induced transients as long as the optical power level is well above the minimum level required without the presence of radiation. This data was used to develop a model for the bit-error rate in the proposed application. The results of the modeling are shown in Fig. 67. They indicate that acceptable performance is expected provided the signal level remains above –25 dBm. This provides significant margin even when considering degradation of components within the receiver from increasing levels of proton irradiation.
III-70
59 MeV protons
10−2
Dashed line shows measured BER in the absence of the proton beam
BER
10−4 10−6
Radiation increases BER for signals > -30 dBm
10−8 10−10 10−12
After Faccio, et al., TNS 2002
−35
−30 −25 −20 Signal Amplitude (dBm)
−15
−10
Fig. 67. Predicted error rate for 80-Mb/s receiver.
An additional optical link has developed for applications at the CERN detector that uses a silicon detector and a VCSEL light source operating at 840 nm. Radiation tests have been done on that optical link using 24-GeV protons and pions with energies above 350 GeV [Greg02]. That optical link has acceptable performance in the anticipated environment, even though it uses a detector with much greater charge collection depth. An optical receiver operating at even higher bit rates was subjected to radiation tests by a different laboratory [Mars94]. That receiver could operate at bit rates up to 1-Gb/s, and also used a wavelength of 1300 nm. The input to the receiver was provided by a laser diode. The receiver used an InGaAs photodiode with a diameter of 80 µm and an active thickness of 3 µm. Radiation tests were done with 63-MeV protons and 18-MeV alpha particles. The optical power was changed during the tests to determine how the error rate depended on input power. Fig. 68 shows the results of their tests for two specific angles of incidence. The error rate clearly increased when the bit rate was increased, consistent with the basic concept of link margin shown earlier in Fig. 66. When tested with protons, the error cross section decreased by several orders of magnitude when the optical power was increased. This was attributed to direct ionization from protons, which was sufficiently high at low optical power to contribute to the cross section, but not high enough when the optical power was increased. Results at other angles in their paper corroborated that interpretation. When the link was tested with alpha particles, the error cross section changed very little as the optical input power was changed. This flat dependence occurs because alpha particles induce charge by direct ionization, not by nuclear reaction products, and the charge in the detector from 18-MeV alpha particles was well above the threshold condition for the receiver. One would expect that the optical power level would have a larger effect on the cross section even with alpha particles if the LET were lower, near the minimum sensitivity. However, it would not decrease nearly as rapidly as for protons because direct ionization is still the dominant mechanism, unlike protons where indirect reactions also contribute to the cross section. III-71
10−3 ETX75
Error Cross-Section (cm 2 )
18 MeV He Ions at 70o 10−5
10−7
10−9 10−11 −30
63 MeV p+ at 50o
1000 MHz 400 MHz 200 MHz −25
After Marshall, et al.,TNS, 1994
−20
−15
−10
Optical Power (dBm) Fig. 68. Error cross section of a high-speed optical data bus operated at various clock rates [Mars94].
It is clear from these examples that errors produced by optical receivers operating at high bit rates are dominated by charge produced by the radiation source in the detector. The bit rate is far lower when the receiver operates with high optical power levels, where the signal during each pulse integration period is well above the minimum required signal level. Degradation of the light source, the detector or the coupling efficiency during extended space missions will reduce the link margin, increasing the error rate. Thus, it is important to make sure that the design considers degradation in link performance in order to maintain satisfactory operation. Errors may also be caused by upsets in the amplifier, integrator, or other components used in high-performance receivers. In some cases those errors may be more important than single-bit errors from “noise” in the detector because they may persist for many clock cycles, producing strings of errors that may be more difficult to accommodate in system design.
III-72
VIII. Summary
This section of the 2004 NSREC Short Course has reviewed the operating principles of photonic sources, optical fibers and optical receivers, emphasizing the underlying physics as well as the physical structure of these devices along with the mechanisms that cause them to respond to radiation. In addition to understanding the operation of these devices, we have to be concerned with two very different effects in space: permanent damage from displacement and ionization damage, and transient effects from heavy ions and energetic protons. Permanent Damage In active devices – optical sources and detectors – permanent damage is generally dominated by displacement damage from protons and electrons, requiring tests with these types of particles instead of (or in addition to) cobalt-60 gamma rays. Displacement damage from protons and electrons depends on energy, and that dependence has to be taken into account in performing radiation tests and in interpreting test results for space applications. For III-V devices, there are discrepancies between theoretical and experimental results for proton energies above 50 MeV that can affect the interpretation of test results. The sensitivity of light-emitting diodes to damage from protons varies over an extremely wide range. The most sensitive LEDs are severely degraded at a fluence of 1010 p/cm2, which is well below the radiation requirement of many earth-orbiting space systems. Those types of devices are used in mainstream optocouplers, causing the optocouplers to degrade very severely at low radiation levels. Other types of LEDs use different fabrication techniques, and can withstand radiation levels that are two to three orders of magnitude higher than amphoterically doped LEDs. When more advanced LEDs are used, degradation in the LEDs is often negligible, and the main system concern then becomes degradation of optical detectors, which are also sensitive to displacement damage. Optical fibers are affected by ionization damage, not by displacement damage effects, and consequently tests of optical fibers can be done using cobalt-60 gamma rays or pulsed X-rays. Most optical fibers have relatively low loss per unit length, so that radiation-induced absorption in optical fibers is usually important only for fiber lengths > 10 m, or for applications where the fiber is required to function at low temperature. Transient Effects We also have to be concerned with transient effects from protons and heavy ions. Most optoelectronic systems are very sensitive to transient effects, and will upset readily from protons (through indirect processes). The optical detector is the key element in these systems. Selecting detectors with small area and thin active regions improves radiation hardness. Thick detectors – such as those used in optocouplers- can be sensitive to direct ionization from protons as well as indirect particles through nuclear reactions, causing the results to depend strongly on particle energy and the incident angle used during testing, and causing high error rates in space applications. Tests of optical receivers show many of the same characteristics exhibited by transient tests of optocouplers, but are more difficult to interpret. The key point is that the error rate in a radiation environment is directly related to the optical receiver bit rate and the optical power. Receivers that operate with high optical power – which implies high margin between minimum sensitivity and the actual operating point – can be designed that show only slight changes in BER when they are exposed to moderate flux levels, but the designs are more challenging at high bit rate. Tests III-73
done with reduced input power show a large increase in BER compared to tests done with nominal operating power levels. By choosing detectors with small thicknesses it is possible to use optical receivers in the radiation environments that are encountered in many types of spacecraft with relatively small increases in bit error rate. Radiation-tolerant optical receivers have even been designed to withstand the more demanding radiation environment in the Large Hadron Collider detector at CERN. Future Trends We have limited this section of the course to older technologies where extensive radiation test results exist, and there is a relatively mature understanding of the mechanisms involved. However, photonic components continue to advance. For example, new designs have been demonstrated that incorporate moveable MEMS-based elements for tunable laser diodes [Vail97]. New applications have been proposed for optical interconnects that use arrays of VSCELs and detector/amplifiers [Olli02] that will use far more compact elements compared to the mainstream components discussed earlier, with reduced operating margins. Hopefully the material presented here will provide sufficient background and understanding to make first-order estimates of potential radiation issues for such technologies as well as providing the background that is needed to understand how radiation affects conventional photonic devices when they are used in spacecraft.
References [Akbu01] M. Akbulut, et al., “Digital Communications above 1-Gb/s Using 890-nm Surface-Emitting LightEmitting Diodes,” IEEE Phot. Tech. Lett., 13(1), pp. 85-87 (2001). [Barr95] A. L Barry, et al., “The Energy Dependence of Lifetime Damage Constants in GaAs LEDs for 1-500 MeV Protons,” IEEE Trans. Nucl. Sci., 42(6), pp. 2104-2107 (1995). [Bart97] J. L. Barth, Part II of the Short Course presented at the 1997 Nuclear and Space Radiation Effects Conference. [Bart03] J. L. Barth, C. S. Dyer and E.G. Stassinopoulos, “Space, Atmospheric and Terrestrial Radiation Environments,” IEEE Trans. Nucl. Sci., 50(3), pp. 466-482(2003). [Beck03] H. N. Becker, T. F. Miyahira and A.H. Johnston, “The Influence of Structural Characteristics on the Response of Silicon Avalanche Photodiodes to Proton Irradiation,” accepted for publication in IEEE Trans.Nucl. Sci., 50(6), pp. 1974-1981 (2003). [Brai85] M. Brain and T.-P. Lee, “Optical Receivers for Lightwave Communication Systems,” IEEE J. Lightwave Tech.., 3(6), pp. 1281-1300 (1985). [Choi04] H. W. Choi, C. W. Jeon and M. D. Dawson, “InGaN Micro-ring Light-Emitting Diodes,” IEEE Phot. Tech. Lett., 16(1), pp. 33-35 (2004). [Choq97] K. D. Choquette and H. Q. Hou, “Vertical-Cavity Surface Emitting Lasers: Moving from Research to Manufacturing,” Proc. of the IEEE, 85(11), pp. 1730-1739 (1997). [Chua95] S. L. Chuang, Physics of Optoelectronic Devices, Wiley Interscience: New York, 1995. [Cold95] L. W. Coldren and F. W. Corzine, Diode Lasers and Photonic Integrated Circuits, New York: Wiley, 1995. [Cole00] J. J. Coleman, ”Strained-Layer InGaAs Quantum-Well Heterostructure Lasers,” IEEE J. Sel. Topics in Quant. Elect., 6(6), pp.1008-1013 (2000). [Comp67] D. M. J. Compton and R. A Cesena, “Mechanisms of Radiation Effects on Lasers,” IEEE Trans. Nucl. Sci., 14(6), pp. 55-61 (1967). III-74
[Devi83] N. Devine and H. B. Garrett, “Charged Particle Distributions in Jupiter’s Magnetosphere,” J. Geo. Phys. Res., 88(A9), pp.6889-6903 (1983). [Diec01] E. Diechsel, et al., “High Power Laser Diodes with Dry-Etched Mirror Facets and Integrated Monitor Photodiodes,” IEEE J. on Selected Topics in Quant. Elect., 7(2), pp.106-110 (2001). [Dodd03] P. E. Dodd and L. W. Massengill, “Basic Mechanisms and Modeling of Single-Event Upset in Digital Microelectronics,” IEEE Trans. Nucl. Sci., 50(3), pp. 583-602 (2003). [Evan93] 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 Broadband LED,” IEEE Trans. Nucl. Sci. 40(6), pp. 16451654 (1993). [Facc01] F. Faccio, et al., “Single-Event Upset Tests of an 80-Mb/s Optical Receiver,” IEEE Trans. Nucl. Sci., 48(5), pp. 1700-1707 (2001). [Fehs02] R. Fehs, et al., “A Quantitative Study of Radiative, Auger, and Defect Related Recombination Processes in 1.3-µm GaInNAs-Based Quantum-Well Lasers,” IEEE J. Sel. Top. In Quant. Elect., 8(4), pp. 801-810 (2002). [Frieb88] E. J. Friebele, et al., “Interlaboratory Comparison of Radiation-Induced Attenuation in Optical Fibers. Part I: Steady-State Exposures,” IEEE J. Lightwave Tech., 6(2), pp. 165-171 (1988). [Germ02] R. Germanicus, et al., “Analysis of the Proton-Induced Permanent Degradation in an Optocoupler,” IEEE Trans. Nucl. Sci. 49(3), pp. 1421-1425 (2002). [Gilb82] R. M. Gilbert, “Photobleaching of Radiation-Induced Color Centers in a Germania-Doped Fiber,” IEEE Trans. Nucl. Sci., 29(6), pp.1484-1488 (1982). [Gill00]
K. Gill, et al., “Radiation Damage and Annealing in 1310 nm InGaAsP/InP Lasers,” 2000 IEEE Radiation Effects Data Workshop, pp.153-157.
[Gira03] S. Girard, et al., “Pulsed X-Ray and γ-Ray Irradiation Effects on Polarization-Maintaining Optical Fibers,” presented at the 2003 RADECS Conference, Noordwijk, The Netherlands, Sept. 15-19, 2003. [Grig03] I.-M. Grigor, “Radiation-Tolerant Optical Links for the ATLAS Inner Detector,” IEEE Trans. Nucl. Sci., 49(3), pp. 1114-1116 (2003). [Iga00]
K. Iga, “Surface-Emitting Laser – Its Birth and Generation of New Optoelectronics Field,” IEEE J. Selected Topics in Quant. Elect., 6(6), pp. 1201-1215 (2000).
[John98] A. H. Johnston, et al., “Single-Event Upset in Optocouplers,” IEEE Trans. Nucl. Sci., 45(6), pp. 28672875 (1998). [John99] A. H. Johnston, et al., “Angular and Energy Dependence of Proton Upset in Optocouplers,” IEEE Trans. Nucl. Sci., 46(6), pp. 1660-1665 (1999). [John00] A. H. Johnston and T. F.Miyahira, “Characterization of Proton Damage in Light-Emitting Diodes,” ,” IEEE Trans. Nucl. Sci., 47(6), pp. 2500-2507 (2000). [John01] A. H. Johnston, T. F. Miyahira and B. G. Rax, “Proton Damage in Advanced Laser Diodes,” IEEE Trans. Nucl. Sci., 48(6), pp. 1764-1772 (2001). [John02] A. H.. Johnston and T.F. Miyahira, “Energy Dependence of Proton Damage in Optical Emitters,” IEEE Trans. Nucl. Sci., 49(6), pp. 1426-1431 (2002). [John03] A. H. Johnston, “Radiation Effects in Light-Emitting and Laser Diodes,” IEEE Trans. Nucl. Sci., 50(3), pp. 689-703 (2003). [LaBe97] K. LaBel, et al., “Proton-Induced Transients in Optocouplers: In-Flight Anomalies, Ground Irradiation Tests, Mitigation and Implications,” IEEE Trans. Nucl. Sci., 44(6), pp. 1895-1902 (1997). [Lede00] N. N. Ledenstov, et al., “Quantum-Dot Heterostructure Lasers,” IEEE J. Sel. Topics in Quant. Elect., 6(3), pp. 439-451 (2000). [Liu92]
Y. Liu, et al., “A Planar InP/InGaAs Avalanche Photodiode with Floating Guard Ring and Double Diffused Junction,” IEEE J. Lightwave Tech., 10(2), pp.182-193(1992). III-75
[Loo81]
R. Loo, R. C. Knechtli and G. S. Kamath, “Enhanced Annealing of GaAs Solar Cell Damage,” in Proc. of the 15th IEEE Photovoltaic Specialties Conference, vol. 33, 1981.
[Mars94] P. W. Marshall, C. J. Dale, M. A. Carts and K. A. LaBel, “Particle-Induced Bit Errors in High Performance Fiber Optic Data Links for Satellite Management,” IEEE Trans. Nucl. Sci., 41(6), pp. 19581965(1994). [Mazu02] J. Mazur, Part II of the Short Course presented at the 2002 Nuclear and Space Radiation Effects Conference. [Miya02] T. F. Miyahira and A. H. Johnston, “Trends in Optocoupler Radiation Degradation,” IEEE Trans. Nucl. Sci., 49(6), pp. 3009-3015 (2002). [Naka91] K. Nakagawa and K. Iwashita, “High-Speed Optical Transmission Systems Using Advanced Monolithic IC Technologies,” IEEE J. on Sel. Areas in Communications, 6(5), pp. 683-688 (1991). [Olli02]
E. Ollier, “Optical MEMS Devices Based on Moving Waveguides,” IEEE J. Sel. Topics in Quant. Elect., 8(1), pp.155-162 (2002).
[Pani76] M. B. Panish, “Heterostructure Injection Lasers,” Proc. of the IEEE, 64(10), pp. 1512-1540 (1976). [Park00] G. Park, et al., “Low-Threshold Oxide-Confined 1.3-µm Quantum-Dot Laser,” Phot. Tech. Lett., 13(3), pp. 230-232 (2000). [Pank76] J. I. Pankove, Optical Processes in Semiconductors, Prentice-Hall, Englewood Cliffs, NJ and Dover, New York (1976) [Rax96]
B. G. Rax, et al., “Total Dose Effects and Hardness Assurance for Optocouplers,” IEEE Trans. Nucl. Sci., 43(6), pp. 3145-3150 (1996).
[Raze00] M. Razeghi, “Optoelectronic Devices Based on III-V Compound Semiconductors Which Have Made a Major Scientific and Technological Impact in the Past 20 Years,” IEEE J. on Selected Topics in Quant. Elect., 6(6), pp. 1344-1354 (2000). [Ree00]
R. A. Reed, et al., “Energy Dependence of Proton Damage in AlGaAs Light-Emitting Diodes,” IEEE Trans. Nucl. Sci., 47(6), pp. 2492-2499 (2000).
[Rose82] B. H. Rose and C. E. Barnes, “Proton Damage Effects on Light Emitting Diodes,” J. Appl. Phys., 53(3), pp. 1772-1780 (1982). [Rose95] T. S. Rose, M. S. Hopkins and R. A. Fields, “Characterization and Control of Gamma and Proton Radiation Effects on the Performance of Nd:YAG and Nd:YLF Lasers,” IEEE J. Quant. Elect., 31(9), pp. 1593-1602 (1995). [Rupp66] H. Rupprecht, et al., Appl. Phys. Lett., 9, pp. 221 (1966). [Schö97] H. Schöne, et al., “AlGaAs Vertical-Cavity Surface-Emitting Laser Responses to 4.5-MeV Proton Irradiation,” IEEE Phot. Tech. Lett., 9(12), pp. 1552-1554, 1997. [Selm01] S. R. Selmic, et al., “Design and Characterization of AlGaInAs-InP Multiple-Quantum-Well Lasers,” IEEE J. on Selected Topics in Quant. Elect., 7(2), pp. 340-349 (2001). [Sext03] F. W. Sexton, “Destructive Single-Event Effects in Semiconductor Devices and IC’s,” IEEE Trans. Nucl. Sci., 50(3), pp. 603-621 (2003). [Srou88] J. R. Srour and J.M. McGarrity, “Radiation Effects on Microelectronics in Space,” Proc. IEEE, 76, pp. 1433-1469, Nov. 1988. [Srou03] J. R. Srour, C. J. Marshall and P. W. Marshall, “Review of Displacement Damage Effects in Silicon Devices,” IEEE Trans. Nucl. Sci., 50(3), pp. 653-670 (2003). [Summ87] G. P. Summers, et al., “Correlation of Particle-Induced Displacement Damage in Silicon,” IEEE Trans. Nucl. Sci., 34(6). pp. 1134-1141 (1987). [Summ93] G. P. Summers, et al., “Damage Correlations in Semiconductors Exposed to Gamma, Electron and Proton Irradiation,” IEEE Trans. Nucl. Sci., 40(6). pp. 1372-1379 (1993).
III-76
[Swan87] E. A. Swanson, E. R. Arnau and F. D.Walther, “Measurements of Natural Radiation Effects in a LowNoise Avalanche Photodiode,” IEEE Trans. Nucl. Sci., 34(6), pp. 1658-1661 (1987). [Sun03] D. Sun, et al., “Sub-mA Threshold 1.5 µm VCSELs with Epitaxial and Dielectric DBR Mirrors,” Phot. Tech. Lett., 15(12), pp.1677-1679 (2003). [Sige81] G. H. Sigel, et al., “An Analysis of Photobleaching Techniques for the Radiation Hardening of Fiber Optic Links,” IEEE Tran. Nucl. Sci., 28(6), pp. 4095-4101 (1981). [Swif03] G. M. Swift, et al., “Annealing Displacement Damage in GaAs: Another Galilio Success Story,” IEEE Trans. Nucl. Sci., 50(6), pp. 1991-1997 (2003). [Tay901] E. W. Taylor, et al., “Interlaboratory Comparison of Radiation-Induced Attenuation in Optical Fibers. Part II: Steady-State Exposures,” IEEE J. Lightwave Tech., 8(6), pp. 967-976, 1990. [Tay902] E. W. Taylor, et al., “Interlaboratory Comparison of Radiation-Induced Attenuation in Optical Fibers. Part III: Transient Exposures,” IEEE J. Lightwave Tech., 8(6), pp. 977-989, 1990. [Vail97] E. C. Vail, et al., “High Performance and Novel Effects of Micromechanical Tunable Vertical-Cavity Lasers,” IEEE Select. Topics Quantum Electron., 3(2), pp. 691-697 (1997). [Vurg97] I. Vurgaftman and J. R. Meyer, “Effects of Bandgap, Lifetime, and Other Nonuniformities on Diode Laser Thresholds and Slope Efficiencies,” IEEE J. Sel. Top.in Quant. Elect., 3(2), pp. 475-484 (1997). [Walt01] R. J. Walters, et al., “Correlation of Proton Radiation Damage in InGaAs-GaAs Quantum-Well LightEmitting Diodes,” IEEE Trans. Nucl. Sci., 48(6), pp. 1773-1775 (2001). [Weat03] T. R. Weatherford and W. T. Anderson, Jr., “Historical Perspective on III-V Devices,” IEEE Trans. Nucl. Sci., 50(3), pp. 704-710 (2003). [West88] R H. West, “A Local View of Radiation Effects in Fiber Optics,” IEEE J. Lightwave Tech., 6(2), pp. 155164 (1988). [Witt96] V. Wittpahl, et al., “A Degradation Model for the Light Output of LEDs Based on Cathodoluminescence Signals and Junction Ideality Factor,” Proc. 1996 International. Reliability Physics Conf., pp.188-194. [Vand97] D. A. Vanderwater, et al., “High-Brightness AlGaInP Light Emitting Diodes,” Proc. IEEE, 85(11), pp. 1752-1764 (1997).
III-77
2004 IEEE NSREC Short Course
Section IV
Optical Detectors and Imaging Arrays
Terrence S. Lomheim The Aerospace Corporation
OPTICAL DETECTORS AND IMAGING ARRAYS ♦
+
+
Terrence S. Lomheim , Arne H. Kalma , Donald E. Romeo , +
+
+
Joseph R. Srour , Warren F. Woodward , and Tracy E. Dutton +
1.0 2.0
3.0
4.0
5.0
♦
The Aerospace Corporation, SAIC
Overview of Optical Detectors and Imagers Overview of Space Radiation Environments of Interest 2.1 Introduction 2.2 Total Accumulated Dose Modeling using AP8 and AE8 2.3 South Atlantic Anomaly 2.4 Modeling the Proton Environment Using the Combined Release and Radiation Effects Satellite (CRRES) Database 2.5 Spacecraft Shielding and Radiation Transport to the Focal Plane Radiation Effects on Visible Imaging Arrays 3.1 Introduction and Overview 3.2 Visible Imager Technologies 3.3 Displacement Damage Effects 3.3.1 Dark Current 3.3.1.1 Mean Dark Current 3.3.1.2 Single Particle Effects, Dark Current Distributions, and Hot Pixels 3.3.1.3 Field Enhancement Effects 3.3.1.4 Random Telegraph Signals 3.3.2 Charge Transfer Efficiency 3.3.3 Displacement Damage Annealing 3.4 Ionizing Radiation Effects on Visible Imagers 3.5 Visible Array Hardening Approaches 3.6 Current Trends for Visible Imagers Radiation Effects on Infrared Detector Arrays 4.1 Introduction 4.2 Infrared Detector Technologies 4.3 Ionization-induced Transient Response 4.3.1 Ionization-induced Pulses and Noise 4.3.2 Low-dose-rate Effects 4.3.3 Prompt-pulse-induced Transients 4.4 Permanent Degradation 4.4.1 Displacement Damage Effects 4.4.2 Total-dose-induced Degradation 4.4.2.1 HgCdTe Devices with ZnS and Other Passivations Susceptible to Total-Dose Exposure 4.4.2.2 Hardened HgCdTe Devices with CdTe Passivation 4.4.2.3 Total-dose Effects in ROICs 4.5 Summary and Future Directions Modeling of Radiation-induced Transients in Focal-plane Arrays
IV-1
5.1
6.0
7.0 8.0 9.0
Simulation of Proton Transient Effects 5.1.1 Proton Environment Synthesis at the Focal Plane 5.1.2 Proton Energy Transfer to Semiconductor Focal Plane 5.1.3 Focal Plane Response Model and its Validation 5.1.3.1 Focal-plane Proton Ionization Ray-Tracing Details 5.1.3.2 Sensor Architecture Modeling 5.2 Simulation Results 5.3 Updates to Proton Effects Modeling and Recent Transient Simulations 5.4 Impact of Electrons, Shielding, and Bremsstrahlung Radiation Effects on MOS Readout Integrated Circuits 6.1 Background Information 6.2 Total Dose Hardening of Visible Imager ROIC Designs 6.3 Approaches to ROIC Design for Infrared Imagers 6.4 Infrared ROIC Circuit Design for Total Dose Hardness 6.5 ROIC Hardening by Design for Single Event Effects 6.5 Conclusions Summary and Concluding Remarks Acknowledgements References
IV-2
1.0
Overview of Optical Detectors and Imagers
Advances in visible and infrared focal-plane technology over the past two to three decades have enabled spectacular improvements in the imaging capability (i.e. better resolution, large field-of-view, enhanced sensitivity) of space-based electro-optical (EO) sensors. Advanced space-based EO sensor systems have been developed with increasingly enhanced performance capabilities. These systems span the ultraviolet to long-wavelength regimes, cover a wide variety of spatial resolution and area coverage capabilities, include scanning and staring instrument designs, and can be tailored for panchromatic, multispectral, and hyperspectral imaging applications. Key system drivers for virtually all space-based EO sensor designs include focal-plane lineor frame-rate flexibility, sensitivity, dynamic range, pixel size and overall array size. For these advanced applications the focal plane has significant system leverage due to the impact, assuming constant system performance, of sub-par focal-plane performance on masssensitive EO instrument design parameters such as optical telescope aperture diameter, required integration time, and focal-plane cooling. While the performance attributes and characteristics of a given sensor are key to its intended application, many applications also require coverage of extensive portions of the earth’s surface and/or the sky on prescribed and often short timescales. Such data may be acquired using multiple copies of a given sensor populating a carefully chosen orbital constellation. With constellation coverage there is a choice of using fewer of these sensors based at higher altitudes or more of them at lower altitudes. For equivalent system performance, the higheraltitude sensors must have larger optical apertures but need only relatively small fields-ofview, whereas for lower-altitude constellations the sensors have smaller apertures but require wider fields-of-view. It is in the area of constellation selection/sensor platform altitude determination where the impact of space radiation effects on focal planes is most apparent. The natural space radiation environment around the earth is structured both spatially and temporally due to the existence of the Van Allen belts. There are orbital regimes wherein the natural radiation environment is stressing and other regimes where it is relatively benign (i.e., above and below the Van Allen belts). The lack of radiation hardness of the focal-plane technology may preclude the use of a constellation deemed superior due to strong economic or operational advantages. There is strong motivation for focal-plane technologies to be able to operate in moderate to severe radiation environments over the mission lifetime. Therefore, this tutorial is aimed at providing both an overview of the limitations of visible and infrared focal-plane technology when subjected to the wide variety of space radiation effects and a summary of promising advances in the state-of-the-art of radiation-hardened focal-plane technology. In the next section, a brief qualitative overview is provided of the radiation environment with which array designers must contend for space applications. Subsequent sections cover radiation effects on visible and infrared focal-plane arrays, modeling of radiation-induced transients in arrays, and radiation effects on readout integrated circuits.
IV-3
2.0
Overview of Space Radiation Environments of Interest
2.1
Introduction
Several excellent reviews of the natural space radiation environment have appeared in recent years ([1]–[5] and references therein). The environment information provided in those reviews is generally applicable to focal planes and their associated electronics. The natural radiation environment of concern for earth-bound space applications from low earth orbit (LEO) through geosynchronous orbit (GEO) is comprised of (1) protons, electrons and heavy ions trapped by the earth’s magnetic field (i.e. the Van Allen belts), (2) protons emitted from semi-periodic solar flares, and (3) galactic cosmic rays. Item (3) is of great concern for single-event effects (e.g., upset and latchup) in space electronics, and substantial research has been carried out in that area over the past several decades. Our primary interest for earthbound space-based sensors that use advanced focal-plane technology is with particle radiation – primarily due to energetic protons that survive penetration through expected amounts of shielding to the level of the focal plane. For interplanetary applications, different considerations would prevail. For example, in the vicinity of the Jovian moons (NASA and JPL are currently formulating mission designs for exploring the icy moons of Jupiter) an extremely high electron flux exists. Modeling radiation dose and transient effects in that environment is immature.
2.2
Total Accumulated Dose Modeling Using AP8 and AE8
The primary tools for modeling the earth-bound trapped radiation environment have been the NASA computer codes AP8 for protons [6] and AE8 for electrons [7]. The AP8 and AE8 codes along with a code for estimating the impact of spherical shielding and an orbital propagation code are available in commercial software such as Space Radiation [8]. We use that program to illustrate some key points about the relative importance of the various radiation components given assumptions about reasonable amounts of shielding expected in typical space-based sensor applications. The estimated effects of solar protons are based on representative empirical data from solar flare activity [9], [10]. Figure 1 shows total annual ionizing dose [in rad(Si)] versus shielding thickness for several orbital altitudes. The range of equivalent spherical shielding thicknesses represents plausible spacecraft and sensor shielding levels. The orbits considered are three highly inclined sun-synchronous orbits at altitudes of 500, 750, 1000 km and a circular 1600-km equatorial orbit. The latter case is chosen to show the impact of moving up toward the inner Van Allen belt. (Similar information can also be presented for nonionizing dose, which gives rise to displacement damage effects in imaging arrays, as discussed in Sections 3 and 4.) Figure 2 shows the results of using the AE8 environment in calculating total accumulated dose per year versus shield thickness for energetic electrons for the same set of orbits used in Figure 1. A comparison of Figures 1 and 2 reveals the substantial attenuation of electron dose by shielding compared to that for protons. At an equivalent aluminum shielding thickness of 300 to 400 mils, protons produce the dominant ionizing dose effect. It is clear that shielding against energetic protons is an issue in space applications.
IV-4
Figure 1. Calculated total annual ionizing dose versus equivalent aluminum spherical shielding thickness for trapped protons at four orbital altitudes.
The required equivalent shielding thicknesses are impractical from a weight viewpoint. It should be noted that the dose calculations shown in Figures 1 and 2 include the effects of bremmstrahlung x-rays generated by the interaction of protons and electrons within the spacecraft and sensor shielding. If extreme shielding is employed using high-Z materials for instance, the interaction of protons with such shielding will generate secondary neutrons, which would need to be accounted for in any dose estimation.
2.3
South Atlantic Anomaly
Figure 3 shows energy-integrated parametric plots of proton flux versus time during one period of a circular 1100 km, 63 degree inclined orbit. The proton flux peaks sharply in the region above the South Atlantic Anomaly (SAA) and is essentially negligible elsewhere. The approximate latitude and longitude location of the SAA places it over the Atlantic Ocean off the eastern coast of southern Brazil /Argentina. This figure indicates the transient nature of the SAA with a duration in the 10 to 20 minute range (depends on proton energy) for the indicated LEO orbit parameters. This duration should be compared to a typical LEO orbit period of the order of 100 minutes. These durations are a strong function of the orbital parameters, as discussed below.
IV-5
Figure 2. Calculated total ionizing annual dose versus equivalent aluminum spherical shielding thickness for trapped electrons at four orbital altitudes.
Figure 3. Trapped proton fluxes external to spacecraft in the South Atlantic anomaly region for the indicated orbit.
IV-6
2.4 Modeling the Proton Environment Using the Combined Release and Radiation Effects Satellite (CRRES) Database Proton flux data are usually obtained by exercising the AP8 model. AP8 is a mature proton flux model based on estimating proton flux entrained in a magnetic field and usually is taken to be the standard model for protons. Variants of AP8 exist in which the original 1965 magnetic field model was updated with more recent information. The Combined Release and Radiation Effects Satellite (CRRES) [11], [12] collected proton flux data while in a geosynchronous transfer orbit during a 14-month period from July 1990 to October 1991, a time of solar maximum. In March 1991, a geomagnetic storm occurred that caused a reconfiguration of the inner magnetosphere, thereby modifying the proton radiation belt. For this reason, the CRRES proton flux data were divided into two components: those collected before the storm (quiet field) and those collected after the storm (active field). The proton flux data collected by the CRRES experiment provides an opportunity to update general proton flux modeling compared to AP8 for orbital altitudes consistent with the CRRES geosynchronous transfer orbit. This is done by extracting information from the database that accompanies the CRRES-PRO utility. The CRRES-PRO utility consists primarily of several coupled programs: CRRES-PRO, LOKANGL, MAGMODEL, TIMBINS, and FLUXDATA [11], [12]. It also contains two sets of flux data files, FLUXDATA.XX, containing proton flux as a function of geomagnetic coordinates. There is one file for each proton energy band measured by the CRRES proton telescope. There are two sets of these files, one corresponding to the proton environment for a quiet geomagnetic field and one corresponding to an active geomagnetic field as experienced by CRRES. The user interface is contained within the CRRES-PRO program and allows the user to input orbital parameters in one of three formats: position and velocity, mean orbital elements, or solar elements. The final output of the CRRES-PRO utility is the integral and differential proton fluence impinging on the spacecraft exterior in one year. Although CRRES-PRO does not provide flux as a function of orbital position, it conveniently allows integration of user-created software with the CRRES-PRO utility. For example, a FORTRAN program can be written to read the LOKANGL (geographic) and MAGMODEL (geomagnetic) ephemeris files to obtain coordinate information necessary to look up data in the FLUXDATA files. The flux is analyzed as a function of time, geographic latitude, geographic longitude, altitude, and true anomaly. Flux histograms are then created showing the maximum, average, and minimum flux encountered as a function of each of these coordinates. Two very different orbits, defined in Table 1, have been analyzed as examples. Figure 4 shows the proton flux for orbit coordinates of true anomaly and altitude averaged over a seven-day period for the MIDLAT orbit. (Figure 5 defines the relevant orbital parameters such as apogee, perigee, true anomaly, etc.) This orbit is inclined at 53° but has apogee (perigee) 11° above (below) the equatorial plane so that the peak flux is seen near 0° and 180° true anomaly (Fig. 4(a)). This is consistent with Figure 4(b), which shows that the proton flux is maximum near apogee and perigee where the orbit dips into the
IV-7
proton belt. The peaks centered about 0° and 190° true anomaly are broader than a single crossing of the proton belt would imply because the earth's magnetic dipole is offset and rotated with respect to the earth's rotation axis. From the point of view of the satellite, the proton belt wobbles over the course of the seven-day integration period, changing the apparent location of the belt each orbit for any given fixed value of true anomaly. Because the orbit has a period of 3.7 hours, the structure that would be seen in the flux as a function of geomagnetic longitude for a single orbit is washed out when the averaging takes place over several days as it is here. The earth rotates approximately 28° between each crossing of the proton belt. Except for slight perturbations caused by apparent belt wobble, flux as a function of geographic latitude derives from an average sampling of the belts at two altitudes. Table 1. Characteristics of Orbits Used in Sample Analyses
Orbit
Apogee (km)
Perigee (km)
Inclination (degrees)
Argument of Perigee (degrees)
Period (hours)
MIDLAT
7460
4100
53
169
3.7
MOLNIYA
39,000
900
63
279
12
Results for the MOLNIYA orbit are shown in Figure 6. At 0° true anomaly (perigee), the spacecraft is near the lower edge of the proton belt and experiences a relatively small proton flux. The notch in the flux spectrum that can be seen at approximately 60° and 300° true anomaly results from the superposition of two flux curves that are slightly offset from one another, one each for the odd and even orbits occurring during one rotation of the earth. This orbit is highly eccentric, so 90° true anomaly occurs at a relatively low altitude. Flux as a function of altitude, shown in Figure 6(b), gives some indication of the radial extent of the proton belt, even though the spacecraft is not quite equatorial during belt traversal. The contrast between results obtained using the CRESS-PRO quiet and active field models can be seen in Figure 7 for the MOLNIYA orbit. The rise in flux at about 104 km for the active field analysis corresponds to the second proton belt discovered by CRRES. Note that CRRES data do not extend beyond 90 MeV and must be supplemented in some way to estimate flux at higher energies. The CRRES-PRO program uses the integrated flux obtained from the NASA AP8 model for this purpose. If higher energies are needed, some method of extrapolation must be used to estimate the proton flux. For the present examples, the CRRES-PRO data are used [11], [12]. Other useful earth-bound natural radiation environment codes and data include: the APEXRAD code, which models total ionizing dose due to trapped protons and electrons at low altitudes and low latitudes behind four shielding thicknesses, and SAMPEX, which provides limited particle flux and pitch angle information on trapped electron and hydrogen isotopes at low altitudes and high inclinations.
IV-8
Figure 4. Integral proton flux of MIDLAT averaged over many orbits as a function of: (a) true anomaly and (b) altitude.
2.5
Spacecraft Shielding and Radiation Transport to the Focal Plane
The spacecraft mass shields internal components from the space radiation environment. Estimation of degradation of impinging radiation by spacecraft shielding is an important step toward analyzing system performance. Simple geometries often are used to describe spacecraft shielding, such as the solid sphere, the spherical shell, the slab, and the semi-infinite slab, in order to perform orbit trade analysis before a spacecraft design exists or just to simplify analysis of an existing system. If high-fidelity results are sought, a detailed model of the spacecraft mass must be made, and calculations must be performed to estimate the transport of radiation through the spacecraft to an internal dose point. IV-9
Figure 5. Orbital parameter definitions
A number of radiation transport codes exist (varying in degree of complexity) that facilitate estimation of total integrated dose levels at the location of the optical sensor (i.e. focal plane). These codes allow insertion of user-defined shielding thicknesses and associated detailed surrounding geometries. To make complete use of such capabilities for space-based electrooptical systems requires that the details of the spacecraft and optical sensor payload physical configuration be known to fairly high fidelity. Such configurations typically evolve and change during the development of those advanced systems, with the final shielding configurations not being known in detail until the end of the development cycle. However, the radiation specifications for focal planes must be fixed very early in the payload development cycle. To accommodate this uncertainty in the final knowledge of the focalplane radiation environment requires the use of use of radiation specification and test methodologies that include appropriate margins to cover these uncertainties.
IV-10
Figure 6. Integral proton flux for a MOLNIYA orbit averaged as a function of: (a) true anomaly and (b) altitude.
The process starts with the determination of the relevant radiation environment external to the spacecraft using tools, codes, and databases mentioned in Section 2.2. This radiation environment is then propagated through a simplified shielding model (i.e. a spherical shell is an example of the simplest geometry) using a standard reference material (i.e. aluminum). The SHIELDOSE code is often used for such simplified shielding analysis. The Al thickness is varied to account for the worst-case expected spacecraft and payload shielding level. For example, the choice of an appropriate minimum shielding thickness represents a conservative IV-11
assumption with respect to total integrated dose at the focal plane. Total ionizing and nonionizing dose levels are computed at the focal plane to specify the required level of radiation tolerance. The unique sensitivities of particular types of focal-plane devices to ionizing and nonionizing dose (i.e., CCD focal planes) and particle transient effects must, must be accounted for in the radiation tolerance specification process. In addition margin factors are used that account for the cascading effect of: (a) the uncertainty in the knowledge of the space environment, (b) the uncertainly in the knowledge of the final spacecraft and payload shielding configuration, and (c) the statistical variability in the response of focalplane sensors to specified radiation exposure levels. The process ends with statistically significant and appropriate radiation testing, undertaken with the focal plane operating in its intended design conditions (i.e. appropriate temperature, clocking speed, etc.) using the relevant post-shielding radiation environment and pertinent pass/fail criteria. Any deviation from these defined radiation and operating test conditions must be shown, by experiment, not to be significant with regard to radiation susceptibility. For example, in earth bound space sensor applications wherein only the natural space environment must be contended with, energetic protons survive as the dominant postshielding source of radiation damage and transient effects in advanced focal planes. Hence proton testing with protons of an appropriate energy is required. Replacing proton testing with simpler surrogate total-dose testing using, for example, cobalt-60 gammas, is only permitted when it is shown experimentally and in a statistically significant manner that the radiation response differences between proton and cobalt-60 testing for equivalent dose are insignificant.
IV-12
Figure 7. The integral proton flux for the MOLNIYA orbit shown in (a) was computed using the CRRES active field model; that in (b) was computed using the quiet field model (same as figure 6(b)).
IV-13
3.0
Radiation Effects On Visible Imaging Arrays
3.1
Introduction and Overview
Radiation effects on visible imaging arrays are discussed in this section. The effects of both displacement damage and ionizing radiation on array properties are described. Following a brief summary of the visible imager technologies in use today, displacement damage effects produced by incident energetic particles, such as protons in space applications, are described. All visible arrays exhibit an increase in thermally generated dark current due to displacement damage. That dark-current increase is manifested in both the mean dark current and the darkcurrent distribution in an array. Some pixels exhibit an increase in dark current that can be much larger than the mean value, and are referred to as hot pixels or dark spikes. The effects of displacement damage on dark current are also enhanced by the presence of relatively high electric fields in imagers. In addition, temporal instabilities occur in the dark current for given pixel. Displacement damage also causes the charge transfer efficiency (CTE) in charge-coupled devices to decrease. Annealing of displacement damage occurs that reduces the dark current and the radiation-induced change in CTE. Ionizing radiation effects on imaging arrays are also discussed in this section. Techniques for improving the radiation tolerance of visible imagers are summarized, and current trends for the use of those imagers in radiation environments are discussed.
3.2
Visible Imager Technologies
Visible imaging arrays are fabricated primarily using silicon technology. Charge-coupled device (CCD) and charge injection device (CID) technologies began in the mid 1970s with the development of modestly sized imagers. Continuous improvements in electro-optical performance, array size, pixel size reduction, and radiation tolerance have occurred since then. Major advances in CCD technology [158], due primarily to a series of scientific spacebased astronomy applications (e.g., for NASA programs such as the Hubble Space Telescope, the Chandra X-Ray Observatory, and Galileo), have resulted in a very high levels of electro-optical performance (high quantum efficiency, ultra-low noise, wide dynamic range, etc.) and very large array sizes. In the late 1980s, the active pixel sensor (APS) concept was introduced (these devices are also known as CMOS imagers) largely for commercial applications [159], [160]. In recent times, CMOS imaging technology has been undergoing development to improve its performance characteristics for the purpose of attaining the same levels achieved with scientific CCD technology [63], [64]. A recent development is the introduction of small-pixel hybrid silicon photodiode technology [161]. For space applications, the use of CCDs dominates due to their low noise and technical maturity. In CCDs, photosignals are converted directly into charge packets at the location of each CCD pixel. By applying time-varying voltage signals to these CCD pixel gates, the photogenerated charge packets are transported (multiplexed) in an essentially lossless manner to one or a few charge-to-voltage conversion circuits along the periphery of the imaging chip. It is notable that the same pixels that collect the photosignals as charge packets are used subsequently to provide charge transport and video information extraction from the imaging array. The charge transport process occurs in a buried channel beneath the pixel surface to IV-14
avoid contact between the charge packets and interface traps at the silicon-gate insulator boundary. This very successful charge transport process is also a drawback of CCD technology in terms of space radiation damage. As discussed in Section 2.0, the dominant space radiation component at the focal plane, and hence at the CCD surface, is energetic protons. Nonionizing energy loss from these particles damages the buried channel and produces bulk charge trapping centers. Those traps interact with the signal charge packets as they move from pixel-to-pixel (gate-to-gate) during multiplexing and result in charge transfer efficiency (CTE) degradation, which causes a loss of signal for a point-source image or a smearing effect when extracting an extended image. Bulk displacement damage also causes an increase in the mean thermally generated dark current and gives rise to dark current spikes for some pixels. Ionizing radiation mainly produces oxide charging in the pixel gate oxide and in the chargeto-voltage output circuit (also known as an electrometer or a source-follower circuit). The charging of the oxide layer in the CCD pixel gates is a rather benign effect since the clocking voltage amplitudes, to first order, impact pixel well depth or charge storage capacity. Thus for charge packet sizes sufficiently below the pixel well capacity, small voltage shifts in CCD pixel voltages will have a minimal performance impact. Changes to the CCD output electrometer due to ionizing radiation effects are more important. Here the threshold voltage shift associated with the gate oxide in the CCD electrometer source-follower field-effect transistor (FET) will produce a loss in dynamic range and an increase in the source-follower 1/f noise. (Usually no system effect is seen due to the common use of correlated-doublesampling circuits, which remove 1/f noise). Depending on the polarity of the output electrometer reset FET (e.g, n-type), radiation-induced damage can cause off-state leakage that adds to the photosignal, thereby reducing the output dynamic range. This subject is discussed in more detail in Section 6. High-performance CCDs can have a backside-thinned structure to enhance quantum efficiency. The back surface that is first seen by the incoming photon flux must be devoid of a potential well to ensure efficient signal collection. Proton-induced displacement damage must not impact this back surface or a loss in quantum efficiency (typically at the blue end of the spectrum) will result. There are similarities and differences between space radiation effects on highly optically sensitive CCD imaging devices and on other analog and digital electronic components present in space instruments and systems. In a CCD, very large dynamic ranges are required (sometimes as high as 80 dB, or 10,000-to-1). Given that the analog voltage swing of the maximum signal is typically on the range of 1 to 2 volts, this requires readout noise floors in the 10 to 100 rms microvolt range. The electrometer electrical response must also be highly linear over its dynamic range (typically less than one percent measured as a maximum departure for a linear least-squares fit over the entire output amplitude range). The photocurrents that are integrated to produce the minimum signals (near the noise floor) are typically in the femptoamp and in some cases attoamp regimes. In contrast, for digital microelectronics only two analog levels (digital 0 and 1) need to be clearly distinguished. As such, radiation-induced FET threshold shifts and gate leakage effects can be quite large when compared to the photocurrent and video voltage ranges over which a CCD sensor must
IV-15
operate. For example, radiation-induced off-state leakage in an electrometer reset FET must not grow beyond the level of a few to a few tens of femptoamps or the imager dynamic range will be impacted. This is significantly different from the equivalent current and voltage scales of concern in digital space-based microelectronics. In addition, there are radiation damage mechanisms that are essentially unique to imaging sensors. For example, radiation-induced increases in dark current have a significant effect on the performance of image sensors due to increases in dark current shot noise, loss of dynamic range, and the introduction of a darkcurrent offset pattern, including the effects of dark-current spikes. Such dark current effects do not impact digital and low-resolution analog microelectronic circuits. Other visible imaging technologies, including CIDs, CMOS imagers and hybrid CMOS imagers, share many of the same radiation susceptibilities experienced by CCDs (e.g., dark current increase, hot pixels, electrometer degradation issues). However these three technologies differ from CCDs in that they do not make use of charge transport, so charge transfer efficiency loss is not an issue. Section 6.0 provides a detailed discussion of radiation effects on imager microelectronics separate from the effects experienced in the detecting silicon pixel layers in imaging arrays, which are described in the present section. APS (or CMOS) pixels contain either photodiodes or photogates (i.e., MOS capacitors). CMOS readout electronics are most commonly on-chip but may also be contained in a separate chip in hybrid forms of CMOS imagers. CID technology also allows individual pixels to be read out, with the only charge transfer involved being that from one electrode to another within the same pixel during the readout process. CID technology is evolving to having a preamplifier with every pixel. In general, all visible imagers are subject to ionizing radiation effects but are hardenable. All types of visible imagers are susceptible to displacement damage effects, which are difficult to mitigate. CCDs are the only type of imager susceptible to CTE degradation, as noted above. Subsequent sections discuss these general and specific radiation effects on visible imagers.
3.3
Displacement Damage Effects
This section describes displacement damage effects on Si visible imaging arrays. Included is a discussion of radiation-induced dark current in any type of array and charge transfer efficiency degradation in CCDs. Annealing of dark current and CTE are also considered. 3.3.1 Dark Current Dark current production is a very important effect in silicon imagers in space applications. Displacement-damage-induced dark (or leakage) current arises from the bulk of semiconductor devices (i.e., not the surface) and is due to thermal generation of carriers in depletion regions via energy levels near midgap. Experimental and analytical studies of radiation-induced dark current in silicon imaging devices conducted over the past thirty years have led to numerous insights, and successful dark-current modeling efforts have resulted. Early work (e.g., [13], [14] and references therein) emphasized the determination of dark-
IV-16
current damage coefficients for specific cases of interest, i.e., a specific particle type and energy, whereas more recent efforts have focused on developing a modeling approach that is independent of particle type and energy [15]-[19]. In addition, detector specialists within the high-energy physics community have performed directly related work on radiation-induced changes in silicon particle detectors (e.g., [20], [21] and references therein). The production of dark current in an imaging array evolves in the following way for space applications. The first incident energetic particle to interact with silicon atoms in the depletion region in a specific pixel will produce displacement damage, and that damage will increase the dark current in that pixel. As more particles strike the array, additional pixels are damaged. At somewhat higher particle fluences, individual pixels will have experienced more than one displacement-damage “event” (i.e., the interaction of an energetic particle with Si atoms in the depletion region of those pixels). In this manner, the imaging array eventually will exhibit radiation-induced dark current increases in all pixels. There is then a distribution of dark-current magnitudes over those pixels. The subsections below describe the mean dark current in a relatively heavily irradiated imager, the dark-current distribution, hot pixels (i.e., dark current spikes), field enhancement of dark current, and temporal darkcurrent fluctuations (random telegraph signals). 3.3.1.1 Mean Dark Current This section addresses the mean dark current produced by radiation-induced generation centers in the depletion region bulk of an imaging array. Such centers are the dominant source of dark current in many silicon devices, although surface generation due to ionizing radiation effects is important in some cases. Emphasis is placed here on effects that occur at or near room temperature for a sufficient time after irradiation (e.g., one week) such that the defects introduced are relatively stable. Five dark-current damage factors, or coefficients, are employed in the literature: 1) K g , the generation lifetime damage coefficient, which is related to the rate of lifetime degradation for a given fluence of a specific particle [13], [14]; 2) K p , the particle damage factor, which is the increase in dark current density per unit fluence of a specific particle [15]; 3) α , the damage factor used by the high-energy physics community, which is the increase in dark current per unit damaged volume per unit fluence of a specific particle [21]; 4) K de , a “damage-energy damage factor,” which is the increase in dark current density per unit energy deposited in the damaged volume [16], [18]; 5) K dark , a universal dark-current damage factor, which is the increase in thermal generation rate per unit deposited displacement damage dose [19]. Emphasis is placed on K dark here, and expressions are given that relate K dark to the other four factors. Radiation-induced dark current density, ∆J d , can be expressed as (e.g., [13], [14]) ∆J d = qni ΦW / K g , (1)
IV-17
where q is the electronic charge, ni is the intrinsic carrier concentration, Φ is particle fluence (particles/cm2), and W is the depletion region width. This expression assumes a uniform thermal generation rate G in the depletion region, where G = ni / τ g . The generation-lifetime damage coefficient for a given particle type and energy is defined through the relation
K g ≡ Φ /[(1 / τ g ) − (1 / τ g 0 )] ,
(2)
where τ g 0 and τ g are pre- and post-irradiation generation lifetimes, respectively, in the depletion region. Equation (1) applies for devices in which bulk thermal generation dominates in producing dark current. The physically relevant process is thermal generation of charge in a depleted volume with dimensions specific to a given device and its operating conditions. Assuming that a single level Et in the silicon bandgap dominates the thermal generation process, generation lifetime is given by
τ g = τ p exp[( Et − Ei ) / kT ] + τ n exp[−( Et − Ei ) / kT ]
(3)
where Ei is the intrinsic Fermi level (midgap). The prefactors are given by τ p = 1 / σ p vth N g and τ n = 1 / σ n vth N g , where σ n and σ p are the electron and hole capture cross sections, respectively, vth is thermal velocity, and N g is the generation center density. Note that (3) does not distinguish between pre- and post-irradiation lifetimes. It is a general single-level expression for any generation lifetime, i.e., either τ g or τ g 0 . The radiation-induced dark current formulation given by (1) - (3) provides a clear link to the basic mechanisms of dark current introduction, i.e., increased thermal generation of carriers at radiation-induced generation centers. The damage coefficient, K g , is a quantitative measure of the effectiveness of generation center introduction by particles of a given type and energy. However, the key limitation of that approach (i.e., (1) and (2)) to determining dark current is that a value for K g is needed for every particle type and energy of interest, which necessitates an experimental determination in each case. Detector specialists in the high-energy physics community commonly use a radiationinduced leakage current damage factor, α , defined as [21]
α ≡ ∆I dv / Φ .
(4)
In this expression, ∆I dv is the radiation-induced increase in dark current per unit volume in a depletion region damaged by a fluence of a specific particle. As in the case of K g , α also depends on the particle type and energy used in the irradiation.
IV-18
Burke [22], Summers et al. [23], and Dale et al. [24] developed a valuable approach to assessing displacement damage effects in a wide variety of devices. Their method involves using the nonionizing dose deposited in a device as the key factor for determining the degradation of a specific device property. The nonionizing dose, or displacement damage dose ( Dd ), is given by the product of the particle fluence and the rate of nonionizing energy loss (NIEL). NIEL has been determined for semiconductor materials and particle types and energies of common interest, and a general approach to determining NIEL for any material and particle was described [25]. Dale et al. [15] studied dark current in irradiated silicon CID (charge injection device) imagers and defined a damage factor as ∆J d / Φ . That factor is referred to here as a “particle damage factor,” K p , which gives the increase in dark current density per unit fluence of a specific particle:
K p ≡ ∆J d / Φ .
(5)
Dale found that K p scaled linearly with NIEL. That scaling can also be expressed as ∆J d ∝ Dd since Dd = ( NIEL)(Φ ) . It is well established experimentally that radiationinduced dark current scales linearly with fluence for irradiation with a given particle. It then follows directly that ∆J d will scale linearly with Dd . Also note that K p and α are related through the depletion width; i.e., K p is equal to Wα . Marshall et al. [16], [18] modeled mean radiation-induced dark current and its pixel-to-pixel distribution, or variation, for specific silicon imaging arrays. Their model employed the damage energy ( E d ) deposited in a depleted volume, which is that fraction of the available particle energy lost through nonionizing processes. Here we express their work in terms of a “damage-energy damage factor,” K de :
K de = ∆J d / E d .
(6)
K de is the mean radiation-induced increase in dark current density per unit amount of damage energy deposited in the depletion region. Dark current can be modeled in a general way in terms of the radiation-induced increase in thermal generation rate, ∆G , per unit displacement damage dose deposited in a depleted volume [19]. A dark-current damage factor can be defined as
K dark ≡ ∆G / Dd .
(7)
This expression is more basic than previous approaches used in the literature (described above) to define a dark-current damage factor. The physical process involved is introduction
IV-19
of generation centers by displacement damage. Those centers cause the generation rate to increase, and the proportionality factor relating that increase to the dose deposited is K dark . Thus, K dark is the number of carriers thermally generated per unit volume per unit time in a depletion region per unit nonionizing dose deposited in that volume. It is straightforward to determine dark current density once K dark is known. The relationship is ∆J d = qWDd K dark ,
(8)
where Dd is the displacement damage dose deposited during a given irradiation of a specific device having a depletion region width W . Similarly, the dark current per unit depleted volume ( ≡ ∆I dv ) is given by qDd K dark . One can relate K dark to the other damage factors discussed above (i.e., K g , α , K p , and
K de ): K dark = [ni /( NIEL)]( K g ) −1 = [1 /{( q )( NIEL)}]α
= [1 /{( qW )( NIEL)}]K p = [ Aρ / q ]K de .
(9)
In these relationships, ρ is the silicon density and A is the cross-sectional area of a specific depletion region (e.g., that portion of the pixel area in an imager corresponding to the depleted volume for that pixel). Note that the expression in (9) that relates K dark to K g only applies for a specific type and energy of irradiating particle since K g is defined in that comparatively restricted manner, as discussed above. A similar statement applies for the relationships in (9) between K dark and both α and K p , as defined by (4) and (5). (It is straightforward to convert α and K p to Dd -based damage factors.) The damage factor definitions for K dark and K de are based on the displacement damage dose deposited by irradiation with any particle type and energy. After reviewing three decades of dark-current data for various types of silicon devices irradiated with various particle types and energies, it was found [19] that much of the data can be described by a single K dark value. The mean and standard deviation for K dark , except for cases in which devices were bombarded with low-energy electrons and 60 Co photons, are (1.9 ± 0.6)x10 5 carriers/cm 3 sec per MeV/g. That damage factor applies for the broad variety of devices and irradiation conditions examined, including bombardment with protons, neutrons, pions, and heavy ions. Specific devices included are CIDs, n- and p-channel CCDs, npn and pnp bipolar transistors, JFETs, MOS capacitors on n-and p-type Si, and diode detectors fabricated on n- and p-type material with a wide range of dopant and impurity concentrations. Figure 8 shows the K dark results of [19] along with subsequent confirming
IV-20
measurements for a CCD [26]. The value shown for K dark applies at 300 K for relatively stable damage, i.e., after approximately one week at room temperature following irradiation. Also, the mean dark-current damage factor holds for devices in all radiation environments except those that produce relatively isolated defects, such as low-energy electrons.
Figure 8. Radiation-induced increase in thermal generation rate per unit fluence (equals the product of K dark and NIEL) versus the displacement damage dose deposited per unit fluence (equals NIEL). Solid and dashed lines are from [19] and data points are from [26]
Using K dark to predict radiation-induced dark current for a specific silicon device is straightforward in many cases, i.e., when the damage distribution in the depletion region is relatively uniform. First, the displacement damage dose deposited in the damaged portion of the depletion region is determined for the radiation environment of interest. Determining Dd for irradiation with penetrating monoenergetic particles requires only the fluence and the NIEL value. Obtaining Dd for an energy spectrum is more involved but relatively straightforward (e.g., see [25], [27], [28]). The next step is to use (8) in conjunction with the universal damage factor and the depletion region width to obtain the dark current density. One can express the dark current density in any manner desired. For example, referencing
IV-21
∆J d to the pixel area is a common practice for imaging arrays. Also, note that the dark current per unit volume, ∆I dv , can be determined directly from K dark and Dd without knowledge of the depletion width. The main caution in using K dark to predict dark current is to be aware of the linear and nonlinear regimes in Fig. 8. The universal damage factor only applies in the linear regime, e.g., for proton and neutron irradiations. For electrons with energies in the MeV range, i.e., for NIEL values on the order of 10 −4 MeV cm 2 /g or lower, a NIEL-dependent K dark is applicable. 3.3.1.2
Single Particle Effects, Dark Current Distributions, and Hot Pixels
A very low particle fluence incident on a silicon device can result in nonuniformly distributed displacement damage. Such effects are especially evident in a visible imaging array, such as a CCD, that may contain millions of individual pixels. For that important example, radiation-induced dark current can vary significantly from pixel to pixel, as discussed below. In the extreme case of a single incident particle that produces damage, only one pixel in a dense array will exhibit an increased dark current. Gereth et al. [29] evidently made the first report, in 1965, of the effects of displacement damage produced by single particles incident on silicon devices. They explored those effects by irradiating avalanche diodes with fission neutrons and with 2-MeV electrons. Notable differences in device behavior were observed between these two cases. Two decades later, numerous studies of single-particle-induced displacement damage effects were conducted [15]-[18], [30]-[33], nearly all of which used visible imaging arrays as test devices. Pickel et al. [34] reviewed the work conducted during that later era. A review by Hopkinson et al. [27] provides further information regarding nonuniform displacement damage effects. Radiation-induced dark current in a visible imager exhibits a variation from pixel to pixel. That distribution is also often referred to as the dark-signal nonuniformity (DSNU). In the absence of field enhancement effects, which are discussed below, the dark-current distribution is due to inherent statistical variations in the number and magnitude of elastic and inelastic interactions between bombarding particles and Si atoms from pixel to pixel. It is those interactions that give rise to displacement damage, which in turn causes thermally generated dark current. Dark-current distributions in Si visible imagers were first measured and modeled by Dale et al. [15], [17] and by Marshall et al. [16], [18]. Robbins [35] proposed modifications to their modeling work. Fig. 9 shows an example of dark-current distributions for a proton-irradiated CCD imaging array. [27] Each of those distributions, of course, has a mean value (discussed in the previous section) and a variance. In addition, there is a relatively long tail to each distribution that includes events that produce dark currents much higher than the mean. Those pixels are typically referred to as “hot pixels” or “dark current spikes”. Figure 10 shows examples of hot pixels in a proton-irradiated CCD. [27] The two primary causes of hot pixels are 1) relatively uncommon inelastic interactions in which a large amount of displacement damage is produced by a single primary silicon recoil, and 2) electric-field
IV-22
enhancement of the dark current produced by the damage caused mainly by relatively common elastic interactions, as discussed below.
Figure 9. Illustration of dark-current distributions in a proton-irradiated CCD [27].
Figure 10. Illustration of hot pixels in a proton-irradiated CCD [27].
IV-23
3.3.1.3
Field Enhancement Effects
Visible imaging arrays can include regions in which the electric field strength is relatively high (e.g., > 1E4 V/cm). The dark current resulting from radiation-induced defects located in such regions will be enhanced compared to that for damage located in lower-field locations. Those local high-field regions may be due to implantations, such as the built-in fields in a virtual-phase CCD, or to the applied voltages. Fifteen years ago, several workers described the enhancing effects of local electric fields on dark current in irradiated imaging arrays. [16], [17], [33], [36], [37]. Field-enhanced emission can occur via the Poole-Frenkel effect and/or by phonon-assisted tunneling, as illustrated in Figure 11. Field enhancement becomes particularly strong for local field strengths on the order of 1E5 V/cm and higher. As an example, at a field of 1E5 V/cm the field enhancement factor for dark current in Si devices is ~10 [17]. EC
ET
Trap level produced by displacement damage (a) No field applied
1 Poole-Frenkel effect Phonon-assisted tunneling 2 ∆E
1
2
Trap level
(b) With field applied 0324-04A
Figure 11. Schematic illustration of electron emission into the conduction band by the PooleFrenkel effect and by phonon-assisted tunneling. The energy required for electron emission with no field applied is denoted by ET. The field-induced barrier lowering, ∆E, reduces that energy.
A characteristic of field-enhanced emission in an irradiated visible imager is that the activation energy derived from the temperature dependence of dark current in a given hot pixel generally decreases as the dark-current amplitude increases. When field enhancement is negligible, that activation energy is ~0.63 eV. When such enhancement is important, the
IV-24
activation energy can be as low as ~0.3 eV. Figure 12 illustrates the dependence of activation energy on the magnitude of the dark current increase for four pixels in a virtual-phase CCD irradiated with 14-MeV neutrons. [33] Various workers have explored field enhancement as an important contributor to dark-current spikes in irradiated Si imagers (e.g., see [38]-[40].) For space applications of visible imagers, field enhancement effects appear to be an important consideration for any type of array. The dominant mechanism for production of hot pixels appears to be due to damage caused by proton-induced elastic interactions. A small fraction of the large number of such interactions that occur, which is due to its relatively large cross section, will reside in local high-field regions, and field enhancement results. Designing arrays to minimize the presence of such regions is one obvious way to reduce the number of hot pixels produced in space applications.
Figure 12. Change in dark current density versus reciprocal temperature for four pixels in a 14MeV neutron-irradiated virtual-phase CCD [33].
3.3.1.4 Random Telegraph Signals
The terminology “random telegraph signal” (RTS) behavior is often used in the radiation effects field today to describe noise arising in visible imagers, including CCDs, CIDs, and APS arrays, due to temporal fluctuations in the radiation-induced dark current amplitude in specific pixels. That amplitude switches between two or more dark-current levels, and the time between observed fluctuations is typically on the order of minutes to one hour at room temperature. As the array temperature is reduced, the switching time increases. RTS behavior can affect the noise performance and in-flight calibration of visible imagers.
IV-25
The published literature on radiation-induced RTS phenomena is summarized here. In a study of hot-pixel formation in proton-irradiated CCDs, room-temperature annealing studies revealed the presence of significant fluctuations in the dark current level for some hot pixels [32]. A small fraction of those pixels exhibited the production, or appearance, of damage long after irradiation. Another small fraction exhibited “precipitous” (i.e., substantial and relatively abrupt) annealing of radiation-induced dark current. In a study of proton effects on CIDs [16], a fraction of the pixels that had a relatively large radiation-induced dark current also exhibited fluctuations in that current by a factor of two in both directions. The damage variations with time observed in those two studies appear to be consistent with RTS behavior explored in detail later by several workers. Temporal instabilities in radiation-induced dark current evidently were first referred to as random telegraph signals in a study by Hopkinson [41] due to a proposed mechanistic similarity with non-radiation-induced RTS noise in small-geometry MOS devices. In that study, field-enhanced emission over a potential barrier was suggested as a possible contributing mechanism for radiation-induced RTS. More detailed studies of RTS phenomena and mechanisms were subsequently performed [38], [42], [43]. RTS observations for various imagers have also been reported for proton [39], [44]-[47] and neutron [48] bombardment. Figure 13 shows example RTS behavior for proton-irradiated imagers at -40 deg C for a CCD [47] and at 27 deg C for an APS [43]. A complete understanding of the physical mechanisms responsible for RTS behavior evidently has not yet been achieved. Here is a summary of key RTS observations and proposed mechanisms: • RTS behavior occurs in all types of irradiated Si visible imagers and is due to bulk displacement damage effects • The introduction of RTS effects correlates with the bombarding particle fluence and appears to correlate with the rate of nonionizing energy loss for elastic interactions [38] • The introduction rate for RTS defects is on the order of 2E-5 per proton per pixel for 10-MeV proton-irradiated imagers [38], [47] • The switching time between RTS dark-current levels increases with decreasing temperature • RTS phenomena appear to be associated with defects produced in high-electric-field regions in imagers [38] • RTS effects in a given pixel appear to be uncorrelated with the magnitude of the pedestal, or background, radiation-induced dark current in that pixel [38] • RTS phenomena have been postulated to be due to the production of reorientable defects [38] and to interacting adjacent defects [43], with field enhancement evidently playing a role in any dominant RTS mechanism
IV-26
Figure 13. Examples of RTS behavior: A) CCD at -40 deg C [47]; B) APS at 27 deg C [43].
IV-27
3.3.2
Charge Transfer Efficiency
Displacement damage degrades the charge transfer efficiency, CTE, in a CCD, resulting in a loss of signal charge. (Note that because charge transfer from pixel to pixel is not involved in CID and APS imagers, CTE is not applicable in those cases.) The technical literature also makes common use of the charge transfer inefficiency, CTI, when addressing radiation effects on CCDs, where CTI = 1 – CTE. Radiation-induced CTI in an irradiated CCD [i.e., CTI (post-irradiation) – CTI (preirradiation) increases linearly with bombarding particle fluence. The mechanism involved is the introduction of temporary trapping centers in the forbidden gap. Those centers trap charge residing in the buried channel and cause a reduction in the signal-to-noise ratio. Many studies have been conducted of radiation-induced CTE degradation in specific CCDs and of the underlying responsible physical mechanisms. (For reviews, see [27], [49]. See also [50].) CTE in an irradiated CCD depends on various parameters, including clock rate, background charge level (fat zero), signal charge level, irradiation temperature, and measurement temperature. Dale et al. [51] found that CTE degradation in a given CCD scales with the deposited displacement damage dose. An application of that scaling for CCDs on the Chandra spacecraft was presented recently [52]. Figure 14 shows an example of charge transfer degradation in a proton-irradiated CCD [50]. The effects of background charge (fat zero) and signal charge on post-irradiation CTI are illustrated. Methods for reducing CTE degradation are summarized in [27].
Figure 14. Example of proton-induced change in CTI for a CCD. The effects of signal charge and background charge on CTI are illustrated. [50]
IV-28
3.3.3
Displacement Damage Annealing
Numerous thermal annealing studies have been conducted for irradiated Si visible imagers to examine their recovery at both room temperature and elevated temperatures. Such studies are of basic importance in terms of understanding damage mechanisms and the defects responsible for observed effects. Annealing investigations are also important from a practical viewpoint because they can reveal the timescale over which recovery occurs at a given temperature and can show what temperature regimes are effective in recovering from the effects of radiation-induced damage. Annealing examples are provided in this section. CTE is considered first. The dominant radiation-induced trapping center in Si CCDs that degrades CTE has been identified as the E center (vacancy-phosphorus defect) by several workers. (See [27] for a summary.) The E center anneals in the temperature range 110-150 deg C [53]. In a study of transient annealing of CTI in a CCD following a burst of fission neutrons [54], it was found that negligible room-temperature annealing occurred over the time regime from 1.5 sec to 1000 hours after the burst. Thus, at least for the CCDs examined in [54], the displacementdamage-induced change in CTI at room temperature can be considered “permanent” for times > 1 sec following a burst of radiation. Elevated temperatures are required to achieve CTE recovery. For example, Hopkinson et al. [47] observed significant CTE annealing in proton-irradiated CCDs above 100 deg C. Various workers have observed significant dark-current annealing in irradiated Si devices. It is well established that room-temperature annealing occurs in such devices following irradiation, including bombardment with neutrons, protons, electrons, and other particles. Examples for particle detectors are given in [20] and [55]. Data for neutron-irradiated devices [14], [31], [54] are used here to illustrate annealing behavior. Measurements of mean dark current in CCDs as a function of time following pulsed fission neutron bombardment [14], [54] exhibited considerable annealing at room temperature over the time regime from 2 sec to 10,000 sec. Subsequent long-term annealing measurements on 14-MeV neutron-irradiated CCDs [19] showed annealing continuing to 1000 hours (3.6x10 6 sec), where the measurements were terminated. CCDs irradiated with 99-MeV protons also exhibited significant annealing, which continued to 1000 hours after bombardment [32]. Fig. 15 shows a composite of observed annealing behavior for neutron-irradiated CCDs [31]. Plotted is dark-current annealing factor (AF) versus time, where
AF = [ J d (t ) − J d (0)] /[ J d (∞) − J d (0)] .
(10)
In this expression, J d (0) and J d (t ) are the dark current densities before irradiation and at any time t after irradiation, respectively. J d (∞) is the dark current density at some sufficiently long time after bombardment when annealing is essentially complete. In Fig. 15, a time of 1000 hours is arbitrarily used for determining J d (∞) , although it is evident that annealing is not yet complete at that time.
IV-29
The above example of dark-current annealing in CCDs is relatively macroscopic in that the mean dark current for many pixels was measured. More microscopic studies have also been conducted [32], [48] in which the annealing behavior of individual pixels was examined. In a study of annealing in 778 pixels in a 99-MeV proton-irradiated CCD [32], several types of “recovery” were observed. (The relationship of that work to RTS behavior was noted above.) 32% of the 778 pixels exhibited monotonic forward (i.e., conventional) annealing over the time period from 1 to 1000 hours after irradiation. Another subset (10%) showed monotonic reverse annealing (i.e., damage production) during that same period. 15% of the cells exhibited negligible annealing, and the remaining 43% showed “mixed” annealing, i.e., a combination of forward and reverse annealing. Those results indicate that displacementdamage annealing is complex on a microscopic level. Such complexities become evident when single pixels in an irradiated imaging array are examined. In a recent annealing study for a neutron-irradiated CCD [48], dark-current instabilities and annealing were observed for specific hot pixels following irradiation. A relationship between those effects and multi-level RTS behavior was noted. In general, visible imagers are very good test vehicles for exploring the microscopic details of displacement damage reordering. Significant annealing of dark current occurs in the temperature regime from 80-150 deg C [47]. It has been established that radiation-induced dark current is not attributable to impurity-related defects (e.g., see [19]). Because the divacancy in Si anneals at approximately 300 deg C (e.g., see [53]), that defect appears to be ruled out as being responsible for the unstable portion (i.e., that damage which readily anneals in the range 25– 150 deg C) of radiation-induced dark current. One possible source for that relatively unstable damage is a higher-order vacancy defect complex (e.g., see [19], [47]). More study is needed to establish the defects responsible for dark current in irradiated Si imagers.
Figure 15. Dark current annealing factor (AF) versus time following neutron bombardment of CCDs [31]. AF is arbitrarily normalized here to unity at 1000 hours.
IV-30
3.4
Ionizing Radiation Effects on Visible Imagers
All Si visible imagers in use today include MOS technology in their various design and fabrication realizations. As a result, all Si arrays are susceptible to ionizing radiation effects. In general, ionizing radiation causes charge buildup in oxide (or, more generally, insulating) layers and interface trap production at oxide-silicon interfaces. Those basic phenomena in turn give rise to increased surface leakage current, due to enhanced thermal generation at interfaces, and to shifts in operating conditions for pixels and readout electronics (i.e., flatband voltage shifts for MOS capacitors and threshold voltage shifts for MOS transistors). In general, these ionizing radiation effects depend on operating voltages and on ionizing dose rate. Shifts in operating conditions due to charge buildup can be reduced significantly by employing design and processing techniques developed to harden MOS electronics for space applications. Surface leakage current increases can be diminished or avoided by keeping interface traps filled with charge, thereby making those traps unavailable to act as thermal generation centers. Keeping the surface in inversion is one approach to accomplishing this reduction in ionizing-radiation-induced leakage current [56]. If that source of leakage current is avoided, then dark current due to bulk displacement damage will dominate in an imaging array (Section 3.3.1).
3.5
Visible Array Hardening Approaches
Approaches to improving the radiation tolerance of CCDs were reviewed by Hopkinson et al. [27], and are summarized here. Methods identified for hardening against the space proton environment include the following: • •
• • • • • •
Shielding – Effective for up to 15 mm of Al. However, high-energy protons still penetrate. Secondary neutron production is an issue for thick dense shielding. Device design, architecture, and processing – Keep internal electric fields low. Use a CID (or APS) if CTE is a severe issue. Use a notched buried channel. Optimize active-region thickness to reduce transient events. Use MPP-mode devices to avoid total-ionizing-dose-induced dark current. Use hardened oxide technology. Cooling – Improves CTE and reduces dark current Clock rate – Influences CTE, dark current, and transient hit rate Software – Discriminate against transient events and hot pixels. Correct for CTE loss. Annealing – Periodic heating to improve CTE and reduce dark current. Background charge – Fat zero improves CTE. Defect engineering – For example, use p-channel CCDs to improve post-irradiation CTE [57] – [59]
Total-dose-hardened CMOS active-pixel-sensor visible arrays have reached a high level of complexity, currently containing as many as 16 Mpixels and beyond for use in selective imaging (i.e., non-astronomy) applications. Due to the high functional complexity and pixel performance uniformity required, many CMOS imager designs are currently fabricated by commercial wafer foundries. These foundries have the capabilities to fabricate very large IV-31
area reticle phototooling using proprietary “stitching” techniques, which permits the photocomposure of arrays well beyond the conventional reticle area limit of 2.2 cm x 2.2 cm. In addition, a number of commercial foundries have extensive experience in achieving acceptable yields with 300 mm diameter wafers, which enhances the probability of achieving at least one low-defect-array die per wafer. Table 2 contains a summary of the major total dose effects for wafer-foundry-produced arrays and the corresponding impacts on CMOS imager performance. The task of developing a total-dose-hardened visible imager design with a commercial wafer foundry is generally relegated to the use of design hardening techniques. These techniques include the following. (Section 6 provides a more complete discussion of radiation effects on and radiation hardening of readout circuitry.) •
•
•
• •
•
•
The use of state-of-art foundry processes with nanometer-thick gate-oxide layers has been shown as the solution to realizing analog transistor operation relatively immune to TID-induced voltage shifts, commonplace with the older foundry product, as shown in Figure 16. Similarly, the imager layout design should adhere to the set of specific layout design rules developed for the visible silicon detector by the candidate wafer foundry, based on prior experience. Foundry layout design rules commonly contain critical dimensional information regarding minimum separation between active detector areas and surrounding implanted n-well, p-well, and source and drain implanted regions. Cooling of the detector and readout array (i.e., mechanical cooler such as a JouleThompson cryostat or solid-state Peltier junction cooler) to reduce critical leakage paths. For example, the leakage current for both visible detectors and field-effect transistors operated in weak inversion vary exponentially with absolute temperature. Selective use of p-channel enhancement (PMOS) field-effect transistors for the pixel electronics applications requiring the lowest leakage (i.e., for the devices operated within weak inversion or actively biased in the “sub-threshold” region of operation). Utilizing design hardening layout techniques to decrease TID-induced leakage paths associated with the inversion or the thick field silicon dioxide-silicon interface regions, especially those regions surrounding n-channel enhancement (NMOS) field effect transistors, as discussed in Section 6. Figures 17 and18 contain pre- and postTID exposure measurements of current vs. voltage for NMOS transistors using 0.5µm-feature-size technology. The relative increased hardness for the transistor in Figure 18 is a result of the use of design hardening layout techniques. (Note that after TID exposure those regions surrounding PMOS devices are generally shifted toward surface accumulation rather than inversion, thereby avoiding-TID induced leakage increases at end-of-life. The selective use of pixel array support circuits which function to subtract the TIDinduced voltage shifts, which contribute to fixed pattern noise (FPN). These circuits, which have reached a high level of development for removal of so-called “kTC” reset noise, have recently been demonstrated to remove as much as 99% of the TIDinduced FPN voltage for TID levels up to 10 Mrad (Si), as discussed in Section 6.
IV-32
Table 2. Total ionizing dose effects on APS imagers TID Effect Gate Oxide Trapped Holes
Effect on Pixel Electronics Trapped Holes Shift Transistor Threshold Voltage—Annealing Reduces or Increases the Effect for Low Dose Rates
Holes Trapped Near JunctionField Oxide Interface (Bird’s Beak)
Trapped Holes Increases NMOS Transistor Leakage
Field Oxide Trapped Holes
Silicon-Dioxide to Silicon Interface Traps
Low Effective Annealing Rates Due to Low P-well Surface Concentration Trapped Holes Increases Leakage Between Pixel Photodiodes, Photodiodes and Transistors, etc. Annealing Rates Lower Due to Field Oxide Thickness Transistor Threshold Voltage Shift
Impact on APS Performance Increased Array Fixed Pattern Noise Increased 1/f noise Potential Significant Increase in Pixel Dark Current
Pixel-Pixel Photocurrent Crosstalk Increased Dark Current
Potential for Increased 1/f noise
Surface Carrier Mobility Decreased
Figure 16. Total-dose-induced flatband voltage shift vs. gate oxide thickness [60].
IV-33
Figure 17. Drain current versus gate voltage for an unhardened 0.5-µm NMOS transistor [61]
Figure 18. Drain current versus gate voltage for a design-hardened 0.5-µm NMOS transistor [61]
IV-34
3.6
Current Trends for Visible Imagers
CCDs are in common use today as imagers in space applications, and will continue to be employed for the foreseeable future, particularly for hardware with low-noise, highreliability, high-performance requirements. CMOS imagers are emerging as a viable contender for selected applications. As noted above, all Si imaging arrays are subject to displacement damage effects, which are difficult to avoid. (The exception is CTE degradation, which is specific to CCDs.) Also, all visible imagers are susceptible to ionizing radiation effects, including leakage current and operating voltage shifts. There are hardening approaches to alleviate those issues. For arrays that employ CMOS circuitry (i.e., CMOS imagers), one must give attention to potential single-event effects, such as latchup. Ionization-induced transients are another issue for all visible imagers, and that subject is treated in Section 5. Reviews of CCD and CMOS arrays that address their use in space [34] and in general applications [62], [63] have been published recently, including comparisons and technology trends. Janesick [63] described CMOS pixel and array designs that have the potential for making that technology more competitive with CCDs for high-performance applications. He also discussed hybrid arrays in which the imaging pixels and the readout circuitry are on separate chips mated together using indium bump bonding. One approach is to mate a CCD to a CMOS readout circuit, and a second is to do the same for a CMOS array. He also notes the potential for an all-SOI (silicon-on-insulator) hybrid array realized in a single chip. Such arrays would have the well-known radiation tolerance of SOI technology.
IV-35
4.0
Radiation Effects on Infrared Detector Arrays
4.1
Introduction
Infrared detector arrays are finding increased use in applications where exposure to radiation environments poses a challenge to their functionality. These arrays are subject to all the familiar radiation effects on microelectronics at room temperature, such as transients, total ionizing dose (TID) damage, displacement damage, as well as some additional effects. Those additional effects are due to the low signal and noise levels, cryogenic operating temperatures, and the unique physics of these devices. In this section, we describe the effects of radiation on infrared (IR) detectors, including a view of the future of the technology from a radiation effects perspective. With the exception of a brief mention of the effects of total ionizing dose on Readout Integrated Circuits (ROICs) in Section 4.4.2.3, the focus of Section 4 is on the detector. Radiation effects on ROICs are discussed in more detail in Section 6.
4.2
Infrared Detector Technologies
Infrared detectors are a key element in many modern optical systems because of their thermal imaging capabilities. Their major advantage is the ability to detect a large fraction of the radiative emission from objects rather than relying on detecting light reflected off objects as is the case for visible detectors. For example, the peak emission from a 300 K blackbody is at ~10 µm, and 99% of the emission from a 1000 K blackbody lies beyond 2.5 µm. Among the important uses of infrared detectors are astronomy, earth surveillance from space, missile detection/tracking/homing, and low-light/night-vision applications. A modern infrared focal plane has the pixel architecture illustrated in Figure 19. Infrared photodetection occurs in an array optimized for that purpose using an infrared-sensitive material. The conversion from signal photocurrent to charge and then to the voltage domain, along with noise suppression functions, occurs in a separate silicon ROIC array located directly beneath the infrared detecting array. The interconnection between these two arrays occurs via ductile indium bumps. The photocurrent passes from the photodiode pixel, through the indium bump and into the interface circuit of the ROIC, as indicated in Figure 20. The interface and signal encoding circuits indicated in that figure usually lie within the same footprint as the overlying detector pixel. In the ROIC this is called the “unit cell”. The terminology that describes this detector/ROIC sandwich structure is not standardized. Some focal-plane vendors call this a “hybrid infrared array”, while others call it a “sensor chip assembly” (SCA). It should be noted that the hybrid or SCA structure is cryogenically cooled to whatever temperature is needed to control the dark current of the photodiode pixels. The indium bumps are ductile and allow for some degree of mismatch in the coefficients of thermal expansion between the infrared detector material and the silicon ROIC, but often the detector or ROIC layer is thinned and constrained to match the thermal expansion of the other layer. It is notable that a large infrared hybrid array or SCA, for example a 2048 by 2048 pixel infrared device, will have in excess of 4 million replications of infrared pixels, indium bumps, and unit-cell interface and signal encoding circuits, each confined to a area in the typical range of 20 to 50 microns on a side, depending on the wavelength regime. The outputs from a large number of detector/unit-cell slices are then multiplexed together to one or a few video outputs. The radiation sensitivities of infrared detector materials and silicon
IV-36
ROICs are very different. As noted above, the radiation effects on the ROIC technology are discussed in detail in Section 6. In the present section, we focus on radiation effects on infrared detectors and arrays.
Figure 19. Typical focal-plane architecture
Figure 20. Cross section of typical indium-bump-bonded hybrid infrared detector array
IV-37
Historically, the first materials used to fabricate infrared detectors were the lead salts: PbS with a cutoff wavelength between 2.5 µm and 4.0 µm and PbSe with a cutoff wavelength between 4.5 µm and 6.0 µm, depending on operating temperature. In the 1960s and 1970s, these materials began to be replaced as other materials were developed, such as InSb, doped germanium, doped silicon (which quickly replaced doped germanium), HgCdTe, PbSnTe (which was only investigated for a brief period until it was determined not to be any better than HgCdTe), and silicides. In the late 1980s and early 1990s, quantum-well and superlattice detectors based on III-V materials were also developed. HgCdTe is currently the most widely used detector material, primarily because detectors with near-theoretical performance can be fabricated and the cutoff wavelength can be tuned between ~1.5 µm and ~20 µm by varying the Hg-to-Cd ratio. The cutoff wavelength of quantum-well and superlattice detectors can also be tuned by varying the superlattice parameters. However, the quantum efficiency of these detectors has been low thus far, so they have not found the wide usage experienced by HgCdTe detectors, despite using a more common semiconductor material. While doped silicon and silicide detectors have the advantage of being siliconbased, doped silicon detectors (which can operate out to >25 µm) require very low operating temperatures and the quantum efficiency of silicide detectors is very low, so use of these materials is also limited. The performance of InSb detectors, which have a cutoff wavelength of ~6.0 µm, is quite competitive with that of similar HgCdTe detectors. However, the cutoff wavelength of InSb is fixed, so it cannot cover as wide a range of applications as HgCdTe. The first infrared detectors (PbS, PbSe, early HgCdTe, early doped Ge, and early doped Si) were photoconductive devices (i.e., a biased photoresistor that exhibits resistance changes when illuminated). Although some photoconductive detectors are still used, they are being replaced by photovoltaic devices (i.e., zero- or reverse-biased diodes that exhibit current increases when illuminated) that have the advantage of drawing much lower current for large array applications. Related to photovoltaic devices are the Impurity Band Conduction (IBC) detectors (which contain a high-low junction as opposed to a p-n junction) fabricated from doped silicon, and Schottky-barrier detectors fabricated from silicides. There have also been some Metal-Insulator-Semiconductor (MIS) detectors fabricated, especially from HgCdTe and InSb, but such devices have not received wide usage. The infrared detection wavelength regime is divided into several sub-regions. Those regions and the detectors that operate in them are shown in Table 3. The detectors shown in bold are the primary detectors in a given wavelength region. Table 3. Wavelength Regions in which Infrared Detectors Operate. Wavelength Region SWIR MWIR LWIR VLWIR
Wavelength (µm) 1–3 3–5 8 – 14 >14
Detector Types HgCdTe, InGaAs, silicides, PbS HgCdTe, InSb, superlattice, silicides, PbSe HgCdTe, superlattice, doped germanium, PbSnTe IBC silicon, doped silicon, HgCdTe, superlattice, doped Ge
In part because of the pervasive nature of HgCdTe detectors, most of the radiation-effects information for infrared detectors has been developed for HgCdTe. Therefore, we emphasize
IV-38
that material in the remainder of Section 4 and in general refer only briefly to the other IR detector materials.
4.3
Ionization-induced Transient Response
4.3.1 Ionization-induced Pulses and Noise Ionization-induced transients often create the most important radiation-response issues in applications of infrared detectors. In order to detect optical photons, infrared detectors must be very sensitive. That is, they must have a high sensitivity to photons and a low noise floor. As a result, they are also extremely effective detectors of ionization associated with radiation environments, including often being able to sense individual particles. This is similar to single-event transients caused by heavy ions in standard integrated circuits, except that the sensitivity of infrared detectors is so great that they often can detect single electrons. In fact, infrared detectors are often designed using the same principles used to design nuclear radiation detectors. The two main applications that drive the concern for transients are those systems that must consider the effects of gamma and electron fluxes and astronomy systems that must deal with the effects of cosmic rays. Discussion of the modeling of proton-induced transients (and, to a lesser extent, electron-induced transients) is covered in Section 5. The present section focuses on gamma-induced transients.
γ
γ
e-
e-
Figure 21. Schematic illustration of the interaction of gamma rays with a detector.
When gamma rays impinge on and interact with materials, they produce electrons by the photoelectric effect (primarily at gamma energies < 0.5 MeV), Compton scattering (at
IV-39
intermediate gamma energies), or pair production (at gamma energies > 1.02 MeV). The most important of these effects for detectors is Compton electron production because the lower-energy photoelectrons do not have a very long range and thus are absorbed near where they are produced. Also, there are few gamma rays with energies sufficient to initiate pair production. The gamma-induced Compton electrons that strike a detector interact with it and produce pulses. Most of those pulses are not the result of direct interaction of gammas in the detector. Rather, the gammas interact with the surrounding material and produce Compton electrons, which in turn interact with the detector, as shown schematically in Figure 21. Approximately 5% to 10% of the pulses are produced by gammas that interact directly with the detector, as also illustrated Figure 21. The pulses produced by the interaction of a flux of gamma rays with a detector have a distribution of amplitudes. The primary reason a distribution of pulse amplitudes is produced is because of the varying path lengths that the electrons traverse through the detector volume. A secondary reason is differences in the energy loss rate of different energy Compton electrons. A typical distribution is shown in Figure 22 [64]. In order to calculate the gamma-induced pulse amplitude distribution in a detector with high fidelity, one needs to perform a radiation transport calculation. Unfortunately, those calculations are not simple easy-to-perform analytic calculations. Further, transport codes often truncate their calculations at a finite electron energy, which ignores low-energy electrons that can be significant in producing the pulse amplitude distribution.
Figure 22. Gamma-flux-induced pulse amplitude distribution in HgCdTe detectors at T = 77 K [64].
IV-40
Fortunately, Pickel and Petroff developed a semi-empirical method for making estimates of the pulse amplitude distribution that is analytic and easy to use [65]. The average chord length ( C ) in a rectangular volume is given by
C =
2 1/ x + 1/ y + 1/ t
(1)
where x, y, and t are the dimensions of the detector volume. Some of the Compton electrons (usually between 5% and 10%) stop or start within the detector. To allow for those electrons, a term is added empirically to the above equation, and one finds the average path length ( Le ) of the Compton electrons in the detector to be Le =
2 1 / x + 1 / y + 1 / t + 2 / Lc
(2)
where L c is the average path length of the average-energy Compton electron in the detector material. The average pulse amplitude n in carrier pairs is then given by n=
E Le Lc ε p
(3)
where E is the average energy of the Compton electrons produced by the gammas and εp is the energy required to create a carrier pair in the material. The internal event rate (Ri) produced by the gammas interacting directly with the detector can be shown to be given by [65]
Ri = µ V ϕγ
(4)
where µ is the linear absorption coefficient of the gammas in the material, V is the detector volume, and ϕγ is the gamma flux. The external event rate (Re) produced by the Compton electrons interacting with the detector is [65] Re =
[
]
µ ( xy + xt + yt ) Lc φγ 2
(5)
The total event rate (R) is just the sum of the internal and external event rates. Although Pickel and Petroff [65] also developed an expression for the pulse amplitude distribution, it is not of analytic form and requires a computer code for solution. Most users
IV-41
of this semi-empirical approach simply rely on the empirical observation that pulse amplitude distributions are approximately exponential, and determine the predicted distribution from the calculated average event amplitude and total event rate. This approach tends to underpredict the number of pulses at low amplitude and over-predict the number at high amplitude. It is generally believed that the exponential approximation gives results that are within an order of magnitude at all amplitudes (except very large ones where an exponential distribution remains finite but the real distribution drops to zero), and that are within a factor of two at most amplitudes. The gamma-induced pulses in a detector result in increased noise. An example of measured gamma-induced rms noise in HgCdTe and PbSnTe detectors is shown in Figure 23 [64]. The equation to calculate the noise squared (Nγ2) produced by events with a distribution of amplitudes is 2
Nγ = 2 R tint n 2
(6)
where tint is the integration time and n 2 is the second moment of the amplitude distribution of the events. For an exponential distribution of amplitudes n2 = 2 n 2
(7)
where n is the first moment of the distribution, and is the same as the average pulse amplitude calculated from Equation (3). Using Equations (3) through (7), one finds (for an exponential distribution of pulses) that ⎡ 2 E C ⎤⎡ ⎧ ⎤ ⎛ xy + xt + yt ⎞ ⎫ Nγ = ⎢ ⎥ ⎢ µ ⎨ xyt + ⎜ ⎟ Lc ⎬ϕ γ t int ⎥ 2 ⎝ ⎠ ⎭ ⎢⎣ Lc ε p ⎥⎦ ⎣ ⎩ ⎦
½
(8)
In a similar manner, one can show that the gamma-induced noise current (iγ) produced by an exponential distribution of pulse amplitudes is ⎡ 2E C e ⎤ ⎡ ⎧ ⎤ ⎛ xy + xt + yt ⎞ ⎫ iγ = ⎢ ⎥ ⎢ µ ⎨ xyt + ⎜ ⎟ Lc ⎬φγ ∆f ⎥ 2 ⎝ ⎠ ⎭ ⎦ ⎣⎢ Lc ε p ⎦⎥ ⎣ ⎩
½
(9)
where ∆f is the bandwidth of the measurement. The prefactor ( n 2 / n 2)½ in this analysis — the prefactor is 2½ in this case of an exponential distribution such as was used to obtain Equations (8) and (9) — has been defined by some [66] as the shot-noise multiplier. That multiplier varies, depending on the shape of the distribution, but can be readily calculated if the shape is known analytically or from experimental data.
IV-42
4.3.2 Low-dose-rate Effects Some extrinsic detectors, especially doped silicon and doped germanium, display another type of response when exposed to a flux of ionizing particles while operated at low temperature and low optical background. The optical response increases while the detector is exposed to the ionization flux, and then decreases when the flux is removed. These effects can occur at low dose rates, and the time constants involved can be quite long. An example of the response of an arsenic-doped silicon detector is shown in Figure 24 [67]. In this case, the time constants were tens of seconds. At even lower optical backgrounds, time constants on the order of hours or even days have been observed. The phenomenon was termed the “gamma response anomaly” when it was initially observed.
Figure 23. Gamma-flux-induced noise in HgCdTe and PbSnTe detectors. T = 77 K [64].
The cause of this effect is that the ionization-induced charge fills empty sites at which optically excited carriers could recombine. For example, in arsenic-doped silicon at an operating temperature of 10 K, all of the donor electrons are frozen out on the arsenic sites. Some of the arsenic sites are compensated (empty) by compensating acceptor sites. Optically generated electrons, which are generated from filled arsenic sites, fall into empty arsenic sites, effectively recombining with a positive charge, as shown in Figure 25. The lifetime of the optically generated charge, and thus the optical response, is inversely proportional to the concentration of empty arsenic sites. Ionization-generated electrons, which are generated from the valence band, also fall into empty arsenic sites, as shown schematically in Figure 26. However, the ionization-generated holes in the valence band are swept out of the detector, leaving the detector in a metastable condition with fewer empty arsenic sites than previously. This increases the lifetime of optically generated carriers and causes an increased response. The long time constants are because there are few carriers available at low temperature and background, and it takes a long time for the device to reach a new equilibrium when a change is made. This is a process is similar to dielectric relaxation. In
IV-43
fact, similar long-time-constant changes were subsequently observed when other changes (e.g., optical level, temperature) were made.
Figure 24. Long equilibration times exhibited by extrinsic photoconductive detectors operated at low temperature and low optical background. Arsenic-doped silicon, operating temperature = 4.3 K, optical background = 8x1012 photons/cm2-s, gamma flux = 10 rad(Si)/min. Gamma flux initiated at t = 0 and halted at t = 52 s [67].
EC
ED
EA
EV
Figure 25. Schematic of optically excited electrons recombining at empty donor sites in n-type doped-silicon detectors.
IV-44
EC
ED
EA
EV
Figure 26. Schematic of ionization-generated electrons filling empty donor sites in n-type doped-silicon detectors, creating a metastable condition and increasing the optical response.
The long time constants and the pervasive nature of the gamma-response anomaly in extrinsic silicon detectors was a major driving force behind the development of IBC silicon detectors. IBC devices are essentially a reverse-biased high-low junction. The doping level is high enough that a band of donor sites is formed that can exchange charge with each other, as shown schematically in Figure 27, instead of the donor sites being isolated and not able to interact with each other, which is the case in doped-silicon detectors. As a result, charge motion can occur along the impurity band and it easier to reestablish equilibrium in IBC detectors when changes are made. Therefore, the time constants are much faster (on the order of µs) [68], [69], as illustrated in Figure 28 [68].
EC
ED
EA
EV
Figure 27. Charge can be swept along the impurity band in IBC detectors, allowing equilibrium to be reestablished easily.
IV-45
Figure 28. IBC silicon detectors operated at low temperature and low optical background exhibit much shorter equilibration times. IBC As-doped silicon, operating temperature = 6 K, electron flux = 4E9 e/cm2-s. Electron flux initiated at t = 5 µs and halted at t = 15 s [68].
4.3.3 Prompt-pulse-induced Transients When an infrared detector is exposed to a high-dose-rate ionization pulse (a prompt pulse), the individual events are no longer distinguishable and a large current pulse is observed [64], [70], [71]. It is easy to calculate the magnitude of the charge produced (Qφ) by the ionizing pulse if the detector volume is known. One starts with the dose rate and multiplies by the pulse width to get the dose deposited (φ) (or uses φ itself if this is what is known). One then uses the definition of ionizing dose (1 rd = 100 ergs/g), performs a units conversion to units of eV/cm3, divides by the energy required to create a carrier pair (εp), and multiplies by the electron charge and detector volume (V) to obtain
⎡φ ρ V ⎤ −5 Qφ = ⎢ ⎥ • 10 ⎢⎣ ε p ⎥⎦
(10)
where ρ is the density of the detector material. The temporal response of infrared detectors to a prompt pulse is generally controlled by circuit parameters and not by the detector itself. The rise time of the response usually follows the ionization pulse, with the decay time determined by how fast the circuit can remove the ionization-generated carriers. An example of prompt-pulse response is shown in Figure 29 [70].
IV-46
Figure 29. Prompt pulse response of a HgCdTe detector. X-ray pulse of 2.5E9 rad(Si)/s, pulse width = 20 ns, T = 80 K [70].
The magnitude of the prompt-pulse response is quite large in any type of infrared detector, even for very-low-dose-rate prompt pulses. As a result, the prompt pulse response of infrared detectors will almost always produce a response pulse that is large compared with the optical signal. Whether or not this large ionization-induced pulse upsets a system depends of other factors such as the recovery time and the sensitivity of subsequent electronics. However, most optical sensors are designed to ignore glitches, and prompt-pulse-induced glitches can be filtered out of the data stream as long as the recovery time is sufficiently fast.
4.4
Permanent Degradation
4.4.1 Displacement Damage Effects In a path very similar to that for silicon integrated circuit (IC) technology, the first infrared detectors were single element devices. They may have been passivated to stabilize the device properties, but they did not include insulators as an integral part of the active device. Therefore, displacement damage dominated the permanent degradation, and total-ionizingdose effects were not important. There have been many studies performed to investigate displacement damage in HgCdTe material [72]-[89].
The basic displacement damage effect of irradiating HgCdTe is the introduction of donors, which are probably Hg vacancies. The donor introduction rate in LWIR HgCdTe at 80 K is IV-47
shown as a function of electron energy in Figure 30 [78]. The theoretical curve shown in the figure is based on the Nonionizing Energy Loss (NIEL) rate of electrons, and has been normalized to the experimental data. Also shown in the figure are the donor introduction rates for fission and 14-MeV neutrons, and some data measured at 10 K. These radiationinduced donors are also Shockley-Read-Hall (SRH) centers that degrade the lifetime. Therefore, the donor introduction rates are also essentially the same rate at which recombination centers are introduced.
Figure 30. Energy dependence of the carrier introduction rate in LWIR HgCdTe [78].
For an n-type photoconductive detector, the optical response (∆V) can be shown to be given by ∆V =
gτ m V , n
(11)
where g is the optical generation rate, τm is the majority-carrier lifetime, V is the applied bias, and n is the majority-carrier (electron) concentration [90]. The parameter found to be most sensitive to displacement damage is n [72]-[89]. Displacement-damage-induced donor introduction will increase the electron concentration, which in turn will degrade the optical response. For high performance, the initial donor concentration is made as low as possible, usually on the order of 1x1015 cm-3. The addition of the same concentration of displacementinduced donors will degrade the optical response by a factor of two.
IV-48
Photovoltaic detectors are diodes and are minority-carrier devices. The best performance metric for photovoltaic detectors is the product of the resistance at zero bias times the detector area (R0A). The best photovoltaic detectors are operated in the diffusion current regime where the zero-bias resistance is dominated by diffusion of minority carriers to the depletion region. Early detectors were all n-on-p devices with R0A dominated by diffusion current from the p-type material. In this case, R0A is given by R0 A =
k T pτ 2 e 2 ni t
(12)
where k is the Boltzmann constant, T is the temperature, p is the hole carrier concentration, τ is the minority-carrier lifetime, e is the electronic charge, ni is the intrinsic carrier concentration, and t is the detector thickness [91]. A similar equation can be developed for pon-n devices. Because the hole concentration on the p side in these devices was relatively high, it would require the introduction of a significant number of displacement-induced donors to decrease this concentration. Therefore, the parameter most impacted by displacement damage is the minority-carrier lifetime. This lifetime is degraded by the introduction of SRH centers. The initial SRH defect concentration in photovoltaic HgCdTe detectors is made as low as possible, often on the order of 1x1015 cm-3. The introduction of the same concentration of SRH centers by displacement damage would degrade R0A by a factor of two. Another damage mechanism that could impact R0A is lifetime degradation on the n side. The maximum minority-carrier lifetime in n-type HgCdTe is set by the Auger lifetime, which varies inversely as the square of the electron carrier concentration [91]. For good performance, this electron carrier concentration is kept low, perhaps as low as 1x1015 cm-3, so the Auger lifetime on the n side can be approximately as sensitive to displacement-damageinduced effects as the minority-carrier lifetime on the p side. The estimated degradation thresholds, defined as the fluence at which the responsivity degrades by a factor of two, for HgCdTe and other detector types are shown in Table 4. In three studies, [64], [70], [71] photovoltaic HgCdTe detector arrays were exposed up to a level of ~2x1013 n/cm2 (fission neutrons), which is less than the estimated degradation threshold, without observing any displacement-induced degradation. At this exposure level, the accompanying total ionizing dose had begun to produce degradation, as discussed in Section 4.4.2. One study of fast neutron (14- or 15-MeV) damage in InSb photoconductive and photovoltaic detectors has been reported [92]. The photoconductive detectors were p-type, and the hole removal rate was found to be 1.1 cm-1. This resulted in a slight increase in the optical response of photoconductive detectors beginning at a fluence of ~5x1012 n/cm2. The photovoltaic detectors were n-on-p devices with a leakage current dominated by generationrecombination current from the more lightly doped n-type base region. The most radiationsensitive parameter in these devices was the minority-carrier lifetime in the n region. The impact of neutron irradiation on R0A was not reported, but the optical response degradation IV-49
threshold was reported to be 3x1011 n/cm2. These degradation thresholds are shown in Table 4. Table 4. Degradation Thresholds in Various Detectors. Detector Type
Radiation Environment
LWIR HgCdTe
Photoconductive InSb Photovoltaic InSb
Photoconductive PbS
Si:As
Fission Neutrons 14-MeV Neutrons 2-MeV Electrons Co60 Gammas 14-MeV Neutrons 14-MeV Neutrons Thermal Neutrons 14-MeV Neutrons 7.5-MeV Protons 12-MeV Protons 133-MeV Protons 450-MeV Protons Fission Neutrons
Irradiation Temperature (K) 78 78
Displacement Damage Threshold ~3x1014 n/cm2 ~1x1014 n/cm2
78 78 78
~6x1014 e/cm2 ~4x107 rd(HgCdTe) ~5x1012 n/cm2
78
~3x1011 n/cm2
300
~5x1015 n/cm2
300
~2x1013 n/cm2
300 300 300
~2x1012 p/cm2 ~7x1012 p/cm2 ~1x1013 p/cm2
300
~2x1013 p/cm2
10
~1x1011 n/cm2
Photoconductive PbS detectors are an example of detectors operated with internal gain (G), which is the ratio of the material lifetime (τ) to the sweepout time (ts) in the device. The optical responsivity (R) in such a detector can be written as R=
η G eλ
(13)
hc
where η is the quantum efficiency, λ is the wavelength, h is Planck’s constant, and c is the speed of light. The sweepout time can be written as ts =
1 µE
(14)
where µ is the mobility and E is the field across the detector. (Although it could lead to confusion, we use the symbol “µ“ for mobility in Equations 14 through 16 and also for the linear absorption coefficient of gammas in Equations 4 and 5 because this is the common symbol for both of these parameters.) Thus,
IV-50
G =τ µ E
(15)
and R=
ητ µ E e λ
(16)
hc
Displacement damage produced by thermal neutrons [93], 14-MeV neutrons [94], [95], or protons [93], [96] in PbS photoconductive detectors has been reported. The only parameters in Equation 16 that can be affected by displacement damage are the mobility and the lifetime, with the lifetime being by far the more sensitive. However, those reports did not provide enough information to determine any material parameters, so we can only summarize the degradation threshold. The fluence at which the responsivity degraded by a factor of two for each particle type investigated is shown in Table 4. Although not always stated, it is likely that all of those PbS tests were performed at room temperature. As noted previously, in doped-silicon detectors, the optical response is inversely proportional to the concentration of recombination centers. In n-type doped-silicon detectors operated at 10 K, the recombination centers are donor sites that are compensated by acceptor sites. Displacement damage introduces additional acceptor sites that in turn compensate additional donor sites, as shown schematically in Figure 31. This degrades the lifetime of optically generated carriers and the optical response. In high-quality detectors, the concentration of compensating acceptor sites is kept as low as possible for long lifetime and high performance. In the one reported study of neutron-irradiated arsenic-doped silicon material [97], the acceptor concentration was ~5x1013 cm-3. The measured acceptor introduction rate was 16 cm-1 in both float-zone and pulled-crystal samples. This neutron-induced response degradation remained approximately the same for annealing temperatures as high as 673 K. This resulted in a factor of two response degradation after exposure to ~3x1013 n/cm2 in the tested material. However, the acceptor concentration in high-quality detectors is usually on the order of ~2x1012 cm-3, which would make the degradation threshold as low as ~1x1011 n/cm2 in detectors fabricated from higher quality material. This lower degradation threshold is shown in Table 4 because it represents what would be expected in high-quality detectors. 4.4.2 Total-dose-induced Degradation 4.4.2.1 HgCdTe Devices with ZnS and Other Passivations Susceptible to Total-dose Exposure As infrared detector development shifted to multi-element detector arrays, surface passivation between the individual detector elements became necessary. In HgCdTe arrays, the initial architecture was n-on-p photovoltaic detectors passivated with a deposited layer of ZnS. ZnS is very effective at trapping charge, so total-dose-induced permanent degradation became much more important than displacement effects. As a result, many studies were reported that investigated the mechanisms of total-dose-induced effects in arrays and how one might harden against them [64], [98]-[108]. Most of the early hardening approaches involved investigation of alternate passivation insulators such as anodic sulfide [109]-[111]
IV-51
or deposited SiO2 [100], [112]. Occasionally, unpassivated devices were also investigated [103]. While some quantitative differences were observed, both MWIR and LWIR devices exhibited similar behavior.
EC
ED
EA
Edef EV
Figure 31. Compensation of donor sites by displacement-induced acceptors in n-type dopedsilicon detectors.
Studies of the C-V characteristics of MIS capacitors demonstrated that both electrons and holes are trapped in ZnS, as shown schematically in Figure 32, with the net charge near the HgCdTe surface depending on the sign of the electric field across the ZnS (especially at the interface) [99]-[102], [109]-[112]. The total-dose-induced trapped charge in the ZnS causes a shift in the surface potential in the HgCdTe. Depending on the sign of the net trapped charge near the interface with the HgCdTe, either accumulation or depletion of the HgCdTe surface is produced. Either condition causes increased leakage currents in the detectors. In addition, if depletion proceeds far enough to cause surface inversion, a conduction path between previously isolated diodes can occur, which increases crosstalk. The magnitude of the applied bias and the surface treatment of the HgCdTe control the amount of charge trapped and thus the size of the potential shift, as shown in Figure 33 [100]. Even at relatively high bias, the net trapped charge is only a few percent of the total charge produced in the ZnS by the irradiation. At zero bias, the sign of the net trapped charge can be positive or negative, depending on the surface treatment of the HgCdTe and perhaps other factors (e.g., stray fields). Therefore, the ionization-induced trapped charge in the ZnS can either cause accumulation or depletion (and eventually inversion, leading to increased crosstalk due to the presence of a conducting path between detector elements) of the HgCdTe surface between the diodes. In either case, this causes an increase in the surface leakage current and degraded array performance, as shown in Figure 34 [64]. ZnS is so effective at trapping charge that HgCdTe arrays passivated with that material exhibited degradation at low TID exposures that ranged from ~3x103 rad(ZnS) to ~5x104 rad(ZnS). In general, detectors with higher initial quality (e.g., higher R0A) exhibited less total-dose vulnerability. However, these high-quality, total-dose-hard detectors could not be reproduced consistently, nor could they be produced uniformly across full arrays. Thus, while there was an existence proof that high-quality, rad-hard detectors could be produced, it eventually became clear that ZnS passivation was not the solution.
IV-52
Surface Accumulation
Surface Depletion/Inversion
ZnS
n
ZnS
n
n
n
n
HgCdTe
n HgCdTe
p-type
p-type
+V
-V
Figure 32. ZnS can trap both ionization-induced electrons and holes, with the sign of the net trapped charge near the interface with HgCdTe depending on the field across the ZnS.
Figure 33. Total-dose-induced flatband shift in HgCdTe MIS capacitors with ZnS insulator and different surface treatments. Total dose = 6.8x103 rad(ZnS), T = 77 K [100].
The finding that detectors with high initial quality generally were also more total-dose tolerant led to the eventually accepted hypothesis that the mechanism that caused high totaldose susceptibility was related to the cause of poor detector performance. Therefore, it was thought that solving the problem of poor detector performance would likely result in totaldose hardness as well. In addition, it was clear that the state of the interface between the HgCdTe and the passivation was key to device performance and hardness, and that IV-53
developing a high-quality interface that could be fabricated reproducibly was the most important issue.
Figure 34. Total-dose-induced increase in leakage current in ZnS-passivated HgCdTe diodes. Total dose = 1E5 rad(HgCdTe), T = 125 K [64].
While use of alternate passivation materials (e.g., anodic sulfide, deposited SiO2, or silicon nitride) sometimes decreased the total-dose vulnerability of HgCdTe arrays, the improvement was not enough to be called a solution. In particular, high performance and hardness could not be obtained consistently. Unpassivated devices were found to be hard to ~3x106 rad(HgCdTe), as shown in Figure 35 [103]. However, passivation for providing device stability is even more important for HgCdTe devices than it is for silicon devices, so elimination of the passivation was not an acceptable option. All of the ionization-induced trapped charge was found to anneal out by 300 K. Thus, periodic heating of HgCdTe arrays can effectively remove all the radiation damage. However, this was not a solution that could be used in all applications. The eventual solution was the development of CdTe passivation, as discussed in Section 4.4.2.2. 4.4.2.2 Hardened HgCdTe Devices with CdTe Passivation
Eventually, most manufacturers of HgCdTe arrays successfully developed CdTe passivation, which is much more compatible with HgCdTe than was ZnS or any of the other passivations used earlier. The driving force behind the development of CdTe passivation was improved detector and array performance, especially for LWIR devices. The CdTe approach proved very successful, as high performance detectors and uniform arrays (with fewer bad elements) were produced. At the same time as the switch to CdTe passivation occurred, most manufacturers also shifted from n-on-p architectures to p-on-n architectures. While use of the p-on-n architecture did lead to revolutionary changes in device performance (especially when
IV-54
combined with CdTe passivation), its impact on total-dose hardness was at most evolutionary. Higher total-dose hardness was primarily the result of using CdTe passivation. Arrays passivated with CdTe were much harder to total-dose exposure [70], [71], [113][118]. It was found that CdTe-passivated HgCdTe detectors would survive exposure to >1 Mrad (HgCdTe), as shown in Figure 36 [70]. This meant that the hardness of a hybrid array was no longer controlled by the response of the detector array, but was controlled by the hardness of the MOS readout circuit.
Figure 35. Total-dose-induced increase in leakage current in unpassivated HgCdTe diodes. T = 77 K [103].
4.4.2.3 Total-dose Effects in ROICs
Since the detector array and readout integrated circuit are hybridized (usually using indium bump bonds) and the detectors must be operated at cryogenic temperatures, the readouts are also operated at cryogenic temperature. Total-dose effects are more severe in MOS devices operated at cryogenic temperatures than they are at room temperature because of enhanced charge trapping in the oxides. Hybrid arrays with unhardened readouts tend to fail after exposure to a few tens of krad(SiO2). It is possible to obtain hardened readouts that will survive up to one Mrad(SiO2) at cryogenic temperature but, as is the case for roomtemperature electronics, the cost of rad-hard readouts is high and the number of process lines that are willing to fabricate them is diminishing. Fortunately, the trend toward use of higher density CMOS processes is also favorable to total-dose hardness since oxides are thinner and inversion thresholds tend to be higher for scaled processes. Use of hardness-by-design IV-55
practices and submicron processes has allowed radiation-tolerant readouts that are acceptable for many uses in the natural space radiation environment to be fabricated in commercial foundries. The radiation response of readouts is discussed in more detail in Section 6.
Figure 36. Improved total-dose hardness in CdTe-passivated HgCdTe arrays. T = 80 K [70].
4.5
Summary and Future Directions
Radiation effects mechanisms in IR detectors and imaging arrays continue to require attention as the technology evolves and applications become more demanding. The transients and noise produced in infrared detectors by individual gamma photons and particles, such as electrons, protons and heavy ion, remain an issue that is as important now as it was when the technology was first developed and the problem recognized. There have been some approaches postulated for separating the optically induced charge from the ionizationinduced charge in extrinsic or superlattice detectors. These approaches have sometimes been termed Intrinsic Event Discrimination (IED). The underlying premise is that the two energy states for the ionization-induced charge are the valence and conduction bands in the material, while one of the states for the optically induced charge is a bound state (e.g., an impurity level in an extrinsic detector or a bound state in a superlattice). However, no IED device concept has been fully demonstrated, and there are significant questions about how well (or even whether) such a device will eliminate the ionization-induced response. Because the amplitude of the ionization-induced pulses in most detectors depends primarily on the smallest dimension (usually the thickness - see Equations 2 and 3), use of thin detectors does result in smaller ionization-induced pulses. However, even these smaller pulses can be larger than the optical signal, and thus can still interfere with device
IV-56
performance. On the other hand, the number of ionization-induced pulses depends on the average projected area of the detector and is controlled primarily by the optical area (see Equations 4 and 5). The average projected area doesn’t decrease much as the detector is thinned, so use of thinned detectors does not produce significantly fewer ionization-induced pulses. There have been some reasonably successful attempts to put microlenses on detectors to shrink the electrical area of detectors while retaining the optical area. One major driver for this activity is to decrease the volume from which leakage current is produced while retaining the full optical response, thus improving detector performance even without considering exposure to ionizing radiation. In addition, since ionizing particles would not be focused by these microlenses, the number of ionization-induced pulses would be decreased by this approach. Thus, microlenses should be a successful means to harden infrared detectors against fluxes of ionizing particles. However, testing of the impact of microlenses on the transient response of infrared detectors to a flux of ionizing particles has not been reported. Many applications handle ionization-induced pulses and noise by recognizing and eliminating the unwanted pulses. If the array is oversampled either temporally or spatially, the ionization-induced pulses, which are temporally and spatially confined, can be eliminated without significantly degrading the optical signal. The recognition and elimination of ionization-induced pulses is usually done electronically off the focal plane [119], and is quite effective if the application can stand the amount of oversampling necessary, which usually requires significantly higher data rates and additional signal processing. Total-dose-induced degradation continues to be an important damage mechanism in modern infrared detector arrays, particularly for space applications. HgCdTe technology has solved the problem in detectors by going to CdTe passivation. The issue has not been addressed in most other detector technologies, so there may be technologies where the detector arrays are still susceptible to total-dose exposure. In addition, all detector array technologies, including those in which the detectors are total-dose hard, are subject to the total-dose susceptibility of the ROIC. While total-dose-tolerant MOS readouts exist, fabrication lines that can process such devices are becoming harder to find as organizations drop out of this small niche business. Fortunately, the commercial trend toward higher-density processes benefits totaldose hardness because oxides are thinner and surface inversion thresholds are higher. By using hardening-by-design techniques and modern processes, readouts with adequate totaldose tolerance can be fabricated in commercial foundries for many applications requiring radiation hardness. This topic is discussed in Section 6. Displacement-damage-induced degradation was not a primary problem in most infrared detector arrays (in particular for HgCdTe technology) because total-dose susceptibility of the detector arrays was the limiting mechanism. However, as the total-dose problem in detector arrays is solved, displacement-damage effects are becoming a more important degradation mechanism. In addition, as material quality improves (e.g., by lowering native defect concentrations), the devices become more susceptible to displacement-damage-induced defects. However, the displacement-damage thresholds of most detector materials are still high enough that displacement damage should not be a major issue in most applications. One exception is silicon-based detectors, which have the advantage of high-quality material developed by the commercial silicon industry. However, this advantage makes silicon
IV-57
detectors susceptible to displacement-damage effects. Another exception may be astronomy applications where even small increases in dark current can be problematic.
IV-58
5.0
Modeling of Radiation-induced Transients in Focal-plane Arrays
When energetic protons traverse the active detector volumes of visible or infrared focal planes, the production of electron-hole pairs by ionization along the traversing proton tracks is almost always sufficient to produce signals above typical thresholds of detection for modern high-performance focal planes. These proton-induced “signals” act as a source of interference to the detection of desired scene information from photoelectrons. In order to better understand the system-level impacts of proton transient effects on visible and infrared imaging devices, proton-induced transient effects simulations have been developed and validated [120]-[122]. These simulations are quite detailed and include most of the important physics associated with the interaction of protons with the pixel volume(s) in visible and infrared focal-plane arrays. At a qualitative level, much can be inferred about the frequency of occurrence of proton transients (i.e. the “hit rate”) by examining parametric proton flux curves such as those found in the figures in Section 2 for the MIDLAT and MOLNIYA orbit examples. For example, Figure 6(b) shows external proton flux versus altitude and proton energy. Spacecraft and payload shielding typically removes low-energy protons for energies below about 20 MeV. However, the high-energy tail of the trapped proton spectrum is energetic enough to penetrate feasible amounts of shielding and impinge on the focal plane of a space-based electro-optical sensor. We expect that protons with energies above approximately the 33MeV curve in Figure 6(b) will make it through the shielding and strike the focal plane. If we select the peak proton flux with energy below about 33 MeV [Figure 6(b)] (this occurs at an altitude of about 5000 km) we get a proton flux at the focal plane of about 20,000 protons cm-2sec-1. Assuming that all of these protons produce focal-plane responses above some detection threshold, we can estimate the total number of hits in a frame of focal-plane data by taking the product of the flux quoted above, the total detector area, and the integration time. For example, for a 1024 by 1024 array with 18-micron pixels operating at an integration time of 33.3 msec (i.e. a 30-Hz frame rate), we expect about 2000 proton-induced transient signals per frame. This represents an image “contamination” level of about 1 out of 500 pixels. Of course, with longer integration times more pixels will be contaminated with these transient events. The simulation example described in this section aims to make the prediction of transient effects due to the post-shielding proton/focal plane interaction as accurate as possible both in terms of frequency of occurrence and the details of the physical interaction.
5.1
Simulation of Proton Transient Effects
Figure 37 shows a block diagram of the end-to-end simulation process. The external proton flux and energy are modified after propagation through various thicknesses of shielding. It is that spectrum and flux that act as input to the Radiation Event generator in the simulation block diagram.
IV-59
. Figure 37. End-to-end focal-plane proton effects simulation process
5.1.1 Proton Environment Synthesis at the Focal Plane
Given that an estimate of the proton spectrum at the focal plane is available, several parameters still must be specified to characterize a proton event, which are shown in Table 5. In this example, the spacecraft is modeled as a solid sphere, so the impact coordinates on the focal plane are uniformly distributed. This would change if a detailed spacecraft model were being analyzed. Table 5. Random proton event parameters and key characteristics for the example described in the text. Parameter Impact position Trajectory - Azimuth - Angle of Incidence Time of Arrival Energy
Symbol x, y φ θ t E
Key Characteristics Uniformly Distributed Isotropic Uniformly Distributed Not Uniformly Distributed Poisson Process Conforms to Energy Spectrum at the Focal Plane
Protons trapped in the earth’s magnetic field have a preferred trajectory in a plane normal to the earth’s magnetic field because protons having a large component of their velocity along the field lines penetrate the atmosphere at their mirror points and have a greater probability of being scattered out. At low attitudes, there also is an east-west asymmetry caused by scattering of protons that penetrate the atmosphere during the lower
IV-60
excursion of their orbit of magnetic field lines. Thus, the expected distribution of proton trajectories depends on orientation of the spacecraft (focal plane) in the orbit as well as on shielding geometry. In the case given here, the proton flux is chosen to be isotropic. For an isotropic proton flux, the probability of a proton striking the focal plane at an angle of incidence between θ and θ + d θ i s P(θ)dθ = 2 sin θ cos θdθ
(5-1)
where 0o< A<90°. The sin θ factor accounts for the projected area of the focal plane, and the cos θ cofactor accounts for the variation in size of the differential solid angle at θ. Thus, the most probable angle of impact for an isotropic flux is 45°. The distribution of azimuth angles is taken to be uniform. The probability of the time between events, τ, being between τ and τ + dτ is calculated using
P ( τ ) dτ =
e −4πAΦτ dτ 4πAΦ
(5-2)
where Φ is the total integral flux at the focal plane and A is the area of the focal plane. This is consistent with arrival times for a Poisson process. A computer program was written to produce as output a file containing random proton events with parameters x, y, ϕ , θ , E, t, and τ behaving statistically according to these probability functions. The program just as easily can produce events with these parameters computed to conform to characteristics of a laboratory proton beam or other anisotropic environment. 5.1.2 Proton Energy Transfer to Semiconductor Focal Plane
The Coulombic (electrical) interaction between the incident particle and the electrons in the semiconducting solid is by far the most significant interaction in terms of relevance to focal-plane transient response. This interaction is responsible for ionizing energy loss for the incident proton and carrier generation in the focal plane. If the interaction takes place in the active region of the semiconductor (the epitaxial layer in most modern devices), then the generated charge is free to diffuse and/or be drawn into the depletion regions of the pixels. If, on the other hand, the charge is generated within oxides, then semipermanent net charging of those oxide layers can result, thereby causing an alteration or degradation of device performance. Device degradation also can occur within the focal plane due to nonionizing energy loss that is associated with lattice displacements, as discussed in Sections 3 and 4. The focus of the present section is on transient response simulation and characterization, so device degradation will be disregarded here. The
IV-61
discussion here is limited to carrier generation within the active (epitaxial) region by ionizing energy loss of the incident protons. It has been shown [123] that for many semiconducting materials, the energy required to generate an electron-hole pair, ε, is given by ε = (14 / 5 ) E g + r ( hν R ) where 0.5 ≤ r ( hν R ) ≤ 1.0 eV
(5-3)
The energy in excess of the bandgap (Eg) and kinetic energy (the first term) is channeled into thermal and optical phonons. The relatively small amount of energy transferred to an electron makes intuitive sense, because conservation of energy and momentum limit the amount of energy transferred between particles of substantially different mass (Mproton ~ 2000 × Melectron).
Figure 38. Proton stopping power in Si and HgCdTe.
With this result, the energy transfer equation can be converted from energy/unit length to free carriers/unit length of interaction expressed as LET: LET =
dE dE (electrons / micron) = (energy / micron) / ε dx dx
(5-4)
The LET (= dE/dx/energy/carrier) varies along the path of interaction. In many instances, it is appropriate to assume that the LET is effectively constant in order to simplify the calculation. Note in Figure 38 that a 10-MeV incident proton will lose less than 10 keV per µm of interaction length in silicon. Since dE/dx is a slowly varying function in this (and higher) energy range, we can safely assume constant LET in the interactions that typically occur. For example, a 20-µm pathlength in silicon would result in only a 2% energy loss for a 10-MeV proton, leaving dE/dx essentially unchanged. The suitability of the constant LET approximation must be evaluated based on the energy spectrum of the incident protons. We have found that the proton energy spectra associated with most IV-62
natural environments investigated are reasonably appropriate for use with the constant LET approximation. As an example of a case in which the approximation would break down, consider a post-shielding proton spectrum that is peaked at 2 MeV and incident on a HgCdTe focal plane. It can be seen in the stopping power figure that a 2-MeV proton will lose about 40 keV (or 2%) of its energy for only 1 µm of interaction length. Since a HgCdTe detector may have a nominal “active” epitaxial thickness of 15 µm, even a normal-incidence penetration is sufficient to invalidate the use of constant LET. Also not considered is the possibility that the HgCdTe is “backsideilluminated” through a CdTe-type substrate with a thickness on the order of 200 µm. In this case, there could be significant energy loss for an incident particle even before reaching the electrically active epitaxial region of the detector. In all of the simulations performed here, the simplest assumptions have been chosen, i.e., constant LET and no significant superstrate. For the simulation of a space environment, the assumptions are not too unreasonable: slight variations in effective shielding due to neglect of energy loss in a substrate is not significant for typical proton spectra, especially given the overall approximation of “thin spherical shell.” For the more controlled laboratory environment with a well-known proton energy distribution, the effects of the substrate on HgCdTe device shielding have been seen by some researchers [124]. 5.1.3
Focal Plane Response Model and its Validation
The earliest version of the presently described focal-plane proton response simulation was developed to explain proton responses observed experimentally when bombarding an area Kodak charge-coupled device [121] and a Fairchild bilinear CCD (odd/even shift registers on opposite sides of the linear photosite array) [122] with a beam of highenergy protons at the Lawrence-Berkeley Laboratory (LBL) 88-inch cyclotron. The orientation of the device with respect to the collimated beam was found to have a profound impact on the structure of the proton-induced signatures seen in device “images”. Figures 39 and 40 illustrate the proton incidence geometry with respect to the surface of the Fairchild bilinear and Kodak area CCDs, respectively. There was no optical image focused on the device; all output was the result of the interaction of the proton beam and the CCD. The first attempt to model the structure of the proton signatures was based solely on the diffusion model of Kirkpatrick [125], the details of which are described below.
IV-63
Figure 39. Definition of the beam incidence geometry used for the Fairchild bilinear CCD proton irradiation
Figure 40. Definition of the beam incidence geometry used for the Kodak area CCD proton irradiation
IV-64
It was found that the Kirkpatrick model did not adequately explain the large central peak seen in the individual proton event signatures. The second iteration of the model included simple modeling of the depletion region underneath the pixel gates. The depletion model assumed that each depletion region under a pixel, transfer gate, or shift register could be modeled by a rectangular parallelepiped. Any charge generated within this region by an incident proton was assumed to be collected with 100% efficiency by the overlying pixel. When the depletion contribution to generated/collected charge was summed with the diffusion contribution, the results of the simulation were found to be in excellent agreement with those of the experiment for the area Kodak CCD (see Figure 41). As additional proof of the accuracy of the depletion/diffusion simulation, the responses of the relatively more complicated photodiode and left/right shift register structure of a bilinear Fairchild CCD were compared with experimental proton irradiation data recorded for this device and for both epitaxial and bulk versions of this device. As seen in Figure 42, excellent agreement is observed between the experimental results and the simulation. The fact that the magnitude of the proton-induced responses (as measured in electrons/pixel) were matched closely for the selected monoenergetic proton beam energies of 18 and 50 MeV validates the constant LET approach and the assumed values of the LET for each proton energy case.
Figure 41. Single 17-MeV proton response for area Kodak CCD with beam a angle of incidence = 70 degrees and azimuth = 0.0 degrees. Comparison of simulated and measured responses.
IV-65
Figure 42. Single 18 MeV proton response for a bilinear Fairchild CCD with (a) beam angle of incidence = 70 degrees and azimuth = 60 degrees for the bulk device and (b) beam angle of incidence = 30 degrees and azimuth = 90 degrees for the epitaxial device. Comparison of simulated and measured responses.
For the epitaxial device (top diagrams in Figure 42) the incident proton entered the CCD hitting near the right shift register traveling perpendicular and away from the center photodiodes and left shift register, with a downward trajectory. Only the odd pixels corresponding to the right shift register respond. In addition, this epitaxial device shows little adjacent pixel response since the diffusion volume of the device is greatly reduced. For the bulk device, the 18-MeV proton enters the right shift register with an angle of incidence of
IV-66
70 degrees and an azimuth of 60 degrees moving in the direction toward the left shift register. The contribution of the charge generated in the large bulk CCD diffusion volume is obvious. The proton enters the right shift register with a substantial downward trajectory, and hence passes under the left shift register at a considerable depth below it in the diffusion volume. The smooth and smaller left shift register response (odd pixels) is due to protoninduced ionization charge collected from the diffusion volume whereas the larger right response (even pixels) is due to both depletion and diffusion components. The peaks of the right and left shift register responses are shifted in the composite video due to the 60 degrees azimuth of the traversing proton trajectory. Although the agreement between model and experiment was excellent, it became obvious that the model still lacked utility for several real-world applications. Because the proton signature was so strongly dependent on the relative orientation of the particle trajectory and the focal plane, it was obvious that the code had to be modified to accept a stream of proton events, each with its own energy and trajectory, as input and produce a series of frames of images as outputs, such that the downstream effects on image processing could be evaluated. Each frame represents the nominal time exposure that the focal plane might experience during one normal image collect. In this way, an actual data collection at a chosen orbital location could be simulated. It also became obvious that the model should be enhanced to allow the simulation of linear scanning, time-delay-and-integration scanning, and staring arrays in addition to the original bilinear configuration. In order to incorporate the mesa structure characteristic of some photodiode arrays, it became necessary to allow the simulation of more than one layer of depletion regions. In this implementation, the mesa structure is approximated by a sandwich of two or more depletion layers and one diffusion region. The diffusion layer is at the bottom of the sandwich; above this lies a thin depletion parallelepiped with lateral dimensions matching the pixel pitch; above this is a depletion region of moderate lateral dimensions appropriate for the fill factor of the mesa. 5.1.3.1
Focal-plane Proton Ionization Ray-Tracing Details
The sensor is modeled as a stack of depletion layers and (typically) one diffusion layer. Each layer consists of one or more rectangular masks that are smaller than or equal to the cell (pixel) size. Particle trajectory segments lying within the mask region of the cell are assumed to result in charge generation and collection; and segments lying within the cell but outside the mask are assumed to make no contribution. Trajectory segment boundaries are set by cell and mask boundaries. Charge generation/collection is confirmed/denied by determining whether or not the midpoint of a segment is within a mask region; if so, an evaluation is made to determine which pixel is collecting the charge. A simple example [121] is shown in Figure 43 where a proton track is incident on a pair of adjacent pixels. This pixel structure has a depletion layer and a diffusion region below all pixels. The proton, following a straight-line trajectory, enters pixel 1 at position PO = (xo, yo, 0), where the origin is defined by the conventional spherical coordinates shown in Figure 43. The proton passes into pixel 2 at position P 1 = (x1, y 1 , z1), where clearly y1 = the pixel pitch. Finally, the proton passes out of the bottom of the depletion region of pixel P2 = (x 2, y2, z2), where z2 = the depletion layer total thickness. From IV-67
position P 0 to P2, the proton is within the depletion region of the CCD, first under pixel 1, then under pixel 2. The electric fields present in the depletion volume under a given pixel collect any charged-particle-induced minority carriers (electrons) produced in that pixel volume. If the particle trajectory through one or more adjacent pixel volumes is known, then the collected charge can be determined by multiplying the particle trajectory length in each pixel by the LET. For example, in Figure 43 the total depletion charge collected by pixel 1 would be the distance from P0 to P 1 times the LET; for pixel 2, it would be the distance from P 1 to P2 times the LET. The number of pixel depletion volumes crossed depends on the particle kinematics; for large angle-of-incidence, more pixels generally would be crossed than for small angle-of-incidence. In the current configuration of the simulation, it is assumed that the proton is not significantly scattered (in direction) by interaction with the crystal lattice of the semiconductor focal plane. After the event trajectory is propagated through all depletion regions, the diffusion component is evaluated. The model for depletion layer charge collection amounts to nothing more than computing line segment lengths through adjacent parallelepipeds (pixel depletion volumes) given an entry point and trajectory definition. Note that this computation is similar to that used by Pickel and Blandford [126] in evaluating heavy-ion-induced single-event upsets in MOS memories. To compute the charge collected from the diffusion layer, shown as the dimension Zdiff in Figure 43, the Kirkpatrick model is applied. In the diffusion volume, there are no electric fields; hence, ionization-induced charge diffuses to adjacent pixels. The Kirkpatrick model starts by solving the three-dimensional diffusion equation in a field-free, semiinfinite medium assuming a point source due to electron-hole-pair (EHP) generation (units of EHP/unit volume) at some depth into the diffusion volume. The effects of recombination are neglected at this stage of the model but later are included using an approximation. The method of images then is used to establish boundary conditions for the point source solution. The resulting current density at the depletion-diffusion boundary interface (the plane given by zdepl in Fig. 43) is computed by taking the negative of the diffusion coefficient times the normal gradient of the point source density evaluated at the surface. That the normal gradient is used assumes that the interface is a perfect absorber. The current density is integrated over the device integration period, which is considered infinite for mathematical simplicity. (The effective integration times of most imaging sensors, on the order of milliseconds, are much larger than the diffusion times; hence this approximation is justified.) This gives the charge per unit area at the depletion-diffusion layer interface obtained from a point source inside the diffusion region: Q ps ( x, y ) =
n 0 z′ 3 2⎤ 2
(5-5)
2π ⎡ ( x − x ′ ) + ( y − y ′ ) + ( z ′ ) ⎣⎢ ⎦⎥ 2
2
where n0 is the EHP density per unit length, ps refers to “point source,” x and y are coordinates at depletion diffusion boundary, and x', y', and z' are the coordinates of the point source generator in the diffusion volume.
IV-68
Figure 43. Proton trajectory through depletion and diffusion volumes of two adjacent pixels.
This method then is generalized to a line of EHPs representing the ionization track of a charged particle corresponding to the trajectory from point P2 to P3 in Figure 43. The next step in the model consists of integrating down this line (from P2 to P3) where the chosen length determines the effective diffusion layer thickness of the device (denoted z d;ff), herein approximating the effects of recombination: L
Qls ( x, y ) = ∫ Qps ( x, y ) dl
(5-6)
0
where ls refers to line source, θ is elevation angle, φ is azimuthal angle, x' is lsinθcosφ, y' is lsinθsinφ, z' is lcosθ, and L is the proton trajectory length in the diffusion volume. This surface charge density can be rewritten as
IV-69
⎡ qn cos θ ⎢ ( bε + 2a ) ⎢ Qls ( x, y ) = 0 − 1 πk ⎢ 2 2 ⎢⎣ a + bε + ε
(
)
⎤ ⎥ ⎥ 1 2 2⎥ a + bL + L ⎥⎦
( bL + 2a )
(
)
(5-7)
where
(
)
b = −2 ( x cos φ + y sinφ ) sin θ
a = x 2 + y2 ;
(
)
k = 4 x 2 + y 2 − 4 ( x cos φ + y sin φ ) sin 2 θ 2
As x and y approach zero, the factor k in the denominator of Eq. (5-12) vanishes and causes a divergence in the collected charge. The diffusion model described in this section has been upgraded to avoid this singularity entirely. This is described below in Section 5.3. It should be noted that this diffusion-effects modeling improvement results in computationally intensive proton-response imaging simulations. The charge collected under each pixel is obtained by integrating over the pixel areas at the diffusion-depletion layer boundary Q n,m =
∫∫ Qls ( x, y ) dxdy
(5-8)
x n ,y m
where Qn,m is the charge collected by pixel n,m of an N by M array. The entire model was assembled by first computing the depletion charge generated from P0 to P2 (see Fig. 44 for details of the coding logic), then using point P 2 as the entry point for the diffusion charge calculation, and finally adding together the two resulting (depletion and diffusion) pixel maps. Since the numerical integration of the diffusion-induced surface charge density requires that each pixel be divided into a few hundred sub-pixels, the procedure is computationally intensive. It is for this reason that the diffusion component is evaluated only in the vicinity of the trajectory intersection with the sensor. Currently, the extent of the diffusion integration is determined by initially evaluating one high-LET event over a very large area. Visual inspection of the array containing electrons collected by each pixel allows a good estimate of the necessary spatial extent of the diffusion integration. The required extent is chosen by determining the distance from proton impact at which the collected charge is comparable to or less than the assumed noise level (in electrons) of the focal plane. This physical extent (in pixels) then is used to restrict the diffusion calculation for the full simulation.
IV-70
Figure 44. Flow chart of simplified software logic for depletion region charge collection calculation.
5.1.3.2 Sensor Architecture Modeling
The full simulation can be used for a variety of imaging sensors, both scanning and staring. In the case of monolinear and bilinear scanning CCDs, the response of the shift register sites (randomized in time to reflect accurately the nature of a CCD's line transfer to serial transfer timing) can be included in addition to the response of the photosites. This was found to be important since the line readout time typically is comparable to the photosite image integration time and the shift registers are essentially identical to the photosites with an additional overlayer to eliminate optical absorption. For staring CCDs, if the readout time is short compared to the integration time, then the readout effects are ignored. If the readout time is significant, then two appropriately time-integrated arrays of data can be overlayed to account for proton interactions during readout. For photodiode arrays, the sensitive area of the readout circuitry for each cell usually is small compared to the area of IV-71
the overlying photodiode, and the readout circuitry is likely to be affected only if the proton interaction occurs during the readout of that particular pixel; so the readout has been ignored for the purposes of this simulation. 5.2
Simulation Results
For a given sequence of proton events (the output data from the environment simulation of Section 5.1.1) and device parameters, such as architecture, integration time, et cetera, a series of files is output. Each sequentially numbered file is a raw binary image of one “frame” of sensor data. Each 4-byte integer contains the value (in electrons) of the charge collected by that pixel. After an appropriate rescaling to 8-bit grayscale or remapping to pseudocolor, the images may be viewed immediately in an image-processing application such as Adobe Photoshop. Figures 45 and 46 are examples of proton signatures as they might appear on a 15- µ m pixel pitch silicon 256 × 256 pixel staring array shielded by 100 mils of aluminum while in a mid-altitude, moderately eccentric orbit and exposed for 10 msec. In one case (Fig. 45), the entire 15-µ m epitaxial layer is assumed to be entirely “depletion-like” in nature; some photodiode arrays and deep depletion CCDs might be expected to display similar signatures. Note that the tracks are sometimes pointlike and sometimes linear in nature. In Figure 46, the results of the identical proton environment incident on a similar silicon device are displayed. This device has a 4µ m depletion thickness of a 15-µm epitaxial layer and the same 15-µ m pixel pitch. The existence of an 11-µ m-thick diffusion region underneath the depletion region allows the collection of charge in neighboring pixels (crosstalk) as is evident in the figures. Because of the large variation in the magnitude of the response signatures, these results were mapped logarithmically onto an 8-bit grayscale image format. Figures 47 and 48 show the results of a simulation of one 50-µm pixel pitch HgCdTe (15 µ m epitaxial layer) in two different formats. This 128 × 128 pixel infrared image assumes the same midaltitude, moderately eccentric orbit, 100 mils of aluminum shielding, and 10 msec integration time. Figure 47 shows a linear 8-bit format, while Figure 48 shows the same results on a logarithmic scale so the diffusion-induced cross-talk can be more easily seen. Since all device response simulations were generated using the identical proton environment, a one-for-one correspondence can be seen in parts of the images. Generally, the proton response signatures have larger lateral extents when the pixel pitch is small compared to the total effective thickness (nominally the thickness of the epitaxial layer). Infrared devices, with their larger pixels, typically show smaller signatures. It is expected that the smallest CCD pixels (5 to 8 µm) will show relatively large spatial extent in their responses to incident protons.
IV-72
Figure 45. Logarithmically scaled 8-bit grayscale image of 256 by 256 pixel, 15 micron pixel pitch silicon imager with 15 micron depletion region and no diffusion region (mid-latitude environment with 100 mils of aluminum shielding integrated for 10 msec).
Simple post-processing, limited to generating histograms of the generated charge, is shown in Figures 49 and 50. In Figure 49, the results of three similar silicon simulations are shown. Each case is the result of binning 256 successive 25-msec image collection simulations of 256 × 256 pixels. In each of the three cases, the identical proton environment was used. Note the sensitivity in the pulse height distribution to the relative thicknesses of the depletion-like and diffusion-like regions. Due to the limited number of events available for statistics, the histogram is cut off at 50,000 electrons; however, the last bin contains the fraction of pixels for all events exceeding 50,000 electrons. Since the magnitude of the proton-induced response can cover the entire dynamic range, simple thresholding algorithms for proton event rejection cannot be implemented successfully in some cases. In such cases, foreknowledge of the spatial structure of typical proton events could be used to construct a more sophisticated filtering algorithm. Figure 50 shows the effect of varying shielding thickness on an otherwise identical HgCdTe simulation. Again, it is obvious that the proton-induced response can cover the entire dynamic range of the device.
IV-73
Figure 46. Logarithmically scaled 8-bit grayscale image of 256 by 256 pixel, 15 micron pixel pitch silicon imager with 4 micron depletion region and 11 micron thick diffusion region (midlatitude environment with 100 mils of Al shielding integrated for 10 msec).
Figure 47. Linearly scaled 8-bit grayscale image of 128 by 128 pixel, 50-micron pixel pitch HgCdTe imager with 4-micron depletion region and 11-micron diffusion region (mid-latitude environment with 100 mils of aluminum shielding integrated for 10 msec).
IV-74
Figure 48. Logarithmically scaled 8-bit grayscale image of 128 by 128 pixel, 50 micron pixel pitch HgCdTe imager with 4 micron depletion region and 11 micron diffusion region (midlatitude environment with 100 mils of aluminum shielding integrated for 10 msec).
Although the simulation assumes constant LET for the purposes of charge collection calculations, the approximate total energy loss (integrated actual dE/dx) for the proton trajectories is calculated to determine what fraction may not be suitably modeled by the constant LET assumption. The results are split into two categories: (a) what fraction of protons actually lost >10% of their initial energy, thereby calling into question the validity of constant LET, and (b) what fraction actually lost all of their energy (i.e., came to a stop within the sensor). The results are listed in Table 3-5. These results demonstrate that the constant LET approximation is tolerable but far from perfect for simulating an actual environment. Table 6. Constant LET approximation.
Silicon Silicon Silicon HgCdTe HgCdTe HgCdTe
Shielding (mils Al) 100 200 500 100 200 500
Fraction Slowed (%) 1.2 2.3 3.3 2.0 4.0 5.3
IV-75
Fraction Stopped (%) 0.25 0.58 0.86 0.40 0.8 1.28
Two obvious post-processing uses of these simulated images are to: (a) overlay the data with optical imagery similar to what might be acquired on-orbit in order to investigate the significance of the proton-induced artifacts on image quality, and (b) evaluate false-alarm suppression algorithms in star trackers or missile detection payloads.
Figure 49. Collected charge per pixel for a 15 micron pixel pitch silicon focal plane in a mid-latitude environment with 100 mils Al shielding and a 25 msec integration time. Histogram bins have 500 electron increments; last bin contains all pixels with charge exceeding 50,000 electrons.
Figure 50. Collected charge per pixel for a 50 micron pixel HgCdTe focal plane in a mid-latitude environment. Histogram bins have 2000 electron increments; last bin contains all pixels with charge exceeding 400,000 electrons.
IV-76
5.3
Updates to Proton Effects Modeling and Recent Transient Simulations
The simulation described in these notes is a useful tool for quantifying the detailed effects of proton transients on visible and infrared focal planes used in advanced spaced-based electrooptical sensors. The tool is not static and model upgrades are continuously implemented. As previously mentioned, the simulation can now be applied to photodetectors other than the original CCD focal-plane pixels. A simulation tool upgrade, which has been implemented, is described below and planned improvements are listed. A previous version of the simulation tool suffered from a deficiency in the proton-inducedtransient charge generation and collection model. The model deficiency manifested itself in proton events that produced rare, anomalous, unphysical results that originated from a divergence in the model. The anomalous results were cast out based on comparisons with experimental and empirical data. The divergence originates from the charge collection model described previously (Eq. (5-7). A new physical charge generation and collection model that does not exhibit the unphysical results of the previous model has been implemented. The proton-induced ionization track is now modeled as a spatially extended solid-state plasma filament of electron-hole pairs. Several additional improvements are planned that are aimed at increasing the fidelity of the predicted results. These include: (a) accounting for the loss in proton energy in the overlying substrate of an infrared photodiode focal plane, since this substrate can modify significantly the energy spectrum of incident protons for glancing angle impacts; (b) accounting for Bragg peak effects in the charge collection model when the incident proton energy is appropriately low (here dE/dx varies rapidly with distance); (c) accounting for small-angle scattering effects, which are more likely to be significant for infrared focal planes; (d) improving the diffusion model so that it properly reflects lifetimes and diffusion lengths typically encountered in infrared arrays; and (e) investigating an analogous simulation based on energetic electrons and x-rays. In addition, we plan to compare the results of this simulation with on-orbit EO sensor data whenever such data are available. Several transient simulations have been performed recently for infrared detector arrays [127][130]. Those simulations have elements in common with the modeling work described here in Sections 5.1 and 5.2 and provide useful extensions. 5.4
Impact of Electrons, Shielding, and Bremsstrahlung
Electrons trapped in the earth's magnetic field can penetrate spacecraft shielding to induce transient charge in focal-plane arrays. Because the electron mass is much smaller than the proton mass, the transport of electrons through shielding and focal-plane detectors differs significantly from that of protons. Most geomagnetically trapped electrons can be stopped with 0.5 in. of aluminum shielding; however, electrons lose a large portion of their energy through the production of secondary x-radiation (bremsstrahlung), which is very penetrating and has transport characteristics [131] completely different from the primary radiation. Figure 51 shows examples of the secondary IV-77
x-rays on spectra produced from a given input electron spectrum. On a per-particle basis, the x-ray flux is lower than the input electron flux by two to three orders of magnitude. This ratio can be made even smaller by the use of compound shielding consisting of layers of low-Z and high-Z materials. Visible and infrared focal-plane detector arrays typically used in space applications have a relatively small cross section for detecting x-rays. Figure 52 shows the attenuation lengths of x-rays as a function of x-ray energy for various detector materials. An x-ray having an absorption length of 1 cm, as might be appropriate for an 80-keV electron traversing silicon, has only a 0.1% probability of being absorbed in a detector that is 10- µ m thick. Just as with protons, focal-plane detector elements are efficient detectors of primary electron radiation. Accurate estimation of the transient effects of electrons and electron-induced x-rays on focal plane signals requires that detailed radiation transport and interaction calculations be performed. (A semi-empirical model to estimate focal-plane response to a gamma flux is discussed in Section 4.3.1.)
Figure 51. Differential x-ray fluxes produced by electron transport through solid Al spheres.
IV-78
Figure 52. Absorption length of x-rays as a function of x-ray energy in typical detector materials.
IV-79
6.0
Radiation Effects on MOS Readout Integrated Circuits
6.1
Background Information
Multiplexer or readout integrated circuits (ROICs) for large visible and infrared arrays for staring (i.e., non-scanning) applications, shown in block diagram form in Figure 19, have experienced a significant increase in functional complexity over the last two decades. Silicon technology feature sizes used to implement mixed-signal analog circuits have been reduced, which has allowed increased circuit layout densities to be achieved. Figure 53 shows the complexity increase versus calendar year for infrared focal-plane arrays [132]. The complexity level of arrays for infrared imaging applications is expected to increase significantly during this decade (i.e., MW MCT in Figure 53). Figure 20 shows a cross– sectional schematic of the indium-bump-bonded construction used for hybrid focal-plane arrays, which consist of an infrared detector array mated to a silicon CMOS array containing the pixel readout electronics (i.e., ROIC).
Number of detectors per chip
107
Typical chip size 1024 x 1024
106
i PtS
IRCCD invented
105
b I nS M MW
104 S Pb
102 1960 PbS IRCCD PtSi PACE
CT
ce
IP QW
T MC LW
Si:X
103
Pa
640 x 480 256 x 256 128 x 128 64 x 64 480 x 4 32 x 32 128 x 2
1970
1980 Calendar year
Lead sulfide linear arrays Infrared charge-doupled device Platinum silicide Schottky Barrier Producable alternative to CdTe for epitaxy (HgCdTe on sapphire)
InSb Siix MW MCT LW MCT QWIP
1990
2000
Indium antimonide Extrinsic silicon Mid-wavelength HgCdTe Long-wavelength HgCdTe Quantum well infrared photodetector
0002-99
Figure 53. Infrared focal-plane array size trends
With the advent of micron feature-sized CMOS transistors, ROIC designers were able to reduce the area occupied by a single pixel circuit to less than 25 µm2 for visible imagers. For the visible imager case, one of the simplest pixel circuits containing only three transistors, namely the source follower per detector (SFD) [133] circuit shown in Figure 54, can be designed to be compatible with a detector pixel pitch as small as a few microns. This socalled “3T” SFD pixel circuit, used primarily for visible imager applications, electrically isolates the detector integration node from stray capacitances allowing for realization of IV-80
integration capacitances of 5-10 fF or less per pixel. A voltage ramp is generated on the integration node as the reverse-biased detector diode current is integrated on this capacitance, CI, and subsequently is sensed at the gate of Q2, the source-follower transistor input. At the “3T” SFD pixel output, the analog voltage sample is read-out through the series transistor switch Q3 to the column interconnect trace when selected by the x-y addressing scheme as shown in Figure 54. SFD pixel simulation results showing these integration node ramp waveforms are included in Section 6.2.
Figure 54. Schematic of source follower detector (SFD) pixel electronics
For submicron-feature-sized CMOS circuits, very small values of pixel output capacitance can be achieved along with high values of photoconversion gain GC(ω), defined as the ratio of the pixel output voltage to photocurrent [134]: GC(ω) = q GV(ω)/C where q is the electronic charge, GV(ω) is the electronic circuit gain, and C is the pixel integration node capacitance. For a pixel integration capacitance in the range of 5-10 fF, the ideal conversion gain can be as large as 16-32 µV/e-, although losses in the pixel readout source follower, due to a less than unity gain characteristic and nonlinearity over the full dynamic range, may reduce pixel conversion gain to 10-20 µV/e- or less. A single transistor current sink Q4, as shown in Figure 54, is used to terminate each column and to enhance column settling time by selection of optimum current as required by the IV-81
frame rate. During pixel readout, a single pixel source follower along a column supplies current to the column sink and the column voltage settles to a source voltage corresponding to the analog sample at the source follower input, which is proportional to the photocurrent integrated on capacitance CI. Typical active power dissipation for visible imager designs can be as low as 30-50 µW per column for a typical array size of 4096 x 4096 pixels operating at 30 frames per second, and can be further reduced by power-gating the column current sink transistor consistent with the imager global timing constraints. An example of 3T pixel layouts for visible image arrays is shown in Figure 55, with the various photomask layers shown. [162] Indicated in the figure are the three NMOS transistors corresponding to Q1, Q2, and Q3 in Figure 54. The pixel dimensions in this layout are 5 µm x 5 µm, with the largest single area occupied by the photodiode for this APS pixel layout. The ratio of photodiode area to total pixel area, known as pixel “fill factor”, is one element which determines the detector effective quantum efficiency for mid-band spectral response in the visible band, typically 0.4 µm to 0.9 µm. As shown in Figure 55(c), thirdlevel interconnection thin-film metal is deposited over the three transistors as a light shield to eliminate spurious photo-response from these devices. A relatively high fill factor is achieved due to the circuit configuration in which transistors Q1 and Q2 share a common drain and Q2 and Q3 share a common source-drain region. Furthermore, these pixel cell layouts generally merge the source implanted region of Q1 with the tailored photodiode cathode region in order to reduce the total implanted area to optimize dark current and integration node capacitance as well as conversion gain. In addition, the layouts shown in this figure can be geometrically arrayed within the same pitch as the pixel dimension due to pixel layout design optimization. Note that for commercial pixel designs, NMOS device technology is favored due in part to the higher surface mobility of electrons as compared to holes, which allows transistors Q2 and Q3 to be reduced to approximately half the area required for a PMOS implementation operated at the same frame rate. More importantly, for commercial CMOS imager applications, NMOS-based pixels contain a photodiode consisting of an n-well implanted into the p-type substrate with typical thickness of several millimeters, resulting in acceptable spectral response. For a PMOS-based pixel photodiode, the corresponding heavily doped ptype implant into an n-well results in a much thinner collection volume, and corresponding loss of spectral response in the red and near infrared. [162]
IV-82
(a)
(b)
(c) Figure 55. Scaled layout drawings of 3T pixel.
IV-83
6.2
Total Dose Hardening of Visible Imager ROIC Designs
The SFD circuit shown in Figure 54 can be implemented with all NMOS or all PMOS devices, or a combination of device polarities. However, an important design consideration is the production by total ionizing dose of trapped positive charge at or near the silicon-silicon dioxide interface in gate regions of the “intrinsic” transistor as well as in the field oxide or, for many deep-submicron processes, LOCOS (i.e., local oxidation of silicon) regions surrounding the intrinsic transistor. This positive oxide charge shifts the NMOS and PMOS transistor turn-on threshold voltages such that NMOS transistors shift from enhancement mode (i.e., normally “off” state) to depletion mode (i.e., normally “on” state). The effect for PMOS devices is to continue to remain operating in the enhancement mode due to an opposite polarity shift in threshold voltage. For small-feature-sized devices, for which the gate oxide thickness is below 10 nm, these threshold shifts are small and on the order of a few percent of the pixel voltage supply. Measured NMOS device flatband voltage shift, which is the primary component in threshold voltage shift for both device types, versus gate oxide layer thickness is shown in Figure 16. That figure clearly demonstrates one of the potential benefits for the selection of smaller feature-size technology for many electronic circuit applications operating in a total-dose environment. An important aspect to understanding total-dose effects at the transistor level involves the NMOS device physical construction within the region in which the polysilicon gate layer is terminated over LOCOS field-oxide regions (i.e., ~500 nm), which is much thicker than the gate oxide (i.e., 5 nm) over the conducting channel. Post-total-dose measurements on transistors of various length and width dimensions and differing layout configurations demonstrate that most of the total-dose-induced NMOS transistor drain current at zero gate bias can be attributed to a drain-to-source leakage at the field oxide-silicon interface where the gate polysilicon is terminated over field oxide, as shown in Figure 56. This is due in part to the fact that holes trapped in LOCOS field-oxide regions prevent the device from full turnoff at zero gate bias. This is the result of trapped holes attracting electrons (i.e., minority carriers in the p-type silicon) to the surface where they can form a conductive channel between transistor source and drain. This effect (“channeling”) occurs at the surface of the lightly doped p-type bulk silicon regions surrounding NMOS transistors in many commercial foundry processes. Note that for PMOS transistors, the trapped holes within the corresponding gate regions terminated over field oxide results in the accumulation of electrons (majority carriers in the n-type bulk silicon) thereby preventing channeling due to the formation of a surface inversion layer of holes to conduct current from source to drain. These trends for total-dose transistor parameter shifts for PMOS and NMOS devices operating at gate voltages less than the threshold voltage (i.e., subthreshold) are illustrated in Figure 57. [135] Measured increased leakage for conventional NMOS transistors is shown in Figure 17 and Figure 58(a) for conventional device layouts. [135] Significant decreases in post-total-dose leakage for enclosed NMOS transistors are shown in Figure 18 and Figure 58(b). [135] The so-called enclosed device layout, as shown in the inset drawing of Figure 58(b), eliminates the polysilicon gate termination over field oxide and the corresponding drain-to-source leakage path due to surface inversion caused by holes trapped in the field oxide, as described previously. [136]
IV-84
Figure 56. Idealized Drawing of Oxide-Isolated NMOS Transistor Showing Nominal and Edge Leakage Current Paths
Figure 57. Effects of Total-dose Generated Oxide Traps (ot) and Interface Traps (it) on Transistor Drain Current vs Gate Voltage
IV-85
(a)
(b) Figure 58. Drain Current vs Gate Voltage For Total Integrated Dose from 50 krad(Si) to 400 krad(Si) for 0.18-µm feature-sized NMOS Transistors with a) Conventional Layout, and b) Enclosed Layout
IV-86
As applied to the SFD circuit shown in Figure 54, total-dose-induced leakage current for the reset transistor, Q1, can add significantly to photodiode dark current at the integration node, resulting in an unacceptable bias voltage shift as well as loss of dynamic range, which is detrimental particularly for non-astronomy imager staring-array applications that typically operate at integration times of tens of milliseconds. As an example, for a typical imager operating at a frame rate of 30 frames per second (fps) with interface to a 12-bit analog-todigital converter with full scale of 1.024 V, the total leakage at a 10 fF integration node is presumed to be dominated in a total-dose environment by the reset transistor operating in very weak inversion (i.e., subthreshold). A concern for visible imagers is that this leakage current, which would otherwise benefit by cryogenic cooling for infrared applications, cannot exceed 42 aA (i.e., 42 x 10-18 amp) in order to avoid adding an offset to the pixel sample voltage of more than ½ LSB for operation at 30 frames per second. The selective use of enclosed NMOS transistors can be an alternative solution for total-dose hardening to avoid the effects of low-level leakage paths provided that photodiode area and the corresponding fill factor can be traded off with the added layout area occupied by the enclosed device. Efforts to achieve hardness-by-design using enclosed-transistor layouts for imaging arrays have been reported for moderately large feature-size devices. [137] However, in general the increased transistor layout area and corresponding increased capacitances appear to be significant disadvantages to their use for small pixel designs. Figure 58(a) shows pre- and post-total-dose drain current versus gate voltage data measured for conventional NMOS devices fabricated by a 0.18 µm commercial wafer fabrication line. [135] The values of drain current at zero gate voltage measured after 50 krad(Si) and 100 krad(Si) are approximately 40 pA and 100 nA, respectively. Typical values of post-total-dose drain current versus gate voltage for commercial NMOS transistors are incorporated into a large-signal circuit simulation of the SFD operating at an active 30 mS integration period. For these simulation results (Figures 59 and 60), the post-total-dose pixel integration node waveform is significantly degraded as compared to the pre-irradiation case due primarily due to the increase in subthreshold current of Q1 (Figure 54) resulting in a pixel waveform offset voltage. In these two figures, the dark current was set to zero, and values of photocurrent were increased from 200 fA to 1 pA in steps of 200 fA. The total-dose-induced transistor leakage model in the Figures 60 and 61 simulations consisted of a nonlinear leakage current source connected in parallel with Q1, with nominal leakage current at low drain-source voltage of approximately 10 fA and increasing (with decreasing integration node voltage) to beyond several hundreds of femtoamps. Figure 61 contains estimates of the effects of total-dose-induced reset NMOS leakage in degrading pixel conversion gain. In this case, the conversion gain is derived from the simulation results over the first 25 mS of a 30-mS integration period. The estimated conversion gain, specified in the conventional units of µV/e-, is shown to approach zero at a “half-full” well integration node signal of 50K photoelectrons at approximately 50 krad(Si). This mode of operation would seriously limit the imager application for ionizing dose levels less than 50 krad(Si) due to the fact that the conversion gain will continue to degrade with total dose.
IV-87
Figure 59. Pre-Total-Dose Pixel Integration Node Waveforms for Various Photodiode Currents for SFD Pixel Implemented with Commercial NMOS
Figure 60. Post-Total-Dose Pixel Integration Node Waveforms for Various Photodiode Currents for SFD
IV-88
Figure 61. Input-to-Output Conversion Gain versus Photodiode Current for SFD Pixel Implemented with Commercial NMOS (T = 220 K)
Another approach that may be used to overcome the problems of NMOS threshold voltage shift with TID utilizes a reverse bias applied to the NMOS transistor p-well region in order to increase the bulk silicon charge component of the device operating turn-on threshold voltage, frequently referred to as threshold voltage body effect (i.e., (1/COX) x (qεSINAVSB)1/2), which occurs in the case of NMOS devices when the source-to-bulk silicon voltage (VSB) becomes positive. [138] However, this approach has the disadvantage of limiting the analog voltage dynamic range for a transistor used as a linear series switch, which is characteristic of many sample-and-hold and analog multiplexer applications within imaging arrays. Because many available CMOS foundries utilize an n-well process with a common p-type epitaxial layer to contain all NMOS transistors, the effect of a substrate reverse bias would be to increase the threshold voltage via the body effect for each NMOS transistor irrespective of its application, which may be unacceptable for the design of peripheral circuits integrated within the readout chip (ROIC). [138] These ROIC imager support circuits include linear amplifiers and digital logic circuits used for on-chip multiplexing, array self-scanning, ADCs, and output buffer circuits. A design modification to selectively replace low-leakage analog switch functions with PMOS transistors can also be incorporated into SFD circuit designs. For certain applications, substituting PMOS transistors in place of NMOS pixel reset transistors avoids the detrimental effects of increased leakage with total dose. This benefit also applies when a PMOS IV-89
transistor is used in place of a source follower or row select switch transistor (Q2 and Q3 in Figure 54), and overcomes another problem associated with total-dose-induced leakage from the large number of deselected row transistors within the image array, as discussed subsequently. It should be pointed out that one potential disadvantage of using PMOS transistors for these applications, in which the transistor source terminal is operated near the mid-point of the pixel output voltage range, is that the pixel linear dynamic range can be reduced more severely than for an NMOS transistor. This is due to the fact that the effect of total-dose threshold shift for a PMOS transistor is additive to the transistor body effect, whereas for NMOS transistor the body effect is reduced by the total-dose-induced threshold voltage shift. In addition, for n-well CMOS processes, the PMOS body effect can be larger due to the higher surface concentration of the n-well as compared to the p-type substrate surface concentration used for NMOS devices. Another potential benefit of a PMOS device used as the integration-node reset transistor Q1 (Figure 54) is that it allows the full reset voltage derived from a precision voltage reference, VRESET, to be transferred through the PMOS drain to the integration node. Note that if an NMOS transistor were substituted in the same circuit with the same set of nominal voltages used (i.e., typically 3.3 volts), the integration-node reset voltage transferred from the NMOS source would at best approach a value of VRESET – VTO, where VTO is the effective turn-on threshold voltage of the NMOS reset transistor, including the body effect term that adds to the pixel output offset voltage distribution. This distribution will generally be the result of imperfect matching of NMOS reset transistor parameters associated with random spatial distribution of impurities implanted into the bulk silicon as well as gate oxide charges. It is expected that total-dose-induced threshold voltage shifts will further complicate the process of pixel offset voltage subtraction during image post-processing. The distribution of this integration node offset voltage, described above, is one example of several terms referred to as fixed pattern noise or FPN. Active FPN cancellation can be achieved with the use of an analog difference circuit known as a correlated double sampler (CDS) [139], which is used while reading the analog pixel samples. The CDS first stores an analog sample of the current pixel during readout immediately after integration node reset, and subtracts this analog level from the integrated signal. For this implementation of FPN subtraction, CDS circuitry is included within the readout circuitry for each column within the array. Figure 62 shows the reported performance of FPN reduction circuitry achieving removal of FPN to less than 0.04% of the full pixel signal with no increase in the residual for operation up to 10.2 Mrad (Si). [140] Figure 63 contains a comparison of a non-hardened array with the measured increase in total-dose-induced dark current for a non-hardened and two total-dose hardened 512 x 512 arrays up to 10.2 Mrad (Si), showing negligible dark current increase up to approximately 200 krad(Si). A dose rate of 2.7 rad(Si)/sec was used to irradiate the samples using a Cobalt-60 source. [140] The pixel array can be degraded by total-dose-induced column offset voltage for the 3T pixel cell implemented with commercial NMOS technology. Figure 64 shows a simplified model of the origin of the leakage path at the column of an array, wherein the effect of the totaldose-induced leakage components of the deselected pixels within a given column contribute to a spurious current that serves to shift the column voltage positive during pixel read. To
IV-90
estimate the magnitude of the leakages, assume that for a 4096 x 4096 array the columns are subdivided in length to contain 2048 pixels per active subcolumn. During an individual pixel read operation, the total-dose-induced leakage current from the remaining deselected 2047 pixel read switches can contribute to a combined current that approaches or exceeds the nominal value of the drain current of Q4 (Figure 54), which has been specified for this “point” design as 10 µA. Compare that design value of current with the sum of the currents for the deselected transistors, which is shown in Figure 58(a) to be 40 pA per transistor at zero gate voltage. [135] Note that this condition is consistent with the worst-case column voltage for a full-well pixel voltage.
Figure 62. Residual Fixed Pattern Noise versus Total Ionizing Dose
For this example, the total column leakage current is estimated to be 82 nA at 50 krad(Si) and in excess of 200 µA at 100 krad(Si). As the leakage current increases between these two limits, typically a current approaching the nominal current sink design value of 10 µA, the pixel sample voltage will be significantly corrupted by this leakage induced offset voltage. Figure 65 contains a simulation result showing that the pixel sample voltage contains approximately 100 mV offset (i.e., 400 counts for 12 bit ADC resolution) at the total-dose level corresponding to column leakage values between 8 µA and 9 µA, well below a totaldose upper limit of 100 krad(Si). The simulation results show that as the column leakage increases beyond a value corresponding to the current sink design point of 10 µA, the column voltage increases the drain-to-source voltage of Q4 to accommodate the added current. Design mitigation to overcome this total-dose effect includes the use of enclosed NMOS transistors within the 3T pixel layout for implementing the source-follower and row-select functions (Q3 and Q4 in Figure 54). Differential voltage sensing at the column line may also
IV-91
be applicable for image arrays, which include a light-shielded (i.e., unilluminated) reference column.
Figure 63. SFD Pixel Dark Current Increase with Total Ionizing Dose
Figure 64. Simplified Diagram Showing Imager Selected and Deselected Pixel Currents
IV-92
Figure 65. Predicted Pixel Sample Offset Voltage versus Deselected Pixel Current in Active Column
6.3
Approaches to ROIC Design for Infrared Imagers
For infrared imagers, the pixel area is generally relaxed to 20 µm x 20 µm or larger based on tailoring to the infrared optical blur circle. The larger pixel pitch (compared to a visible pixel pitch of less than 10 µm) allows increased functional complexity within the pixel electronics and potentially the selective use of enclosed NMOS transistor layouts to avoid increased total-dose-induced leakage. A frequently used infrared-imager pixel circuit, commonly referred to as a capacitance transimpedance amplifier (CTIA) [134], is shown in Figure 66 and consists of a reset integrator implemented with an active linear CMOS transimpedance amplifier and smallvalued feedback capacitor used for photocurrent integration. A sample-and-hold circuit at the transimpedance amplifier output is used to provide a snapshot of the pixel analog signal at the end of the integration period such that one entire image frame can be transferred to the output of the imager during the integration time of the next frame. As shown in Figure 66, the combination of an NMOS source follower transistor and an NMOS series analog switch transistor at the pixel output interface is typically used to transfer the sampled pixel voltage to the output column bus line when the row address in enabled. A PMOS source-follower implementation can also be used to achieve total-dose hardening without the use of enclosed transistor layouts to avoid the effects of increased leakage from deselected row transistors within the image array, as discussed in Section 6.2.
IV-93
Figure 66. Schematic of CTIA Pixel Electronics
For the CTIA, pixel conversion gain at the integration node, GC(ω), is given by qGV(ω)/C, as noted in Section 6.1, where C is typically dominated by the integrator feedback capacitor, C1. in Figure 66. Practical values of pixel conversion gain can exceed 10-20 µV/e- because the active CMOS feedback amplifier input behaves as a virtual ground for photodetector current, thereby greatly reducing the effects of stray input capacitance in parallel with the detector, a feature that is lacking in the SFD pixel configuration. This feature is an advantage for hybridized infrared arrays, which are illustrated in Figure 20, since the added stray capacitance of the indium bump interconnection is largely eliminated. The active CMOS amplifier also provides improved output dynamic range for the CTIA pixel, as compared to the SFD pixel circuit. However, while the SFD circuit dissipates power only when accessed during the pixel read cycle, the CTIA active amplifier dissipates continuous power. However this can be as low as 75 nW, as demonstrated for a low-power design. Testing of a 16 x 16 readout array designed with enclosed NMOS transistors showed minimal performance shifts for total-dose exposure up to 1 Mrad(Si). [141]
6.4 Infrared ROIC Circuit Design for Total-Dose Hardness For the CTIA pixel electronics, the formation of leakage paths, which degrade the analog signal voltage stored on small-valued capacitor, remains the most significant problem for total-dose hardening. As discussed previously, with the use of submicron feature-sized CMOS for implementation of imagers, a typical value of CTIA feedback capacitor C1 in IV-94
Figure 66 is 10 fF or less. As discussed previously for staring arrays operated at rates of 10 fps to 30 fps, total-dose-induced leakage currents on the order of 10 aA to 40 aA can result in significant degradation of the integrated pixel signal. In Figure 66, the CTIA integration capacitor reset switch Q3 represents a potential leakage current path, which would result in a loss of integrated signal. However, in most CTIA applications, this reset transistor can be implemented using a small-geometry PMOS device. When the gate of this device is pulsed to a voltage sufficiently low to turn-on the PMOS channel, the integration capacitor can be discharged. Thereafter, its gate voltage can be returned to a positive voltage level, forcing the device into cutoff. For this device, total-dose exposure will result in a transistor threshold shift toward enhancement mode, thereby eliminating increases in subthreshold current as discussed in Section 6.2. The CTIA pixel sample-and-hold electronics presented in Figure 66 shows a conceptual design including a two transistor NMOS (Q4) and PMOS (Q5) transmission gate analog switch used to implement the sample-and-hold circuit series switch function and an NMOS shunt switch (Q6) used to reset the hold capacitor. For the sampler switch, an NMOS transistor (Q4) is used to extend the sampled CTIA output voltage dynamic range. This is because the use of a PMOS switch alone would limit the sampled voltage to a value greater than about twice the absolute value of its threshold voltage. However, many practical CTIA designs eliminate the NMOS transistor (Q4) for this analog switch, since for low-level sensing applications the voltage level transferred is sufficiently positive such that the PMOS transistor (Q5) does not limit or “clip” the signal. Similarly, NMOS transistor Q8 may be eliminated from the pixel series output switch, although the gate-source voltage of the source follower transistor Q7 will tend to reduce the dynamic range margin for signal levels approaching “full well”. Similarly, signal charge stored on the hold capacitor can be corrupted by total-dose-induced leakage of the hold capacitor reset transistor Q6. Post-total-dose leakage currents for either of these NMOS transistor applications should be limited to 10 aA to 40 aA for imager operation at 10 fps to 30 fps. For many infrared applications, the use of NMOS transistors utilizing design-hardening techniques, including enclosed or other tailored layout approaches, can potentially be used to achieve moderate levels of total-dose hardness for submicron-featuresize technology. These design-hardening layout techniques are used to mitigate the effects of field-oxide hole trapping and the significantly reduced thermal annealing rates at cryogenic temperatures, as is briefly described subsequently. Furthermore, for staring infrared array imagers operated at 100 K or below, the use of radiation-hardened foundry capabilities for ROIC development provides several advantages, including the use of nonenclosed NMOS transistor designs, which provide an arbitrary range of channel aspect ratios to select from during design optimization. For example, Figure 67 shows subthreshold drain current versus gate voltage for an enclosed NMOS transistor with 0.9-µm gate length and a total area approximately 4 times larger than a minimum channel length transistor. Pre- and post-total-dose measurements of subthreshold current for this transistor are shown in Figure 67. The measurement technique is as follows. The enclosed NMOS transistor was operated functionally as Q6 in Figure 66 as the hold
IV-95
capacitor reset switch. However, for this test structure, the gate of Q6 was available for external adjustment and was used to generate the data in Figure 67, which displays the subthreshold drain current versus the gate voltage for Q6. The drain current was derived from the discharge slope of the hold capacitor resulting from the subthreshold current of Q6. Hence, the data points in Figure 67 were generated from measured values of discharge slope over time periods up to 100 ms, with the discharge waveform captured by a digital processing oscilloscope using waveform averaging. The test samples measured were biased with a steady-state voltage of 2 volts applied to the gate of Q6 as a worst-case condition. [141] These results verify the performance of enclosed NMOS devices fabricated with a 0.5µm process in maintaining the pre-irradiation subthreshold current-voltage slope and in avoiding significant increases in subthreshold current for total-dose exposure up to 100 krad(Si) at ambient temperatures. This technique also provides a methodology for subthreshold current measurement in the fA range.
Figure 67. Subthreshold Current versus Gate Voltage for Enclosed NMOS Transistor
For operation at cryogenic temperatures, it might be assumed that a significant improvement in reducing total-dose-induced increases in subthreshold current would be associated with its exponential temperature dependence. [142] However, it is expected that this improvement only applies to enclosed NMOS transistors. For nonenclosed devices, the subthreshold current is dominated by the component due to electron channeling at the field-oxide interface under the polysilicon gate. However, another factor in mitigating the effect of total dose on subthreshold current is that, for a typical imager application, the hold-capacitor reset-transistor gate voltage would normally be pulsed at a duty cycle less than 0.1% during array operation. Consequently, for this mode of operation, which is typical of many imager capacitor reset functions, these IV-96
transistors would be expected to exhibit a reduced subthreshold current increase during operation than predicted by the measured data for the previously discussed enclosed NMOS device under continuous-voltage bias during exposure. [143] During the CTIA design optimization process, to reduce the effects of total-dose threshold shifts as well as those effects on transimpedance amplifier performance associated with operation at cryogenic temperatures, current-mirror biasing is frequently used. [144] This bias technique, commonly used to bias transistors for image array peripheral circuits such as low-power operational amplifiers used in the end-to-end signal chain, utilizes a reference voltage to produce a fixed current that is relatively independent of the device parameters, including the transistor threshold voltages. The reference voltage, which generates the current-mirror reference current, is distributed to each of the transimpedance amplifiers within the array, resulting in a common bias current for each amplifier. This technique has been successfully used to generate ROIC array bias currents as low as 25 nA per amplifier for a 0.5-µm-feature-sized 16 x16 pixel test array using enclosed NMOS devices operated at ambient temperature to 1 Mrad(Si). [141] Fortunately, current-mirror bias schemes are almost always used for pixel amplifiers in order to provide temperature compensation at cryogenic temperatures. In practice, for these circuits biased at subthreshold current levels, the drain-to-source current is dominated by diffusion current and is exponentially related to the ratio of applied gate voltage to the thermal voltage, kT/q. To first order, the primary total-dose effects are related to the interface trap effects on surface mobility. However, the design of pixel array peripheral circuits, including linear buffer amplifiers and comparators used to implement analog-to-digital converters, must consider the effects of transistor threshold voltage shifts with total-dose. For some ambient temperature designs in which a stable voltage reference is needed to establish a reference current independent of threshold voltage shifts, a substrate pnp transistor is used. Because the pnp emitter-base voltage trends with temperature and current can be modeled, stable current references can be designed. [144] With proper hardened-by-design layout applied to the pnp device design, this device can be made insensitive to total-dose-induced surface effects. As an example of various aspects of the use of enclosed NMOS transistor hardness-by-design techniques for a 16 x16 element CTIA test array, the pre-and post-total-dose results for measured dynamic range and linearity are presented. These results for measurements made at ambient temperatures are shown in Figures 68 and Figure 69, respectively. [141] The CTIA pixel amplifier current sink (Q2 in Figure 66) was operated at 25 nA during irradiation and subsequent testing. Photoconversion gain, dynamic range, and linearity remained nearly constant with exposure up to 1 Mrad(Si). Linearity was seen to degrade at 500 krad(Si), but this occurred beyond the designed maximum well-capacity design of 100 kiloelectrons. A Cobalt-60 source was used to irradiate the samples with power applied at a dose rate of approximately 72 rad(Si)/sec. Note that the readout array testing was accomplished with the use of 40 fF capacitors incorporated into the inputs of each of the 256 CTIA cells and used as photocurrent simulators when driven with a voltage ramp.[141] While these results for the test array were satisfactory, it should be noted that the enclosed transistors were each surrounded by a heavily doped p-type implanted guard ring to avoid
IV-97
any channeling between its source terminal and another n-type implanted region, particularly for operation at cryogenic temperatures. However, the added guard ring area increased the pixel layout area from an initial estimate of less than 32 µm x 32 µm to a final pixel area of greater than 50 µm x 50 µm. In the absence of a concern for total-dose-induced electron channeling in the field areas, clearly a concern at cryogenic temperatures, the guard rings would have been eliminated with a significant reduction in pixel area. With the use of a totaldose-hardened field-oxide wafer foundry process to reduce electron channeling in field regions and to reduce subthreshold current for switches, the pixel layout area could likewise have been reduced. Additional aspects of radiation hardening for ROIC operation at cryogenic temperatures are now considered. While the general trends for total-dose hardening of circuits operating at cryogenic temperatures (< 120 K) are consistent with those measured at ambient temperature, some oxide hole trapping mechanisms at cryogenic temperatures result in reduced pixel circuit operating margins. For example, the transport of holes within the silicon dioxide gate dielectric is greatly reduced at cryogenic temperatures, resulting in a relatively uniform hole distribution over the gate dielectric thickness. As a result, the use of an oxide growth process that is tailored to achieve hardness by avoiding the formation of traps at deep energy levels at or near the silicon dioxide-silicon interface appears to be less effective in reducing transistor threshold voltage shift for cryogenic use, partly because fewer holes are transported to this interface. Furthermore, the annealing rate of holes at room temperature due to hole transport and tunneling at the silicon dioxide-silicon interface is also believed to be greatly reduced at cryogenic temperatures, except for very thin gate oxides, typically less than 2 nm. Note that 2-nm gate-oxide technology is not currently utilized for infrared imager development to any large extent due to the breakdown voltage limits. For a typical imager application, the pixel output dynamic-range requirement of approximately 1 volt implies an electric field in the transistor gate oxide dielectric region in excess of 1 MV/cm for a supply voltage greater than 2 volts. Also note that the typical range of pixel supply voltage is 3-4 volts. For pixel circuits with amplifier transistors (Q1 in Figure 66) biased in the subthreshold regime, the total-dose-induced device threshold shift (i.e., translation along the x-axis for plots similar to Figure 57) is larger than for transistors biased into strong inversion. This is due to the competing mechanisms of electron-hole recombination at low electric fields within the device channel versus enhanced hole transport at high electric fields. Experimental results show that for practical device geometries, the maximum transistor channel electric field can be quite high, approaching 1 MV/cm, which is typical for subthreshold operation. [145] Therefore, the effects shown in Figure 57 are more significant for the typical amplifier transistors used for pixel electronics than for imaging-array peripheral circuitry with transistors biased into strong inversion.
IV-98
Figure 68. Mean Value of Output Voltage for 16 x 16 CTIA Test Readout Array
Figure 69. Mean Value of Nonlinearity for 16 x 16 CTIA Test Readout Array
Increased low-frequency transistor noise must be considered for pixel circuits operating at cryogenic temperatures. The voltage-noise power spectral density for NMOS transistors has IV-99
been often modeled as proportional to the number of carriers in the MOSFET channel for operation in strong inversion [146]: Sv = (K/fα) [VD2/(VG-VTO)2] Relatively recent experimental results support the hypothesis that low-frequency MOSFET noise is dominated by a thermally activated process rather than by tunneling. [147] For NMOS devices with a gate oxide thickness of 4.5 nm irradiated to 500 krad(Si) using a 10keV x-ray source, the low-frequency noise increased approximately by tenfold over an exposure range of 85 K to 285 K. Increases in both SV and α at a reference frequency of 1 Hz were observed. However, significant differences among devices measured from the same wafer were seen at low temperature, leading to the preliminary conclusion that roomtemperature noise screening may not be reliable for cryogenic applications. [147] For SFD pixel cells used for ambient and cryogenic temperature applications, the lowfrequency noise charge (i.e., noise electrons) can be estimated using the following [148]: 1/ 2
N amp
⎡ ∆f ⎤ 1 − cos 2πft ) ⎥ ( 2⎢ 2 ≈ Vn ( f ) df ⎥ SV ⎢⎢ ⎡1 + ( 2πfT )2 ⎤ ⎥ D ⎢⎣ ⎦⎥ ⎦ ⎣
∫
where SV is the SFD conversion gain (V/e-), Vn2(f) is the 1/f noise voltage PSD, ∆f is the pixel cell frequency bandwidth, and TD is the sampling time constant for a post-pixel CDS circuit used to reduce reset noise. Generally, total-dose-induced increases in low-frequency noise at cryogenic temperatures may be more important for the on-chip image array peripheral circuits, such as linear amplifiers and comparators, than for the transistors within the SFD or CTIA pixel cell electronics. This is due to the fact that with the small integration-node capacitors used to achieve photocurrent conversion gain, the mean square noise voltage component, kT/C, associated with the reset transistor (Q1 in Figure 54 and Q3 in Figure 66) typically dominates the noise level, including the amplifier transistor low frequency noise voltage integrated over the SFD or CTIA frequency bandwidth. For a CTIA pixel layout design with a 20-µm pitch or larger, amplifier transistor geometry can generally be designed to minimize 1/f noise. An equivalent expression for equivalent noise charge in noise electrons for a CTIA pixel cell for the reset-noise-limited case is given as [149]
( Cdet + Cfb ) nkTCfb = 2 CL ( Cfb + Cdet ) + Cfb Cdet q 2
N read
where n is a design parameter (typically 2), Cfb is the CTIA feedback capacitor, Cdet is the photodetector capacitance, and CL is the frequency band-limiting capacitor.
IV-100
Large imaging arrays contain significant functional complexity associated with the peripheral circuitry (i.e., circuits outside the two-dimensional array containing only pixel cells). These peripheral circuits contain functions including the following: dc bias generation, image array scanning and timing pattern generation, and analog multiplexing. Analog-to-digital conversion (ADC) is another function which may be included on the array, and the design options vary from the use of a limited number of converters operated at a high sampling rate to an approach referred to as “massively parallel” in which large numbers of ultra-low-power ADC circuits are abutted to the pixel array columns and each ADC processes only a limited number of columns. Total-dose hardening of ADC circuits included within contemporary ROIC designs is beyond the scope of the present discussion. Bias generation circuits are based on the design of a precise and stable current reference, and many CMOS designs have appeared in the literature.[144] For both visible and cryogenic infrared image arrays, it is critical to establish the target current value at either an ambient or cryogenic operating temperature depending on the application. However, the combined effects of total-dose-induced threshold voltage shift and temperature coefficients of threshold voltages increase the difficulty of the current reference design. Most practical solutions to developing a precise and stable current reference resort to its implementation with discrete total-dose-hardened, temperature-compensated circuitry external to the ROIC. External totaldose-hardened digital-to-analog converters (DAC) may also be used to provide fine adjustment to the current reference within the ROIC by means of calibration data downloaded to the DAC from an external data port. Image array scanning circuitry consists of digital logic elements that can be total-dose hardened using enclosed NMOS transistors. Typical digital logic cells containing enclosed NMOS layouts have been described. [135] The 16 x 16 element CTIA readout array discussed previously made extensive use of enclosed NMOS transistors within the scanning circuitry. This array was tested to 1 Mrad(Si) with no functional failures and minimal degradation of CTIA performance. [141] Timing pattern generator design for a total-dose environment can utilize hardened digital logic elements implemented with enclosed NMOS transistors. Array scanning and timing pattern-generator applications can be satisfied with typical total-dose-hardened cell library elements. Analog multiplexing is used for two primary applications. The first is to format the serial analog pixel video stream and to input this stream to a set of analog output amplifiers for designs that do not include an ADC on the ROIC. The second is to format and store analog pixel samples for the input to an ADC within the ROIC. Both applications make extensive use of low-loss analog switches and polysilicon-gate capacitors for the formatting and storage functions. Low-power operational amplifiers, generally used as voltage followers, provide buffering to the ADC input capacitance. CMOS operational amplifier design is well covered in the technical literature. Typical input offset voltage shifts with total dose must be accommodated, but many techniques used to provide operational amplifier offset cancellation have also been discussed and are relevant to total-dose hardening. [144] Clearly the use of enclosed NMOS transistors used for low-level analog switches can be an advantage in avoiding spurious leakage, which would degrade analog pixel samples stored on polysilicon-gate capacitors. Analog output buffer circuitry is similar in function to the
IV-101
internal operational amplifier circuits. Transfer of the video stream from the imaging array to the image processing electronics can be simplified using a differential approach. In this approach, external differential signal sensing compares each pixel sample with a lightshielded reference channel, with the total-dose-induced common-mode voltages eliminated by the external receiver circuitry.
6.5
ROIC Hardening by Design for Single Event Effects
Mitigation of single event effects (SEE) [150] as related to ROIC designs fabricated using currently available submicron feature sized bulk-silicon foundry CMOS has been largely limited to reducing the effects of single-event upset (SEU) rates for various latch designs used in the image array peripheral circuitry. These peripheral circuits primarily include the digital logic latches used to implement on-chip timing pattern generators and digital logic state machines used in conjunction with on-chip digital image processing. Other functions frequently included within the ROIC are analog-to-digital converters (ADC) for pixel conversion and, in some cases, digital-to-analog converters (DAC) embedded within highperformance pipelined ADC converters. Generally, the upset rates for unhardened bulk CMOS latches are considered inadequate for many critical digital logic array functions. Hardening-by-design can be employed to reduce latch upset rate, and one of the most efficient SEU-hardened latch designs is based on a circuit topology that has a very low probability of logic state upset due to a single heavy-ion strike. A latch design of this type contains internally redundant circuitry within the data retention path to avoid the upset of the latch due to a single node state transition. [151] These latches can be used to implement critical digital counters (i.e., binary or grey-code counters) used primarily for on-chip ADCs, for array timing pattern generators, and in certain cases to SEU harden latches used to implement shift-registers used to implement image array scanning. Another approach to SEU hardening consists of a technique commonly referred to as triple modular redundancy (TMR) [152] used with digital logic majority voting circuitry, indicated as “MAJ”, as shown in Figure 70. Also illustrated in this figure is another technique, known as temporal data sampling, used to avoid capture of any data that may have been corrupted by a single-event transient (SET), which may occur during the clock period that occurs during the data (or counter bit) transfer to the voting logic. [153], [154] Hence, the critical data or binary counter state can be protected from an SEE transient event as well as steadystate latch upset. As applied to the SEU hardening of the master counter function used for a timing pattern generator, each of the output bits of three identical counters are compared and corrected on a “bit-slice” basis by the voting or “majority” logic circuitry designated as the block “MAJ” in Figure 70, which is operating synchronously with the input data clock. The corrected master count is provided as a set of input bits to the decoding logic used to generate the various image array timing pulses. For ROIC designs that incorporate a high level of image processing, partial or full pixel frame buffers may be implemented on the ROIC. Pixel data are more efficiently stored in cross-coupled static RAM (SRAM) cells as compared to general-purpose digital logic
IV-102
registers. Figure 19 contains a schematic of an SEU-hardened SRAM cell containing two cross-coupled latches, as compared to the single cross-coupled latch used for non-hardened SRAM cells [155]. The nominal operation of the SEU-hardened cell prevents data corruption due to latch upset from a single-particle strike at a critical latch node. The SEU hardness is achieved due to a unique aspect of this latch design, including redundancy in the form of a dual latch, which, if uncorrupted, will restore the initial state of the upset latch node. In addition, due to the known opposite polarity voltage transitions in response to a heavy ion strike of the n-implanted versus the p-implanted transistor regions, this knowledge can be used to implement a latch that retains a nearly incorruptible zero state and a corresponding latch retaining a nearly incorruptible one state. In Figure 71, the upper cross-coupled latch retains an incorruptible one state and the lower latch retains an incorruptible zero state. Since these two latches form the essential incorruptible set to achieve data retention, this cell design can be used to avoid particle-induced upsets due to high-energy protons and heavy ions. [155] While this SRAM cell requires approximately twice the number of transistors as the conventional non-hardened SRAM, its use can reduce the hardware complexity associated with the alternative approach consisting of hardware-based error detection and correction (EDAC), including the elimination of the memory to store the parity bits used for EDAC implementation.
Figure 70. SEU Mitigation Circuit Implemented with Temporal Sampled Latches and Synchronous Majority Voting
For imaging arrays used in low earth orbits, continuous imaging may not be required. Furthermore, increased SEU rates may be limited only to locations in the orbit that pass over IV-103
the region of the South Atlantic Anomaly (SAA) [150], which contains increased proton flux levels known to result in higher SEU rates for some CMOS technologies. In such cases, SEU mitigation can be achieved after transiting through the SAA by updating critical latch data within the ROIC, during each orbital pass, from external SEU-hardened SRAM. Thereafter, the probability of latch SEU may be limited to a lower rate associated with galactic-cosmicray-induced upsets, which are less frequent compared to protons within the SAA affected region of the orbit. Single-event latch-up (SEL) can occur in technologies containing a 4-layer pnpn path, as in the bulk CMOS cross-sectional drawing shown in Figure 72. During nominal CMOS operation, the pnpn junctions would each be either at zero bias or reverse bias. However, under certain extreme bias conditions, including those of a heavy ion strike, during which local current densities develop junction forward bias, the formation of one or more active conducting pnpn structures can occur.
Figure 71. Schematic of SEU Hardened SRAM Cell
IV-104
Figure 72. Cross Sectional Drawing of PNPN Structure Within Bulk CMOS
The cross-sectional drawing in Figure 72 contains a simplified transistor model of this pnpn structure, with the equivalent npn device emitter formed by an NMOS transistor source terminal which is permanently connected to the negative supply voltage. The equivalent pnp device emitter is formed by a PMOS transistor source connected to the positive supply voltage. The base terminal of the equivalent npn consists of the p-type contact to the pepitaxial or p-type bulk substrate and can become locally forward biased due to a current transient flowing through the epitaxial layer or substrate bulk spreading resistance. The base terminal of the equivalent pnp is nominally tied to the positive supply voltage as connected through the n-well spreading resistance, and can likewise become forward biased due to a voltage developed by a current transient flowing through the n-well spreading resistance. In general, for conventional bulk silicon CMOS, pnpn conduction can be initiated by electrical triggering, pulsed laser photocurrent, and by a heavy ion strike and in some cases by a proton strike. Figure 73 contains a typical pnpn I-V characteristic, in which the device can sustain the high current state in part due to thermally induced regeneration within the structure, given that the npn and pnp transistor current gain values increase with increasing temperature and that the junction voltages decrease with increasing temperature generated locally by the pnpn high current density. Figure 73 also shows the critical pnpn latching voltage, VL, the holding voltage, VH, and holding current, IH. For a given CMOS technology primarily utilized for digital logic applications, these values of VL, VH, and IH can be determined from device characterization data for a specific set of layout design rules. Consequently, if the ROIC supply voltage is less than VL, SEL will generally not occur in the absence of significant stray series inductance. Similarly, if the external power supply operating current is limited to a value below IH at which the voltage drops below VH, the SEL will not be sustained. The CMOS foundry pnpn characterization data may not be fully applicable to ROIC designs, especially those which exceed the nominal die size used for digital logic. For example, the layout design rules used for digital logic generally assume that substrate and implanted well regions contacts have been incorporated into an extensive gridwork within array of digital logic cell layouts, but for ROIC designs a similar gridwork may not be present due to pixel IV-105
area limitations. Furthermore, for ROIC packaging the elimination of the backside ohmic contact to the bulk silicon substrate is not an uncommon practice, which further increases the likelihood of pnpn triggering. Hence, for ROIC designs, due to the lack of an extensive gridwork of low-resistance metal interconnects nominally used to shunt the substrate and implanted well region spreading resistance, SEL margins developed from characterization of digital logic test structures should not necessarily be applied, at least for those imagers operated at non-cryogenic temperatures. For infrared and visible imagers operated below temperatures at which significant thermal carrier generation occurs, cryogenic temperature operation results in reduced bipolar device (i.e., npn or pnp) current gain, commonly referred to as carrier freeze-out, and consequently SEL cannot be sustained. This effect is a consequence of an additional criterion for pnpn regeneration that requires at the pnpn operating point, IH and VH, the sum of the npn and pnp common-base current gains must exceed unity. For n-well CMOS technology, typically the wide-basewidth npn device current gain is very low at cryogenic temperatures. However, readout wafers are routinely screened for initial functionality during testing at ambient, and pnpn latchup can damage die during this screening. During testing, damage to the readout die can be avoided by proper powersupply sequencing and current limiting of voltage supplies used under the assumption that localized latchup sites can be electrically triggered.
Figure 73. Typical Current vs Voltage Characteristic of CMOS Parasitic PNPN Structure
Detector response to proton and heavy ion strikes has been studied for visible APS as well as for infrared imagers. [156], [157]. For the APS imagers, it was found during image array testing that the number of pixels collecting charge induced by heavy ions is smallest for an angle normal to the surface of the imager die and greater for shallow grazing angles. In some cases, the heavy-ion-induced charge along a 1-mm track within the bulk silicon appears to IV-106
spread over a roughly circular area with a diameter of 250 µm. These results for heavy ion testing are consistent with testing of ROIC circuitry designed for infrared applications, and this reported effect may be related to the spurious charge collection efficiency of the critical junctions within CTIA circuitry. Apparently, similar results for charge spreading are also obtained when certain digital memory circuits are tested with heavy ions. Charge collection times extending to integration times between 5 mS to 25 mS are attributed to a slow collection mechanism with heavy-ion-induced carrier diffusion deep into the in the field-free regions of the lightly doped bulk silicon substrate. For ROIC wafer fabrication, the use of conventional micron-thin lightly doped expitaxial layers on heavily doped substrates exhibiting shorter minority carrier lifetimes may be preferred over bulk silicon die on the order of several millimeters in thickness. The use of bulk silicon thinning of the ROIC die after wafer processing may also reduce the effects of charge spreading associated with a particle strike.
6.6
Conclusions
ROIC development for visible and infrared imaging has reached levels of complexity, in terms of transistor count, that are approaching that for CMOS digital logic processors for space applications. This is an even more surprising outcome when one considers that ROIC designs typically include very large sets of both minimum and non-minimum geometry transistors used for various applications within the imager readout electronics. At present, commercial foundries remain a preferred fabrication source for ROIC designs due to factors including the availability of photostepper level reticle photocomposure of very large arrays, which can be designed to exceed the currently available reticle field limits of approximately 2.2 cm x 2.2 cm. Furthermore, the commercial foundry history of achieving predictable yields for large die, and the routine use of large-diameter wafers allows for maximizing the number of large-area ROIC die sites per wafer. With the advent of deep-submicron feature-sized CMOS, the task of total-dose hardening at the transistor level has largely been reduced to employing hardening-by-design techniques to suppress edge-effect leakage paths. Commercially available nanometer gate oxide transistors typically exhibit threshold shifts at 100 krad(Si) and beyond of on the order of a few percent of the analog supply voltage. Additionally, cancellation of total-dose-induced shifts in threshold voltage and dark current at the pixel level has been successful using analog background subtraction techniques. Also, the use of current-mirror techniques for biasing pixel amplifier and peripheral image array analog functions results in first-order compensation of threshold voltage shifts due to total dose as well as operation at cryogenic temperatures. Design considerations and performance up to 1 Mrad(Si) and beyond for both SFD and CTIA pixel types have been discussed. For example, 512 x 512 pixel image arrays implemented with SDF pixels have been irradiated up to 10.2 Mrad(Si) with no apparent change in the operation of the FPN reduction circuitry contained within the array column readout electronics. [140] In another case, 0.5-µm-feature-sized circuits containing a 16 x 16 element CTIA test array that uses enclosed NMOS devices have successfully been operated at IV-107
ambient temperature up to the 100 kiloelectron well capacity with no appreciable loss of photoconversion gain, dynamic range, or linearity after 1 Mrad(Si) Cobalt-60 exposure. [141] Mitigation of heavy-ion single-event upset effects is likewise managed using hardening-bydesign techniques. The use of hardened latch topologies results in greatly reduced upset rates as compared to unhardened latch circuit designs from cell families available for use with commercial foundry offerings. Reduced upset rates using triple modular redundancy and majority voting logic circuitry provides further mitigation when applied either at the latch level or at the functional level, such as for implementing critical timing pattern generators. It is believed that with increased use of on-chip deep-submicron-feature-sized digital logic functions to implement more powerful digital image processing within an ROIC chip (including digital subtraction of scene background and fixed pattern noise sources as well as image frame averaging), the end-to-end sensor performance in space radiation environments will continue to improve.
IV-108
7.0
Summary and Concluding Remarks
Over the past two decades, advanced focal-plane arrays operating in the visible and infrared regions of the electromagnetic spectrum have undergone remarkable technological advancement in terms of array size, sensitivity, thermal performance, operating speed, mode programmability, and ease of use. This progress has enabled the design and development of new classes of highly capable electro-optical (EO) space-based sensor systems. The interaction of advanced focal planes found in these EO sensors (and their associated microelectronics) with the natural space radiation environment has been a subject of concern and often a central theme in their development. Careful consideration of the pertinent space radiation environments, including their quantitative spatial/temporal characteristics, coupled with understanding the details of radiation transport through sensor shielding and assessing sensor performance degradation induced by radiation-induced transient and permanent damage, are essential steps for EO sensor designers. That process in turn provides clear insight into the requirements for radiation-hardened focal-plane technology. It also serves as a guide to the development of system-level strategies for mitigating transient and permanent radiation effects. This tutorial provides an overview of key aspects of the process outlined above. In Section 2, the principal models used for predicting the structured earth-bound space-based radiation environment (such as NASA’s AP-8 and AE-8 codes) are discussed. More recent (and in many cases more accurate) modeling approaches, based on the CRRES and APEX databases, are also considered. The implications of spacecraft shielding and the models used to accurately perform shielding analysis are discussed. Examples are provided of useful parametric plots, such as annual dose-depth curves and external proton/electron flux versus particle energy parameterized by orbital position. This information is presented for wellknown example orbits. In addition, the effects of solar flares are discussed. In Section 3, radiation effects on all of the key types of silicon-based visible focal planes are reviewed, including charged-coupled devices, active pixel sensors, charge-injection devices, and silicon hybrid technologies. The effects of ionizing and nonionizing radiation that are common to all of these sensors (e.g., the increase in average dark current and the formation of dark-current spikes) are detailed. The displacement damage effect that is unique to CCD technology - the increase in charge transfer inefficiency (CTI) - is described. Other topics covered include dark-current enhancement due to high-electric-field regions, annealing effects, random telegraph noise, and potential hardening approaches. In Section 4, radiation effects on infrared focal planes are reviewed. The discussion in that section focuses on effects in the focal-plane layer that serves as the infrared detecting material (i.e., indium antimonide, mercury cadmium telluride, doped silicon, etc.), which is generally operated at cryogenic temperatures. (The radiation susceptibility of cryogenic silicon CMOS readout circuits used to extract small signals from infrared detector arrays is described separately in Section 6.) The roles of displacement and surface damage are described, as are the systematic effects of ionization-induced transients. An aspect unique to
IV-109
infrared detector arrays – damage to the infrared material surface passivants (i.e., ZnS and CdTe) - is described. Section 5 provides detailed information on both empirical measurements and simulationbased modeling of focal-plane responses to transient ionizing radiation (primarily due to incident protons). Simulation results are given for visible (silicon) and infrared (HgCdTe) focal planes as well as for devices with complex pixel/multiplexing architectures. The focus is on understanding the spatial and amplitude characteristics of these radiation-induced transient effects - in an image format - to facilitate the understanding of their system-level implications and potential mitigations. Section 6 describes radiation susceptibility and hardening strategies associated with the readout-integrated-circuit portion of visible and infrared detector arrays. Whereas in the earlier sections of this tutorial the emphasis was on the impact of radiation on detector pixels, here the on-focal-plane microelectronic circuitry responsible for converting pixel photocurrents into usable signals is addressed. The impact of radiation on field-effect transistors that act as building blocks for these signal conversion circuits, such as the sourcefollower-per detector and the capacitive-feedback transimpedance amplifier, are described as well as selected design and process mitigation strategies. The dependence of radiation susceptibility and permanent damage on cryogenic operating temperature is also reviewed. This tutorial aims to provide a systematic review of the important aspects of space radiation effects on visible and infrared focal planes, spanning from the system end of the problem (i.e. environments for specific orbits and shielding) to the technology details of the focal-plane devices and their microelectronic components. Hopefully, the background information provided in this course will be useful to space-system instrument designers and to technology planners for future applications and the associated need for radiation-hardened focal-plane technologies.
8.0
Acknowledgements
One of us (AHK) thanks Robert E. Mills of Raytheon Vision Systems for providing two figures from one of his papers.
IV-110
9.0
References
[1] E. G. Stassinopoulous and J. P. Raymond, “The space radiation environment for electronics”, Proc. IEEE, vol. 76, pp. 1423-1442, Nov. 1988. [2] E. G. Stassinopoulous, “Radiation environments of space”, 1990 IEEE NSREC Short Course. [3] J. L. Barth, “Modeling space radiation environments” 1997 IEEE NSREC Short Course. [4] J. Mazur, “The radiation environment outside and inside a spacecraft”, 2002 IEEE NSREC Short Course. [5] J. L. Barth, C. S. Dyer, and E. G. Stassinopoulos, “Space, atmospheric, and terrestrial radiation environments,” IEEE Trans. Nucl. Sci., vol. 50, no. 3, pp. 466-482, June 2003. [6] D. M. Sawyer and J. I. Vette, “AP8 Trapped Proton Environment for Solar Maximum and Solar Minimum”, Report NSS DC 76-06, National Space Science Data Center, Greenbelt, MD, December 1976. [7] J. I. Vette, “The AE8 Trapped Electron Model Environment”, Report NSS DC/WDC-A-RS-9124, National Space Science Data Center, Greenbelt, MD, 1991. [8] Space Radiation, Severn Communications Corporation, Millersville, MD. [9] J. B. Blake et al., “On the Access of Solar Protons to the Synchronous Altitude Region”, Journal of Geophysical Research, Volume 79, pp. 1345-1348, 1974. [10] J. H. King, “Solar Proton Fluences for 1977 – 1983 Space Missions”, Journal of Spacecraft and Rockets, Volume 11, pp. 401 – 408, 1974. [11] M. S. Gussenhoven, E. G. Mullen, M. D. Violet, C. Hein, J. Bass, and D. Madden, “CRRES high energy proton flux maps,” IEEE Trans. Nucl. Sci., vol. 40, pp. 1450-1457, Dec. 1993. [12] J. D. Meffert and M. S. Gussenhoven, “CRRESPRO Documentation”, Phillips Laboratory Report, PL-TR-94-2218 ERF, No. 1158, 1994. [13] J. R. Srour, S. C. Chen, S. Othmer, and R. A. Hartmann, “Neutron Damage Mechanisms in Charge Transfer Devices,” IEEE Trans. Nucl. Sci., vol. 25, pp. 1251-1260, December 1978. [14] J. R. Srour, S. C. Chen, S. Othmer, and R. A. Hartmann, “Radiation Damage Coefficients for Silicon Depletion Regions,” IEEE Trans. Nucl. Sci., vol. 26, pp. 4784-4791, December 1979. [15] C. J. Dale, P. W. Marshall, E. A. Burke, G. P. Summers, and G. E. Bender, “The generation lifetime damage factor and its variance in silicon,” IEEE Trans. Nucl. Sci., vol. 36, no. 6, pp. 1872-1881, December 1989. [16] P. W. Marshall, C. J. Dale, E. A. Burke, G. P. Summers, and G. E. Bender, “Displacement damage extremes in silicon depletion regions,” IEEE Trans. Nucl. Sci., vol. 36, no. 6, pp. 18311889, December 1989. [17] C. J. Dale, P. W. Marshall, and E. A. Burke, “Particle-induced spatial dark current fluctuations in focal plane arrays,” IEEE Trans. Nucl. Sci., vol. 37, no. 6, pp. 1784-1791, December 1990. [18] P. W. Marshall, C. J. Dale, and E. A. Burke, “Proton-induced displacement damage distributions in silicon microvolumes,” IEEE Trans. Nucl. Sci., vol. 37, no. 6, pp. 1776-1783, December 1990. [19] J. R. Srour and D. H. Lo, “Universal damage factor for radiation-induced dark current in silicon devices,” IEEE Trans. Nucl. Sci., vol. 47, no. 6, pp. 2451-2459, December 2000. [20] G. Lindstrom, M. Moll, and E. Fretwurst, “Radiation Hardness of Silicon Detectors – A Challenge from High-Energy Physics,” Nuclear Instruments and Methods in Physics Research, Section A, vol. 426, pp. 1-15, 1999. [21] H. W. Kraner, Z. Li, and K. U. Posnecker, “Fast Neutron Damage in Silicon Detectors,” Nuclear Instruments and Methods in Physics Research, Section A, vol. 279, pp. 266-271, 1989. [22] E. A. Burke, “Energy Dependence of Proton-Induced Displacement Damage in Silicon,” IEEE Trans. Nucl. Sci., vol. 33, pp. 1276-1281, December 1986.
IV-111
[23] G. P. Summers, E. A. Burke, C. J. Dale, E. A. Wolicki, P. W. Marshall, and M. A. Gehlhausen, “Correlation of Particle-Induced Displacement Damage in Silicon,” IEEE Trans. Nucl. Sci., vol. 34, pp. 1134-1139, December 1987. [24] C. J. Dale, P. W. Marshall, G. P. Summers, and E. A. Wolicki, “Displacement Damage Equivalent to Dose,” Appl. Phys. Lett., vol. 54, pp. 451-453, 1989. [25] S. R. Messenger, E. A. Burke, G. P. Summers, M. A. Xapsos, R. J. Walters, E. M. Jackson, and B. D. Weaver, “Nonionizing Energy Loss (NIEL) for Heavy Ions,” IEEE Trans. Nucl. Sci., vol. 46, pp. 1595-1602, December 1999. [26] S. Kuboyama, H. Shindou, T. Hirau, and S. Matsuda, “Consistency of bulk damage factor and NIEL for electrons, protons, and heavy ions in Si CCDs,” IEEE Trans. Nucl. Sci., vol. 49, no. 6, pp. 2684-2689, December 2002. [27] G. R. Hopkinson, C. J. Dale, and P. W. Marshall, “Proton Effects in Charge-Coupled Devices,” IEEE Trans. Nucl. Sci., vol. 43, pp. 614-627, April 1996. [28] S. R. Messenger, M. A. Xapsos, E. A. Burke, R. J. Walters, and G. P. Summers, “Proton Displacement Damage and Ionizing Dose for Shielded Devices in Space,” IEEE Trans. Nucl. Sci., vol. 44, pp. 2169-2173, December 1997 [29] R. Gereth, R. H. Haitz, and F. M. Smits, “Effects of single neutron-induced displacement clusters in special silicon diodes,” J. Appl. Phys., vol. 36, no. 12, pp. 3884-3894, December 1965. [30] J. R. Srour, Z. Shanfield, R. A. Hartmann, S. Othmer, and D. M. Newberry, “Permanent damage introduced by single particles incident in silicon devices,” IEEE Trans. Nucl. Sci., vol. 30, no. 6, pp. 4526-4532, December 1983. [31] J. R. Srour and R. A. Hartmann, “Effects of single neutron interactions in silicon integrated circuits,” IEEE Trans. Nucl. Sci., vol. 32, no. 6, pp. 4195-4200, December 1985. [32] J. R. Srour, R. A. Hartmann, and K. S. Kitazaki, “Permanent damage produced by single proton interactions in silicon devices,” IEEE Trans. Nucl. Sci, vol. 33, no. 6, pp. 1597-1604, December 1986. [33] J. R. Srour and R. A. Hartmann, “Enhanced displacement damage effectiveness in irradiated silicon devices,” IEEE Trans. Nucl. Sci., vol. 36, no. 6, pp. 1825-1830, December 1989. [34] J. C. Pickel, A. H. Kalma, G. R. Hopkinson and C. J. Marshall, “Radiation effects on photonic imagers – a historical perspective,” IEEE Trans. Nucl. Sci., vol. 50, no. 3, pp. 671-688, June 2003. [35] M. S. Robbins, “High-energy proton-induced dark signal in silicon charge coupled devices,” IEEE Trans. Nucl. Sci., vol. 47, no. 6, pp. 2473-2479, December 2000. [36] G. R. Hopkinson and Ch. Chlebek, “Proton damage effects in an EEV CCD imager,” IEEE Trans. Nucl. Sci., vol. 36, no. 6, pp. 1865-1871, December 1989. [37] J. Janesick, T. Elliott, and F. Pool, “Radiation damage in scientific charge-coupled devices,” IEEE Trans. Nucl. Sci., vol. 36, no. 1, pp. 572-578, January 1989. [38] I. H. Hopkins and G. R. Hopkinson, “Further measurements of random telegraph signals in proton irradiated CCDs,” IEEE Transactions on Nuclear Science, vol. 42, no. 6, pp. 2074 – 2081, December 1995. [39] G. R. Hopkinson, “Radiation effects in a CMOS Active Pixel Sensor,” IEEE Trans. Nucl. Sci., vol. 47, pp. 2480-2484, December 2000. [40] J. Bogaerts, B. Dierickx, and R. Mertens, “Enhanced dark current generation in protonirradiated CMOS Active Pixel Sensors,” IEEE Trans. Nucl. Sci., vol. 49, pp. 1513-1521, June 2002. [41] G. R. Hopkinson, “Cobalt60 and proton radiation effects on large format, 2-D, CCD arrays for an earth imaging application,” IEEE Trans. Nucl. Sci., vol. 39, no. 6, pp. 2018-2025, December 1992. [42] I. H. Hopkins and G. R. Hopkinson, “Random telegraph signals from proton-irradiated CCDs,” IEEE Transactions on Nuclear Science, vol. 40, no. 6, pp. 1567-1574, December 1993.
IV-112
[43] J. Bogaerts, B. Dierickx, and R. Mertens, “Random Telegraph Signals in a RadiationHardened CMOS Active Pixel Sensor,” IEEE Trans. Nucl. Sci., vol. 49, pp. 249-257, February 2002. [44] T. L. Miller, D. A. Thompson, M. B. Elzinga, T.-H. Lee, B. C. Passenheim, and R. E. Leadon, “Experimental evaluation of high speed CCD imager radiation effects using Co60 and proton irradiation,” IEEE Radiation Effects Data Workshop Record, July 1993, pp.56-63 (published by IEEE in 1994). [45] K. P. Klaasen et al., “Inflight performance characteristics, calibration, and utilization of the Galileo solid-state imaging camera,” Opt. Eng., vol. 36, pp. 3001-3027, November 1997. [46] D. R. Smith, A. D. Holland, M. S. Robbins, R. M. Ambrosi, and Ian B. Hutchinson, “Protoninduced leakage current in CCDs,” Proceedings of SPIE, vol. 4851, pp. 842-848, March 2003. [47] G. R. Hopkinson and A. Mohammadzadeh, “Comparison of CCD damage due to 10- and 60MeV protons, IEEE Trans. Nucl. Sci., vol. 50, pp. 1960-1967, December 2003. [48] M. Chugg, R. Jones, M. J. Moutrie, J. R. Armstrong, D. B. S. King, and N. Moreau, “Single particle dark current spikes induced in CCDs by high energy neutrons,” IEEE Trans. Nucl. Sci., vol. 50, pp. 2011-2017, December 2003. [49] C. J. Marshall and P. W. Marshall, “Proton effects and test issues for satellite designers, part B: Displacement effects,” in 1999 IEEE NSREC Short Course Notes, July 1999, pp. IV-50 – IV110. [50] G. R. Hopkinson, “Proton-induced changes in CTE for n-channel CCDs and the effect on star tracker performance,” IEEE Trans. Nucl. Sci., vol. 47, no. 6, pp. 2460-2465, December 2000. [51] C. Dale, P. Marshall, B. Cummings, L. Shamey, and A. Holland, “Displacement effects in mixed particle environments for shielded spacecraft CCDs,” IEEE Trans. Nucl. Sci., vol. 40, no. 6, pp. 1628-1637, December 1993. [52] D. H. Lo and J. R. Srour, “Modeling of proton-induced CCD degradation in the Chandra Xray Observatory,” IEEE Trans. Nucl. Sci., vol. 50, no. 6, pp. 2018-2023, December 2003. [53] A. Holmes-Siedle and L. Adams, Handbook of Radiation Effects, Second Edition, New York: Oxford University press, 2002, p. 81. [54] J. R. Srour, R. A. Hartmann, and S. Othmer, “Transient and Permanent Effects of Neutron Bombardment on a Commercially Available N-Buried-Channel CCD,” IEEE Trans. Nucl. Sci., vol. 27, pp. 1402-1410, December 1980. [55] M. Moll, E. Fretwurst, and G. Lindstrom, “Leakage Current of Hadron Irradiated Silicon Detectors – Material Dependence,” Nuclear Instruments and Methods in Physics Research, Section A, vol. 426, pp. 87-93, 1999. [56] 1. N. S. Saks, “A technique for suppressing dark current generated by interface states in buried channel CCD imagers,” IEEE Electron Device Letters, vol. 1, n0. 7, pp. 131-133, July 1980. [57] G. R. Hopkinson, “Proton damage effects on p-channel CCDs,” IEEE Trans. Nucl. Sci., vol. 46, no. 6, pp. 1790-1796, December 1999. [58] J. Spratt, B. C. Passenheim, and R. E. Leadon, “The effects of nuclear radiation on P-channel CCD imagers,” in 1997 IEEE Radiation Effects Data Workshop Record, pp. 116-121. [59] C. Bebek et al., “Proton radiation damage in P-channel CCDs fabricated on high-resistivity silicon,” IEEE Trans. Nucl. Sci., vol. 49, no. 6, pp. 1221-1225, June 2002. [60] N. S. Saks, M. G. Ancona, and J. A. Modolo, “Radiation Effects in MOS Capacitors with Very Thin Oxides at 80 K,” IEEE Transactions on Nuclear Science, vol. 31, no. 6, pp. 1249-1255, December 1984. [61] D. E. Romeo and C. S. Paul, unpublished report, The Aerospace Corporation, 2001. [62] J. Janesick and G. Putnam, “Developments and applications of high-performance CCD and CMOS imaging arrays,” Annual Review of Nuclear and Particle Science, vol. 53, pp. 263-300, 2003.
IV-113
[63] J. Janesick, “Charge coupled CMOS and hybrid detector arrays,” Proc. SPIE, vol. 5167, pp. 1-18, 2004 [64] A. H. Kalma and R. A. Cesena, “Radiation Testing of Trimetal Infrared Detectors,” IEEE Trans. Nucl. Sci., vol. 26, 4833 (1979). [65] J. C. Pickel and M. D. Petroff, “Nuclear Radiation Induced Noise in Infrared Detectors,” IEEE Trans. Nucl. Sci., 22, 2456 (1975). [66] E. A. Burke, J. J. Boyle, and H. J. Hummler, “Gamma-Induced Noise in CCDs,” IEEE Trans. Nucl. Sci., 28, 4068 (1981). [67] M. D. Petroff, J. C. Pickel, and M. P. Curry, “Low-Level Radiation Effects in Extrinsic Infrared Detectors,” IEEE Trans. Nucl. Sci., 26, 4840 (1979). [68] J. T. Montroy, R. Baron, G. C. Albright, J. Boisvert, and L. D. Flesner, “Energetic ElectronInduced Impurity Ionization in Si:As IBC Detectors,” IEEE Trans. Nucl. Sci., 1307 (1988). [69] R. E. Clement, J. C. Boisvert, and P. J. Sullivan, “Characterization and Modeling of Neutron Induced Transient Response Changes in Si:As IBC Detectors,” IEEE Trans. Nucl. Sci., 38, 1377 (1991). [70] E. L. Divita, M. J. Holtzman, R. E. Mills, and D. T. Walsh, “Methodology for Testing IR Detectors in Simulated Nuclear Radiation Environments,” Proc. SPIE, 1108, 274 (1989). [71] E. L. Divita, R. E. Mills, T. L. Koch, M. J. Gordon, R. A. Wilcox, and R. E. Williams, “Methodology for testing infrared focal plane arrays in simulated nuclear radiation environments,” Proc. SPIE, 1686, 50 (1992). [72] A. G. Foyt, T. C. Harman, and J. P. Donnelly, “Type Conversion and n-p Junction Formation in HgCdTe Produced by Proton Bombardment,” Appl. Phys. Lett., 18, 321 (1971). [73] J. Melngailis, J. L. Ryan, and T. C. Harman, “Electron Radiation Damage and Annealing of Hg1-xCdxTe at Low Temperature,” J. Appl. Phys., 44, 2647 (1973). [74] C. E. Mallon, J. A. Naber, J. F. Colwell, and B. A. Green, “Effects of Electron Irradiation on the Electrical and Optical Properties of HgCdTe, IEEE Trans. Nucl. Sci., 20, 214 (1973). [75] J. A. Naber, R. E. Leadon, H. T. Harper, B. A. Green, and C. E. Mallon, “Effects of Electron Irradiation at 80 K on Ternary Compounds Containing Tellurium,” Lattice Defects in Semiconductors, p. 321, Inst. Of Phys., London (1974). [76] F. D.Shepherd, Jr., “Radiation Effects on the Spectral Response of HgCdTe,” IEEE Trans. Nucl. Sci., 21, 34 (1974). [77] R. E. Leadon and C. E. Mallon, “Model for Defects in HgCdTe Due to Electron Irradiation,” Infrared Phys., 15, 259 (1975). [78] C. E. Mallon, B. A. Green, R. E. Leadon, and J. A. Naber, “Radiation Effects in Hg1xCdxTe,” IEEE Trans. Nucl. Sci., 22, 2283 (1975). [79] B. A. Green, R. E. Leadon, and C. E. Mallon, “Mobility Changes Produced by Electron Irradiation of n-type Hg1-xCdxTe,” J. Appl. Phys., 47, 3127 (1976). [80] R. E. Leadon, C. E. Mallon, and J. A. Naber, “Effects of Radiation on the Low-Temperature Carrier Recombination Lifetimes in HgCdTe,” Proc. of the International Conference on Radiation Effects in Semiconductors, 514, Inst. Of Phys., London (1977). [81] V. N. Brudnyi et al., “Electrical and Recombination Characteristics of CdxHg1-xTe with electrons at T=300 K,” Sov. Phys.-Semiconductors, 11, 905 (1977). [82] A. V. Voitsehovski, V. N. Brudnyi, Yu. V. Lilenko, M. A. Krivov, and A. S. Petrov, “High Temperature Defects in Electron-Irradiated Semiconductors HgCdTe, PbSnTe,” Solid State Comm., 31, 105 (1979). [83] A. V. Voitsekhovskii, A. P. Kokhanenko, Yu. V. Lilenko, and A. S. Petrov, “Carrier Lifetime in Electron-Irradiated p-Type HgCdTe Crystals,” Sov. Phys.-Semiconductors, 15, 386 (1981). [84] A.V.Voitsekhovskii, A.P.Kokhanenko, Yu.V.Lilenko, and A.S.Petrov, “Changes in the Carrier Lifetime in HgCdTe as a Result of Electron Irradiation and Subsequent Annealing,” Sov. Phys.-Semiconductors, 15, 930 (1981).
IV-114
[85] A.V.Voitsekhovskii, Yu.V.Lilenko, A.P.Kokhanenko, and A.S.Petrov, “High Temperature Electron Irradiation and Isochronal Annealing of p-Type HgCdTe Crystals,” Rad. Eff., 66, 79 (1982). [86] A. V. Voitsekhovskii, A. P. Kokhanenko, S. F. Koverchik, Yu. V. Lilenko, and A. S. Petrov, “Characteristic Features of the Behavior of Electron-Irradiated Hg1-xCdxTe Crystals Subjected to Various Heat Treatments,” Sov. Phys.-Semiconductors, 17, 1119 (1983). [87] A. V. Voitsekhovskii et al, “Positron Annihilation in Electron-Irradiated HgTe and Hg1xCdxTe (x=0.2) Crystals,” Sov. Phys.-Semiconductors, 20, 514 (1986). [88] A. V. Voitsekhovskii, A.P.Kokhanenko, A.S.Petrov, Yu.V.Lilenko, and A.D.Pogrebnyak, “Investigation of Radiation Defects in Electron Irradiated Hg1-xCdxTe Crystals Using Positron Annihilation,” Cryst. Res. Technol., 23, 237 (1988). [89] V.I.Ivanov-Omskii, N.V.Kutekhov, V.A.Smirnov, Sh.U.Yuldashev, and O.A.Gadaev, “Neutron Irradiation of CdxHg1-xTe,” Sov. Phys.-Semiconductors, 26, 238 (1992). [90] R. M. Broudy and V. J. Mazurczyk, “(HgCd)Te Photoconductive Detectors,” Semiconductors and Semimetals, Vol. 18, pp157-111 (Edited by R.K.Willardson and A.C.Beers, Academic Press, 1981). [91] M. B. Reine, A. K. Sood, and T. J. Tredwell, “Photovoltaic Infrared Detectors,” Semiconductors and Semimetals, Vol. 18, pp 201-311 (Edited by R.K.Willardson and A.C.Beers, Academic Press, 1981). [92] N. D. Wilsey, C. S. Guenzer, B. Molnar, and W. J. Moore, “A Comparison of Fast Neutron Irradiation Effects in Photoconductive and Photovoltaic InSb Infrared Detectors,” IEEE Trans. Nucl. Sci., 22, 2448 (1975). [93] R. R. Billups and W. L. Gardner, “Radiation Damage Experiments on PbS Infrared Detectors,” Infrared Physics, 1, 199 (1961). [94] B. Molnar, “Fast Neutron Irradiation Damage on Room Temperature PbS Detectors,” IEEE Trans. Nucl. Sci., 21, 103 (1974). [95] Sh.B.Atakulov, F.A.Zaitov, Yu.V.Matershev, K.E.Onarkulov, and A.E.Shavrov, “Diffusive Nature of Radiation-Induced Degradation of Photoconducting Lead Sulfide Films,” Sov. Phys.Semiconductors, 19, 1288 (1985). [96] R.P.Day, R.A.Wallner, and E. A. Lodi, “An Experimental Study of Energetic Proton Radiation Effects on IRI lead sulfide Infrared Detectors,” Infrared Physics, 1, 212 (1961). [97] A. H. Kalma and C. J. Fischer, “Neutron Damage Mechanisms in Silicon at 10 K,” IEEE Trans. Nucl. Sci., 24, 2158 (1977). [98] A. H. Kalma, ‘Nuclear and Space Radiation Effects in Infrared Detectors,’ Proc. SPIE, 217, 186 (1980). [99] F.A.Junga, W.W.Anderson, & R.B.Emmons, “Effects of Gamma Irradiation on Surface Properties and Detector Properties of HgCdTe Photoconductors,” IEEE Trans. Nucl. Sci., 25, 1274 (1978). [100] A. H. Kalma and M. A. Hopkins, ‘Ionizing Radiation Effects in HgCdTe MIS Capacitors,’ IEEE Trans. Nucl. Sci., 28, 4083 (1981). [101] J. R. Waterman and J. M. Killiany, “2 MeV Electron Irradiation Effects in (Hg,Cd)Te CCDs,” IEEE Trans. Nucl. Sci., 30, 4209 (1983). [102] J. R. Waterman, “Radiation Induced Interface Trap Limited Storage Times in 10 Micron Cutoff Wavelength (Hg,Cd)Te MIS Capacitors ,” IEEE Trans. Nucl. Sci., 35, 1313 (1988). [103] D. W. Domkowski, D. G. eller, L. R. Johnson, C. I. Westmark, C. B. Norris, C. T. Fuller, and J. Bajaj, “Effects of 6 MeV Electron Irradiation on the Electrical Characteristics of LPE Hg0.7Cd0.3Te Mesa Photodiodes,” IEEE Trans. Nucl. Sci., 33, 1471 (1986). [104] I. Arimura, J. N. Polky, and W. E. Maher, “Radiation Effects Analysis of a Coherent Laser Detection System,” IEEE Trans. Nucl. Sci., vol. 24, 2315 (1977).
IV-115
[105] G. M. Williams, A. H. B. Vanderwyck, E. R. Blazejewski, R. P. Ginn, C. C. Li, and S. J. Nelson, “Gamma Radiation Response of MWIR and LWIR HgCdTe Photodiodes,” IEEE Trans. Nucl. Sci., 34, 1592 (1987). [106] J. Favre, M. Konczykowski, and C. Blanchard, “Irradiation Defects in Hg1-xCdxTe Alloys,” Ann. Phys., 14 colloq., 193 (1989). [107] G. Sarusi, D. Eger, A. Zemel, N. Mainzer, R. Goshen, and E. Weiss, “Degradation Mechanisms of Gamma Irradiated LWIR HgCdTe Photovoltaic Detectors,” IEEE Trans. Nucl. Sci., 37, 2042 (1990). [108] J. E. Hubbs, G. A. Dole, M. E. Gramer, and D. C. Arrington, “Radiation Effects Characterization of Infrared Focal Plane Arrays Using the Mosaic Array Test System,” Opt. Eng., 30, 1739 (1991). [109] J. R. Waterman and R. A. Schiebel, “Ionizing Radiation Effects in n-Channel (Hg,Cd)Te MISFETs with Anodic Sulfide Passivation,” IEEE Trans. Nucl. Sci., 34, 1597 (1987). [110] M. M. Moriwaki, J. R. Srour, L. F. Lou, and J. R. Waterman, “Ionizing Radiation Effects on HgCdTe MIS Devices,” IEEE Trans. Nucl. Sci., 37, 2034 (1990). [111] M. M. Moriwaki, J. R. Srour, and R. L. Strong, “Charge Transport and Trapping in HgCdTe MIS Devices,” IEEE Trans. Nucl. Sci., 39, 2265 (1992). [112] A. H. Kalma, R. A. Hartmann, and B. K. Janousek, ‘Ionizing Radiation Effects in HgCdTe MIS Capacitors Containing a Photochemically Deposited SiO2 Layer,’ IEEE Trans. Nucl. Sci., 30, 4146 (1983). [113] E. Finkman and S. E. Schacham, “Interface Properties of Various Passivations of HgCdTe,” Proc. SPIE, 1106, 198 (1989). [114] R.E.Mills, “Novel Methods for Cryogenic Testing of Infrared Focal Plane Arrays in a Co60 Radiation Environment,” Proc. SPIE, 1686, 65 (1992). [115] E.R.Blazejewski, G.M.Williams, A.H.Vanderwyck, D.D.Edwall, E.R.Gertner, J.Ellsworth, L.Fishman, S.R.Hampton, and H.R.Vydynath, “Advanced LWIR HgCdTe Detectors for Strategic Applications,” Proc. SPIE, 2217, 278 (1994). [116] T.Wilhelm, R.K.Purvis, A.Singh, D.Z.Richardson, R.A.Hahn, J.R.Duffey, and J.A.Ruffner, “Status of Focal Plane Arrays (FPAs) for Space-Based Applications,” Proc. SPIE, 2217, 307 (1994). [117] X. Hu, X. Li, H. Lu, J. Zhao, H. Gong, and J. Fang, “Effect of Gamma Irradiation on RoomTemperature SWIR HgCdTe Photodiodes,” Proc. SPIE, 3553, 85 (1998). [118] X. Hu, K. X. Li, and J. Fang, “Influence of Gamma Irradiation on the Performance of HgCdTe Photovoltaic Devices,” Proc. SPIE, 3698, 920 (1999). [119] J. C. Pickel, R. A. Reed, R. Landbury, B. Rauscher, P. W. Marshall, T. M. Jordan, B.Fondness, and G.Gee, “Radiation-Induced Collection Charge in Infrared Detector Arrays,” IEEE Trans. Nucl. Sci., 49, 2822 (2002). [120] T.E. Dutton, W.F. Woodward, and T.S. Lomheim, “Simulation of Proton-Induced Transients on Visible and Infrared Focal Plane Arrays in a Space Environment”, Proceedings of the SPIE, Vol. 3063, pp.77-101, April, 1997. [121] T.S. Lomheim, R.M. Shima, J.R. Angione, W.F. Woodward, D.J. Asman, R.A. Keller, and L.W. Schumann, "Imaging Charge-Coupled Device (CCD) Transient Response to 17 and 50 MeV Proton and Heavy-Ion Irradiation," IEEE Transactions on Nuclear Science, Vol. 37, Part I, pp. 1876-1885, December, 1990. [122] T.S. Lomheim, et al., "Simulation and Verification of Proton-Induced Transient Responses in Linear CCD Imaging Arrays with Integral Bilinear Shift Registers", presented at the IEEE Charge-Coupled Devices (CCD) and Advanced Image Sensors Workshop (1993), University of Waterloo, Waterloo, Ontario, Canada, June 1993. [123] C. A. Klein, Journal of Applied Physics, Vol. 39, No. 4, 1968, pp. 2029-2038. [124] R. Lowell et al., SAIC, private communication
IV-116
[125] S. Kirkpatrick, “Modeling Diffusion and Collection of Charge from Ionizing Radiation in Silicon Devices," IEEE Trans. Electronic Devices, Vol. ED-26, 1979, pp. 1742-1753. [126] J. C. Pickel and J. T . Blandford, “Cosmic Ray Induced Errors in MOS Memory Ce l ls," IEEE Trans. Nuclear Science, Vol. NS-25, 1978, pp. 1166-1171. [127] J. C. Pickel et al., “Radiation-induced charge collection in infrared detector arrays,” IEEE Trans. Nucl. Sci., vol. 49, no. 6, pp. 2822-2829, December 2002. [128] R. Ladbury et al., “Characteristics of the Hubble Space Telescope’s radiation environment inferred from charge-collection modeling of near-infrared camera and multi-object spectrometer darkframes,” IEEE Trans. Nucl. Sci., vol. 49, no. 6, pp. 2765-2770, December 2002. [129] J. C. Pickel et al., “Proton-induced secondary particle environment for infrared sensor applications,” IEEE Trans. Nucl. Sci., vol. 50, no. 6, pp. 1954-1959, December 2003. [130] P. W. Marshall et al., “Proton-induced transients and charge collection measurements in a LWIR HgCdTe focal plane array,” IEEE Trans. Nucl. Sci., vol. 50, no. 6, pp. 1968-1973, December 2003. [131] E. G. Stassinopoulos, “Charged Particle Radiation Exposure of Geocentric Satellites," American Institute of Physics Conference Proceedings, Vol. 186, High Energy Radiation Background in Space, 1987, pp. 3-63. [132] T. S. Lomheim, Infrared Systems and Technology Course, 2002. [133] M. J. Hewitt, et al., “Infrared Readout Electronics: A Historical Perspective,” SPIE, vol. 2226, 1994. [134] L. J. Kozlowski, “Low-noise capacitive transimpedance amplifier performance versus alternative IR detector interface schemes in submicron CMOS,” Proc. SPIE vol. 2745, pp. 2-11, June 1996. [135] R. C. Lacoe, NSREC Short Course, July 2003. [136] R. C. Lacoe, J. V. Osborn, R. Koga, S. Brown, and D. C. Mayer, “Application of Hardnessby-Design Methodology to Radiation –Tolerant ASIC Technologies, “ IEEE Trans. on Nuclear Science, Vol. 47, No.6, pp. 2334-2341, December 2000. [137] W.J. Snoeys, T.A. Palacios Guiterrez, G. Anelli, “A New NMOS Layout Structure for Radiation Tolerance, “ IEEE Trans. on Nuclear Science, Vol. 49, No. 4, pp. 1829-1833, August 2002. [138] J. K. Shreedhara, et al., “Circuit Technique for Threshold Voltage Stabilization Using Substrate Bias in Total Dose Environments,” IEEE Trans. on Nuclear Science. Vol. 47, No. 6, pp. 2557-2560, December 2000. [139] M. H. White, et al., “Characterization of Surface Channel CCD Imaging Arrays at Low Light Levels,” IEEE Transactions on Solid State Circuits, vol. SC-9, pp. 1-13, February 1974. [140] J. Bogaerts, et al. “Total Dose and Displacement Damage Effects in a Radiation-Hardened CMOS APS,” IEEE Transactions on Electron Devices, Vol. 50, pp. 84-90, No. 1, January 2003. [141] D. E. Romeo and C. S. Paul, unpublished report, The Aerospace Corporation, 2001. [142] S. M. Sze, Physics of Semiconductor Devices, 2nd Edition, Wiley, New York, 1981. [143] J. A. Felix, et al., “Bias and Frequency Dependence of Radiation-Induced Charge Trapping in MOS Devices,” IEEE Transactions on Nuclear Science, pp. 2114-2120, vol. 48, no. 6, December 2001. [144] J. Baker, H. W. Lee, D. E. Boyce, CMOS, Circuit Design, Layout, and Simulation, pp. 427457, IEEE Press, New York, 1998. [145] J. C. Pickel, IEEE NSREC Short Course, July 1993. [146] D. M. Fleetwood, T. L. Meisenheimer, J. H. Scofield, “1/f-noise and Radiation Effects in MOS Devices,” IEEE Transactions on Electron Devices, vol. 41, pp 1953-1964, December, 1994. [147] H. D. Xiong, et al., “Temperature Dependence and Irradiation Response of 1/f-Noise in MOSFETs,” IEEE Transactions on Nuclear Science, vol. 49, no. 6, December 2002. [148] M. Loose, et al., “HAWAII-2RG: a 2K x 2K CMOS Multiplexer for Low and High Background Astronomy Applications,” SPIE Vol. 4850.
IV-117
[149] L. Kozlowski, et al., “Performance Limits in Visible and Infrared Imager Sensors,” IEDM, December 1999. [150] G. C. Messenger, M. S. Ash, The Effects of Radiation on Electronic Systems, Second Edition, pp, 416-494, Van Nostrand Reinhold, New York, 1992. [151] R. Velazco, et al., “SEU-Hardened Storage Cell Validation Using A Pulsed Laser,” IEEE Trans. on Nuclear Science, vol. 43, no. 6, pp. 2843-2848, December 1996. [152] R. Katz, et al., “SEU Hardening of Field Programmable Gate Arrays (FPGAS) for Space Applications and Device Characterization,” IEEE Trans. on Nuclear Science, vol. 41, no. 6, pp. 2179-2186, December 2000. [153] D.G. Mavis and P.H. Eaton, “SEU and SET Mitigation Techniques for FPGA Circuit and Configuration Bit Storage Design,” Proc. 2000 Mil. Aero. Appl. Prog. Dev. Tech. Conf., 2000. [154] D.G. Mavis and P.H. Eaton, “Soft Error Rate Mitigation Techniques for Modern Microcircuits,” Proc. of 2002 Intl. Rel. Phys. Symp., pp. 216-225, Apr. 2002 [155] M. N. Liu and S. Whitaker, “Low Power SEU Immune CMOS Memory Circuits,” IEEE. Trans. Nuc. Sci., vol. 39, no. 6, pp. 1679-1684, Dec 1992. [156] X. Belredon et al., “Heavy Ion-Induced Charge Collection Mechanisms in CMOS Active Pixel Sensor,” IEEE Trans. On Nuclear Science, Vol. 49, No. 6, pp. 2836-2843, December 2002. [157] C. J. Marshall et al., “Heavy Ion Transient Characterization of a Hardened-by-Design Active Pixel Sensor Array,” IEEE NSREC Radiation Effects Data Workshop, pp. 187-193, July 2002. [158] J. R. Janesick, Scientific Charge Coupled Devices, SPIE Press, Bellingham, WA, 2001. [159] E. R. Fossum, “Active pixel sensors: Are CCDs dinosaurs?,” Proc. SPIE, vol. 1900, pp. 2-14, 1993. [160] H-S Wong, “Technology and device scaling considerations for CMOS imagers,” IEEE Trans. Electron Devices, vol. 43, pp. 2131-2142, 1996. [161] Y. Bai et al., “Hybrid CMOS focal plane array extended UV and NIR response for space applications,” Proc. SPIE, vol. 5167, pp. 83-93, 2004. [162] C-Y Wu, Y-C Shih, J-F Lan, C-C Hsieh, C-C Huang, and J-H Lu, “Design, optimization, and performance analysis of new photodiode structures for CMOS Active-Pixel-Sensor (APS) imager applications,” IEEE Sensors Journal, vol . 4, no. 1, pp.135-144, February 2004.
IV-118
2004 IEEE NSREC Short Course
Section V
Solar Cell Technologies, Modeling, and Testing
Robert J. Walters Naval Research Laboratory
2004 NSREC Short Course
Solar Cell Technologies, Modeling, and Testing Robert J. Walters, S. R. Messenger, G. P. Summers, and E. A. Burke Naval Research Laboratory 1 2
Introduction ...........................................................................................................................................................2 Radiation Effects on Solar Cells............................................................................................................................3 2.1 Introduction .....................................................................................................................................................3 2.2 Solar Cell Device Physics ...............................................................................................................................3 2.3 Radiation-Induced Degradation Mechanisms .................................................................................................7 3 Modeling Techniques ..........................................................................................................................................12 3.1 Background ...................................................................................................................................................12 3.2 The JPL Method ............................................................................................................................................14 3.3 NRL Method .................................................................................................................................................17 3.4 Correlating the RDCs with NIEL ..................................................................................................................25 4 On-Orbit Solar Cell Performance Predictions .....................................................................................................26 4.1 Environment Calculations .............................................................................................................................26 4.2 Shielding Calculations...................................................................................................................................27 4.2.1 JPL Shielding Calculations..................................................................................................................27 4.2.2 NRL Shielding Calculations ................................................................................................................28 4.3 Solar Cell Performance Predictions...............................................................................................................32 4.4 Mission Examples .........................................................................................................................................33 5 Specific Solar Cell Technologies.........................................................................................................................38 5.1 Single-junction, Crystalline Semiconductor Solar Cells ...............................................................................38 5.2 Multijunction Solar Cells ..............................................................................................................................38 5.2.1 Mechanisms for Multijunction Solar Cell Radiation Response ...........................................................38 5.2.2 Modeling Multijunction Solar Cell Radiation Response .....................................................................47 5.3 Thin Film Photovoltaics ................................................................................................................................49 5.3.1 Amorphous Si ......................................................................................................................................49 5.3.2 CuIn(Ga)Se2 .........................................................................................................................................54 6 On-Orbit Solar Cell Performance Predictions .....................................................................................................62 7 Special Topics in Solar Cell Radiation Response................................................................................................69 7.1 Solar Cell Response at High Degradation Levels..........................................................................................69 7.1.1 Region II ..............................................................................................................................................70 7.1.2 Region III.............................................................................................................................................71 7.2 Case of Nonuniform Damage Deposition .....................................................................................................72 8 Testing Approaches .............................................................................................................................................79 9 Summary..............................................................................................................................................................81 10 References .....................................................................................................................................................82
V-1
1 Introduction This short course presents a study of the effects of exposure to the space radiation environment on the electrical performance of a variety of solar cell technologies and approaches used to model them. The discussion begins with a review of the basic physics of the photovoltaic effect and the operation of a solar cell. The basic mechanisms controlling the solar cell radiation response are then described for the case of a simple, single-junction device. Building on this background, two techniques for modeling solar cell performance in a radiation environment are presented, the equivalent fluence method developed by the US Jet Propulsion Laboratory and the displacement damage dose method developed by the US Naval Research Laboratory. The primary goal of both procedures is the correlation of degradation data taken after irradiation by different particles at various energies. The main difference between the two methods is that in the JPL method the energy dependence of the relative damage coefficients is experimentally determined, whereas in the NRL approach the energy dependence is calculated. The procedures for accounting for shielding are studied in detail. The two methods are then exercised to produce performance predictions for a single-junction GaAs solar cell in an Earth orbit, and the two methods are shown to give the same results. With the basic physics of a simple solar cell device understood and methods for modeling and predicting the solar cell performance in a space radiation environment established, the discussion expands to include more complex solar cell technologies. In particular, the radiation response mechanisms of multijunction and thin film solar cell technologies are discussed. In the case of the thin film solar cells, the annealing characteristics are also investigated in detail. The result is a comprehensive understanding of the response of these technologies to exposure to the space radiation environment. Using this understanding, a section of the short course is dedicated to making on-orbit solar cell performance predictions for state-of-the-art multijunction and thin film technologies. Predictions are made assuming both a typical rigid honeycomb solar array and a lightweight, flexible array. Orbits representative of LEO, MEO and GEO are considered. The results are used to highlight the strengths and weakness of each solar cell technology in realistic operational scenarios. The short course also includes a discussion of two aspects of solar cell radiation response termed "special topics". One such topic is the response of a solar cell to irradiation to high fluence levels. The damage mechanisms operative in this high fluence regime are described to produce a more comprehensive understanding of solar cell radiation response. The other special topic is the case of a solar cell response to non-uniform radiation-induced damage. A calculational method is presented for accurately modeling the solar cell response in this case, which also leads to a method for predicting the response of an arbitrary solar cell given knowledge of the cell structure and an estimate of the minority carrier diffusion length in the material. This method is applied to both single-junction Si and GaAs solar cell as well as more complicated structures such as multijunction InGaP2/GaAs/Ge solar cells. The discussion concludes with a look at approaches for ground based radiation testing of solar cells. It is shown how knowledge of the basic physics governing the solar cell radiation response can be used to craft a radiation experiment that produces maximum results while being both time
V-2
and cost efficient. Specific aspects of solar cell radiation testing that require special attention are highlighted. 2
Radiation Effects on Solar Cells
2.1 Introduction A solar cell consists essentially of a p-on-n junction in a semiconductor material that, when under illumination by solar photons, produces a voltage, called the photovoltage, by the photovoltaic effect. If an ohmic contact is placed on p and n sides of the junction, then the photogenerated charge carriers can be extracted from the device, thus generating a photocurrent. If the solar cell is connected across an electronic load, then power can be extracted and used to drive a system. Photovoltaics are, by far, the primary space power source for earth orbiting satellites and many interplanetary spacecraft. The space environment is often characterized by a harsh radiation environment; therefore, to be used in space, the radiation response of a solar cell must be well understood. In this section, the radiation response of a solar cell is described. The section begins with a brief review of the device physics of a solar cell and then describes the mechanisms by which particle radiation affects solar cell operation. 2.2 Solar Cell Device Physics A schematic drawing of a basic, single-junction solar cell is shown in Figure 1. This figure depicts an n+p structure where the n-type dopant concentration is one or two orders of magnitude higher than that of the p-type region so that a one-sided, abrupt junction exists. When the solar cell is illuminated, the photons penetrate the material, and those photons with energy above the semiconductor bandgap are absorbed and create electron-hole pairs. Each solar cell material has specific absorption characteristics. Examples of the absorption coefficient for Si and GaAs are shown in Figure 2. As the wavelength increases, the absorption coefficient decreases so that the longer wavelength photons are absorbed deeper into the solar cell. The absorption coefficient for GaAs is seen to be over an order of magnitude larger than that of Si for most of the wavelength shown. This is the case because GaAs is a direct band-gap material, so the photon can be directly converted into an electron-hole pair. Si, on the other hand, being an indirect band-gap, requires the absorption of a phonon during photon absorption in order to conserve both energy and momentum thereby reducing the photon absorption efficiency. Because of this, full absorption can be achieved in a GaAs based solar cells that is as thin as 2 µm while a Si solar cell must be 100 µm thick or more to achieve full photon absorption.
V-3
Incident sunlight
depletion n-type emitterregion ~ 10 18- W cm -3
LOAD
Depletion region - W
p-type base ~ 10 16 cm -3
p-type back surface field ~ 10
18
cm -3
Figure 1: This is a schematic drawing of a single junction solar cell.
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
-1
Absorption Coefficient (cm )
The absorption coefficient data for a given material as shown in Figure 2 must be combined with the incident photon spectrum to determine the solar cell efficiency. Solar cell efficiency refers to the efficiency with which a solar cell converts the incident photons to electrical energy. For space solar cells, the incident photon spectrum is the air mass zero (AM0) solar spectrum, i.e. the spectrum encountered outside the Earth’s atmosphere (Figure 3). For reference, the AM1.5 spectrum which is representative of the spectrum on the Earth’s surface is also shown in Figure 3. Calculations of ideal solar cell efficiency as a function of band-gap are shown in Figure 4 [1].
Si GaAs
400
600
800
1000
Wavelength (nm)
Figure 2: Absorption coefficient data for Si and GaAs.
V-4
1200
2.5
2.0
1.5
2
Irradiance (mW/cm /nm)
AM0
1.0
0.5
0.0 AM1.5 -0.5 0
500
1000
1500
2000
2500
Wavelength (nm)
Figure 3: AM0 and AM1.5 solar spectra. AM0 is the spectrum encountered in space. AM1.5 is representative of the spectrum on the Earth’s surface.
Figure 4: Ideal solar cell electrical conversion efficiency as a function of bandgap for both the AM0 space spectrum and the terrestrial AM1.5 spectrum. [1]
While Figure 4 gives ideal solar cell efficiency values, efficiencies achieved in practice are typically lower due to internal loss mechanisms within the absorber material. Once a photon is absorbed and the electron-hole pair is created, the free charge carriers must be separated and collected. Depending on the polarity of the semiconductor material in the region where the photon is absorbed, one of the charge carriers will be a minority carrier. It is the extraction of this minority carrier that gives rise to the photocurrent; therefore, a solar cell is a minority carrier
V-5
device. If the electron-hole pair is created away from the junction in a field-free region, then the minority charge carrier must diffuse through the material until it reaches the junction. Once at the junction, the minority charge carrier is swept across the depletion region by the junction electric field and collected. If the photon is absorbed within the depletion region, then the photogenerated charges are immediately separated by the field and collected. Since collection by drift along the electric field is much faster than diffusion through the bulk, carrier collection is much more efficient in the depleted region. The response of a solar cell to illumination by monochromatic light is referred to as the quantum efficiency (QE) [2]. There are two types of QE measurement. An external QE measurement does not account for reflections from the cell surface; whereas, an internal QE measurement accounts for surface reflections. External QE measurements made on several single-junction solar cell technologies are shown in Figure 5. For reference, the AM0 solar spectrum, normalized to the maximum value, is superimposed on the QE data. Referring back to the absorption coefficient data (Figure 2), the longer wavelength photons are absorbed deeper into the cell, so as the wavelength increase, the QE data are representative of the response of the emitter, depletion region and then the base of the solar cell. The cutoff in the QE data at the longer wavelength values corresponds to the band-edge of the semiconductor material. Si, with the smallest band gap of 1.12 eV has the widest spectral response, InP with a band-gap of 1.34 eV cuts off at about 920 nm, while GaAs with the largest band gap of 1.43 eV has the narrowest spectral response. Note how the GaAs and InP data show a sharp cutoff at the band-edge while the Si data show a more gradual decrease. This is again due to the fact that GaAs, and InP, are direct band-gap materials in contrast to Si. 1.0
External Quantum Efficiency
0.8
0.6
0.4
Si InP GaAs AM0 Spectrum (norm.)
0.2
0.0 400
600
800
1000
1200
Wavelength (nm)
Figure 5: External QE data measured on several single-junction solar cell technologies. For reference, the AM0 spectrum is also shown where the AM0 data has been normalized to the maximum value.
It is clear that the efficient operation of a solar cell is critically dependent on the mobility of minority charge carriers, and the primary physical parameters that influence solar cell
V-6
performance are the minority carrier lifetime (t) and diffusion length (L). The other physical attribute of the n/p junction that strongly affects the photovoltaic performance is the dark current characteristic which is the current generated across the junction under applied forward bias while the device is in the 4dark. The forward bias dark current flows in opposition to the photogenerated current thereby draining away some of the photovoltage. This can be seen in Figure 6, which gives an example of a current vs. voltage (IV) curve measured on a solar cell in the dark and under simulated solar light. In the following sections, the effect of particle irradiation on the minority carrier diffusion length and junction dark current will be investigated. The electrical output of a solar cell is typically characterized by parameters extracted from the IV curve. Three common parameters are illustrated in Figure 6. The current measured at short circuit is called the short circuit current (Isc). The voltage at open circuit is called the open circuit voltage (Voc). The maximum electrical power generated by the solar cell is given the symbol Pmp. The efficiency at which the solar cell converts the incident solar energy to electrical energy is given by dividing Pmp by the appropriate solar constant, which, for the space air mass zero (AM0) spectrum, is 136.7 mW/cm2. The shape of the IV curve is characterized by the fill factor (FF). The FF is defined as the ratio of Pmp to the product of Isc and Voc. The FF gives an indication how well the IV curve fills the maximum power rectangle, which is the rectangle formed in the forth quadrant by a vertical line passing through Voc and an horizontal line passing through Isc as shown by the dotted lines in Figure 6. 150
100
Dark current Under illumination voltage at open circuit, V oc
Current (mA)
50 0
-50
maximum power, P mp
current at short circuit, I sc
-100
0.0
0.2
0.4
0.6 Voltage (V)
0.8
1.0
Figure 6: This is a plot of typical IV curves measured on a single-junction solar cell. Illuminating the solar cell pulls the IV curve down into the forth quadrant so that power can be extracted from the device. The dark current is seen to oppose the photogenerated current. Typical parameters used to characterize solar cell performance are shown. The dotted lines define the “maximum power rectangle”.
2.3 Radiation-Induced Degradation Mechanisms From the point of view of photovoltaic operation, the primary effect of particle irradiation of a solar cell is displacement damage where atoms in the semiconductor lattice are moved from their equilibrium position to form point defects like vacancies and interstitials or defect complexes like vacancy-impurity clusters. These defects can form energy levels within the forbidden gap of
V-7
the semiconductor forming charge trapping centers. The existence of these defect centers affects charge transport in essentially five basic ways as shown schematically in Figure 7 [3].
1 +
-
Generation
2
3
Recombination
Trapping
Conduction Band
4
EC
Compensation
Valence Band
EV
5
Tunneling Figure 7: This is a schematic representation of the effects that radiationinduced defect levels can have on current transport in a solar cell [3].
Carrier generation (labeled #1 in Figure 7) occurs when the existence of the defect energy level makes it statistically favorable for a charge carrier to move from a bound to a free energy level. The liberated charges are then swept away by the junction field, which produces a current. This causes the dark IV characteristic to increase which degrades the photovoltage (i.e., Voc). Recombination (labeled #2 in Figure 7) occurs when it is statistically favorable for an electronhole pair to recombine at the defect site. When this occurs, free charge carriers are lost resulting in degradation of the photocurrent (i.e. Isc). Trapping (labeled #3 in Figure 7) occurs when a defect level is able to capture and temporarily localize free charge carriers which are then thermally reemitted. The forth mechanism illustrated in Figure 7 is referred to as compensation. Compensation occurs when a defect level permanently localizes a free charge carrier supplied by the dopant atoms. This reduces the majority charge carrier density and is referred to as carrier removal. The fifth mechanism is trap-assisted tunneling where the position of the defect level effectively lowers the tunneling potential. The discussion will now relate the effect of the radiation-induced defects to degradation of the solar cell photovoltaic output. This discussion will focus on single-junction, crystalline technologies. Other solar cell technologies like multi-junction and thin film cells will be addressed later. The effects of irradiation by 1 MeV electrons on the maximum power output of several technologies are shown in Figure 8. The crystalline Si data are from [4], and these cells are representative of the standard Si solar cells that served as the workhorse of the solar power industry for both terrestrial and space applications for many years. The GaAs/Ge data are from [5]. The GaAs/Ge devices consist of GaAs layers grown epitaxially on Ge substrates. These were the first III-V based devices to gain wide use as a photovoltaic device, and the GaAs/Gebased technology has since about 1990 replaced Si as the baseline for space power systems. The InP data are from [6], and these cells consist of InP layers grown expitaxially on InP substrates. V-8
The InP technology is another III-V based technology that was considered for space use due to its high efficiency and radiation resistance, but this technology is not currently in use. 30
25
2
Pmp (mW/cm )
20
15
10
Si InP GaAs
5
0 10
12
10
13
14
15
10 10 -2 1 MeV Electron Fluence (cm )
10
16
10
17
Figure 8: The radiation response of single-junction crystalline semiconductor solar cell technologies that have been developed for space use.
The general shape of the degradation curves for the different technologies is similar. When plotted against the log of the particle fluence, as in Figure 8, the PV parameter data remain essentially constant up to a certain fluence level beyond which the data degrade nearly linearly with the log of the fluence. This fluence level can be used to give an evaluation of the relative radiation hardness of the technology. The fluence level for the Si data, for example, is about 1x1012 cm-2 while that for the InP technology is about 5x1014 cm-2, indicating the InP solar cells to be much more radiation resistant that the Si cells. The curves in Figure 8 have the same general shape because the radiation-induced degradation mechanisms are similar in the different technologies. It has been shown in InP, GaAs, and Si cells that the primary degradation mechanism is the decrease in L due to radiation-induced recombination centers, and the implication is that this mechanism applies generally to all crystalline semiconductor-based photovoltaic devices [7,8,9]. Degradation of L reduces the carrier collection efficiency since those photogenerated carriers created in the cell base are less likely to reach the junction. This effect is seen in the QE data where most of the radiationinduced degradation appears in the response to longer wavelengths of light which are absorbed deeper in the cell (Figure 9). Degradation in the shorter wavelength response is apparent at higher fluence levels, and the mechanism for this degradation will be discussed in a later section. Degradation of the solar cell QE results in degradation of Isc which is reflected in the decrease in Pmp seen in Figure 8.
V-9
1.0 InP
External Quantum Efficiency
0.8
0.6 Fluence of 3 MeV protons
0 11 1x10 12 1x10 13 1x10 14 1x10
0.4
0.2
0.0 400
600
800
1000
Wavelength (nm)
Figure 9: The degradation in QE of an InP solar cell due to 3 MeV proton irradiation. The particle fluence is given in the legend in units of cm-2. The long wavelength response degrades due to diffusion length degradation.
Using the formalism of Hovel [2], an estimate of L can be extracted from analysis of the QE data, which has been done for the data of Figure 9 [6]. The decrease in L with the introduction of defects is given by [4]: Equation 1 σ υI 1 1 1 = 2 + ∑ i ti φ = 2 +K L φ D L (D d ) L 0 L0 2
where Lo is the pre-irradiation value of L, σi is the minority carrier capture cross section of the ith recombination center, Iti is the introduction rate of the Ith recombination center, ν is the thermal velocity of the minority carriers, D is the diffusion coefficient, and φ is the particle fluence. As shown in Equation 1 the specific parameters for each defect are typically lumped into a constant, KL, referred to as the damage coefficient for L. By determining L at various fluences, KL may be determined (Figure 10). The introduction of recombination/generation centers also causes an increase in the dark IV characteristic of the solar cell which degrades Voc. An analysis of dark IV data measured in an irradiated InP solar cell is shown in Figure 11 [6]. The irradiation is seen to cause an increase in the dark IV characteristic. The solid lines in the figure represent fits of the measured data to a theoretical expression for diode dark current. As observed in Figure 6, it is the current at the higher voltages (> ~ 0.65 V in this case) that most strongly affects the photovoltaic output. This portion of the dark IV curve is the diffusion current as given by Shockley [10], in which the magnitude of the diffusion current is shown to be inversely proportional to L. Thus, the introduction of recombination/generation centers by irradiation causes L to degrade which
V - 10
degrades Isc and the dark IV characteristic to increase which degrades Voc. These effects combine to reduce the solar cell Pmp. data determined from QE measurments least squares fit
40
KL = 2.2x10-7 1/L2 (µm-2)
30
20
10
0
n+p InP/Si
0
2
4
6 8 10 12 14 16 18 1 MeV Electron Fluence (x1015 cm-2)
20
22
Figure 10: Minority carrier diffusion length data determined from analysis of QE data as a function of particle fluence (Figure 9). The line represents a linear regression of the data from which the diffusion length damage coefficient, KL, can be determined according to Equation 1.
10-1 10-2 10
-3
10-4
InP/InP solar cell Irradiation with 3 MeV protons Fluence = 1x1013 cm-2
Current (A)
10-5 10-6 10-7 10-8 before irradiation fit to before irradiation data after irradiation fit to after irradiation data
10-9 10-10 10-11 10-12 0.0
0.2
0.4 Voltage (V)
0.6
0.8
Figure 11: Dark current data measured in the InP solar cell from the preceding figures. The proton irradiation causes an increase in the dark current. Analysis of these data shows that this increase is due an increased diffusion current brought on by radiation-induced recombination/generation centers.
The dark IV curve at lower voltages also shows an increase due to the irradiation. This is due to an increase in the recombination/generation current as described in Figure 7, which is the dark current produced by defects within the depleted region of the diode junction [11,12]. The magnitude of this current is directly proportional to the defect concentration and exponentially dependent on the difference between the trap energy level and the intrinsic Fermi level. The closer the trap level lies to mid-bandgap, the more efficient it will be as a
V - 11
recombination/generation center and the larger the recombination/generation current will be. This portion of the dark current is typically several orders of magnitude less than the diffusion dark current, so it has a proportionally smaller effect on the illuminated IV curve. However, at large enough defect concentration levels, the recombination/generation current can become significant, which results in a degradation of the FF. 3 3.1
Modeling Techniques Background
In the previous section, the basic mechanisms controlling the radiation response of crystalline, single-junction solar cells were described. In this section, the techniques employed to model solar cell performance in a space radiation environment will be reviewed. Solar cell degradation in space is caused primarily by incident protons and electrons either trapped in the earth’s radiation belts (the Van Allen belts) or ejected in solar events. These particles have energies that range from near zero up to several hundred MeV. In planning a space mission, engineers need a method of predicting the expected cell degradation in the space radiation environment. This is not a simple calculation because the rate of degradation for a given type of cell depends on the energies of the incident protons and electrons. In addition, the front surface of the cell is usually shielded by coverglass, and the back surface by the substrate material of the cell and the supporting array structure, so that the incident particle spectrum is ‘slowed down’ before it impinges on the active regions of the cell. Finally, as observed in Figure 8, different cell technologies have their own specific radiation response characteristics depending on the solar cell material, the thickness of the active regions, and the types and concentrations of dopants employed. In any method to predict solar cell response in a space radiation environment, several steps are necessary. First, a way is needed to correlate the degradation caused by particles of different energies, i.e. the energy dependence of the cell damage coefficients must be determined. Second, the radiation environment needs to be accurately specified, including the effects of any shielding materials present. Finally, a method must be found to convolve the energy dependence of the damage coefficients with the radiation environment for the duration of the mission in a way that allows comparison with ground test data. The discussion in this section will describe methods for correlating solar cell radiation damage. In the section to follow, it will be shown how these data are used to make on-orbit performance predictions for solar arrays. There are currently two main approaches being used to model solar cell degradation in space. The first method was developed at the U.S. Jet Propulsion Laboratory (JPL) and has been described in four extensive NASA publications [4,5,13,14]. The goal of this approach is the determination of the normal-incidence 1 MeV electron fluence, which produces the same level of damage to the cell as a specified space radiation environment. The JPL method has been used successfully over a number of years by many workers in the space community and serves as the present industry standard. JPL first published a comprehensive description of the equivalent fluence method in 1982 in the Solar Cell Radiation Handbook [4]. The model was applicable to several kinds of Si solar cells having average initial efficiencies of ~13% (AM0, 1 sun, 25oC). An appendix was included in the handbook containing a FORTRAN program entitled EQFLUX, which enabled predictions of Si cell degradation in a space environment to be made for the short V - 12
circuit current (Isc), open circuit voltage (Voc), and maximum power (Pmp). The user was required to enter an input file, which specified the incident omnidirectional radiation spectrum for the mission in question. The program incorporated experimentally determined relative damage coefficients (RDCs) for bare cells for normally incident electrons and protons. The calculations were updated in 1984 and 1989 to take into account newer types of Si cells as well as LPE GaAs cells, which were just coming into use [13,14]. The results for Si cells were deemed to be in substantial agreement with the earlier work, with only minor modifications being required. In 1996, JPL published the GaAs Solar Cell Radiation Handbook [5], which contained a similar analysis for the widely used GaAs/Ge solar cells (1990 vintage, grown by Applied Solar Energy Corporation). These cells had initial efficiencies of ~18% (AM0, 1 sun, 25oC). An appendix contained a FORTRAN code entitled EQGAFLUX, which was similar to EQFLUX but with results applicable to GaAs/Ge cells. EQFLUX and EQGAFLUX have been used widely by the space power community to make satellite end-of-life (EOL) power predictions. The second method was developed over the past decade at the U.S. Naval Research Laboratory (NRL) and has been described in a series of publications and conference proceedings [15,16,17,18,19,20,21]. The essence of the NRL method is the calculation of the displacement damage dose (Dd) for a given mission. The Dd is calculated from the nonionizing energy loss (NIEL) for protons and electrons traversing the cell material. This analysis draws on studies in which calculations of the NIEL were used to correlate displacement damage effects in microelectronic devices. The so-called ‘NIEL approach’ to correlating displacement damage effects is now widely used, especially in the particle detector field. For solar cell radiation response analysis, NRL and NASA have been working together [22] under NASA funding [23] to produce the Solar Array Verification and Analysis Tool (SAVNT), which is WindowsTM based program that implements the NRL method [24]. The NIEL is the rate of nonionizing energy loss of an irradiating particle in a material. As such, the NIEL indicates the rate at which energy from the incident particle is transferred to displacements in the target material and is expressed in terms of units of energy per unit of material (MeV/g). It is a calculated quantity involving interaction cross sections and recoil kinematics for different kinds of incident particles and target materials. It is the direct analog for displacement damage of the stopping power (or linear energy transfer, i.e., LET) used to describe ionization effects [16]. The NRL work showed that the radiation response of crystalline solar cells is determined only by the energy deposited into atomic displacements, just as the response of many biological and other systems to ionization depends only on the deposited ionizing dose. (The distinction of crystalline is used here since, as will be seen in a later section, ionization and not displacement damage may be the dominant degradation mechanism in non-crystalline devices.) The analogy was taken further with the establishment of a new quantity, i.e., the displacement damage dose (usually represented by Dd). Although the basic units of dose are energy per unit mass, ionizing dose is usually measured in rads(X), where X is the material in question. However, there is as yet no special unit for displacement damage dose and the usual unit employed is MeV/g. Displacement damage dose enables a characteristic degradation curve for crystalline solar cells to be determined, which is independent of the incident particle.
V - 13
Particle-induced degradation in GaAs/Ge cells will now be used to illustrate both the JPL and NRL methods, beginning first with the JPL approach. Note that the present discussion, for the most part, assumes the irradiating particle produces uniform damage as it passes through the solar cell, which is the case when the irradiating particle energy remains essentially constant as it passes through the solar cell active region. For electrons, this is true for all of the cases of interest here. The situation is much different for protons. The lower the proton energy, the greater the energy loss as the proton traverses the cell, so there is a lower energy limit below which care must be taken in applying the analyses to be presented. The value of this limit is dependent on the solar cell material and the thickness of the solar cell. A detailed discussion of the analysis of lower energy proton irradiation will be addressed in Section 7.2. 3.2 The JPL Method A flow chart describing the steps involved in the JPL method is shown in Figure 12. In the JPL method the RDCs for shielded cells are calculated from experimentally determined values for incident particles impinging perpendicularly on uncovered cells. It is, therefore, first necessary to measure degradation curves for all the typical photovoltaic parameters over a wide range of energies for both protons and electrons. In the case of the GaAs/Ge cells reported in the GaAs Solar Cell Radiation Handbook [5,25], these measurements included four electron energies (0.6, 1, 2.4, and 12 MeV) and eight proton energies (0.05, 0.1, 0.2, 0.3, 0.5, 1, 3, and 9.5 MeV). For each incident particle, eight fluence steps were typically required to generate a sufficiently detailed degradation curve and, to obtain good statistics for each fluence level, several cells needed to be irradiated and measured. It can be seen that many hundreds of current-voltage measurements were required to generate the necessary data. As an example, the results for Pmp are shown in Figure 13 [25]. It should be noted that the symbols shown in Figure 13 are not original data points as measured by Anspaugh. They are points read off the lines drawn on the figures shown in Reference [25].
JPL Equivalent Fluence Method Measure PV Degradation Curves (~4 electron and ~8 proton energies)
Determine Incident Particle Spectrum (e.g. AP8)
Determine Damage Coefficients for Uncovered Cells
1 MeV Electron Degradation Curve
Calculate Damage Coefficients for Isotropic Particles w/ Coverglasses of Varied Thickness
Calculate Equivalent 1 MeV Electron Fluence for Orbit (EQGAFLUX)
Read Off EOL Values
Figure 12: Flowchart describing the JPL equivalent fluence model for space solar cell end-of-life (EOL) prediction.
V - 14
The RDCs are determined from the measured data in the following way. For each incident particle type, the fluence at which a particular photovoltaic parameter is reduced to 75% of its initial value is determined from the respective degradation curve. These so-called ‘critical fluences’ are found for normally incident protons and electrons at all measured energies. The ratio of the critical fluence for 10 MeV protons to the critical fluence for other proton energies is then taken as a measure of the RDCs for protons. Similarly for the case of normally incident electrons, the critical fluences for different electron energies are normalized against the effect of 1 MeV electrons. The normalized values for all proton and electron energies are then plotted versus particle energy, giving the energy dependence of the RDCs (Figure 14). Using these RDC values, the proton fluence data from Figure 13 can be converted to an equivalent 10 MeV proton fluence, and the electron data can be converted to an equivalent 1 MeV electron fluence as shown in Figure 15. All of the radiation data are now seen to be correlated, as the electron and proton data have been reduced to a single curve for each dataset. To bring the electron and proton datasets into alignment, an empirically determined parameter is used. This is referred to as the ‘proton to electron damage equivalency ratio’, and is here given the symbol Dpe. This is an experimentally determined parameter that converts the 10 MeV proton fluence to an equivalent 1 MeV electron fluence. In the case of Si cells, Dpe has the same value for each of the photovoltaic parameters, i.e. ~3,000 [4]. However, for GaAs cells, Dpe has a different value for each parameter, and the result of the calculation is, therefore, different in each case (Isc = 400, Voc = 1400, Pmp = 1000 [25]). It should be noted that the degradation curves for different proton and electron energies are not exactly parallel (Figure 13), so that slightly different RDCs would be obtained if the critical fluences were taken when the degradation was different than 75%. However, this effect has a relatively small influence on the final results of the calculations.
Normalized Pmp
1.0
0.8
Proton Energy (MeV)
0.6
0.2 0.3 0.5 1 3 9.5
0.4
Electron Energy (MeV) 12 2.4 1 0.6
0.2 10
8
10
9
10
10
10
11
12
13
10 10 10 -2 Particle Fluence (cm )
14
10
15
10
16
10
17
Figure 13: Electron and proton Pmp degradation of GaAs/Ge solar cells as a function of fluence [5,25]. The points shown on the curves were read off the lines drawn on the figures shown in Reference [25].
V - 15
Electron Proton
1
10
0
RDCs
10
0.1
1
10
Energy (MeV)
Figure 14: Relative damage coefficients for normal incidence proton and electron irradiation of bare GaAs/Ge solar cells [25]. The data point at 50 MeV proton was taken from GaAs cells of an earlier vintage (1987) [5]. -2
Equivalent 1 MeV Electron Fluence (cm )
Normalized Pmp
1.0
0.8
Proton Energy (MeV)
0.2 0.3 0.5 1 3 9.5
0.6
0.4
Electron Energy (MeV) 12 2.4 1 0.6
0.2 10
9
10
10
11
12
13
14
10 10 10 10 10 -2 Equivalent 10 MeV Proton Fluence (cm )
15
10
16
Figure 15: GaAs Pmp degradation data from Figure 13 plotted as a function of equivalent fluence (1 MeV for electrons and 10 MeV for protons) determined using the RDCs of Figure 14.
With the RDCs so determined, the response of a given solar cell to irradiation by a spectrum of particles can be calculated. The particle spectrum must be expressed in differential form (dΦe(E)/dE and dΦp(E)/dE). The differential particle spectrum is then integrated with the appropriate damage coefficients (RDCe(E) and RDCp(E)) to determine an equivalent 1 MeV electron fluence for the spectrum in question. These integrals are shown in Equation 2 and Equation 3 for the cases of electrons and protons, respectively. The results of the integrals are
V - 16
the fluences of 1 MeV electrons that would cause the same damage as irradiation by the particle spectrum. The total equivalent 1 MeV electron fluence in a mixed proton and electron environment is obtained by adding the results. Once the total 1 MeV electron fluence equivalent to the particle spectrum is calculated, the predicted degradation can be obtained for each photovoltaic parameter from the experimentally determined 1 MeV electron degradation curves. In EQGAFLUX and EQFLUX, the 1 MeV electron degradation curves are fit to a power expression in natural logarithms out to fifth order. Equation 2
Φ 1MeV electrons, electrons = ∫
dΦ e ( E ) RDC e (E) dE dE
Equation 3
Φ 1MeV electrons, protons = D ep ∫
dΦ p (E) dE
RDC p (E) dE
The next step is apply this analysis to a solar array in a space environment which involves adapting the RDCs for normally incident particles on bare cells to the case of omnidirectional particles incident upon cells with coverglass and mounted on supporting solar array substrates. This will be performed in Section 4. 3.3 NRL Method A flow chart describing the steps involved in the NRL method for predicting radiation-induced solar cell degradation is shown in Figure 16. In this case, the first step is to calculate the NIEL for electrons and protons incident on the solar cell material of interest. The steps involved in the calculations are described in detail in References 16, 17, 18, 26, and 27, and only a relatively brief outline of the calculations will be given here. The calculations determine the energy transferred to the target atoms as a result of several possible interactions with incident protons or electrons. In order to perform the calculations, accurate values for the differential cross sections for Rutherford, nuclear elastic, and nuclear inelastic interactions are needed. For incident protons, Rutherford scattering dominates over the energy range extending from the threshold for atomic displacement (usually about several hundred eV) up to a few MeV. These differential scattering cross sections are quite accurately known from analytical calculations. The nuclear elastic component becomes important above about 1 MeV. Several models for nuclear elastic cross sections have been developed. The NRL group used experimentally determined values in the early calculations of the NIEL, but more recently an optical model approach has been employed. For energies >60 MeV, nuclear inelastic interactions become important. These have to be treated using one of several approximate models. At the moment, the nuclear inelastic interactions are probably the least satisfactory part of the overall calculation. Fortunately, for most cell types shielded with coverglass in typical space proton environments, only transmitted protons with energies up to ~10 MeV make significant contributions to the overall damage to the cell [16,17,18]. This is fortuitous since Rutherford scattering is generally the most accurate part of the overall calculation.
V - 17
NRL Displacement Damage Dose Method Determine Incident Particle Spectrum (e.g. AP8)
Calculate Nonionizing Energy Loss (NIEL) (Energy Dependence of Damage Coefficients)
Calculate Slowed-Down Spectrum (SDS) (Shielding) Measure Characteristic Degradation Curve vs. Dd (Dd=NIELxFluence) (2 e- and 1 p+ energy)
Calculate Dd for Mission (Integrate SDS with NIEL)
Read Off EOL Value
Figure 16: Flowchart describing the NRL displacement damage dose model for space solar cell damage prediction. Only ground-based data at three particle energies are needed.
Over the range of energies important for electron interactions in space, only Coulombic interactions are important and these can be calculated very accurately using the differential scattering cross section developed originally by Mott [28]. Nuclear contributions to the electron NIEL only become important for much higher energies than are typically important in space spectra. In order to calculate the energy dependence of the NIEL, the differential scattering cross sections are combined with the total recoil energy of the target atoms, which is partitioned into ionizing and nonionizing components by the so called Lindhard factor. The result is then integrated over solid angle. As an example, Figure 17 shows the calculated energy dependence of the electron and proton NIEL for GaAs over the energy range from zero up to 200 MeV. In these calculations, the threshold energies for atomic displacements for both Ga and As were taken as 10 eV [5]. It can be seen in Figure 17 that the electron NIEL increases with energy, whereas the proton NIEL generally decreases with increasing energy except close to the threshold. This is energy dependence can be seen qualitatively in the data shown in Figure 13.
V - 18
RDCs
10
3
10
2
10
1
10
0
10
-1
10
-2
10
-3
10
-4
10
-5
10
-6
10
-7
Proton
Electron
10
-5
10
-4
10
-3
10
-2
10
-1
10
0
10
1
10
2
10
3
10
4
Energy (MeV)
Figure 17: Proton and electron NIEL in GaAs.
It is a notable feature of Figure 13 that all the curves, except those for protons with energy <0.1 MeV, appear to have the same shape, with the curves for protons shifted to higher fluence and those for electrons shifted to lower fluence as the energy increases. This suggests that the degradation of the cell Pmp occurs as a result of the introduction of a similar type of damage to the cell, with a particular level of damage occurring at a different fluence depending on the incident particle and energy. The same behavior is also found for the other photovoltaic parameters, although these are not shown here. However, for very low energy protons, the damage is qualitatively different because the proton is slowing down as it passes through the cell active region hence its energy is decreasing. Therefore, damage produced by these particles must be treated separately as will be done in Section 7.2. Fortunately, as will be shown in Section 7.2, the contribution to the total damage from protons with energies <0.1 MeV is small in a typical space environment, so their behavior modifies only slightly the discussion which now follows. Since the NIEL is a calculation of the rate at which displacement damage is transferred to the lattice, the discussion above suggests that the curves in Figure 13 should collapse to a single curve if the particle fluence data are individually multiplied by the NIEL for the appropriate incident particle. This superposition, of course, will only occur if there is a linear dependence between the damage coefficient and the NIEL. The quantity that is produced by the product of particle fluence and the NIEL has the units of energy divided by mass, i.e. MeV/g. The conversion of fluence to displacement damage dose for protons, Dp, can therefore be achieved using Equation 4: Equation 4
D p = Φ p (E) NIEL(E) Figure 18 shows the results obtained by converting the proton fluence data of Figure 13 to Dd using (Equation 4). It can be seen in Figure 18 that where there is measurable degradation, the
V - 19
data collapse to a single, characteristic curve. This is a very important result that has both theoretical and experimental implications. First, it shows that there is a linear dependence of the proton damage coefficients on the calculated NIEL for protons on GaAs cells. Second, it shows that the degradation of Pmp (and other photovoltaic parameters), due to proton irradiation can be described by a single curve for all proton energies. Furthermore, this curve can be obtained in principle from the data for a single incident proton energy. This curve of Pmp against Dp is therefore a characteristic curve for proton degradation of GaAs cells in general.
Normalized Pmp
1.0
0.8 Proton Energy (MeV)
0.2 0.3 0.5 1 3 9.5
0.6
0.4
7
10
8
10
9
10 Dd (MeV/g)
10
10
11
10
Figure 18: ‘Collapse’ of the multi-proton energy fluence data from Figure 13 into a single characteristic curve. The proton data from Figure 13 are replotted as a function of displacement damage dose (Dd).
Although only GaAs solar cells have been considered here, there are many similar results for other cell technologies that show a linear dependence of proton damage coefficient on the NIEL. In fact, the only exception found to date is for crystalline Si cells, where measured damage coefficients deviate from the NIEL for reasons resulting from the thickness of the cells as will be discussed in Section 7.2. Considering the electron irradiation data, it has been found that, in contrast to the proton data, the solar cell damage coefficients do not always vary linearly with NIEL. Electron damage coefficients for n-type active regions appear to be linearly dependent on the NIEL, but electron damage coefficients for cells in general do not show a linear behavior. In fact, there is strong evidence that for most types of p-type active regions the relationship is quadratic. This behavior has been observed, for example, in crystalline InP, GaAs and Si cells [15,29,30,31]. Since in a typical cell technology the photocurrent is collected from both the emitter and the base regions, i.e., from both p- and n-type material, it is quite common to find a dependence somewhere between linear and quadratic. This is actually the case for the electron data shown in Figure 13, as will now be shown. The electron data from Figure 13 are plotted as a function of Dd in Figure 19, where the value of Dd was determined by multiplying the fluence by the appropriate NIEL according to Equation 4. The fact that these curves do not overlap indicates that there is a nonlinear dependence of the electron damage coefficients on the NIEL. Under these circumstances, it is useful to redefine the
V - 20
actual dose De(E) in terms of the effect of an equivalent dose for a particular representative electron energy. Given the importance of 1 MeV electron irradiation in the space solar cell community, it seems reasonable to use an effective 1 MeV electron equivalent dose, De,eff(1.0), as the standard. The relationship between the actual dose and the 1 MeV electron equivalent dose is given by Equation 5:
Normalized Pmp
1.0
0.8 Electron Energy (MeV)
12 MeV 2.4 MeV 1 MeV 0.6 MeV
0.6
0.4
0.2 10
7
10
8
9
10 10 Dd (MeV/g)
10
10
11
Figure 19: Electron Pmp degradation data from Figure 13 replotted as a function of Dd. The fact that the data do not collapse to a single curve as a function of Dd indicates a nonlinear dependence of the electron damage coefficients on NIEL.
Equation 5 ( n −1)
⎡ NIELe (E) ⎤ D e ,eff (1.0) = D e (E) ⎢ ⎥ ⎣ NIELe (1MeV) ⎦ where NIELe(E) is the NIEL for electrons of energy E and NIELe(1.0) is the NIEL for 1 MeV electrons. The value of n in Equation 5 that makes the measured degradation curves superpose gives the dependence of the electron damage coefficients on the NIEL. For the GaAs data shown in Figure 19, an exponent of n=1.7 gives a reasonable correlation, as can be seen in Figure 20.
V - 21
Normalized Pmp
1.0
0.8 Electron Energy (MeV) 0.6
12 MeV 2.4 MeV 1 MeV 0.6 MeV
0.4
assuming n = 1.7 in Equ. 5
0.2 7
10
8
9
10
10
10 10 Dde,eff (MeV/g)
11
10
Figure 20: Electron Pmp degradation data from Figure 13 replotted as a function of the effective 1 MeV electron Dd. A value of n=1.7 in Equation 5 is seen to cause the electron data to collapse to a single characteristic curve.
Although the definition of an equivalent displacement damage dose as defined in Equation 5 may seem somewhat arbitrary, it is actually quite well known in the case of the biological effectiveness of ionizing dose [16]. In this case, the ‘effectiveness’ of the same dose deposited by various particles is often found to be different and the particles are therefore assigned a certain ‘quality factor’. This is similar to defining an effective dose, as shown in Equation 5. From a practical viewpoint, the fact that electron damage coefficients often do not vary linearly with NIEL means that measurements made using at least two electron energies are needed to define the characteristic curve. This is in contrast to what was found above for protons (Equation 4), and Equation 5 allows the equivalent Dd to be calculated for a given monoenergetic particle fluence. The equivalent Dd for irradiation by a particle spectrum can also be determined. A discussed in the development of Equation 2 and Equation 3, the spectrum must be expressed in differential form, which is then integrated with the corresponding NIEL as shown in Equation 6 and Equation 7. These expressions are the analogue of Equation 2 and Equation 3 in the JPL method. Equation 6
Dp = ∫
dΦ p (E) dE
NIEL p (E )dE
Equation 7
dΦ e ( E ) 1 NIEL e (E) n dE ( n −1) ∫ dE NIEL e (1.0) Now that the electron and proton data have been reduced to single degradation curves, i.e. the single superposed curves shown in Figure 18 and Figure 20, the data can be parameterized. Referring to the JPL method, this is done within the EQFLUX (EQGAFLUX) program through a multi-parameter fit. In the NRL method, this is typically accomplished by a fit to the function D e ,eff =
V - 22
shown in Equation 8 where D is the dose (or effective dose), and C and Dx are the fitting parameters. This is the equation given in references 4 and 5 except here it is written in terms of Dd instead of particle fluence. While Equation 8 describes the solar cell data reasonably well, it is a semi-empirical expression, thus it has limited physical meaning. A discussion of alternative expressions drawn from reliability and survival theory can be found in [32]. Table 1 shows the calculated values of C and Dx for the Pmp curves shown in Figure 18 and Figure 20. For completeness, Table 1 also shows fitting parameters for the degradation curves for the Isc and the Voc for the same GaAs/Ge cells, using data taken from Ref. [25]. Note that the same value of C is used for both electron and proton irradiation because the electron and proton degradation curves have the same shape. The significance of the parameter Rep, which is the ratio Dex/Dpx, will be discussed below. Equation 8
D⎤ P ⎡ = 1 − C log ⎢1 + ⎥ Po ⎣ Dx ⎦ Table 1: Fitting parameters determined from fits of the GaAs data from Figure 18 and Figure 20 to Equation 8.
Isc Voc Pmp
C 0.31 0.099 0.322
Dex (x109 MeV/g) 8.7 2.8 4.7
Dpx (x109 MeV/g) 5.6 0.9 1.86
Rep 1.55 3.11 2.53
The solid and dotted lines in Figure 21 show the results of the fits for both proton and electron irradiation data, respectively, using the parameters shown in Table 1. As is the case in the JPL method, the reduced electron curve does not necessarily directly align with the reduced proton curve. Similar to the Dep parameter in Equation 3, the Rep factor is used to convert De,eff(1.0) values to equivalent Dp values. Rep is given by the ratio of Dex to Dpx. The electron curve can be made to coincide with the proton curves by dividing De,eff(1.0) values by Rep. The result is presented in Figure 22, which shows that, indeed, all the measured electron and proton data shown in Figure 13 can be represented by a single curve using the displacement damage dose approach.
V - 23
Normalized Pmp
1.0
0.8 Proton Energy (MeV)
Electron Energy (MeV) 12 2.4 1 0.6 meV fit to electron data
0.2 0.3 0.5 1 3 9.5 fit to proton data
0.6
0.4
0.2 10
7
8
9
10
10 10 10 10 Dd (MeV/g); Electron data - Dde,eff(MeV/g)
11
10
12
Figure 21: Data from Figure 13 replotted as a function of Dd for protons and 1 MeV electron effective Dd for electrons. The curves represent fits of the data to Equation 8.
Normalized Pmp
1.0
0.8 Proton Energy (MeV)
Electron Energy (MeV) 12 2.4 1 0.6 meV
0.2 0.3 0.5 1 3 9.5 fit to proton data
0.6
0.4
0.2 10
6
10
7
8
9
10
10 10 10 Dd (MeV/g); Electron data - Dde,eff(MeV/g)
10
11
Figure 22: Data from Figure 21 where the Rep factor has been applied to the electron data to bring them into alignment with the proton data. All of the data from Figure 13 can be therefore represented by a single curve.
Care must be taken in interpreting Figure 22, however, because of the manipulation of the electron data that was needed. The importance of Figure 22 is that it gives a way for handling the case where degradation occurs as a result of combined electron and proton irradiation. It is useful to describe this process using a specific example. The first step is to calculate the effect of the proton environment using Equation 6. Suppose the result of this calculation is 6.05x109
V - 24
MeV/g. The next step is to calculate the 1 MeV electron equivalent dose for the incident spectrum using Equation 7 recalling that a value of 1.7 for n has been found to fit the GaAs/Ge electron data. Suppose the result of this calculation is 1x1010 MeV/g. This value is then divided by Rep for Pmp (i.e., 2.53, from Table 1) to find the actual dose that would cause similar cell degradation. The result is 3.95x109 MeV/g. This dose is then added to the proton dose to find the total dose for the mission, i.e., 1x1010 MeV/g. The expected EOL cell performance for the mission is then read from Figure 22 to be 0.75. The procedure for calculating the total dose for a mixed proton and electron environment can be summarized by Equation 9. Equation 9
D total = D p +
D e ,eff (1.0 )
R ep The next step is to apply this analysis to a solar array in a space environment. For this, methods for accounting for the omnidirectional nature of the space spectrum and the shielding provided by the coverglass and solar array substrates must be employed. The methods employed by the NRL method will be described in Section 4. 3.4 Correlating the RDCs with NIEL While the preceding discussion presented two methods for modeling solar cell radiation response data, it is important to note that the two models are essentially the same since they operate from the same basic principles of solar cell displacement damage. It is only in the method of damage correlation that they differ. This can be seen explicitly when the RDCs generated by the JPL method are correlated in terms of NIEL. This is done in Figure 23 for the GaAs/Ge RDCs of Figure 14. Shown are the Pmp proton damage coefficients determined relative to 10 MeV along with the calculated NIEL normalized to 10 MeV for comparison. Also shown are the Pmp electron damage coefficients along with the calculated NIEL values, normalized to 1 MeV and raised to the 1.7 power according to the value of n determined for this technology. Except for the lowest proton energy, the RDCs are seen to correlate directly with the NIEL. In other words, the energy dependence of the RDCs is given by that of the NIEL. It is this fact that allows the characteristic curve to be determined from measurements at a single proton energy, since the cell response at any other proton energy can be calculated using the ratio of the appropriate NIEL values. The electron data in this graph demonstrate how at least two electron energies are needed to generate a characteristic curve for a given cell technology so that a value of n may be determined. However, once n is known, then the cell response to irradiation by electrons at any energy can be calculated from measurements at any other energy using the appropriate NIEL ratios. The results shown in Figure 23 demonstrate that the JPL and NRL method are based on the same physical principles of displacement damage. The solar cell degradation is caused by displacement damage effects; so the rate of solar cell degradation under a specific irradiation is given by the NIEL, and the variation of the damage rate with particle energy is given by the energy dependence of the NIEL. Therefore, the energy dependence of the RDCs is given by the NIEL, and moreover, the RDCs can be determined from the NIEL. Indeed, it is the very fact that the solar cell degradation can be correlated in terms of NIEL that the equivalent fluence concept in the JPL method is applicable. It follows directly, then, that when considered in terms of
V - 25
displacement damage dose, solar cell degradation measured under various particle irradiations will correlate directly.
Norm NIEL and RDCs
10
4
10
3
10
2
10
1
10
0
10
-1
10
-2
10
Electron NIEL (norm to 1 MeV, rasied to 1.7 power) Proton NIEL (norm to 10 MeV) Electron RDCs Proton RDCs
-2
10
-1
10
0
10
1
10
2
10
3
Energy (MeV)
Figure 23: GaAs/Ge solar cell electron and proton RDC values from Figure 14 plotted along with the calculated NIEL values. The electron NIEL has been normalized to 1 MeV and raised to the 1.7 power. The proton NIEL has been normalized to 10 MeV.
4 On-Orbit Solar Cell Performance Predictions In this section, the procedure for making predictions of the performance of a given solar cell performance in a space environment will be described. As in the preceding section, the methodology employed by the JPL and the NRL methods will be presented, compared, and contrasted. In general, this procedure consists of determining the radiation environment, accounting for any shielding that exists for the solar cells, and then actually calculating the solar cell performance. Each of these topics will be discussed in turn, and at the end, some examples taken from space experiments and orbiting space craft will be given.
4.1 Environment Calculations The prediction of the response of a solar cell to exposure to radiation must begin with a definition of the radiation environment. It is important to note that this is external to the choice of analysis technique, e.g. JPL vs. NRL methods. It is also important to note that analysis results can only be as accurate as the determination of the radiation environment. In this context, then, the JPL and NRL codes are equivalent. For a space application, the industry standards are the NASA-developed AP8 [33] proton and AE8 [34] electron environment models, and derivatives of them that take into account effects such as geomagnetic shielding. There are also more recent models that include the data from missions such as CRRES [35]. In the specific case of the JPL method, the user is required to input the particle spectrum in integral form and EQFLUX (or EQGAFLUX) converts the electron and proton spectra into differential form, which can be integrated with the RDCs (Equation 2 and Equation 3). For the NRL method, the SAVANT code includes the AP8 and AE8 programs, so the user need only enter the mission orbital
V - 26
parameters. In either case, the key point is that the calculations begin from the same point, i.e. calculation of the radiation environment by AP8 and AE8. 4.2 Shielding Calculations Once the incident particle spectra have been calculated, the next step is to account for the omnidirectional nature of the space spectrum and the shielding of the solar cells by the coverglass and array substrate. This is done differently in the two methods, so each will be treated separately. 4.2.1 JPL Shielding Calculations In the JPL method, the omnidirectional nature of the radiation environment and shielding effects are accounted for through modification of the RDCs. The first step is to calculate the corresponding RDCs for omnidirectional particles on bare cells from the measured RDCS for normally incident particles. For simple geometric reasons, if only particles incident on the top surface of the cells are considered, this step reduces the measured damage coefficients by a factor of two. The results are shown in Figure 24 and Figure 25. The effect of coverglass on the cell is taken into account by a series of calculations based on the ‘continuous slowing down’ approximation and the range-energy tables for protons and electrons in several different thicknesses of coverglass material. This calculation is relatively straightforward for electrons and protons that traverse the cell without significant change in energy. However, for protons that stop in the cell, the calculation must take into account that fact the relative damage increases rapidly at the end of the proton range. Consequently approximations have to be used to account for the variation in damage along the proton track. These approximations are discussed in detail in the NASA/JPL publications [4,5,13,14]. The net result of the calculations is the series of curves shown in Figure 24 and Figure 25.
RDCs
Measured RDCs
10
1
10
0
1 mil cover 6 mil cover 10
-1
20 mil cover RDCs corrected for omni-directional irrad.
0.01
0.1
1
60 mil cover
10
100
Energy (MeV)
Figure 24: Relative damage coefficients for proton irradiation of GaAs/Ge solar cells. The discrete points represent the measured data from [5,25]. The curve labeled “RDCs corrected for onni-directional irrad.” represent the adjustment of the RDCs to account for the omni-directional nature of the space radiation environment. The series of curves labeled with increasing thickness of covers represent the RDCs adjusted to account for the shielding due to that thickness of coverglass.
V - 27
RDCs
10
2
10
1
10
0
10
-1
10
-2
10
-3
10
-4
10
-5
10
-6
10
-7
Measured RDCs RDCs corrected for omni-directional irrad.
60 mil cover
1 mil cover 0.1
20 mil cover 6 mil cover 1
10 Energy (MeV)
Figure 25: Relative damage coefficients for electron irradiation of GaAs/Ge solar cells. The discrete points represent the measured data from [5,25]. The curve labeled “RDCs corrected for onni-directional irrad.” represent the adjustment of the RDCs to account for the omni-directional nature of the space radiation environment. The series of curves labeled with increasing thickness of covers represent the RDCs adjusted to account for the shielding due to that thickness of coverglass.
With the family of RDCs shown in in Figure 24 and Figure 25 determined, the equivalent 1 MeV electron fluence for a given mission can be calculated. This is done by integrating the differential spectrum with the appropriate set of RDCs for the given shielding thickness using Equation 2 and Equation 3. It should be noted that by addressing the shielding effects through modification of the RDCs means that the results are technology specific, and an entire new family of RDC curves must be generated for each new technology or new generation of an existing technology. Also, EQFLUX and EQGAFLUX only allow for specific, predefined coverglass thickness values. 4.2.2 NRL Shielding Calculations In the NRL method, the omni-directional spectrum and shielding are accounted for in a single calculation. The calculation modifies the incident spectrum and determines the “slowed-down” spectrum emerging from the shielding material and thus is incident directly upon the solar cell active region. For the case of backside shielding, the solar cell substrate is included in the shielding material. The slowed-down spectrum can then be inserted into Equation 6 or Equation 7 to determine the equivalent value of Dd. In contrast to the JPL method, addressing the shielding effects through a modification of the particle spectrum means that the calculations are technology independent. Also, SAVANT is configured to perform the calculations for any coverglass thickness, and the thickness is a user inputted parameter. The first step in the calculation is to determine the slowed-down differential spectrum in the active regions of the cell. This calculation involves modeling the transport of the isotropic, space spectrum through a slab of coverglass material. For the rear surface shielding, the array substrate material is considered as an equivalent thickness of coverglass determined using the material density. The approach used by the NRL researchers [17] and implemented in the SAVANT code V - 28
is based on the proton transport model described by Haffner [36] and Burrell [37]. It is assumed that the particle follows a straight-line path through a shield, with the energy at any point defined by an analytic expression for the range. A comprehensive compilation of experimentally determined values for the ranges of different energy protons in silica is tabulated in SRIM [38]. Haffner [36] used a simple power function for the range which fits the data well down to ~ 1 MeV. However, since lower energy protons are important for solar cell damage, the ranges are better represented by the sum of two such functions, i.e., Equation 10
R ( E ) = A E a + BE b where R is the range in g/cm2, E is the proton energy in MeV, and the other terms are constants. Equation 10 is found to fit the experimental data very well down to at least 200 eV (Figure 26). 105 104
Protons in SiO2
Range (µm)
103 102 101 100
SRIM data Fit (two-power law)
-1
10
10-2 10-3 10-4
10-3
10-2
10-1
100
101
102
103
Energy (MeV)
Figure 26: Proton range data for SiO2, a typical solar cell coverglass material. The data are seen to be modeled well by a two-power law fit (Equation 10).
An expression can be obtained from Equation 10 for the slowed-down energy of a proton after passing through a specified thickness of coverglass, which for this example will be assumed to be silica. In the NRL model, there is a one-to-one correspondence between the incident proton energy (E) and the energy emerging through the silica shield (e). Therefore, if the incident, isotropic differential spectrum is represented by g(E) and the spectrum emerging into the cell is given by f (e), it is possible to write Equation 11
f ( e) = g ( E )
V - 29
dE de
where dE/de can be obtained from Equation 10. Once the incident spectrum has been calculated, Equation 11 can be solved for f(e), treating the coverglass as a semi-infinite slab of material of thickness, t. As an example, a differential fluence spectrum from AP8 [33] was employed, corresponding to five years in a circular orbit with an altitude of 5000 km and an inclination of 60°. Although this is a relatively severe radiation environment, the results are qualitatively the same for other orbits and also for the effects of solar proton events. The calculation is somewhat involved since protons emerging through the coverglass with a particular energy will initially have possessed a range of energies depending on the angle of incidence. There is a lower, cutoff energy which is determined by those protons that travel perpendicularly to the coverglass. Equation 11 is therefore solved numerically within the SAVANT code by summing all protons in the incident distribution g(E) that contribute to each energy increment in the slowed-down distribution f(e), taking into account all possible angles of incidence. This can be visualized in Figure 27, and a representative calculation is shown in Figure 28. The figure shows schematically how protons with various incident energies contribute to a single “slowed-down” energy increment. The figure also shows example slowed-down spectra for various shield thicknesses. E5
~100 µm
E3
E1
E0
E1
E3
E5
glass
~3 µm
GaAs cell
~100 µm
substrate
E’
Figure 27: This is an illustration of the slowed-down spectrum calculation implemented within the SAVANT code for the NRL method. The slowed-down spectrum is calculated by summing all the protons in the incident spectrum that contribute to each energy increment in the slowed-down distribution at a given point within in the cell active region taking into account all possible angles of incidence.
V - 30
Differential Fluence (cm-2MeV-1)
1016
E1
Uncovered
1015
E3
1014 1013 10
12
1011
108 107 10-4
E4
SiO2 coverglass
E5
3 mil 12 mil 30 mil
1010 109
E2
E' 5000 km, Circular Orbit 60° Inclination 5 year mission 10-3
10-2
10-1
100
101
102
103
Proton Energy (MeV) Figure 28: These are results of an example calculation of the slowed-down spectrum incident upon a solar cell assuming a 5000 km, circular Earth orbit. The labels E1 through E6 refer to the incident proton energies depicted in Figure 27. Each curve corresponds to a different thickness of coverglass as indicated.
Figure 28 shows that shielding ‘‘hardens’’ the slowed-down spectra, i.e., the peaks in the distributions move to slightly higher energies with increasing shield thickness. Figure 28 also shows that the energy dependence of the calculated distributions below ~ 1 MeV is approximately independent of shield thickness. The reason for this is that in this energy region, the distributions follow the functional form of the reciprocal of the total stopping power, which is a result of employing the ‘‘continuous slowing down approximation’’ implied by Equation 10. The total displacement damage dose deposited in GaAs solar cells in the 5000 km orbit is obtained by integrating the product of the spectra shown in Figure 28 and the GaAs NIEL (Figure 17) over energy according to Equation 6 and Equation 7. Results are shown in Figure 29, which also includes the results of similar calculations for five year missions in orbits of 1000 and 2000 km, respectively. As expected, the thicker shields have the effect of significantly reducing the value of Dd deposited in a device. Figure 29 shows, for example, that increasing the thickness of the coverglass from 0.075 to 0.5 mm in the 5000 km orbit reduces Dd by approximately an order of magnitude. However, a reduction of two orders of magnitude can only be achieved by further increasing the thickness to 1.5 mm. Such a thick coverglass over an area of several square meters, typical of a solar array, would have the effect of increasing the weight of the spacecraft by tens of kilograms.
V - 31
1013
GaAs
5 Year Mission 1012
Dd (MeV/g)
1011 1010
5000 km
2000 km
109
1000 km 10
8
107 0.0
0.5
1.0
1.5
SiO2 Thickness (mm)
Figure 29: These are equivalent values of Dd calculated for three circular Earth orbits with the indicated orbital radius assuming a range of coverglass thicknesses.
4.3 Solar Cell Performance Predictions The calculations in the preceding sections reduced the radiation environment of a given space mission to a single number. For the JPL method, this is an equivalent fluence of 1 MeV electrons. For the NRL method, this is an equivalent Dd level. The solar cell performance prediction amounts, then, to simply reading the solar cell output off the appropriate graph, be it a plot of solar cell output as a function of 1 MeV electron fluence (Figure 8) or the characteristic degradation curve expressed in terms of Dd (Figure 22). In the EQFLUX and EQGAFLUX, this is implemented using the multi-term fit to the 1 MeV electron degradation curve. In the SAVANT code, Equation 8 is used. Examples of such predictions of the EOL normalized Pmp for GaAs/Ge cells for a 1 year mission in a circular, 60o inclination, 5093 km (2750 nm) orbit are shown in Figure 30. Since this orbit lies in the heart of the earth’s proton belt, the effects of electron irradiation were neglected. The proton spectrum for this orbit was taken from the commercially available computer code Space Radiation [39]. It can be seen that the JPL and NRL approaches give almost identical results over the whole range of coverglasses used. The data produced by the JPL method are shown as descrete points corresponding to the specific coverglass thicknesses included in EQFLUX and EQGAFLUX. The data from the NRL method, on the other hand, are shown as a continuous curve because SAVANT allows for any coverglass thickness, which can be considered an advantage of the NRL method.
V - 32
Normalized Maximum Power
1.0
GaAs/Ge 0.8
5093 km, circular, 60o orbit (1 year duration)
0.6
0.4 Displacement Damage Dose (NRL) Model Equivalent Fluence (JPL) Model
0.2
0.0
0
10
20
30
40
50
60
SiO2 Coverglass Thickness (mils)
Figure 30: These are predicted solar cell EOL performance values for the specified orbit and the range of coverglass thicknesses shown. Calcualtions produced by both the NRL and JPL methods are shown, and the data are seen to agree well.
4.4 Mission Examples As a conclusion to the discussion of on-orbit solar cell performance predictions, some actual space mission examples will be presented. In these examples, predictions made using the SAVANT code will be shown. The Photovoltaic Array Space Power Plus Diagnostics (PASP Plus) space experiment produced data on several solar cell technologies [40,41]. This experiment was launched on August 3, 1994 and continued to operate until August 11, 1995, giving over one year of data. The satellite was in an elliptical orbit, with an apogee of 2442 km, a perigee of 362 km, and a 70o inclination. Data from SJ GaAS/Ge cells with 6 mil SiO2 coverglass that were body mounted (so infinite back shielding can be assumed) have been analyzed using SAVANT [19]. The results are shown in Figure 31. The calculated values agree well with the measured data. The curvature in the data is caused by a varied rate of Dd accumulation during the mission due to the orbit precession. This is accounted for in the SAVANT code by allowing the orbit to precess and calculating the accumulated Dd on a daily basis.
V - 33
1.01
2,552 km x 363 km; 70o orbit
Normalized Maximum Power
1.00
(Data from 8.03.94 to 8.11.95)
0.99 0.98
GaAs/Ge
0.97 0.96 0.95 0.94
SAVANT Calculation
0.93 0.92 0.91 0.90
0
100
200
300
400
Days in Orbit
Figure 31: These are solar cell data from GaAs/Ge cells in orbit onboard the PASP Plus experiment. The solid curve is a prediction produced by the SAVANT program.
A space solar cell experiment launched by the European Space Agency was part of the Equator-S mission [79]. This mission was designed to study the Earth’s magnetosphere out to geosynchronous distances. The spacecraft was launched on 2 December 1997 into a 500 km (apogee) x 67,300 km (perigee) elliptical, near-equatorial orbit. The mission duration was intended to be two years, but data from the spacecraft ceased after 2 May 1998 due to the failure of the redundant processor. Data from several solar cell technologies were recorded during the operating period. Two of these have been studied [22]: GaAs/Ge from EEV (a company in the UK) with a 4 mil CMG coverglass and CIS from Siemens with a 6 mils CMX coverglass. (CIS is CuInSe2, a thin film solar cell technology that will be discussed in Section 5.3.2). Figure 32 shows the on-board measurements compared with the predictions of SAVANT. In the calculations, back shielding was assumed to be infinite due to the body-mounted nature of the solar panels. The SAVANT calculations agree well with the space data to within a few percent throughout the mission. The CIS results are extremely satisfying given the quite limited amount of ground data available. Some uncertainty in the calculations arises from uncertainty in the beginning-of-life (BOL) data, due to the fact that the spacecraft did not start recording data until 17 days after the launch.
V - 34
1.05
Normalized Maximum Power
500 km x 67,300 km; near equatorial orbit (Data from 12.2.97 to 5.1.98) 1.00
CuInSe2
0.95
SAVANT Calculation 0.90
GaAs/Ge 0.85 0
20
40
60
80
100
120
140
Days in Orbit
Figure 32: Comparison between the SAVANT analysis and the on-board solar cell measurements for GaAs and CIS solar cells onboard the Equator-S mission [22].
Another opportunity to study on-orbit solar cell data came from a failure of the Japanese H-11 rocket. The Communication and Broadcasting Engineering Test Satellite (COMETS) was meant to achieve a geostationary orbit, but the rocket failure place it in an elliptical orbit [42]. As the mission was intended to exist in an orbit where only electrons and solar proton events were expected to dominate, the GaAs/Ge-powered solar panels, which were flexible arrays, received much more radiation damage than intended, as only 4 mil CMG coverglasses were used. Several attempts were made to correct the orbit, after which the satellite settled into a 472 km x 17,700 km, 30o orbit. Data from the GaAs/Ge solar power panels from 21 February 1998 (launch date) to 21 August 1999 were supplied by the National Space Development Agency of Japan (NASDA). Ground measurements on these particular cells were also supplied by NASDA from which the characteristic curve was determined for use in SAVANT [22]. In this case, the data to be analyzed are from the deployed power array, so backside irradiation had to be considered as well as that from the front. The solar array consisted of the solar cells being mounted on 2 mil kapton sheets by 6 mils of adhesive. The back surface shielding, therefore, consisted of the 5.6 mil GaAs substrate, 6 mil adhesive layer, and 2 mil kapton sheet, so an equivalent GaAs shielding thickness of 8.2 mils was assumed. The final orbit of 472 km x 17,700 km was not attained until the 97th day of the mission. A total of 7 corrections were made during these first 97 days to correct the orbit. As noted with the PASP Plus analysis, the SAVANT code is designed to easily account for these changes in orbit with time. Figure 33 shows the results comparing the SAVANT calculation and the on-orbit measurements for the COMETS mission. SAVANT calculations are seen to agree with the on-board data to within a few percent for the first 350 days in orbit. Considering the complexity of the orbit, the ease with which the SAVANT model handled the multiple orbital changes, and the severity of the solar cell degradation (~ 35%), this is quite good agreement. In particular, the structure of the measured data and calculated curve agree well. This structure arises from the non-uniform accumulation of Dd with time, which the SAVANT code accurately models.
V - 35
Normalized Maximum Power
1.05 1.00
472 km x 17,700 km; 30o orbit
0.95
(Data from 2.21.98 to 8.21.99)
0.90 0.85
On-board power data SAVANT Calculation
0.80 0.75 0.70 0.65
GaAs/Ge
0.60 0.55 0.50 0
50
100
150
200
250
300
350
400
450
500
550
Days in Orbit
Figure 33: Comparison between the SAVANT analysis and the on-board solar cell measurements for GaAs solar cells onboard the COMETS mission [22].
The final example to be shown comes from another GaAs/Ge solar array. The solar array under study is dedicated to providing power to the Microelectronic and Photonic Testbed (MPTB) experiment [43]. The array operates at a fixed voltage of 31.2 V, but the current is not controlled. Therefore, any changes in the photocurrent of the array can be measured and analyzed. The array consists of 27 strings of cells with 54 cells in each string. The solar cell technology is Spectrolab (SPL) 8 mil single-junction (SJ) GaAs/Ge. The solar array consists of a ¾ inch think honeycomb Al substrate with 10 mil graphite epoxy face-sheets. A schematic diagram of the solar array structure is shown in Figure 34. This is a good example of the process involved in determining the backside shielding due to the solar array substrate. The analysis of the individual layers of the solar array is shown in Table 2, and it is seen that the substrate is equivalent to just over 31 mils of coverglass. Based on the published report of Dyer, et al. [44], an orbit consisting of 39,158 x 1,212 km 63.43o inclination for 225 days, and then at 38,868 x 1,502 km 63.43o inclination for the remainder of the mission was assumed. The telemetered data along with the SAVANT calculations are shown in Figure 35. The SAVANT calculations predict less degradation than observed. Reasons for this are currently under investigation [43], but a likely candidate is the environmental model used, i.e. AP8. On board dosimeter data [44] suggest that the proton environment is larger by a factor of 2 than predicted by AP8 for the MPTB mission. Multiplying the SAVANT predicted Dd values by 2 does bring the predicted data into reasonable alignment with the telemetered data (Figure 35), but more research is required to truly validate this result. Nevertheless, this example illustrates the difficulties that can be encountered in solar array performance predictions and the need for accurate and reliable environment prediction tools.
V - 36
OCI Coating 3 Mil Coverglass 20 Mil Adhesive 2 Mil GaAs Solar Cell 8 Mil Interconnect/Cell Adhesive 4 Mil FM 73 Dielectric 10 Mil Graphic Epoxy Face Sheet 10 Mil FM73 Epoxy/Scrim Cloth 2 Mil
AL Honeycomb 750 Mil
FM 73 Epoxy 2 Mil Graphic Epoxy Face Sheet 10 Mil White Paint 5 Mil
Figure 34: This is a schematic diagram of the MPTB solar array. To determine the solar cell shielding, within the SAVANT code, each layer is accounted for and expressed as an equivalent thickness of coverglass material. Table 2: Table of thickness and density of the layers comprising the MPTB solar array. The values are reduced to an equivalent thickness of coverglass material at the bottom of the table. These data are used to determine the radiation shielding of the MPTB solar cells on orbit.
Total Equivalent thickness Density of Coverglass (g/cm3)
Equiv. thickness (mil g/cm3) 2.56 11.6 16.5 2.32 24 2.32 16.5 4 ========== 79.8 2.554
Total/CMG = equiv coverglass (mils)
31.25
Adhesive/Interconnect FM 73 dielectric Graphite Epoxy Face Sheet FM 73 dielectric Al honeycomb FM 73 dielectric Graphite Epoxy Face Sheet White Paint
Thickness (mils) 4 10 10 2 750 2 10 5
V - 37
Density (g/cm3) 0.64 1.16 1.65 1.16 0.032 1.16 1.65 0.8
1 0.98 SAVANT calculation
0.96
Telemetered data
Normlaized Imp
0.94 0.92 0.9 0.88 0.86 0.84 0.82
SAVANT calculation assuimng twice the calculated Data
0.8 11/22/97 6/10/98 12/27/98 7/15/99
1/31/00 8/18/00 Date
3/6/01
9/22/01
4/10/02 10/27/02
Figure 35: These are the telemetered MPTB data compared to predictions produced by SAVANT. The SAVANT calculation predicts less degradation than observed. The onboard dosimeters suggest that this may be due to under prediction of the proton environment by the AP8 code.
5 Specific Solar Cell Technologies The preceding discussion described the basic radiation response mechanisms for a crystalline semiconductor, single-junction (SJ) solar cell. This section investigates specific solar cell technologies that are of interest for space applications. In each case, the available radiation data will be presented and the operative damage mechanisms will be studied.
5.1 Single-junction, Crystalline Semiconductor Solar Cells While the radiation response of single junction crystalline semiconductor solar cell technologies was covered in detail in the preceding section, the subject is included here briefly for completeness. The crystalline Si solar cell is of major historical significance as it was the first solar cell used in space, and it powered nearly all spacecraft until the early 1990’s. At that time, single-junction GaAs solar cells grown on Ge substrates (GaAs/Ge) came to maturity with the first commercial communications satellite powered by GaAs/Ge cells being launched in January 1996 [45]. One other technology of note is the InP technology. InP was developed as a potential solar cell because of its superior radiation resistance in comparison to Si and GaAs [6,46,47]. InP wafers, however, are more expensive, heavier, and more fragile than Si and Ge wafers, so InP solar cells grown heteroepitaxially on Si substrates were developed [6]. A comparison of the radiation response of these three technologies has been shown in Figure 8. 5.2
Multijunction Solar Cells
5.2.1 Mechanisms for Multijunction Solar Cell Radiation Response Presently, the highest efficiency solar cell technology for space power is a more complex technology based on multiple n/p junctions deposited epitaxially in a single device. The present state-of-the-art of multijunction (MJ) solar cells is the triple-junction (3J) InGaP2/GaAs/Ge technology. A schematic diagram of a typical cell is shown in Figure 36. The advantage offered by MJ solar cells is more efficient absorption of a wider wavelength range of solar photons. This V - 38
can be seen in QE measurements made on one of these devices. In Figure 37, quantum efficiency data from a 3J InGaP2/GaAs/Ge solar cell are shown, superimposed with the air mass zero (AM0) solar spectrum, normalized to the maximum irradiance value. Comparing these data with that of a SJ cell (Figure 9), it can seen how the 3J device has an extended wavelength response range that matches well with the AM0 spectrum. Contact AR n+-GaAs n-AlInP n-GaInP
Top Cell
p-GaInP p-AlGaInP GaAs n-GaInP
n-GaAs Middle Cell p-GaAs
p-GaInP p++-GaAs n++-GaAs
n-GaAs Bottom Cell n-Ge p-Ge substrate Contact
Figure 36: This is a schematic diagram of a three junction InGaP2/GaAs/Ge solar cell. Figure courtesy of Spectrolab, Inc.
The increased wavelength response of a triple junction device results in increased efficiency. Since the triple junction solar cell is in a series connected configuration, the voltages generated by each sub-junction add so that the Voc of a MJ cell is typically much larger than that of a SJ cell. However, the series configuration limits the MJ overall solar cell Isc to the smallest Isc of the sub-junctions. In other words, one of the sub-junctions limits the current of the overall MJ device. In the InGaP2/GaAs/Ge device under AM0 illumination, the Ge sub-junction typically produces about a factor of 2 more photocurrent than the upper two junctions, so it is usually not the current limiter. Depending on the thickness of the InGaP2 material, the top and middle subjunctions can be close to current-matched, which is the condition for maximum conversion efficiency. This is a physical property that is exploited in the design of the 3J cells for maximum efficiency and, moreover, improved radiation resistance as will be discussed below.
V - 39
1.0
External Quantum Efficiency
0.8 AM0 Spectrum(norm.) InGaP2 top junction
0.6
GaAs middle Junction Ge bottom junction open symbols are after 1 MeV electorn irrad., 15 -2 1x10 cm
0.4
0.2
0.0 400
600
800
1000
1200
1400
1600
1800
Wavelength (nm)
Figure 37: These are quantum efficiency measurements made on a triplejunction InGaP2/GaAs/Ge solar cell. Also shown is the AM0 solar spectrum, normalized to the maximum irradiance value. Each sub-junction has a specific wavelength response range. By combining these three response curves into a single device, more of the incident solar light can be efficiently converted into electricity. The open symbols represent measurements made after 1 MeV electron irradiation to a fluence of 1x1015 cm-2.
Efforts to develop higher powered devices include the development of an appropriate 1-eV bandgap material for a forth junction as described in [48]. Other efforts are focused on applying the concept of bandgap engineering to achieve a bandgap combination that is better tuned to the AM0 spectrum (see Figure 37). Leveraging on the success of the InGaP2/GaAs/Ge technology, the InxGa1-xP/InyGa1-yAs/Ge system is being developed at several laboratories [22,49]. This material system is showing very good results as ongoing improvements in lattice mismatched and strained layer growth techniques are significantly relaxing the lattice matching requirement [50]. QE data measured in InxGa1-xP/InyGa1-yAs/Ge two-junction devices at two stoichiometries are shown in Figure 38. Increasing the In content within each subcell, i.e. increasing x and y, decreases the bandgap thus extending the absorption range. This results in increased photoabsorption and, hence, photocurrent, but it also results in reduced voltages. It is balancing these competing effects that yields the stoichiometry that is optimized for a given input photon spectrum. In addition, the effect of an increased In concentration on the cell radiation response must be considered. This aspect will now be investigated.
V - 40
1.0
External Quantum Efficiency
0.8
0.6
0.4 In0.65Ga0.35P/In0.17Ga0.83As In0.49Ga0.51P/In0.03Ga0.97As open symbols are after irradiation by 4.25 MeV 12 -2 protons, 1x10 cm
0.2
0.0 400
600
800
1000
1200
Wavelength (nm)
Figure 38:. Example QE data for dual-junction InxGa1-xP/InyGa1-yAs solar cells at two stoichiometries.
When a MJ solar cell is irradiated, the basic degradation mechanisms are similar to those described for a SJ device. However, the response of a MJ cell is different than that of a SJ cell because of the mutual interaction of the sub-junctions. Typically in a MJ cell, one sub-junction will be more radiation sensitive than the others such that the response of that sub-junction will control the overall response of the MJ cell. For the InGaP2/GaAs/Ge 3J device, the GaAs subjunction is the most radiation sensitive. This can be demonstrated by analyzing the degradation of Voc in a SJ and MJ solar cell. Such an analysis is shown in Figure 39 where the measured Voc of a SJ and 3J device are plotted as a function of irradiation fluence. In this figure, the data have been normalized to the pre-irradiation value for comparison, and the 3J Voc response is seen to be nearly the same as that of the SJ cell.
V - 41
1.0 1.0 0.9
Normalized Voc
0.9 0.8 0.8 0.7
SJ GaAs/Ge 3J InGaP2/GaAs/Ge
0.7 0.6 0.6 0.5 9
10
10
10
11
10
12
-2
10
13
10
1 MeV Proton Fluence (cm )
Figure 39: These are Voc data measured in a SJ GaAs/Ge solar cell compared with that of a 3J InGaP2/GaAs/Ge cell. The data have been normalized to the pre-irradiation value for comparison. The 3J cell response is essentially the same as that of the SJ device indicating that the GaAs sub-junction controls the 3J cell radiation response.
The radiation response of the photocurrent will also be dominated by the GaAs sub-junction as evident in the radiation response of the spectral response of the 3J device (Figure 37). The degradation of the GaAs sub-junction response is far worse than that of the other two subjunctions. The sensitivity of the photocurrent response can be controlled, however, by altering the device structure. Thinning the InGaP2 top layer will result in reducing the current from that sub-junction and increasing the current from the GaAs sub-junction. If the InGaP2 layer is made thin enough that it limits the 3J cell photocurrent, then the radiation response of the 3J cell will track that of the more radiation resistant InGaP2 sub-junction (Figure 40, the SJ GaAs/Ge data are from [5] and the 3J InGaP/GaAs/Ge data are from [51]). This condition will persist under irradiation until the GaAs sub-junction photocurrent is degraded to a level below that of the thinned InGaP2 sub-junction current. At this radiation level, the response will begin to track the degradation of the GaAs sub-junction. The alteration of the sub-junction thicknesses is referred to as current-matching. The 3J cell will have its maximum efficiency when the sub-junctions are current matched. Thinning the InGaP2 sub-junction to improve the device radiation response represents current-matching for the conditions at the end of a specific mission, and it can be seen how a 3J cell structure can be optimized for a specific operating environment. It should be noted that current matching for the end of a mission sacrifices some of the initial efficiency.
V - 42
1.1
1.0
Normalized Isc
0.9
0.8
SJ GaAs/Ge BOL 3J InGaP2/GaAs/Ge
0.7
EOL 3J InGaP2/GaAs/Ge 0.6
0.5 10
10
11
10
12
10
13
10
-2
1 MeV Proton Fluence (cm )
Figure 40: These are Isc data from a SJ GaAs [5] and two different 3J InGaP2/GaAs/Ge solar cells [51] normalized to their pre-irradiation values. The 3J InGaP2 cell labeled BOL (beginning-of-life) was engineered to be currentmatched before irradiation. The 3J InGaP2 cell labeled EOL (end-of-life) was designed to be current-matched after irradiation. This is evident in the superior radiation resistance of the Isc measured in the EOL 3J device as the fluence increases. At the highest fluences, the 3J Isc data degrade more rapidly indicating that the 3J Isc has transitioned from being limited by the InGaP2 subjunction Isc to that of the GaAs sub-junction.
Since the GaAs sub-cell has been shown to be the most radiation sensitive within the InGaP2/GaAs/Ge stack, significant research is being dedicated to understanding the radiation response of single-junction InyGa1-yAs devices for application in the InxGa1-xP/InyGa1-yAs/Ge system. [52,53,54]. The radiation response of several configurations of this cell type is shown in Figure 41 (the p/n GaAs data are from [5] and the InGaAs data are from [53]). For Dd levels up to ~ 1010 MeV/g, the Pmp degradation for all of the InyGa1-yAs cells is nearly equivalent to that of a conventional GaAs cell (i.e. with y=0). This is quite a high exposure level, being roughly equivalent to a one-year mission in the heart of the Earth's proton belts. This indicates the radiation resistance of the InyGa1-yAs cells to be relatively insensitive to changes in the value of y. Some variation in cell response amongst the different technologies is observed at higher Dd levels. In particular, the 22% In cells showed a rapid decrease in voltage for Dd > 5x109 MeV/g (Figure 41). Measurements showed the Voc degradation to be due to a more rapid radiationinduced increase in dark current in the 22% In cells. This may be a direct result of the higher In content in those cells and, hence, larger lattice mismatch. However, the Isc response suggests that differences in cell structure also significantly impact the radiation response. The 17% In cells, which were of a p-i-n and n-i-p structure, displayed a much better blue response before irradiation, which was nearly insensitive to the irradiation, and after irradiation, those cells showed a better spectral response at nearly all wavelengths (Figure 42 [53]). This can be
V - 43
explained by the enhanced collection efficiency afforded by the intrinsic layer of these cells and to a better front and rear interface passivation scheme. From these results, it can be concluded that, within the range of In concentrations studied, the response of these cells are more strongly controlled by the cell structure than the In concentration. 1.0
Normalized Pmp
0.9
0.8
n/p In0.22Ga0.78As (a) p-i-n In0.17Ga0.83As (b) n-i-p In0.17Ga0.83As (c)
0.7
(d)
p/n GaAs (d)
(c) (b) (a)
0.6
0.5 8
9
10
10
10
10
Dd (MeV/g)
Figure 41:. Radiation-response of single-junction InyGa1-yAs solar cells with different stoichiometries. The p/n GaAs data are from [5] and the InGaAs data are from [53]). 1.0
External Quantum Efficiency
0.8
0.6
0.4
n/p In0.22Ga0.88As n-i-p In0.17Ga0.83As open symbols are after 10 irradiation Dd ~ 4x10 MeV/g
0.2
0.0 400
600
800
1000
1200
Wavelength (nm)
Figure 42: Radiation-response of the QE of two single-junction InyGa1-yAs solar cells [53].
The radiation response of several dual junction InxGa1-xP/InyGa1-yAs devices are shown in Figure 43 [54]. The n/p InGaP2/GaAs device is an EOL optimized cell developed under the ManTech program [51]. Except for the In0.51Ga0.49P/In0.03Ga0.97As cell in the n/p structure, the cells show generally similar radiation characteristics, independent of the In concentration. The n/p V - 44
In0.51Ga0.49P/In0.03Ga0.97As cell was optimized for terrestrial use under AM1.5 illumination, so the middle cell base thickness and dopant level were larger than optimal for good radiation resistance. These results are similar to those of the single junction GayIn1-yAs cells, and again suggest that the cell structure may have considerably more affect on the radiation response than the In concentration 1.0
1.0
(a)
(a) 0.9
0.9
Normalized Isc
p-i-n In0.17Ga0.83As (b)
0.8
(d)
n-i-p In0.17Ga0.83As (c)
(c)
p/n GaAs (d) 0.7
Normalized Voc
(b)
n/p In0.22Ga0.78As (a)
(b) (c) 0.8
(d)
n/p In0.22Ga0.78As (a) p-i-n In0.17Ga0.83As (b) n-i-p In0.17Ga0.83As (c)
0.7
p/n GaAs (d)
0.6
0.6
0.5
0.5 10
8
10
9
10
8
10
10
10
9
10
10
Dd (MeV/g)
Dd (MeV/g)
(a)
(b)
1.0
(d) Normalized Pmp
(b) (a) 0.8
(c) n-i-p In0.65Ga0.35P/In0.17Ga0.83As (a) n-i-p In0.51Ga0.49P/In0.03Ga0.97As (b) n/p In0.51Ga0.49P/In0.03Ga0.97As (c)
0.6
n/p InGaP2/GaAs (d) 7
10
8
10
10
9
10
10
11
10
Dd (MeV/g)
(c) Figure 43: Comparison of the radiation-response of several dual junction InxGa1-xP/InyGa1-yAs solar cells [54] (The n/p InGaP2/GaAs data are from [51].)
The advantages of the EOL optimized InxGa1-xP/InyGa1-yAs 2J cell can be seen through a study of the Isc response of the cells (Figure 44). At low Dd levels, the Isc of the EOL optimized n/p InGaP2/GaAs cell is limited by the top cell, and as such, degrades little since the top cell is quite resistant to irradiation. As the bottom cell degrades at higher Dd levels, the dual-junction device transitions to being bottom cell limited. At this point, the dual-junction Isc degradation curve turns over and rapidly degrades down to the level of the other cells. These results clearly demonstrate that, like the InGaP2/GaAs technology, the GayIn1-yAs sub-cell cell primarily controls the radiation response of the MJ GaxIn1-xP/GayIn1-yAs devices. In Figure 45, the normalized degradation of the Isc of the GaxIn1-xP top and GayIn1-yAs bottom cells are
V - 45
shown independently [54]. These data were calculated by integrating the spectral response of each sub-cell over the energy dependence of the AM0 spectrum. The degradation of all the GaxIn1-xP cells can be seen to be small up to high damage levels, independent of the stoichiometry. The GayIn1-yAs cells, on the other hand, degrade significantly. Also, a significant difference is observed in the degradation of the different GayIn1-yAs bottom cells. Since two cell structures, each with y = 0.03, show significantly different behavior, this difference cannot be attributed only to the cell stoichiometry. Again, the cell response appears to be more strongly controlled by the cell structure. The n-i-p Ga0.97In0.03As sub-cell benefits from the increased carrier collection efficiency afforded by the extended electric field of the intrinsic region. Also, in contrast to the n-i-p Ga0.97In0.03As, which was designed for AM0 operation, the n/p Ga0.97In0.03As cell was designed from AM1.5 operation, so the base dopant level was relatively high in that cell (~ 2x1017 cm-3), which results in a lower radiation resistance.
1.0
Normalized Isc
(d)
(a)
0.8
(b) (c) n-i-p In0.65Ga0.35P/In0.17Ga0.83As (a) n-i-p In0.51Ga0.49P/In0.03Ga0.97As (b) n/p In0.51Ga0.49P/In0.03Ga0.97As (c)
0.6
n/p InGaP2/GaAs (d) 7
10
10
8
10
9
10
10
10
11
Dd (MeV/g)
Figure 44: Comparison of Isc radiation-response in InxGa1-xP/InyGa1-yAs solar cells [54] (The n/p InGaP2/GaAs data are from [51].)
V - 46
Calculated Photocurrent (norm.)
1.0 0.9 0.8 0.7 0.6
(Fraunhofer ISE) In0.65Ga0.35P top cell In0.17Ga0.83As bottom cell In0.49Ga0.51P top cell In0.03Ga0.97As bottom cell
0.5 (J-Energy Corp.) 0.4 0.3 108
In0.49Ga0.51P top cell In0.03Ga0.97As bottom cell
109
1010
1011
1012
Displacement Damage Dose (MeV/g)
Figure 45: Comparison of the radiation-response of photocurrent of the top and bottom cells of the InxGa1-xP/InyGa1-yAs solar cells of Figure 44 [54]. The legend indicates the group that produced the solar cells. The data were calculated from the QE data of these DJ solar cells.
This analysis of the radiation response of MJ GaxIn1-xP/GayIn1-yAs devices suggests that, contrary to initial speculation, the radiation-response of the GayIn1-yAs-based devices is quite good and essentially independent of In content. Instead, it is the cell structure that more significantly controls the radiation-response, and it has been shown how the cell structure may be optimized for maximum BOL and EOL performance that is equal to or better than conventional Ga0.49In0.51P/GaAs cells. 5.2.2 Modeling Multijunction Solar Cell Radiation Response In Section 3, methods for modeling the radiation response of a SJ GaAs/Ge solar cell were presented. This model will now be extended to include the MJ GaxIn1-xP/GayIn1-yAs/Ge devices. It has already been demonstrated that proton irradiation data at various energies measured in the GaxIn1-xP/GayIn1-yAs devices discussed in the preceding section can be correlated in terms of Dd [22,52,53,54]. Indeed, most of the data presented in the preceding section were expressed in terms of Dd. Therefore, both the JPL and the NRL modeling techniques will apply to those devices, by definition. Nevertheless, it is instructive to demonstrate the data correlation explicitly. This can be done in a simple fashion by adding RDC data for the MJ cells to Figure 23. This is done in Figure 46 where the RDC data generated by Marvin for the MJ cells developed in ManTech program [51] have been used. The SJ data are those of Figure 23, which are taken from [5].
V - 47
Norm NIEL and RDCs
10
2
10
1
10
0
10
Proton NIEL SJ RDC 2J RDC 3J RDC Electron NIEL (1.7 power) SJ RDC 2J RDC 3J RDC
-1
10
-2
10
-1
10
0
10
1
10
2
10
3
Energy (MeV)
Figure 46: RDCs for SJ GaAs/Ge [5], 2J InGaP2/GaAs, and 3J InGaP2/GaAs/Ge [51] solar cells plotted along with the GaAs NIEL. To be consistent with the RDCs, the proton NIEL has been normalized to the value at 10 MeV and the electron data to 1 MeV. The electron NIEL was raised to the 1.7 power according to the n value determined for the SJ cells. The correlation of the RDCs with NIEL indicates that the Dd analysis methodology is applicable to the MJ cell technologies.
The fact that the RDCs correlate with the NIEL in Figure 46 indicates that the Dd methodology is applicable to these MJ devices. It should be noted that in Figure 46, the electron NIEL has been raised to the 1.7 power in accordance with the n value determined for the SJ GaAs/Ge dataset (Table 1). For the 2J and 3J data, values of 1.09 and 1 were determined for n, respectively [55]. The correlation of the electron RDCs with the normalized NIEL would be closer if these n values had been applied for each individual dataset. The behavior of the RDCs at very low proton energies can be explained in terms of the finite range of these protons in the solar cell devices under study. This will be investigated in detail in Section 7.2. To facilitate on-orbit performance predictions for these MJ cell technologies, values of C and Dx have been determined. These were found by fitting the radiation data presented in [51] to Equation 8. Table 3: These are the fitting parameters found from fitting the3J solar cell data [51] to Equation 8.
N C Dxe (MeV/g) Dxp (MeV/g) Rep BOL Eff (%)
3J InGaP/GaAs/Ge 1.6 0.295 9.83x109 3.08 x109 3.19 24.7
V - 48
5.3 Thin Film Photovoltaics Thin-film photovoltaics (TFPV) on flexible substrates are considered to be an enabling technology for several next generation space power systems. These technologies offer the advantages of low cost, light-weight, and radiation resistance. TFPV devices have been shown to be much more radiation resistant to proton and electron irradiation than their crystalline material counterparts, because TFPV devices have shown extremely low annealing temperatures suggesting that in actual operation, the net radiation degradation will be minimal. However, the technologies remain at a low level of maturity, and significant research and development is required to optimize the device performance and take full advantage of the advanced material properties. The two primary TFPV materials currently under development are amorphous Si (aSi) and CuIn(Ga)Se2 (CIGS). In this section, an understanding of the basic radiation degradation mechanisms and the processes by which the devices recover from the radiation damage will be developed. 5.3.1 Amorphous Si 5.3.1.1 Amorphous Si Radiation Response Amorphous Si solar cell technology was patented [56] and described by Carlson [57] in 1977. A good review of this technology was given by Guha and Yang [58]. Amorphous Si solar cells are typically grown by decomposition of silane at a temperature between 100 and 300oC using a variety of substrates with stainless steel and polyimide being the primary substrates of interest for space applications to enable lightweight, flexible arrays. An a-Si solar cell is typically a p-i-n structure, where most of the photocurrent generation occurs in the intrinsic layer. The a-Si material, as expected, is characterized by a high degree of disorder which creates wide bandtails in the forbidden gap and structural defects like dangling, strained, and weak bonds. These bandtails and defects act as recombination centers. In addition, it has been observed that illumination of a-Si solar cells leads to the creation of metastable defects that degrade the photovoltaic output (the so-called Staebler-Wronski Effect [ 59 ]). Thus, the challenge for attaining good-quality a-Si solar cell devices lies in the minimization of the concentration of recombination centers in the as-grown material and devising methods for stabilizing the material. The key factor that has been found to reduce the effect of as-grown defects is the use of hydrogen dilution in the gas mixture during growth [60]. The current understanding is that hydrogenated material is more ordered because the excess hydrogen passivates dangling bonds at the growing surface allowing the impinging species to bond in more energetically favorable sites. Stability issues have been addressed by making the absorber layers thinner thereby reducing the distance over which the charge carrier must migrate to be collected. The absorber layers needed to be so thin, however, that the cells suffered from incomplete photon absorption, so a multijunction approach has been adopted. The highest efficiencies have been achieved with a triple junction device (Figure 47) where the cell active layers are less than 1 µm thick in total. QE measurements made on a 3J a-Si cell are shown in Figure 48 with the AM0 spectrum shown for reference. The bandgap of the individual junctions are controlled by varying the growth parameters of the a-Si alloy and by forming a-Si – Ge alloys with varying Ge concentrations. The top junction is typically a-Si with a bandgap ~1.8 eV. The middle junction is typically a 1020% Ge a-Si(Ge) alloy with a bandgap of ~1.6eV. The bottom junction is typically a 40-50% Ge
V - 49
a-Si(Ge) alloy with a bandgap ~1.4eV. Stable, AM0 efficiencies around 10% can be routinely achieved with these devices, and a representative IV curve is shown in Figure 49. ITO - TCO p-i-n a-Si p-i-n a-Si(Ge), 20% p-i-n a-Si(Ge), 40% stainless steel – 5 mils
Figure 47: This is a schematic diagram of a triple-junction a-Si solar cell grown on a stainless steel substrate. The entire 3J stack above the stainless steel substrate is less than 1 um thick. 1.0 Irradiation was 0.5 MeV electrons up to 6x1015cm-2 Anneal was 24 hrs at 70oC in the dark Before Irradiation After Irradiation After Annealing AM0 spectrum (normalized)
0.8
External QE
a-Si(Ge), ~1.6eV
0.6 a-Si, ~1.8eV
0.4 a-Si(Ge), ~ 1.4eV
0.2
0.0
300
675 Wavelength (µm)
1050
Figure 48: These are external quantum efficiency (QE) data measured on a TJ a-Si solar cell. The response of each individual sub-cell is evident. The AM0 solar spectrum, normalized to its maximum value is shown for reference. This cell was irradiated with 0.5 MeV electrons up to a fluence of 6x1015 cm-2 and then annealed for 24 hours at 70oC in the dark and unbiased, and data measured at each stage are shown.
V - 50
0 a-Si
Current (mA)
-3
Area - 1.1 cm2 Isc - 10.1 mA/cm2 Voc - 2.30 V
-6
Imp - 8.6 mA/cm2 Vmp - 1.78 V Pmp - 15.3 mW/cm2 FF - 0.66 Eff - 11.21%
-9
-12 0.0
0.5
1.0
1.5 Voltage (V)
2.0
2.5
Figure 49: This is an IV curve measured on a 3J a-Si solar cell that is generally representative of this technology.
The first evaluations of a-Si solar cells for use in space appeared in the early 1980’s [61, 62]. These early studies were performed on single-junction (SJ) cells. Both electron and proton irradiations were performed, and degradation in short circuit current (Isc) was found to be the primary effect with little degradation observed in open circuit voltage (Voc). The Isc degradation was correlated with a degradation of the cell QE which was attributed to degradation in the minority carrier diffusion length (L) and possibly to surface deterioration. Significant annealing capacity of the a-Si cells was reported although the annealing temperatures employed in these early studies were rather high (>150oC). The degradation of L was said to be most likely due to displacements of hydrogen atoms while the annealing was attributed to the mobility of the hydrogen within the a-Si material. It was speculated that the annealing was likely occurring during the irradiation and that the annealing was controlled by a distribution of thermal barriers characterized by a distribution of annealing coefficients. During the latter half of the 1980’s, a-Si solar cells continued to be developed for terrestrial use such that, by around 1990 the a-Si conversion efficiencies were approaching that of the crystalline Si solar cells being used on space solar arrays. Therefore, the US Air Force funded research to again evaluate a-Si solar cells for space use. As part of this effort, Mueller and Anspaugh produced the first comprehensive study of the radiation response of a-Si solar cells [63]. They studied the electron and proton radiation response and annealing characteristics of 3J a-Si solar cells produced by Solarex. In agreement with the earlier work, this study showed electron irradiation to cause degradation in Isc and Voc that was readily removed by annealing. Muller and Anspaugh went further by showing significant annealing to occur at temperatures as low as 20oC. In contrast to the earlier work, however, Muller and Anspaugh showed very little proton irradiation-induced degradation except for the case where the protons stopped within the active region of the solar cell. In all cases, the proton degradation annealed at low temperatures. Muller and Anspaugh did point out the very interesting result that the damage induced by electron irradiation increased with decreasing electron energy, which is in disagreement with displacement damage theory. This suggests that more than just displacement damage is responsible for the a-Si solar cell response.
V - 51
Beyond the work of Muller and Anspaugh [63], progress in the understanding of the radiation response of a-Si solar cells was lead primarily by research performed by TRW [ 64 and references therein] and Wayne State University under TRW and NASA funding [65], as well as the ongoing joint AFRL/NRL research reported in [66,67] and being presented here. From analysis of the evolution of the junction dark current in SJ a-Si solar cells under proton and electron irradiation, Lord [65] and Wang [68] suggested that the degradation was primarily due to a reduction in the electric field in the intrinsic layer. This can explain the radiation-induced decrease in the dark IV characteristic of 3J a-Si cells (Figure 50) because the decrease can be modeled by a reduction in depletion region dark current due to a reduction in the junction electric field. In terms of the solar cell PV parameters, the reduction in the junction field can also explain the reduction in Isc after irradiation. Lord proposed the mechanism for this to be compensation of the material by radiation-induced defects, i.e. displacement damage effects [65]. However, further research by the Wayne State group observed similar degradation after 40 keV electron irradiation, which is too low an energy to produce displacements [68]. Indeed, returning to the data of Muller and Anspaugh [63], the fact that the degradation increases with decreasing electron energy cannot be explained by displacement damage effects alone. 10-1 Pre-rad
Current (A)
10-2 10-3 10-4 10-5
Post-anneal
10-6 10
Post-rad
a-Si solar cell
-7
500 keV electrons, 4x1015 cm-2
10-8 10
24 hrs, 70oC anneal
-9
0.0
0.5
1.0 1.5 2.0 Wavelength (nm)
2.5
Figure 50: These are dark current data measured in a 3J a-Si solar cell before and after electron irradiation and after annealing. The irradiation causes a decrease in the junction dark current. The annealing restores a significant percentage of the dark current.
A resolution of this result has been proposed by Srour et al. based on a comparison of proton, electron, and x-ray irradiation of 2J a-Si solar cells [64]. Their study showed that 10 keV x-ray irradiation produced the same degradation of the a-Si cells as electron and proton irradiation. This suggests that it is ionization effects that mainly control the device response, as they were able to correlate the damage produced by the various irradiations in terms of the equivalent ionizing dose. This is not to say that displacement damage effects could be completely ignored especially in cases where the irradiation causes large amounts of displacement damage, e.g. low energy protons that stop in the device active region [64], but their analysis showed that for essentially all realistic space environments, the a-Si solar cell degradation will be dominated by ionizing energy deposition. This analysis fits with the degradation model of Lord et al. [65] if one assumes that the ionization-induced charge is captured by pre-existing defects in the intrinsic region and that radiation-induced displacement damage defects play a role only at high damage levels.
V - 52
While the analysis of Srour et al. [64] puts forth a compelling argument, the understanding of the mechanisms controlling the a-Si radiation response is not complete. This can be seen in the present data of the response of 3J a-Si solar cells to electron irradiation at three different energies plotted as a function of absorbed ionizing dose (Figure 51). If the damage were due to absorbed ionizing dose, then the data would be expected to fall on a single curve when plotted against total dose, but they do not. Because the space environment consists of a spectrum of electrons, this has significant implications on the ability to predict on-orbit a-Si solar cell performance. The lack of correlation may be attributed to varied ionization rates within each sub-cell. The middle and bottom sub-cells have increasing amounts of Ge, and the sub-cells showed varied amounts of radiation-induced degradation as can be seen in the QE data (Figure 48). It is interesting to note that the degradation increased with increasing Ge content. As a final note, the present theory has been developed based primarily on analysis of the Isc degradation, but it is fill factor (FF) degradation that dominates the a-Si solar cell response, especially under proton irradiation [66,67]. To fit with the model as presented, the FF degradation might be attributed to an increase in bulk resistivity due to the decrease in conductivity through the intrinsic region, but more study is required to substantiate this.
Electron Irradiation a-Si Pmp (mW/cm2)
15
10
0.5 MeV 1 MeV 1.5 MeV 0.5 MeV after anneal
5
0
106
107 108 Ionizing Dose (rad(Si))
109
Figure 51: These are Pmp data measured in a-Si solar cells under irradiation by electrons of different energies plotted in terms of absorbed ionizing dose. The closed symbols represent data from cells that were not subjected to annealing. The open symbols represent data from cells that were subjected to a 24 hr, 70oC exposure after each incremental fluence. At the highest fluence, the annealing time was 43 hrs.
5.3.1.2 Amorphous Si Annealing Characteristics In Figure 48 and Figure 50, the effects of a 24 hr anneal in the dark at 70oC on an irradiated a-Si solar cell are shown. The annealing is seen to efficiently remove the radiation-induced degradation in both the QE and dark current. Similar recovery was observed in the FF such that substantial recovery was observed in maximum power (Pmp) (Figure 51). The data shown in Figure 51 were obtained by irradiating a single set of solar cells with incrementally increasing fluences of electrons. After each irradiation, some cells were exposed to a 24 hour anneal which
V - 53
resulted in a reduction in the amount of degradation. Furthermore, at the highest fluence, the anneal time was extended to 43 hrs, and more recovery was observed suggesting that full recovery may be attainable after sufficient long annealing times or higher annealing temperatures. Given these results, it was postulated that for an a-Si array in a real space environment, which is characterized by a low flux and an operating temperature of about 70oC, the a-Si solar cells would show essentially no degradation on orbit. United Solar Systems Corporation and Fokker Space teamed to test this theory [69]. They exposed 3J a-Si solar cells to a very low flux, 1 MeV electron irradiation at 70oC, and indeed, the cells showed very little degradation. Their data are shown in Figure 52 along with the data from Figure 51 and data from crystalline Si [4] and GaAs [5] for comparison. From this comparison, it is clear that when the annealing behavior is considered, the a-Si technology displays extreme radiation hardness. As pointed out by Srour et al. [64], this behavior can be explained in terms of the degradation model described above since significant recovery at relatively low annealing temperatures is consistent with the liberation of trapped charge as opposed to the annealing of displacement damage defects. However, just as the case with the development of the degradation model, more research is needed to fully understand the mechanisms for this annealing behavior. 1 MeV Electron Irradiation a-Si
Pmp (mW/cm2)
20
15
10 High Flux, 28oC High Flux, 28oC annealed Low Flux, 70oC crystalline Si GaAs/Ge
5
0
1014
1015 Fluence (cm-2)
1016
Figure 52: These are data from a-Si solar cells compared to data from crystalline Si [4] and GaAs/Ge [5]. Note that the data are plotted as a function of 1 MeV electron fluence. The two data sets labeled “High Flux” are from Figure 51. The data labeled “Low Flux” were taken under very low flux irradiations while the cells were held at 70oC [69].
5.3.2 CuIn(Ga)Se2 5.3.2.1 CuIn(Ga)Se2 Radiation Response CIGS solar cells are typically heterojunction devices formed when CdS is deposited on a layer of CuIn(Ga)Se2 (Figure 53). The CIGS layer is typically 1 – 2 µm thick while the CdS is typically on the order of 50 nm thick with a ~ 50 µm layer of ZnO serving as an antireflective coating and a passivation layer. The ZnO is usually coated with a layer of Indium Tin Oxide (ITO) which serves as a transparent conducting oxide (TCO). Sunlight enters the cell through the ZnO and is absorbed almost exclusively in the CIGS layer. One of the primary advantages of CIGS solar
V - 54
cells is that they may be grown on a variety of substrates. The highest efficiencies have been achieved with cells grown on soda-lime glass substrates [70]; however, for space applications, the focus has been on the growth of cells on lightweight and flexible substrates like stainless steel and Kapton [71,72, 73]. ITO TCO top contact (0.6 um) ZnO buffer layer (0.04 um) CdS (0.1 um) CuIn(Ga)Se2 (2 um) Molybdenum back contact (0.2 um) Substrate
Figure 53: This is a schematic diagram of a typical CIGS solar cell. The thickness given are approximate and vary amongst growth techniques and manufacturers. Various substrates are used ranging from glass to stainless steel to polyimide.
The radiation response of CIGS solar cells has been studied by several groups over the past two decades. In the mid to late 1980’s, Boeing carried out a program to develop CIGS solar cells for space applications [74], which resulted in a space flight experiment on board the Naval Research Laboratory’s LIPS-III Satellite [75]. In the 1990’s, a project was conducted in Japan, lead by NASDA (now JAXA), to demonstrate commercial-of-the-shelf (COTS) electronics for space applications, which included CIGS solar cells designed for terrestrial applications. These cells were launched on The Mission Demonstration Test Satellite-1 (MDS-1) on February 4, 2002 [76,77,78]. The Europeans have an extensive effort underway to develop CIGS solar cells (e.g. see [72]), and the European Space Agency (ESA), has funded efforts to develop CIGS for space use including a space flight experiment, Equator-S [22,79]. All of these projects and flight experiments have been based on laboratory samples, and no space qualified CIGS solar cells are currently available in any production volumes. Because of this, the overall CIGS radiation database is incomplete and somewhat incoherent. Nevertheless, this series of experiments have served to demonstrate the radiation hardness of CIGS solar cells, especially under electron irradiation, and to produce a basic understanding of the CIGS solar cell radiation response. A representative set of radiation degradation data for CIGS solar cells are shown in Figure 54. This figure presents data measured after 1 MeV proton irradiation of CIGS cells grown on flexible metal substrates and other cells grown on glass substrates. These cells were provided to NRL by the Institute of Physical Electronics (IPE) at the University of Stuttgart [73]. Data measured by IPE on similar cells after 4 MeV proton and 0.5, 1, and 3 MeV electron irradiation are also shown [80]. Also shown are the ground test data generated for the MDS-1 mission [78]. The evolution of the PV parameters with increasing particle fluence and energy appears to be in agreement with displacement damage effects. This is an important result because CIGS is a polycrystalline material, so an analysis based on displacement damage effects may not be applicable to a material with such a high degree of disorder. This will be investigated in detail in the next section. Studies of CIGS cells exposed to 50 keV x-rays showed very little degradation [81], which indicates that CIGS cells are not sensitive to ionization effects. Instead, the study by Jasenek and Rau showed the degradation under electron and proton irradiation to be due to the
V - 55
introduction of an acceptor-like defect approximately 300 meV above the valence band that acts as a recombination and a compensation center [80]. 1.0
Normalized Pmp
0.8
0.6
Metal Subst. 1 MeV prot. Glass Subst. 1 MeV prot. IPE data, 4 MeV prot MDS data, 0.38 MeV prot. MDS data, 1 MeV prot. MDS data, 3 MeV prot. IPE data, 0.5 MeV elec. IPE data, 1 MeV elect. IPE data, 3 MeV elec.
0.4
0.2
0.0 10
10
11
10
12
10
13
10
14
10
15
10
16
10
17
10
18
10
19
10
-2
Particle Fluence (cm )
Figure 54: Radiation data measured in CIGS solar cells under proton and electron irradiation. The particle and energy are indicated in the legend. The data labeled metal subst. and glass subst. Are CIGS cells grown on metal substrates and glass substrates, respectively, and both datasets were measured after 1 MeV proton irradiation. The 4 MeV proton and the electron irradiation data labeled IPE were generated on similar cells and are from [80]. The data labeled MDS-1 are the ground test data generated for the MDS-1 mission and are from [73].
While the study of Jasenek and Rau [80] was quite thorough, the exact nature of the radiationinduced defect structure is not completely understood. The more recent publication by IPE suggests that their model remains incomplete [82]. In particular, there appears to be degradation mechanisms operative under proton irradiation that are not evident under electron irradiation. Under proton irradiation, degradation is observed in all of the PV parameters, but under electron irradiation, only Voc shows significant degradation until very high fluence levels. In particular, no Isc degradation is observed under electron irradiation. This results in electron irradiation being significantly less damaging than proton irradiation [71,80]. While proton irradiation is expected to be more damaging than electron irradiation, the observed behavior results in a separation of electron and proton data that is considerably larger for CIGS cells than for other technologies. This can be seen explicitly in Figure 55 where the CIGS data are compared to InP/Si solar cells data. InP/Si solar cells are chosen for comparison because this technology has been shown to be very radiation resistant [83]. The response of the two technologies under proton irradiation is comparable, but under electron irradiation, the resistance of the CIGS cells appears to be superior to that of the InP/Si cells by over an order of magnitude. While the degradation model remains incomplete, a general understanding of the damage mechanisms operative in these CIGS cells has been developed. It can be concluded that the presently available CIGS database has shown CIGS solar cells to have good radiation resistance,
V - 56
especially under electron irradiation, and that the cell degradation can be modeled in terms of displacement damage effects as has been done for crystalline soar cell technologies like GaAs. 1.0
0.8 1 MeV electron Pmp (norm.)
4 MeV proton 0.6
0.4 CIGS InP/Si
0.2
0.0 1E11
1E12
1E13
1E14 1E15 1E16 Particle Fluence (cm-2)
1E17
1E18
1E19
Figure 55: CIGS solar cell degradation data compared to that of InP/Si. The response of the two technologies to proton irradiation is comparable, but the CIGS cells show enhanced resistance to electron irradiation.
5.3.2.2 CIGS Analysis in Terms of Displacement Damage Dose In this section, an explicit analysis of CIGS solar cell radiation damage in terms of displacement damage dose (Dd). The NIEL has been calculated for CIGS and is shown in Figure 56. The data of Figure 54 are plotted in terms of Dd in Figure 57. In determining the electron Dd values, Eref was set to 1 MeV and a value of n=2 was found to fit the data well. The data are seen to correlate to within a reasonable margin considering that the data were obtained from three different sets of radiation experiments (Figure 54). A factor of 156 was found to work well for the present data (Figure 58). Fitting the data to Equation 8 yielded the C and Dx parameters given in Table 4.
V - 57
1.E+02
1.E+01
1.E+00
protons NIEL (MeV cm2/g)
1.E-01
1.E-02
1.E-03
1.E-04
electrons
1.E-05
1.E-06
1.E-07 1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
1.E+01
1.E+02
1.E+03
Energy (MeV)
Figure 56: These are nonionizing energy loss (NIEL) values calculated for protons and electrons incident upon CuInSe2.
1.0
Normalized Pmp
0.8
0.6 Metal Subst. 1 MeV prot. Glass Subst. 1 MeV prot. IPE data, 4 MeV prot MDS data, 0.38 MeV prot. MDS data, 1 MeV prot. MDS data, 3 MeV prot. IPE data, 0.5 MeV elec. IPE data, 1 MeV elect. IPE data, 3 MeV elect.
0.4
0.2
0.0 8
10
9
10
10
10
11
10
12
10
13
10
14
10
15
10
Displacement Damage Dose (MeV/g)
Figure 57: These are the data of Figure 54 plotted as a function of Dd. The data measured at different energies are seen to collapse to a single curve.
V - 58
1.0
Normalized Pmp
0.8
0.6 Metal Subst. 1 MeV prot. Glass Subst. 1 MeV prot. IPE data, 4 MeV prot MDS data, 0.38 MeV prot. MDS data, 1 MeV prot. MDS data, 3 MeV prot. IPE data, 0.5 MeV elec. IPE data, 1 MeV elect. IPE data, 3 MeV elect.
0.4
0.2
0.0 7
10
8
10
9
10
10
10
11
10
12
10
13
10
14
10
Displacement Damage Dose (MeV/g)
Figure 58: These are the data of Figure 57 where the electron data has been brought into alignment with proton data by dividing the electron Dd values by the Rep factor of 156.
Table 4: Fit parameters determined by fitting the CIGS data to Equation 8.
CIGS 2 0.224 1.31 x1012 8.37 x109 156 10% and 15%
n C Dxe (MeV/g) Dxp (MeV/g) Rep BOL Eff (%)
5.3.2.3 CIGS Annealing Characteristics CIGS solar cells have displayed the ability to recover from radiation damage at temperatures as low as room temperature [63,84,85]. Moreover, CIGS solar cells have been observed to recover under an applied light bias [86,87,88,89]. This is extremely attractive behavior for space applications since it indicates that the cells will anneal while under normal operating conditions and thus show an effectively higher radiation resistance. A demonstration of this behavior can be seen in Figure 59 where data from [86] are shown that compare the recovery of Voc in a 1 MeV electron irradiated CIGS solar cell after storage in the dark with the recovery of a similarly irradiated cell after that cell had been illuminated at open circuit. The BOL Voc was ~ 525 mV, and the electron fluence was 1x1018 cm-2. The Voc loss decreases from 125 mV to 45 mV after less than 3 hours of illumination at room temperature. The work of Kawakita et al. [88,89] has shown similar annealing behavior after proton irradiation, and their work has demonstrated the annealing to occur while the solar cells were under irradiation. Furthermore, the data from the solar cell experiment onboard the MDS-1 spacecraft which is flying in a geostationary transfer orbit (GTO), as analyzed by Kawakita et al. [89], have demonstrated the annealing effects in the
V - 59
space environment (Figure 60). These results certainly suggest an extreme radiation resistance for CIGS solar cells.
Figure 59: These are data from [86] that demonstrate the enhanced annealing of irradiated CIGS solar cells under illumination. The cells were irradiated with 1 MeV electrons up to a fluence of 1x1018 cm-2. One cell was stored in the dark while the other was illuminated with white light (100 mWcm-2) at open circuit.
Figure 60: These are data from [89] showing CIGS solar cell data from the MDS-1 space solar cell experiment, which is flying in a high radiation environment. Also shown are predictions based on ground test data with and without annealing effects included. These data show no change in the CIGS solar output. The predictions made without including annealing significantly over estimate the cell degradation.
While the literature data clearly demonstrate the annealing capabilities of CIGS solar cells, the basic mechanisms governing the annealing of the radiation damage are not yet fully understood. Kawakita et al. [88,89] have produced a thorough analysis of the kinetics of the CIGS annealing V - 60
under both thermal stimulation and illumination. Their analysis was based on the formalism derived for crystalline materials such as InP [ 90 , 91 , 92 ] and InGaP [ 93 , 94 ] where the performance recovery is due to electron-hole pair recombination at defect sites that enhance defect migration and hence defect annealing [95]. The success of that analysis to accurately model both ground test and the MDS-1 space data would appear to indicate that the CIGS devices behave as their crystalline counterparts. However, Jasenek et al. have produced results that do not agree with this model. In contrast, their results showed that forward biasing an irradiated CIGS solar cell in the dark does not cause permanent recovery [87]. Thus, the formalism developed to describe the injection-annealing behavior of the crystalline solar cells must be adapted to apply to the CIGS cells, and the physical origins of the CIGS cell response must be at least somewhat different from those of the mechanisms operative in crystalline cells. Although the basic annealing mechanisms are not fully understood, the effects of annealing are substantial and must be considered when analyzing the overall CIGS radiation response. In particular, when making on orbit performance predictions, as will be done in a later section, the cell recovery must be included in the calculations. This can be done by modifying Equation 8. Using the formalism of Heinbockel [96], this is done by determining an effective value of Dd (labeled - Ddeff) which is the actual Dd value that has been reduced to account for annealing. Heinbockel has shown that Ddeff is given by the following expression:
Equation 12
Dd′ A( t ) In Equation 12, Dd' is the dose rate. The f parameter represents the fraction of defects that are expected to anneal, which has been assumed to be unity in the present analysis. A(t) is the defect annealing rate given by: Dd eff = f
Equation 13
A( t ) = A o e
−
Ea
kT
In Equation 13, Ao is a constant, Ea is the activation energy for the annealing process, k is Boltzmann’s constant, and T is the temperature. Using the data of Kawakita et al. [88], values of Ao and Ea have been determined to be 3.4x10-5 s-1 and 0.8 eV, respectively. Note that these annealing parameters are based on the illuminated annealing data from [88] as opposed to data of thermal annealing in the dark. Inserting Equation 13 into Equation 12 and then inserting Ddeff into Equation 8 allows the prediction of the degradation level of a CIGS cell in a given radiation environment that includes the effect of simultaneous annealing. A last issue to be addressed in the analysis of the annealing characteristics of the CIGS cells is the question of whether significant annealing of the cell occurs during the ground test measurements. If true, then the C value (Equation 8) might be considered incorrect since it would not represent a “pure” degradation rate. Kawakita et al. investigated this by performing 10 MeV irradiations at 270 K [88]. The data obtained after irradiation at low temperature is compared with the room temperature irradiation data from Figure 54 in Figure 61. The low V - 61
temperature irradiation data may be exhibiting a slightly higher degradation rate, but the differences are rather small in comparison the over scatter in the data. Considering that the 1 MeV proton irradiations required less than 10 minutes total irradiation time, annealing effects are not expected to be significant for those datasets. Indeed, it is the electron irradiation data that would be expected to show significant annealing effects since those irradiations require on the order of days to complete, but low temperature electron irradiation data are not presently available. For the present analysis, the degradation parameters given in Table 3 will be used. 1.0 0.9 0.8
Normalized Voc
0.7 0.6 0.5 0.4
Metal Subst. 1 MeV prot. Glass Subst. 1 MeV prot IPE data, 4 MeV prot. MDS data, 0.38 MeV prot. MDS data, 1 MeV prot. MDS data, 10 MeV prot. MDS data, 10 MeV prot. measured at low temperature
0.3 0.2 0.1 0.0 8
10
9
10
10
10
11
10
12
10
13
10
Displacement Damage Dose (MeV/g)
Figure 61: These are normalized Voc data from the solar cells of Figure 54 plotted with data measured by Kawakita et al. [89] after 10 MeV proton irradiation at 270 K (labeled “MDS data, 10 MeV prot. measured at low temperature” in the legend). The data measured after irradiation at low temperature does appear to display a slightly increased degradation rate, but the difference is only slightly outside the range of the overall dataset.
6 On-Orbit Solar Cell Performance Predictions This section will take a brief look at a trade study used to design a space solar array. In addition to the choice of solar cell technology, there are many aspects important for such a study such as solar cell circuit layout, array substrate design, and choice of coverglass. In keeping with the focus of the present discussion, this trade study will investigate how the solar cell radiation response drives the array design. To accomplish this, the EOL performance of solar arrays based on 3J InGaP/GaAs/Ge, SJ GaAs/Ge, and CIGS solar cells will be calculated for low-Earth orbit (LEO), medium-Earth orbit (MEO), and geostationary Earth orbit (GEO). The shielding of the solar cells will be varied by choosing a range of coverglass thicknesses and two different solar array designs. The analysis will be made using the Solar Array Verification and Analysis Tool (SAVANT) computer code, which implements Dd analysis. It must be stressed, however, that these calculations are for illustrative purposes only. The parameter space for solar array design is quite large and mission specific, so this brief look cannot be expected to cover all aspects.
V - 62
The performance parameters for the 3J InGaP/GaAs/Ge, SJ GaAs/Ge, and CIGS used in this analysis are given in Table 1,Table 3, and Table 4. The BOL efficiency for the SJ GaAs/Ge cells was 18%. Two BOL efficiencies for the CIGS cells were considered, 10% to be representative of currently available cells and 15% to represent what is expected to be available within five years. Annealing has been considered for the CIGS cells according to the analysis presented above. No annealing has been considered for the crystalline devices. The 3J and SJ cell weights were taken from [97] to be 0.72 kg/m2. The CIGS weight was taken from [98] to be 0.4 kg/m2 where a 1.5 mil steel substrate is assumed for the solar cell substrate. As discussed in 4.2, the coverglass used on the solar cell surface, the substrate on which the solar cell active layers are grown, and the solar array substrate material all provide shielding for the solar cell from the incident irradiation. In the SAVANT calculations, this shielding is accounted for by converting the shielding material to an equivalent thickness of coverglass material. The effect of the equivalent thickness of coverglass on the incident particle spectrum is then calculated by applying the continuous slowing down approximation [21]. Here, the 1.5 mil thick steel substrate for the CIGS cells is estimated to be equivalent to 5 mils of coverglass. The Ge substrate for the SJ and 3J cells was assumed to be 5 mils thick and to be equivalent to 2 mils of coverglass. Two specific array designs have been chosen for the present study – a rigid honeycomb panel like the MPTB array (Figure 34), which might be considered a “standard” array, and a lightweight, flexible array consisting of 2 mil thick sheet of Kapton stretched on a frame. The lightweight array is considered in order to investigate the possible advantages of the thin film CIGS technology. For this study, the parameters of interest are the array substrate weight, expressed in Kg/m2 and the shielding the array provides. The rigid array is assumed to be 1.8 kg/m2 based on the analysis of [97] with an equivalent coverglass thickness of 30 mils. The lightweight array is assumed to weigh 0.69 kg/m2, which was determined by replacing the Al honeycomb weight in the rigid array with that of a 2 mil sheet of Kapton. In terms of shielding, a 2 mil sheet of Kapton is equivalent to about 1 mil of coverglass. The orbits chosen here are all circular. The representative LEO orbit was chosen to have a 1000 km radius, a 90o inclination, and 7 year duration. The MEO orbit was defined as an 8000 km orbit, 0o degree inclination, and 7 year duration. The GEO orbit radius is 35,876 km at a 100o W inclination for 15 years. Inputting these orbital parameters into SAVANT produced estimates of the equivalent Dd values for the electron and proton radiation environment. This was done for a range of coverglass thicknesses. These values give the total Dd due to electrons or protons that a solar cell with the given thickness coverglass will experience. By expressing the backside shielding provided by the solar array substrate as an equivalent thickness of coverglass, these data also provide the backside irradiation equivalent Dd, which is then summed with the frontside value to give the total Dd. The electron and proton Dd are then combined using the Rep value to give the total mission Dd. The simultaneous annealing of the CIGS cells was included by altering the total mission Dd values according to Equation 12 and Equation 13. The array temperature was assumed to be 70oC. The Dd' value was determined by dividing the total Dd by the total mission time. The effect of including the annealing in the CIGS calculations is shown in Figure 62 where the data
V - 63
are plotted as a function of the thickness of the frontside coverglass. In these calculations, the 15% BOL efficiency and the flexible solar array were assumed. As expected, including annealing effects has a significant impact in the situations where there is a significant radiation environment, namely when the coverglass is thin and in the MEO orbit. Indeed, in the MEO orbit, a gain of nearly a factor of 2 is observed due to annealing. Note that the mission duration in the MEO orbit was reduced from 7 years to 1 year because the radiation degradation was so severe for all the technologies on the lightweight array. 200 LEO Orbit o 1000 km, 90 incl., 7 years lightweight array CIGS solar cells, 15% BOL
100 EOL Specific Power (W/kg)
EOL Specific Power (W/kg)
150
100
Analysis without annealing Analysis including annealing effects
50
MEO Orbit o 8000 km, 0 incl., 1 year lightweight array CIGS solar cells, 15% BOL
50
Analysis without annealing Analysis including annealing effects 0 0
2
4
6
8
10
12
0
14
0
10
20
Coverglass Thickness (mils)
30
40
50
60
70
Coverglass Thickness (mils)
(a)
(b) 250
EOL Specific Power (W/kg)
200
150 GEO Orbit o 35,876 km, 100 W incl., 15 years lightweight array CIGS solar cells, 15% BOL
100
50
Analysis without annealing Analysis including annealing effects 0 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Coverglass Thickness (mils)
(c) Figure 62: These are predicted EOL specific power values as a function of front coverglass thickness for a CIGS solar cell mounted on a lightweight array for a LEO (a), MEO (b), and GEO (c) orbit where the calculations have been performed with and without including annealing effects. A significant effect is observed for the thinner coverglasses especially for the high radiation MEO orbit where nearly a factor of 2 gain is observed due to annealing.
For thin film technologies, one of the primary advantages driving the technology development is the potential for flexible solar arrays. However, if the radiation degradation of the solar cell is severe enough, the cell may require a rigid coverglass for shielding. The data in Figure 62 show that in the LEO and GEO orbits, the annealing bolsters the EOL performance enough that flying cells with extremely thin covers may be feasible. For the array to be flexible, the cover would actually need to be a flexible coating. More careful analysis is required to determine if the cells V - 64
would be survivable with only a coating on the surface, but these preliminary data are at least encouraging. Comparisons of the results for the various cell technologies in the LEO orbit for the rigid array are shown in Figure 63. The data are plotted as a function of the thickness of the frontside coverglass. The backside shielding was constant at 30 mils equivalent for the rigid array, and the backside radiation exposure was included in the total Dd calculations. The data are expressed in terms of both power density (W/m2) and specific power (W/kg). The higher BOL efficiency of the 3J solar cells results in higher EOL performance. Also, the data show a clear peak in the specific power at about 2-3 mils of coverglass such that increased shielding results in decreased EOL performance due to increased weight. Similar results are observed for the GEO orbit assuming the rigid honeycomb array (Figure 64) except that the peak in specific power occurs at a slightly lower coverglass thickness. The same results are obtained if the rigid array is replaced by the lightweight array (Figure 65). 130
350
120 300
LEO Orbit o 1000 km, 90 incl., 7 years Annealing incl. in CIGS data
EOL Specific Power (W/kg)
110
2
EOL Pmp (W/m )
250
200
150
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
100
LEO Orbit o 1000 km, 90 incl., 7 years Annealing incl. in CIGS data
100 90 80 70 60 50
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
40 30 20
50 0
2
4
6
8
10
12
14
Coverglass Thickness (mils)
0
2
4
6
8
10
12
Coverglass Thickness (mils)
Figure 63: These are the calculated EOL performance data for the LEO orbit expressed in terms of power density (W/m2) and specific power (W/kg). The coverglass thickness value refers to the glass on the front of the cell. These calculations assumed the solar cells are mounted on a rigid honeycomb panel that provides an equivalent backside shielding of 30 mils, and the backside irradiation exposure was included in the total Dd calculations. In this orbit, the superior BOL efficiency of the 3J InGaP/GaAs/Ge cells results in the best EOL performance except for the thinnest of coverglasses.
V - 65
14
130 GEO Orbit o 35876 km, 100 W incl., 15 years Annealing Incl. in CIGS data
120
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge 0
2
GEO Orbit o 35876 km, 100 W incl., 15 years Annealing included in CIGS data
110 EOL Specific Power (W/kg)
2
EOL Pmp (W/m )
140
360 340 320 300 280 260 240 220 200 180 160 140 120 100 80 60 40 20 0
100 90 80 70 60 50 40
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
30 20 10 0
4
0
Coverglass Thickness (mils)
2
Coverglass Thickness (mils)
Figure 64: These are the calculated EOL performance data for the GEO orbit expressed in terms of power density (W/m2) and specific power (W/kg). The coverglass thickness value refers to the glass on the front of the cell. These calculations assumed the solar cells are mounted on a rigid honeycomb panel that provides an equivalent backside shielding of 30 mils of coverglass, and the backside irradiation exposure was included in the total Dd calculations. As was the case for the LEO orbit, the superior BOL efficiency of the 3J InGaP/GaAs/Ge cells results in the best EOL performance except for the thinnest of coverglasses.
V - 66
4
350
LEO Orbit o 1000 km, 90 incl., 7 years lightweight array Annelaing Incl. in GIGS data
200 300
2
EOL Pmp (W/m )
250
EOL Specific Power (W/kg)
LEO Orbit o 1000 km, 90 incl., 7 years lightweight array Annealing Incl. in CIGS data
200
150
100
CIGS, 10% BOL CIGS, 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
50
150
100
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
50
0
0 0
2
4
6
8
10
12
0
14
2
4
6
8
10
12
14
Coverglass Thickness (mils)
Coverglass Thickness (mils)
(a) 240
350
220 GEO Orbit o 35876 km, 100 W incl., 15 years lightweight array Annealing Incl. in CIGS data
180
2
EOL Pmp (W/m )
250
200 EOL Specific Power (W/kg)
300
200
150
100
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
50
2
140 120 GEO Orbit o 35876 km, 100 W incl., 15 years lightweight array Annealing Incl. in CIGS data
100 80 60
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
40 20
0 0
160
0 0.0
4
Coverglass Thickness (mils)
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Coverglass Thickness (mils)
(b) Figure 65: These are the calculated EOL performance data for the LEO (a) and GEO (b) orbits assuming a lightweight array expressed in terms of power density (W/m2) and specific power (W/kg). The coverglass thickness value refers to the glass on the front of the cell. The lightweight array was assumed to provides an equivalent backside shielding of 1 mil of coverglass, and the backside irradiation exposure was included in the total Dd calculations. As was the case for the LEO and GEO orbits assuming a rigid array, the superior BOL efficiency of the 3J InGaP/GaAs/Ge cells results in the best EOL performance except for the thinnest of coverglasses.
It is interesting to note that, for the most part, the CIGS cells provide little or no advantage in either the LEO or the GEO orbits when mounted on a rigid array. Even assuming a 15% BOL efficiency and including the weight savings, the SJ cells with 50% higher BOL efficiency and the 3J cells with twice the BOL efficiency give better EOL performance. It should be noted, however, that cell cost has not been considered. Actual cost information, i.e. $/W, to perform these calculations are not available. It may well be that the significantly lower manufacturing cost of the CIGS material may place the technology at the top of an EOL $/W plot. The results shown in Figure 63, Figure 64, and Figure 65 show the CIGS cells to be superior to their crystalline counterparts if an extremely thin coverglass is used. As mentioned in the
V - 67
discussion of the effect of including annealing effects in the EOL calculations (Figure 62), it is the potential of the CIGS cells to function well with only a thin surface coating instead of a relatively thick, rigid coverglass that makes the concept of a flexible solar array feasible. The present results suggest that a CIGS solar cell with only a 0.5 to 1 mil thick surface coating and mounted on a 2 mil thick sheet of kapton will produce an EOL specific power of about 180 W/kg which is only about 20% less than that achievable with the best 3J InGaP/GaAs/Ge technology. Given this potential, AFRL is working to develop coating materials that display the proper thermal and optical properties for the cells. Part of the joint AFRL/NRL study is focused on measuring the effect of radiation on the coating properties and measuring the shielding that the coatings provide to the solar cells. This work is ongoing, and the results of this study will be presented in future publications. The MEO orbit, because of the extreme radiation environment, may be an environment where the advantages of the CIGS solar cells would be expected to dominate. The calculated EOL performance in the MEO orbit for both the rigid and lightweight arrays is shown in Figure 66, and indeed, a dramatic advantage is realized from the CIGS solar cells, especially in the case of the 15% BOL efficiency on a lightweight array. Note that the mission duration was reduced to 1 year from 7 years for the lightweight array case because none of the technologies could survive the 7 year mission. In the calculations for the lightweight array case, the SJ and 3J devices were assumed to have an equivalent backside shielding of 1 mil from the Kapton array layer and 2 mils from the Ge wafer used as the solar cell substrate itself. For the CIGS solar cells, the equivalent backside shielding was assumed to be 1 mil for the Kapton sheet and 5 mil for the 1.5 mil steel solar cell substrate. With this significantly reduced shielding, the better radiation resistance and simultaneous annealing of the CIGS technology allows for much better EOL power densities even with the lower BOL efficiency, and the lower weight of the CIGS cells allows for significantly improved EOL specific power values. It is this capability that makes the CIGS technology attractive. It should be noted, however, that for the rigid array, a 12 mil coverglass is required to achieve maximum EOL specific power and for the lightweight array, a 3 mil cover was required. To achieve a flexible array will require either a significant advance in the coating materials or a sacrifice in EOL performance.
V - 68
250 MEO Orbit 0 8000 km, O incl., 7 years Annealing Incl. in CIGS data
200
70
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
60
EOL Specific Power (W/kg)
EOL Pmp (W/m )
50 2
150
100
CIGS, 10% BOL CIGS, 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
50
10
20
30
40
50
60
30 20 MEO Orbit o 8000 km, 0 incl., 7 years Annealing incl. in CIGS data
10 0
0 0
40
-10
70
0
10
Coverglass Thickness (mils)
20
30
40
50
60
70
Coverglass Thickness (mils)
(a) 200
130 120
EOL Specific Power (W/kg)
100
2
EOL Pmp (W/m )
150
MEO Orbit 0 8000 km, O incl., 1 year lightweight array Annealing Incl. in CIGS data
100
CIGS 10% BOL CIGS 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
110
CIGS, 10% BOL CIGS, 15% BOL SJ GaAs/Ge 3J InGaP/GaAs/Ge
50
90
MEO Orbit o 8000 km, 0 incl., 1 year lightweight array Annealing Incl. in CIGS data
80 70 60 50 40 30 20 10
0 0
10
20
30
40
50
60
0
70
0
Coverglass Thickness (mils)
10
20
30
40
50
60
70
Coverglass Thickness (mils)
(b) Figure 66: These are the calculated EOL performance data for the MEO orbit example for cases of the rigid (a) and lightweight array (b) expressed in terms of power density (W/m2) and specific power (W/kg). The coverglass thickness value refers to the glass on the front of the cell. The rigid array was assumed to provide an equivalent backside shielding of 30 mils of coverglass, and the lightweight array was assumed to provide an equivalent backside shielding of 1 mil of coverglass. The backside irradiation exposure was included in the total Dd calculations. The duration of the mission for the lightweight array case was reduced to1 year. In this severe radiation environment, the CIGS show an advantage.
7
Special Topics in Solar Cell Radiation Response
7.1 Solar Cell Response at High Degradation Levels This section will investigate the response of a solar cell to irradiation levels higher than were considered in Section 2.3. Although the radiation levels involved are much larger than those encountered in a typical space mission, the investigation is worthwhile because in addition to diffusion length degradation, other degradation mechanisms emerge that lead to interesting effects on the solar cell performance. V - 69
The degradation of an InP/Si solar cell measured over a wide proton fluence range is shown in Figure 67. The data have been divided into three regions along the fluence axis. Region I corresponds to the fluence range where diffusion length degradation primarily controls the cell response as described in Section 2.3. Region II corresponds to higher fluences where there is an increase in Isc while Voc continues to decrease. These competing effects result in Pmp remaining essentially constant with increasing fluence. Region III corresponds to very high fluences where both Isc and Voc decline significantly and Pmp degrades to near zero. The following discussion will investigate the degradation mechanisms operative in Regions II and III. While this discussion uses InP/Si as an example, the same degradation mechanisms have been observed in Si and GaAs solar cells [7,8,9]. 40 Isc (mA/cm2), Pmax (mW/cm2), Eff (%)
Region
Region
35
Region
30 25
n+p InP/Si
20 15 Isc Pmax Voc
10 5 0
10 0 10
1011
1012
1013
1014
1015
3 MeV Proton Fluence (cm-2)
Figure 67: This graph depicts the evolution of the PV parameters of an InP/Si solar cell under irradiation by 3 MeV protons. The degradation can be separated into 3 fluence regions. In region I at lower fluences, the cell response is controlled mainly by reduction in diffusion length while the response in regions II & III at high fluences is due mainly to carrier removal effects.
7.1.1 Region II The onset of region II is marked by the emergence of the second major degradation mechanism, carrier removal. As described in conjunction with Figure 7, carrier removal occurs when radiation-induced defects capture majority carriers and thereby compensate the material. Since solar cells are typically one-sided abrupt junctions, carrier removal almost exclusively affects the base. Using capacitance vs. voltage (CV) measurements, the base carrier concentration can be determined. Carrier concentration data have been measured in this way in the InP/Si solar cell of Figure 67 as a function of 3 MeV proton fluence (Figure 68). As the carrier concentration decreases, the depletion region width increases (Figure 1), which means more light will be absorbed in that region. Because of the junction electric field, carrier collection is more efficient in the depletion region. Therefore, Isc increases, even above the pre-irradiation value (Figure 67).
V - 70
Majority Carrier Concentration (cm-3)
101 9 8 7
Measured Data fit of data
6 5
n+p InP/Si
4 3
2
100
0
50
100 150 3 MeV Proton Fluence (x1011cm-2)
200
Figure 68: This figure shows the decrease in the base majority carrier concentration due to carrier removal in an InP/Si solar cell under 3 MeV proton irradiation. Radiation-induced defects capture majority charge carriers and thereby compensate the material.
While extension of the depletion region results in enhanced carrier collection efficiency and thus improved Isc, it also results in an increased dark current which degrades Voc. Because the recombination/generation current originates with defects in the depleted region, the magnitude of this current increases linearly with depletion region width. Also, L continues to degrade in this fluence regime resulting in a continuing increase in diffusion dark current. The increase in Isc and decrease in Voc are competing effects, which results in Pmp remaining essentially constant in Region II (Figure 67). 7.1.2 Region III The third degradation region occurs at very high fluence levels, and it is marked by a decrease of the photovoltaic output to zero. It should be noted that while the decrease in Pmp appears abrupt in Figure 67, the logarithmic scale on the fluence axis masks an otherwise extended fluence range. In Region III, the effects of carrier removal become so severe that the p-type base is driven n-type, which destroys the original emitter/base junction of the cell. There can no longer be any photogenerated current or voltage, so Isc and Voc decrease bringing Pmp down to near zero. This evolution can be clearly seen in the cell QE response (Figure 69). As the 3 MeV proton fluence increases from 1x1013 to 7x1013 cm-2, the QE increases corresponding to the increase of Isc in region II. However, as the fluence increases to beyond 1x1014 cm-2, the QE degrades as the base transitions from p- to n-type.
V - 71
1.0
External Quantum Efficiency
0.9
0 1x1013 7x1013 1x1014 2x1014
n+p InP/Si
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 300
400
500
600 700 800 Wavelength (nm)
900
1000
1100
Figure 69: This graph shows the degradation in QE of the InP/Si cell of Figure 67 under 3 MeV proton irradiation. The particle fluence is given in the legend in units of cm-2. For fluences in Region I, the long wavelength response degrades due to diffusion length degradation. At higher fluences, the QE increases due to an increase in the depletion region width caused by carrier removal. This corresponds to the increase of Isc in Region II. At very high fluences, carrier removal destroys the junction and the QE degrades rapidly (Region III).
There is a residual QE response evident in Fig. 5 after the highest fluence. The origin of this photoresponse has been identified by Taylor et al. [99] as the base/back surface field (BSF) interface (Figure 1). The BSF layer is included in the cell structure to create a drift field that drives minority charge carriers back toward the junction. Since the BSF is much more heavily doped than the base, it remains p-type even after the base has turned n-type. Therefore, an n+p junction forms at the rear of the cell which gives rise to a small QE response at the longest wavelengths. However, since the cell is so heavily damaged, L is very short and the dark current is large, so the cell can produce essentially no power. 7.2 Case of Nonuniform Damage Deposition As seen in Figure 17, the proton NIEL increases with decreasing energy. Because of this, lower energy protons loose energy at an increasing rate as they slow down and eventually come to rest in the cell. Increased energy deposition means greater defect concentration, so low energy proton irradiation induces a nonuniform damage profile, with a higher defect concentration at the end of the proton track. This is shown graphically in Figure 70 where the Monte Carlo particle transport program SRIM [100] has been used to calculate the damage tracks produced by protons with various incident energies in Si. The end of track damage peak is seen to become smaller and deeper into the material at lower energies. Since Si solar cell diffusion lengths are on the order of 100 to 200 µm, it is not until the energy exceeds about 10 MeV that uniform damage can be expected in the cell active region.
V - 72
#Vacancies/µm
101
Si
400 keV 1 MeV 4 MeV 10 MeV
100
10-1
10-2 10-1
100
101
102
103
Depth (µm)
Figure 70: These are proton damage tracks in Si for various incident proton energies calculated using SRIM [100]. The proton transfers energy to the crystal lattice as it traverses the solar cell creating vacancies. As the proton energy decreases, the rate of energy transfer (i.e. NIEL) increases, so the induced defect concentration increases. A peak in defect production occurs at the end of the track.
In the case when the proton energy is changing significantly as it passes through the solar cell active region, damage predictions based on the incident proton energy are not expected to be accurate. Because the NIEL is calculated based on the incident proton energy, the Si solar cell damage coefficients [14] correlate with NIEL over only limited energy range (Figure 71). It is important to understand that this is not indicative of a breakdown of the solar cell degradation model based on displacement damage effects. It simply indicates that the damage energy calculation is performed with a higher energy than the proton actually possesses when causing the damage. It follows, then, that if the correct proton energy were used, then the damage calculations, i.e. NIEL, would correlate with the measured RDCs. Messenger et al. have developed a formalism for performing such calculations [101]. This formalism is based on the Kinchin-Pease model, which states that the number of vacancies produced is directly proportional to the damage energy. Using SRIM [38], the vacancy distribution produced by protons of a given incident energy in the solar cell structure of interest is calculated (e.g. Figure 70), which is directly proportional to NIEL. Integrating this distribution over the device depth gives the total displacement damage deposited. Normalizing this value at all energies to the value determined for incident 10 MeV protons produces a set of RDCs that can be compared to the measured dataset from [14]. This is done in Figure 72 and the agreement is seen to be quite good over a wide energy range.
V - 73
Relative Damage Coefficients
102 *All data normalized to 10 MeV protons
Si
Proton NIEL JPL Damage Coefficients 101
100
ESD<EINC 10-1 10-1
100
ESD≈EINC
101
102
Proton Energy (MeV)
Figure 71: A plot of measured Si solar cell damage coefficients relative to 10 MeV (i.e. RDCs) along with the Si NIEL also normalized to 10 MeV. The RDCs correlate with the NIEL over only a very small energy range because for incident energies less than about 6 MeV, the proton looses significant energy as it passes through the cell active region.
Relative Damage Coefficients
102 *All data normalized to 10 MeV protons Proton NIEL JPL Damage Coefficients SRIM Results (80 µm) SRIM Results (100 µm) 101
100 10-1
100
101
Proton Energy (MeV)
Figure 72: This is a comparison of the measured Si solar cell RDCs [14] with those calculated using SRIM [38] according to the formalism of Messenger et al. [101]. Assuming a solar cell active region 80 mm wide produces good agreement between
Two calculated datasets are shown in Figure 72. The difference between these is the length over which the vacancy distribution was integrated. The integration should be performed over the active region of the solar cell, so the integration limits are based on the value of the minority carrier diffusion length. For the data shown in Figure 72, diffusion lengths of 100 and 80 µm were chosen, and the data calculated assuming 80 µm appears to give the better fit. Thus, with knowledge of the minority carrier diffusion length of the material and the structure of the solar cell, the entire set of RDCs can be calculated using SRIM [38]. Furthermore, with the calculated
V - 74
Normalized Maximum Power (P/Po)
RDCs and a single degradation curve measured at the reference energy, in this case 10 MeV, a complete calculation of the solar cell performance under irradiation by a spectrum of particles can be made. A comparison of EOL performance predictions made with the measured RDCs to those made using the calculated RDCs is shown in Figure 73 for a 60o circular Earth orbit as a function of orbital altitude. The calculations shown in Figure 73 were made using the JPL methodology. The data demonstrate that application of the JPL method using either the measured of the calculated RDCs produces the same results to within a few percent. 1.0 Si Solar Cells (1982) 0.9
*solid lines: JPL dashed lines: SRIM
0.8
20 mils
0.7
12 mils 6 mils
0.6
3 mils
0.5 0.4 1 mils
0.3 0.2
0.0
0 mils
*60o, 1 year missions
0.1 0
2000
4000
6000
8000
10000 12000 14000 16000
Orbital Altitude (km)
Figure 73: These are end-of-life performance predictions made using the JPL methodology for the specified Earth orbit as a function of orbital altitude. The coverglass thickness assumed for each altitude is indicated on the graph. The calculations were performed using the measured RDCs from [14] and the RDCs calculated by the methodology of Messenger et al. [101], and the agreement is seen to be quite good.
Low energy proton irradiation induced damage in the MJ InGaP/GaAs/Ge solar cells produces very interesting results due to the fact that the active layers of the MJ devices are extremely thin in comparison to the proton range. Damage track calculations from SRIM for protons incident on a typical 3J InGaP/GaAs/Ge solar cell are shown in Figure 74. Protons with incident energies of about 0.5 MeV and below stop within the active cell region, so the RDCs in this lower energy range are expected to deviate from the NIEL calculated assuming the incident energy. The plot of SJ [5] and MJ cell [51] RDCs with the NIEL (Figure 46) is reproduced in Figure 75. The RDCs decrease below the NIEL for incident energy around 0.1 MeV, but the RDCs are seen to increase for an incident energy of 50 keV. This behavior can be explained by investigating the details of the damage induced by protons with incident energies in this range.
V - 75
Vacancy Production Rate (#/µm/p+)
103
Proton Energy
InGaP
GaAs
Ge
15.8 keV 25.1 keV 39.8 keV 63.1 keV 126 keV 199 keV 316 keV 501 keV 794 keV 1.26 MeV 1.99 MeV 3.16 MeV 5.01 MeV 7.94 MeV 12.6 MeV
102
101
100
10-1
10-2 10-2
10-1
100
101
Penetration Depth (µm)
102
103
Figure 74: These are proton irradiation-induced damage tracks calculated by SRIM assuming an InGaP/GaAs/Ge solar cell structure.
Norm NIEL and RDCs
10
2
10
1
10
0
10
Proton NIEL SJ RDC 2J RDC 3J RDC Electron NIEL (1.7 power) SJ RDC 2J RDC 3J RDC
-1 -2
10
-1
10
0
10
10
1
2
10
3
10
Energy (MeV)
Figure 75: RDCs for SJ GaAs/Ge [5], 2J InGaP2/GaAs, and 3J InGaP2/GaAs/Ge [51] solar cells plotted along with the GaAs NIEL. To be consistent with the RDCs, the proton NIEL has been normalized to the value at 10 MeV and the electron data to 1 MeV. The electron NIEL was raised to the 1.7 power according to the n value determined for the SJ cells. The correlation of the RDCs with NIEL indicates that the Dd analysis methodology is applicable to the MJ cell technologies.
Figure 76 shows QE measurements made on 3J InGaP/GaAs/Ge solar cells after irradiation by protons with incident energies in the range in question. The 0.05 and 0.1 keV protons cause degradation in the top cell only. The 0.4 and 1 MeV protons damage the middle cell almost exclusively. Therefore, in agreement with the measured RDCs, the damage produced by protons with energies of about 0.4 MeV and above is controlled by the GaAs sub-cell, and the RDCs correlate directly with the GaAs NIEL. In the very low energy range, the damage is strongly
V - 76
dependent on the specific cell structure. Depending on the thickness of the top cell and the exact bandgap of the top cell (the top cell bandgap can be varied through control of growth parameters and slight changes in stoichiometry), the 0.1 MeV proton damage can result in relatively more or less damage. The same conclusions hold 0.05 MeV proton damage. However, these RDCs are always less than the NIEL due to the increased radiation hardness of InGaP over GaAs [93]. 1.0
1.0
InGaP/GaAs/Ge
InGaP/GaAs/Ge 0.8
Quantum Efficiency
Quantum Efficiency
0.8
0.6
0.4
50 keV Protons Solid lines: Unirradiated Dashed lines: 1x1012 p+/cm2
0.2
0.0 300
500
700
900
1100
1300
1500
0.6
0.4
100 keV Protons Solid lines: Unirradiated Dashed lines: 1x1012 p+/cm2
0.2
1700
0.0 300
1900
500
700
Wavelength (nm)
900
1100
1.0
1700
1900
InGaP/GaAs/Ge
0.8
0.8
Quantum Efficiency
Quantum Efficiency
1500
1.0
InGaP/GaAs/Ge
0.6
0.4
400 keV Protons Solid lines: Unirradiated Dashed lines: 1x1012 p+/cm2
0.2
0.0 300
1300
Wavelength (nm)
500
700
900
1100
1300
1500
0.6
0.4
1 MeV Protons Solid lines: Unirradiated Dashed lines: 1x1012 p+/cm2
0.2
1700
1900
0.0 300
Wavelength (nm)
500
700
900
1100
1300
1500
1700
1900
Wavelength (nm)
Figure 76: These are QE measurements made on 3J InGaP/GaAs/Ge solar cells after irradiation by protons of different energies. 0.05 and 0.1 MeV protons cause damage only in the top cell while 0.4 and 1 MeV protons damage the middle cell only.
Using the SRIM generated vacancy profile data of Figure 74, the RDCs can be calculated for the MJ InGaP2/GaAs/Ge devices according to the methodology of Messenger et al. as was done above for Si solar cells. This has been done and the data are shown in Figure 77. The calculated RDCs agree with the measured RDCs reasonably well. In particular, the “double-hump” structure and the decrease in RDCs for incident energies below 0.1 MeV is accurately modeled in the calculated RDCs. The calculated RDCs do not exactly agree with the measured data RDCs due to the extreme sensitivity of the calculations to the exact 3J solar cell structure and top cell performance parameters, i.e. the diffusion length and absorption coefficient. Nevertheless, these results show that the MJ radiation response can be accurately modeled by this formalism.
V - 77
Pmax Relative Damage Coefficients
102
MJ
101
100
10-1
NIEL JPL RDCs (SPL 2J) SRIM RDCs JPL RDCs (SJ GaAs)
10-1
100
101
102
Proton Energy (MeV)
Figure 77: This is a comparison of the measured solar cell Pmp RDCs [5] with those calculated using SRIM [38] according to the formalism of Messenger et al. [101] for SJ GaAs/Ge [5] and 2J InGaP2/GaAs/Ge [51]. The proton NIEL, normalized to 10 MeV, is also shown.
This analysis appears to highlight the importance of low energy proton effects in space solar cell radiation response analyses. In this context, it is instructive to consider quantitatively the fraction of the solar cell damage resulting from these low energy protons in a space environment. Take the example of a 5093 km circular orbit at an inclination of 60o, which is a particularly harsh environment in the heart of the earth’s trapped proton belts. The proton spectrum has been calculated using the NASA AP8 model. With this spectrum, the equivalent Dd experienced by a GaAs/Ge solar cell after one year in orbit has been calculated using Equation 6. The calculations were performed assuming three different coverglass thicknesses to show the effects of shielding. The results are shown in Figure 78 as the cumulative fraction of Dd as a function of proton energy where the energy refers to that of the protons as they emerge from the backside of the coverglass and are directly incident upon the cell. The results show that after the incident proton spectrum has passed through the coverglass, protons with energies as low as the displacement damage threshold and ranging to greater than 100 MeV are incident upon the solar cell. However, the large majority of the solar cell degradation, i.e. the largest fraction of cumulative Dd, comes from protons with energies between about 0.l and 10 MeV. Therefore, while monoenergetic, normal incidence irradiation by low energy (< ~ 0.1 MeV) protons can be quite damaging, the effect of such protons in a real space environment is much less dramatic because of the effects of shielding and the content of space proton spectrum. This result has significant impact on choice of ground test protocols as will be discussed in next section.
V - 78
Cumulative Fraction of Dd
1.0
GaAs
0.8
3 mil
0.6
12 mil 30 mil
0.4
5000 km, Circular Orbit 60° Inclination 5 year mission
0.2
0.0 10-4
10-3
10-2
10-1
100
101
102
103
Slowed-Down Proton Energy (MeV)
Figure 78: Cumulative fraction of the total Dd as a function of proton energy through coverglasses of 3, 12, and 30 mils. As an example, this plot shows, using 20 mil fused silica coverglasses in the harsh proton environment of this Earth orbit, that ~75% of the total displacement damage arises from protons having energies 0.1 MeV<E<10 MeV.
The results of Figure 78 indicate that the contribution of low energy protons to the overall solar cell damage is relatively small, but this is not to say that they can be completely ignored. Indeed, the lower energy limit for integral in Equation 2 for the JPL method or Equation 6 for the NRL method must extend down to essentially zero. This is actually a strength of the NRL method because, assuming that the proton spectrum does not change as it passes through the solar cell active layers which is true for the thin III-V MJ devices, the integral of the proton spectrum with the NIEL gives an accurate representation of the total Dd absorbed by the solar cell. The JPL method, on the other hand, is limited to the energy range for which RDCs have been experimentally determined, so extrapolation is required to extend their energy range. The conclusion is, then, that in Earth orbit, the contribution of low energy proton damage to a shielded solar cell is relatively small and that the damage is accurately modeled by the NRL methodology. 8 Testing Approaches The solar cell radiation response characterization and analysis that has been presented here provides useful information and insight for the planning of solar cell ground test approaches. At present, the product qualification regimen for a new solar cell technology is based on the JPL methodology, so the qualification requires a large number of radiation experiments including as many as 8 proton irradiations at energies ranging from 0.05 up to 10 MeV and at least 4 electron irradiations with 1 MeV electrons being a prerequisite. Such an exhaustive study requires significant resources in terms of man-power, irradiation facility time and availability, measurement time, and quantity of samples. The analysis presented here show that such a study is not always required. Once the correlation between the proton and electron RDCs and NIEL
V - 79
has been established, then the full radiation characterization can be achieved with one electron and one proton energy irradiation. This liberates significant resources that can then be applied to make the analysis more accurate. In particular, instead of expending the effort to produce many IV curves measurements before and after a large number of irradiations, the effort can be focused on producing one, more accurate degradation curve at a single particle energy, which can potentially lead to significantly reduced error bars on EOL performance predictions. The present discussion also has important implications on the choice of the specific electron or proton energies for testing. The present methodology employed by the industry based on the JPL method requires that 10 MeV proton and 1 MeV electron irradiations be included in the study. While Van DeGraaff accelerators that produce 1 MeV electrons are reasonably plentiful, there are only a limited number of facilities that can accommodate 10 MeV proton irradiations of solar cells. As a result, generating the 10 MeV proton degradation curve can be difficult to schedule. Furthermore, 10 MeV protons are energetic enough to activate the solar cell material so that the samples may not be available for post-irradiation testing for several months. It is often more convenient to perform the irradiations at a lower proton energy such as 1 or 3 MeV. Performing the data analysis based on Dd allows the flexibility to choose an alternate proton energy. Specifically, the data in Figure 78 indicates that to simulate space effects as closely as possible, the optimum accelerator energy should be in the range 1 – 8 MeV. Specifically, little is to be gained from testing at energies <1 MeV or >10 MeV. In addition to the particle energy, there are several aspects of solar cell irradiation experiments that are worth discussion here. First is the irradiation spot size. Because it is desirable to build solar cells as large as possible, solar cell radiation test samples are typically at least 2x2cm2 or larger. Full size 3J InGaP/GaAs/Ge solar cells are typically 4x6 cm2. Accelerators are often tuned to produce spot sizes only ~ 2.5 cm in diameter. Therefore, some method must be implemented to expand the beam coverage area. If the beam spot size is large enough to cover on dimension of the solar cell, then one method that can be used is to place the solar cells on a wheel that rotates through the beam. This extends the effective coverage area, but care must be taken in making dosimetry measurements in this configuration. Not only does the duty cycle of the rotating wheel need to be accounted for, but moreover, the proper placement of the faraday cup must be ensured. Another method for expanding the irradiation coverage area is to raster the particle beam over a prescribed area. This can enable quite a large coverage area, but again dosimetry must be considered. For a rastered beam, dosimetry is often achieved through charge collection from the end-station of the beam line. This is an accurate method provided the exposure area is well defined and accurately measured and that all leakage currents are minimized. As highlighted in the discussion thus far, care must be taken to ensure accurate dosimetry. In contrast to ionizing radiation, there are few choices for dosimetry choices for displacement damage studies. A Faraday cup is the most widely used means, but care must be taken in the design and implementation of the cup. In particular, back-scattered and secondary particles must be suppressed and/or collected through proper geometrical design and grounding of the cup. Other than the Faraday cup, however, there is no solid state device for dosimetry for displacement damage studies. Therefore, control devices should always be included in each
V - 80
radiation experiment. Alternative concepts for a displacement damage dose dosimeter are under study, e.g. GaAs LEDs or pin diodes, where the theory of operation is based on the damage correlation in terms of NIEL presented here for GaAs solar cells. 9 Summary A comprehensive review of radiation effects in solar cells for space applications has been presented. The primary damage mechanisms in crystalline solar cell materials have been identified to be displacement damage effects that result in the formation of free charge carrier recombination and compensation centers. The presence of these defect centers causes diffusion length degradation and carrier removal, which, in turn, degrade the solar cell performance by increasing the junction dark current and decreasing the photocurrent and photovoltage. The radiation response of crystalline, single-junction InP, GaAs, and Si solar cells and multijunction InGaP/GaAs/Ge solar cells has been studied in detail. The radiation response of thin-film amorphous Si and polycrystalline CuIn(Ga)Se2 solar cells has been presented. The response of the CuIn(Ga)Se2 solar cells was shown to be controlled by the displacement damage effects identified in the crystalline cell case. The amorphous Si solar cells, on the other hand, were seen to be more strongly controlled by ionization effects, although the thin-film cell analysis is not complete and more research is required to fully understand the response of those devices.
Techniques for modeling solar cell radiation damage were explored. The goal of the models is to enable predictions of solar cell performance in a complex radiation environment based on monoenergetic ground test measurements. The two modeling techniques currently available, the “equivalent fluence” technique developed by JPL and the displacement damage dose technique developed by NRL, have been studied in depth. These techniques are based on the same fundamental physical principle that the solar cell degradation can be correlated in terms of the amount of displacement damage deposited in the solar cell material by the irradiation. The full correlation of a large data set measured under electron and proton irradiation at various energies was demonstrated using both methods, and an exercise was performed to demonstrate both methods to produce equivalent solar cell performance predictions. Using the displacement damage dose modeling technique, a trade study was performed to illustrate the procedure for obtaining end-of-life solar array performance predictions in specific Earth orbits. The effect of coverglass thickness and solar array structure was investigated as well as effect of changing the actual solar cell technology. The analysis showed that for most cases, the extremely high beginning-of-life efficiency of the 3J InGaP/GaAs/Ge solar cells makes it the technology of choice. Alternate technologies, namely the CuIn(Ga)Se2, display advantages in the extreme MEO radiation environment where the better radiation resistance and reduced weight of the CuIn(Ga)Se2 cells allows for a lighter weight array. Annealing effects were not included in this study, so this advantage is expected to be enhanced further when annealing is considered. In addition, the cell cost has not been considered, and it is expected that the reduced manufacturing cost of the CIGS solar cells in comparison to the crystalline technologies will enable higher end-of-life $/W values to be obtained. The understanding of the radiation response mechanisms operative in solar cells gained from the present analysis allowed for general comments to be made on the testing approaches adopted for solar cell ground testing. It was demonstrated that most of the solar cell degradation caused by
V - 81
exposure to a typical space spectrum is caused by protons with energies between 0.1 and 10 MeV so that ground testing will be most appropriate when performed in this energy range. Special attention was paid to the effect of low energy protons. It was shown that in a typical space environment, the degradation produced by protons with incident energies low enough that they slow down and possibly stop in the solar cell active region is relatively small. Also, it was shown how this damage is accurately modeled by the displacement dose modeling technique. Therefore, it is not necessary to perform ground tests at such low energies unless there is some specific perceived need due to the planned mission of particular technology under test. 10 References [1] S. M. Sze, Physics of Semiconductor Devices, John Wiley and Sons, New York (1981) [2] H.J. Hovel, “Solar Cells”, Semiconductors and Semimetals (R.K. Willardson and A.C. Beer, eds.) 11, Academic Press, New York, 1975, p 17-20. [3] J.R. Srour, “Displacement Damage Effects in Electronic Materials, Devices, and Integrated Circuits”, Tutorial Short Course Notes, IEEE Nuclear and Space Radiation Effects Conference, Portland, OR, July 1988. [4] H.Y. Tada, J.R. Carter, Jr., B.E. Anspaugh, and R.G. Downing, Solar Cell Radiation Handbook, 3rd Edition, JPL Publication 82-69, 1982. [5] B. E. Anspaugh, GaAs Solar Cell Radiation Handbook, JPL Publication 96-9, 1996. [6] R. J. Walters, S. R. Messenger, H. L. Cotal, G. P. Summers, and E. A. Burke, “Electron and Proton IrradiationInduced Degradation of Epitaxial InP Solar Cells”, Solid-State Electronics, 39 (6), 797 (1996) [7] C. J. Keavney, R. J. Walters, and P. J. Drevinsky, “Optimizing the Radiation Resistance of InP Solar Cells: Effect of Dopant Level and Cell Thickness”, J. Appl. Phys 73, 60 (1993) [8] K. A. Bertness, B. T. Cavicchi, S. R. Kurtz, J. M. Olson, A. E. Kibbler, C. Kramer, “Effect of base doping or radiation damage in GaAs single-junction solar cells”, Proc. IEEE Photovoltaics Spec. Conf. 1991, p1582-1587 [9] M. Yamaguchi, S. J. Taylor, M. Yang, S. Matsuda, O. Kawasaki, T. Hisamatsu, “High-energy and high-fluence proton irradiation effects in silicon solar cells”, J. Appl. Phys. 80, 4916 (1996) [10] W. Shockley, Electrons and Holes in Semiconductors, D. Van Nostrand, New York (1950) [11] Sah, Noyce, and Shockley, Proc. IRE, 45, 1228 (1957) [12] S. C. Choo, Soild State Electronics, 11, 1069 (1968) [13] B.E. Anspaugh and R.G. Downing, “Radiation Effects in Silicon and Gallium Arsenide Solar Cells Using Isotropic and Normally Incident Radiation”, JPL Publication 84-61, 1984. [14] B.E. Anspaugh, “The Solar Cell Radiation Handbook: Addendum 1 1982-1988”, JPL Publication 82-69, Addendum 1, 1989 [15] G. P. Summers, R. J. Walters, M. A. Xapsos, E. A. Burke. S. R. Messenger, P. Shapiro, and R. L. Statler, IEEE Proc. 1st World Conf. on Photo. Energy Conversion, 1994, p. 2068 [16] G.P. Summers, E.A. Burke, and M.A. Xapsos, Radiat. Meas., 24(1), 1-8 (1995). [17] G.P. Summers, S.R. Messenger, E.A. Burke, M.A. Xapsos, and R.J. Walters, Appl. Phys. Lett., 71(6), 832-834 (1997). [18] S. R. Messenger, M. A. Xapsos, E. A. Burke, R. J. Walters, and G. P. Summers, IEEE Trans. Nucl. Sci., 44(6), 2169-2173 (1997). [19] T. L. Morton, R. Chock, K. Long, S. Bailey, S. R. Messenger, R. J. Walters, and G. P. Summers, Tech. Digest 11th Intl. Photovoltaic Science and Engineering Conference, Hokkaido, Japan, pp. 815-816, 1999 [20] S.R. Messenger, E.A. Burke, G.P. Summers, M.A. Xapsos, R.J. Walters, E.M. Jackson, and B.D. Weaver, IEEE Trans. Nucl. Sci., 46(6), 1595-1602 (1999). [21] S. R. Messenger, G. P. Summers, E. A. Burke, R. J. Walters, and M. A. Xapsos, Progress in Photovoltaics: Research and Applications 9, 103-121 (2001) [22] S. R. Messenger, R. J. Walters, G. P. Summers, T. L. Morton, G. LaRoche, C. Signorini, O. Anzawa, and S. Matsuda, Proc. 16th European Photovoltaic Solar Energy Conference and Exhibition, Glasgow, Scotland, May (2000) [23] The development of the SAVANT code was supported by the NASA Living with a Star Space Environment Testbeds Program (http://lws-set.gsfc.nasa.gov)
V - 82
[24] To obtain a beta version of the program, please contact the Space Environments Effects Office at NASA Marshall Space Flight Center (http://see.msfc.nasa.gov/) [25] B.E. Anspaugh, Proc. 22nd IEEE Photo. Spec. Conf., Louisville, KY, (1991), pp. 1593-1598 [26] S.R. Messenger, E.A. Burke, M.A. Xapsos, G.P. Summers, R.J. Walters, I. Jun, and T.M. Jordan, "NIEL for heavy ions: An analytical approach", IEEE Trans. Nucl. Sci., Vol. 50, 1919-1923, 2003. [27] I. Jun, M.A. Xapsos, S.R. Messenger, E.A. Burke, R.J. Walters, G.P. Summers and T. Jordan, "Proton nonionizing energy loss (NIEL) for device Applications", IEEE Trans. Nucl. Sci., Vol 50, 1924-1928, 2003. [28] N.F. Mott, Proc. Roy. Soc. Lond., A124, 425-442 (1929). [29] G. P. Summers, E. A. Burke, P. Shapiro, S. R. Messenger, and R. J. Walters, IEEE Trans. Nucl. Sci., 40(6), 1372 (1993). [30] M.A. Xapsos, G.P. Summers, C.C. Blatchley, C.W. Colerico, E.A. Burke, S.R. Messenger, and P. Shapiro, IEEE Trans. Nucl. Sci., 41(6), 1945 (1994). [31] S. R. Messenger, M. A. Xapsos, G. P. Summers, and E. A. Burke, IEEE Proc. 1st World Conference on Photovoltaic Energy Conversion, Waikoloa, Hawaii, pp. 2153-2156, 1994. [32] S. R. Messenger, E. A. Burke, G. P. Summers, R. J. Walters, and J. H. Warner, Proc. 3rd World Conference on Photovoltaic Energy Conversion, Osaka, Japan, (2003) [33] D.M. Sawyer and J.I. Vette, AP8 Trapped Proton Environment for Solar Maximum and Solar Minimum, NASA Publication NSSCE 76-06, 1976. [34] J. I. Vette, The AE-8 Trapped Electron Model Environment, NASA Publication NSSDC91-24, 1991. [35] M. S. Gussenhoven, E. G. Mullen, M. D. Violet, C. Hein, J. Bass, and D. Madden, IEEE Trans. Nucl. Sci., 40(6), 1450-1457 (1993). [36] J. W. Haffner, Radiation and Shielding In Space (Academic, New York, 1967) [37] M. O. Burrell, Proceedings of the Second Symposium on Protection against Radiations in Space, NASA Publication SP-71, (1964). [38] J. F. Ziegler, J. B. Biersack, and U. Littmark, The Stopping and Range of Ions in Solids, Pergammon, New York, 1985, Vol. 1. [39] Space Radiation (Version 4.5), developed by J. Letaw of Space Radiation Associates. (Available through the World Wide Web at http://www.spacerad.com ) [40] H. Curtis, M. Piszczor, P. Severance, D. Guidice, and D. Olson, Proc. 1st World Conference on Photovoltaic Energy Conversion, Waikoloa, HI (1994), p. 2169 [41] H. Curtis and D. Marvin, Proc. 25th IEEE Photovoltaic Specialist Conference, Washington, DC (1996), p. 195 [42] O. Anzawa, K. Aoyama, T. Aburaya, K. Shinozaki, S. Matsuda, T. Ohshima, I. Nashiyama, H. Ito, S. Okada, T. Nakao, and Y. Matsumoto, Proceedings of the 16th Space Photovoltaic Research and Technology Conference, NASA Glenn Research Center, Cleveland, OH, USA, 1999. [43]R. J. Walters, G. P. Summers, S. R. Messenger and T. L. Morton, “SAVANT (Solar Array Verification and Analysis Tool) Analysis of the Microelectronics and Photonics Testbed (MPTB) Space Solar Cell Data”, Proc. Space Photo. Res. and Tech. Conf., NASA Glenn Research Center (2003). [44] C. S. Dyer, P. R. Truscott, C. Sanderson, C. Watson, C. L. Peerless, P. Knight, R. Mugford, T. Cousins, and R. Noulty, IEEE Trans. Nuc. Sci., 47, 2208, Dec 2000 [45] M. R. Brown, C. A. Garcia, G. S. Goodelle, J. S. Powe, J. A. Schwartz, Prog. in Photo. Res. and Appl., 4 (2), 129 (1996) [46] R. J. Walters, Defect and Diffusion Forum, Vols. 157-159, 125 (1998) [47] M. Yamaguchi and K. Ando, J. Appl. Phys. 63(11), 5555 (1988) [48] J. F. Geisz and D. J. Friedman, “III–N–V Semiconductors for Solar Photovoltaic Applications”, Semicond. Sci. Technol., 17, 769 (2002) [49] R. R. King, N. H. Karma, J. H. Ermer, M. Haddad, P. Colter, T. Isshiki, H. Ion, H.L. Cotal, D.E. Joslin, D.D. Krut, R. Sudharsanan, K. Edmondson, B.T. Cavicchi, and D.R. Lillington, Proc. 28th Photovoltaic Specialists Conference, Anchorage, AK, September, 2000, p. 998. [50] J.A. Carlin, S.A. Ringel, and E.A. Fitzgerald, “Impact of GaAs Buffer Thickness on Electronic Quality of GaAs Grown on Graded Ge/GeSi/Si Substrates”, Appl. Phys. Lett. 76, 1884 (2000). [51] D. C. Marvin, Aerospace Report NO. TOR-00(1210)-1 [52] S.R. Messenger, R.J. Walters, G.P. Summers, A.W. Bett, F. Dimroth, C. Baur, M. Meusel, T. Takamoto, T. Agui, M. Imaizumi, and S. Matsuda, Proceedings of the 17th European Photovoltaic Science and Engineering Conference, Munich, Germany, October, 2001.
V - 83
[53] R.J. Walters, S.R. Messenger, G.P. Summers, A.W. Bett, F. Dimroth, R.W. Hoffman, Jr., M.A. Stan, T. Takamoto, E. Ikeda, M. Imaizumi, O. Anzawa, and S. Matuda, “Radiation Response of GaxIn1-xAs Solar Cells”, Proceedings of the 16th European Photovoltaic Science and Engineering Conference, Glasgow, United Kingdom, May, 2000. [54] F. Dimroth, A.W. Bett, R.J. Waletrs, G.P. Summers, S.R. Messenger, T. Takamoto, E. Ikeda, M. Imaizumi, O. Anzawa, and S. Matsuda, “Radiation Response of Dual-Junction GayIn1-yP/Ga1-xInxAs Solar Cells”, Proceedings of the 28th Photovoltaic Specialists Conference, Anchorage, AK, September, 2000, p. 1110 [55] R. J. Walters, S. R. Messenger, and G. P. Summers, “Analysis and Modeling of Multijunction Solar Cells for Space Solar Cells”, Proceedings of the 28th Photovoltaic Specialists Conference, Anchorage, AK, September, 2000, p. 1097 [56] D. E. Carlson, U.S. Patent No 4,064,521 (1977) [57] D. E. Carlson, IEEE Trans. Electron Devices, vol. ED-24, 449 (1977) [58] S. Guha and J. Yang, Proc. IEEE 29th Photo. Spec. Conf., (2002) p. 1070 [59] D. L. Staebler and C. R. Wronski, Appl. Phys. Lett. 31, 292 (1977) [60] S. Guha, K. L. Narasimhan, and S. M. Pietruszko, J. Appl. Phys. 52 (2), 859 (1981) [61] C. E. Byvik, W. S. Slemp, B. T. Smith, and A. M. Buoncristiani, Proc. 17th IEEE Photo. Spec. Conf., Orlando, Florida (1884), p. 155 [62] R. Schwarz, J. Kolodzey, S. Aljishi, and S. Wagner, Proc. 18th IEEE Photo. Spec. Conf., Las Vegas, Nevada (1885), p. 903 [63] R. L. Mueller and B. E. Anspaugh, Proc. Space Photo. Res. and Tech. Conf XII, NASA Conf. Pub. #3210 (1992), p. 108 [64] J. R. Srour, G. J. Vendura, D. H. Lo, C. M. C. Toporow, M. Dooley, R. P. Nakano, and E. E. King, IEEE Trans. Nuc. Sci. 45(6), 2624 (1988) [65] K. R. Lord, M. R. Walters, and J. R. Woodyard, Proc. 23rd Photo. Spec. Conf., (1993), p. 1448 [66] R. J. Walters, S. R. Messenger, G. P. Summers, J. Nocerino, and K. C. Reinhardt, Space Power Workshop 2001. [67] P. Tlomak, K. C. Reinhardt, R. J. Walters, and S. R. Messenger, Proc. NASA Space Photovoltaic Research and Technology Conference, NASA Glenn Research Center (2001) [68] Q. Wang, K. R. Lord, and J. R. Woodyard, Proc. 28th IEEE Photo. Spec. Conf., Anchorage, AK (2000), p. 1057 [69] G. Oomen, R. van der Heijden, S. Guha, J. Yang, and A. Banerjee, Proc. 16th European Photo. Solar Energy Conf., Glasgow, UK (2000), paper VA3/23. [70] K. Ramanathan, M. A. Contreras, C. L. Perkins, S. Asher, F. S. Hasoon, J. Keane, D. Young, M. Romero, W. Metzger, R. Noufi, J. S. Ward, and A. Duda, Prog. Photovolt: Res. Appl., 11, 225, (2003) [71] B. M. Basol, V. K. Kapur, C. R. Leidholm, and A. Halani, Proc. 25th IEEE Photo. Spec. Conf., Washington, DC (1996), p. 157 [72] M. Hartmann, M. Schmidt, A. Jasenek, H. W. Schock, F. Kessler, K. Herz, and M. Powalla, IEEE Proc. 18th Phot. Spec. Conf., 2000, p. 683 [73] R. J. Walters, G. P. Summers, S. R. Messenger, A. Jasenek, H. W. Schock, U. Rau, J. Nocerino, J. Tringe, and K. Reinhardt, Proc. 17th European Photovoltaic Solar Energy Conference, Munich, Germany, 2001,VD3.25 [74] R. M. Burgess, W. S. Chen., W. E. Devaney, D. H. Doyle, N. P. Kim and B. J. Stanbery, Proc. IEEE Photo. Spec. Conf. (1988), p. 909 [75] R. M. Burgess, W. E. Devaney, and W. S. Chen, Proc. IEEE Photo. Spec. Conf. (1993), p. 1465 [76] Hisamatsu, T. Aburaya, and S. Matsuda, Proc. of the 2nd World Conference on Photovoltaic Energy Convention, Vienna, Austria (1998), p. 3568. [77] S. Matsuda, K. Shinozaki, T. Aburaya, K. Aoyama, and O. Anzawa, Proc. 11th Intern. Photo. Sci. and Eng. Conf., Sapporo, Japan (1999) pp. 155-158 [78] R. J. Walters, G. P. Summers, S. R. Messenger, T. Aburaya, S. Matsuda, Proc. 11th Intern. Photo. Sci. and Eng. Conf. Sapporo, Japan (1999) [79] G. La Roche, B. Hoesselbarth, and K. Bous, Proc. 16th European Photovoltaic Solar Energy Conference and Exhibition, Glasgow, Scotland, May (2000) [80] A. Jasenek and U. Rau, J. Appl. Phys., 90(2), 650 (2001) [81] J. Tringe, J. Nocerino, R. Tallon, W. Kemp, W. Shafarman, and D. Marvin, J. Appl. Phys., 91(1), 516 (2002) [82] K. Weinert, A. Jasenek, and U. Rau, Thin Solid Films, 431 –432, 453 (2003) [83] R. J. Walters, S. R. Messenger, H. L. Cotal, M. A. Xapsos, S. J. Wojtczuk, H. B. Serreze, G. P. Summers, J. Appl. Phys., 82, 2164 (1997)
V - 84
[84] A. Boden, D. Braunig, J. Klaer, F. H. Karg, B. Hosselbarth, and G. La Roche, Proc. 28th IEEE Photo. Spec. Conf., Anchorage, Alaska (2000), p. 1038 [85] A. Jasenek, H. W. Schock, J. H. Werner, and U. Rau, J. Appl. Phys. 79(18), 2922 (2001) [86] A. Jasenek, U. Rau, K. Weinert, H. W. Schock, and J. H. Werner, Proc. 29th IEEE Photo. Spec. Conf., New Orleans, Louisiana (2002), p. 872 [87] A. Jasenek, U. Rau, K. Weinert, H. W. Schock, and J. H. Werner, Appl. Phys. Lett. 82(9), 1410 (2003) [88] S. Kawakita, M. Imaizumi, M. Yamaguchi, K. Kushiya, T. Ohshima, H. Itoh, and S. Matsuda, Proc. 29th IEEE Photo. Spec. Conf., p. 978 (2002) [89] S. Kawakita, M. Imaizumi, M. Yamaguchi, K. Kushiya, T. Ohshima, H. Itoh, and S. Matsuda, Proc. 3rd World Conf. Photo. Solar Energy Conf., Osaka, Japan (2003) [90] R.J. Walters, Defect and Diffusion Forum, Vols. 157-159, 125 (1998) [91] M. Yamaguchi and K. Ando, J. Appl. Phys. 63(11), 5555 (1988) [92] R. J. Walters and G. P. Summers, J. Appl. Phys. 78(12), 7368 (1995) [93] R. J. Walters, M. A. Xapsos, H. L. Cotal, S. R. Messenger, G. P. Summers, P. R. Sharps, and M. L. Timmons, Solid-State Electronics, 42(9), 1747 (1998) [94] A. Khan, M. Yamaguchi, J. C. Bourgoin, N. de Angelis, and T. Takamoto, Appl. Phys. Lett. 76, 2559 (2000) [95] D. V. Lang and L. C. Kimmerling, Phys. Rev. Lett., 33, 489 (1974) [96] J. H. Heinbockel, E. J. Conway, and G. H. Walker, "Simultaneous Radiation Damage and Annealing of GaAs Solar Cells", Proc. 14th IEEE PVSC, p. 1085 (1980) [97] E. L. Ralph, Proc. 1st World Conf. on Photo. Energy Conversion, Waikiloa, HI (1994), p. 1998. [98] D. M. Murphy, M. I. Eskenazi, S. F. White, and B. R. Spence, Proc. 29th IEEE Photo. Spec. Conf., New Orleans, LA (2002), p. 782. [99] S. J. Taylor, M. Yamaguchi, M. Yang, M. Imaizumi, S. Matsuda, O. Kawasaki, and T. Hisamatsu, Appl. Phys. Lett. 70, 2165 (1997) [100] J. F. Ziegler, J. B. Biersack, and U. Littmark, The Stopping and Range of Ions in Solids, Pergammon, New York, 1985, Vol. 1. [101] S. R. Messenger, E. A. Burke, G. P. Summers, and R. J. Walters, IEEE Trans. Nuc. Sci., 49, 2690 (2002)
V - 85
