1999 IEEE NSREC IEEE Nuclear and Space Radiation Effects Conference Short Course
Radiation Effects in the Space Telecom...
368 downloads
1920 Views
5MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
1999 IEEE NSREC IEEE Nuclear and Space Radiation Effects Conference Short Course
Radiation Effects in the Space Telecom Environment
July 12, 1999 Sponsored by: IEEE/NPSS Radiation Effects Committee Supported by: Defense Threat Reduction Agency Sandia National Laboratories Air Force Research Laboratory Jet Propulsion Laboratory NASA – Goddard Space Flight Center Approved for public release; distribution is unlimited.
1999 IEEE Nuclear and Space Radiation Effects Conference
Short Course
Radiation Effects in the Space Telecom Environment July 12, 1999 Norfolk Virginia
Copyright© 1999 by the Institute of Electrical and Electronics Engineers, Inc. All rights reserved. Instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. For all other copying, reprint, or replication permission, write to Copyrights and Permissions Department, IEEE Publishing Services, 445 Hoes Lane, Piscataway, NJ 08555-1331.
Table of Contents
SECTION I …………………………………………………………... I 1-5 INTRODUCTION Daniel M. Fleetwood Sandia National Laboratories
SECTION II ……………………………………………………….. II 1-85 BASIC MECHANISMS FOR SINGLE-EVENT EFFECTS Paul E. Dodd Sandia National Laboratories
SECTION III …………………………………….……………… III 1-114 TOTAL-DOSE EFFECTS: MODELING FOR PRESENT AND FUTURE Jean-Luc Leray CEA/DAM Ile-de-France
SECTION IV ………………………………..…………………… IV 1-110 PROTON EFFECTS AND TEST ISSUES FOR SATELLITE APPLICATIONS Paul W. Marshall, Consultant Cheryl J. Marshall, NASA Goddard Space Flight Center
SECTION V ………………………………………………………... V 1-68 SYSTEM LEVEL MITIGATION STRATEGIES William F. Heidergott Motorola, Inc., Satellite Communications Group
AFTERWORD …………. Order information for Short Course CD ROM
1999 IEEE NSREC SHORT COURSE
SECTION I
INTRODUCTION
Daniel M. Fleetwood Sandia National Laboratories
Approved for public release; distribution is unlimited.
INTRODUCTION This Short Course covers in a tutorial fashion selected topics of relevance to space telecommunications systems. This is the 20th year in which the Short Course has been offered in conjunction with the IEEE Nuclear and Space Radiation Effects Conference (NSREC). The themes of the short course are selected each year to reflect the varying interests and requirements of the attendees of the IEEE NSREC. This year’s theme “RADIATION EFFECTS IN THE SPACE TELECOM ENVIRONMENT” is especially appropriate given the emerging interest in large-scale commercial space telecommunications systems. However, the information contained within the four segments should be useful to all systems that must operate reliably in the challenging radiation environment of space. The greatest challenge to commercial success in space is the integration of commercial and custom components into a system that can achieve its mission reliably and affordably. Rising to this challenge requires knowledge of the space radiation environment external to and within the satellite or spacecraft of interest, the effects of the environment on electronic and photonic devices, and system engineering techniques to mitigate these effects where possible. These requirements present significant constraints on component selection, and on system design and operation. The space telecommunications systems that best meet these challenges through proper assessment of the environment, disciplined parts selection and testing, judicious use of design margin, and efficient system design and integration will be the ones that best exploit commercial opportunities in space. The outline of the course is as follows: In Section II “BASIC MECHANISMS FOR SINGLE-EVENT EFFECTS” Paul Dodd presents an overview of the mechanisms responsible for single-event effects (SEE), with a particular eye toward the use of physics-based modeling and simulation to shed light on the fundamental processes involved. After a brief review of the space radiation environment responsible for SEE, nondestructive and destructive SEE failure modes are discussed. Techniques for mitigating SEE are reviewed, as well as newer topics such as particle energy effects and gate rupture in thin oxides. Future trends in SEE susceptibility are also addressed, including growing concerns for SEE in terrestrial microelectronics. In Section III “TOTAL-DOSE EFFECTS: MODELING FOR PRESENT AND FUTURE” Jean-Luc Leray reviews the basic mechanisms of total ionizing dose effects on semiconductor devices in the natural space environment. Time dependent effects on radiation response are discussed, as are new issues of especial interest to commercial off-the-shelf (COTS) parts. Detailed examples are presented that incorporate numerical modeling of total dose effects in several cases of interest. In Section IV “PROTON EFFECTS & TEST ISSUES FOR SATELLITE APPLICATIONS” Paul and Cheryl Marshall briefly review the proton environments, and
I-1
discuss single event, total ionizing dose, and displacement damage issues specific to proton effects in space. Case studies are presented that illustrate typical applications of single event rate predictions and displacement damage analysis tools. Current issues pertaining to emerging technologies, on-orbit predictions, and test fidelity are also discussed. In Section V, “SYSTEM LEVEL MITIGATION STRATEGIES” Bill Heidergott addresses the commercial space telecommunications industry and satellite systems, space environment effects on spacecraft payloads, and design for radiation environment compatibility. The presentation includes brief discussions of the environment, models, and effects on devices; the primary focus is on single event upset and transient effects, and fault tolerance techniques for mitigating their impact to system operations. I want to personally thank the five Short Course authors, Paul Dodd, Jean-Luc Leray, Paul Marshall, Cheryl Marshall, and Bill Heidergott for their efforts in preparing this Short Course. It is the diligence and expertise of the authors that has continued the tradition of excellence in NSREC Short Courses through the years, and that makes it a highlight of the conference week and a resource for the remainder of the year. This year, we are especially pleased that this tradition has been recognized with a CD-ROM that contains previous NSREC Short Courses from 1980-1998, provided to each course registrant. We thank the Radiation Effects Steering Group (especially Dale Platteter) and the 1999 NSREC Conference Committee for making this possible. I also thank Lew Cohn for his efforts in reviewing the Short Course and ensuring that the notes were printed on schedule, and the DTRA printing office for printing the notebooks.
Daniel M. Fleetwood Albuquerque, New Mexico
I-2
Biographies Daniel M. Fleetwood Short Course Organizer Sandia National Laboratories Daniel M. Fleetwood received his B. S. in Physics and Applied Mathematics from Purdue University in 1980, and his M. S. and Ph. D. degrees in Physics from Purdue in 1981 and 1984. He is a Distinguished Member of the Technical Staff in the Radiation Technology and Assurance Department at Sandia National Laboratories. He has been active in the field of radiation effects in microelectronics since joining Sandia in 1984, is the author or co-author of more than 175 publications, and holds two patents. In 1997 and 1998 he received R&D 100, Industry Week, and Discover Magazine awards for co-invention of a nonvolatile memory based on hydrogen-annealed SiO2. Dr. Fleetwood has served the radiation effects community as a guest editor of the December 1988-1990 and April 1996 special issues of the IEEE Transactions on Nuclear Science, as technical program chair and short course instructor for the IEEE NSREC, and as vice-chair/publications on the Radiation Effects Steering Group. He has received Outstanding Paper Awards for the 1985, 1988, and 1995-1998 IEEE NSREC’s and 1988, 1990, and 1995 HEART Conferences. He was awarded the International Correspondence Chess Master title in 1997, and is a member of The American Physical Society, Phi Beta Kappa, Sigma Pi Sigma, and a Fellow of IEEE. Dr. Fleetwood has accepted the position of Professor of Electrical Engineering at Vanderbilt University beginning August 1999.
Paul E. Dodd Sandia National Laboratories Paul E. Dodd received his B.S. and M.S. in Electrical Engineering from Purdue University in 1988 and 1989. He received his Ph.D. from Purdue in 1993 for research on novel cryogenic InAs bipolar transistors and experimental and theoretical studies of GaAs-based heterojunction bipolar transistors. He joined Sandia National Laboratories in 1993, and is a Senior Member of the Technical Staff in the Radiation Technology and Assurance Department. He is actively involved in the development of Sandia’s 0.5-µm and 0.35-µm bulk and SOI CMOS technologies, and the computer simulation of singleevent, total-dose, and transient radiation effects on microelectronics. Dr. Dodd has served the radiation effects community as publicity chairman and session chairman for the IEEE NSREC, and has been a session chairman for the Single-Event Effects Symposium. He has also served the IEEE International Electron Devices Meeting as a member of the Modeling and Simulation technical subcommittee and session chairman. Dr. Dodd is the author or co-author of more than 30 publications and is a member of the IEEE.
I-3
Jean-Luc Leray CEA/DAM Ile-de-France Jean-Luc Leray received his Engineering Degree from the “Ecole Centrale des Arts et Manufactures de Paris” in 1978, and the “Docteur d'État es Sciences Physiques” Degree from the University of Orsay, Paris, in 1989. Meanwhile, he was successively research engineer, project leader, and group leader at "Commissariat A l'Energie Atomique" (CEA), the French Agency for Atomic Energy. In 1992, he became head of the Radiation Hardening Section at CEA. In 1994, Dr. Leray was awarded "Grand Prix de l’Electronique Général Ferrié" by SEE (Société des Electriciens et des Electroniciens, Paris) and FIEE (French Federation of the Electronic Industries) for works in design and hardening of integrated technologies for military, space, and high-energy physics applications. Dr. Leray has served as Session Chair for the IEEE NSREC, and as Short Course Instructor and Technical Program Chair for RADECS. He is the author or coauthor of more than 110 publications and 4 book chapters, and holds one patent. In 1998, he was awarded the medal "Chevalier des Palmes Académiques" by the Ministry of Education and Research. Dr. Leray is now a scientific assistant and program advisor to the director of the department in charge of hardening matters at CEA. He is a "Membre Sénior" of SEE and a Senior Member of the IEEE.
Paul W. Marshall Consultant Paul W. Marshall received his B. S. in Physics from James Madison University in 1980, his M. S. in Radiation Biophysics from the Medical College of Virginia in 1982, and his Ph. D. from the Department of Nuclear Engineering and Engineering Physics at the University of Virginia in 1985. Since 1985, he has been employed by SFA, Incorporated, under contracts supporting the Naval Research Laboratory’s Radiation Effects Branch. His activities there have included development of proton test capabilities for microelectronic and photonic components of interest to satellite developers. Basic mechanisms of proton interactions have been a major emphasis in his studies of displacement damage and single event effects from protons. Since 1991 Dr. Marshall has been engaged, first as a collaborator and more recently as a consultant, with the NASA Goddard Space Flight Center’s Radiation Effects Group where he supports component and subsystem evaluations for numerous flight projects and continues investigations into basic mechanisms of proton and other radiation effects in emerging technologies. Dr. Marshall has chaired sessions and served on several committees for the NSREC, and he is a member of the IEEE and NPSS with over 80 published papers.
I-4
Cheryl J. Marshall NASA Goddard Space Flight Center Cheryl J. Marshall received her B. S. degrees in Physics and Chemistry from Georgetown University in 1979, and her Ph.D. in Physics from the University of North Carolina at Chapel Hill in 1986. From 1986 until 1998, she worked as a research physicist for the Naval Research Laboratory, investigating basic mechanisms of radiation damage in microelectronic and optoelectronic technologies critical to satellite imaging and communications. She also served as Section Head of the Emerging Technologies Section in the Radiation Effects Branch at NRL, and provided flight program support. Dr. Marshall has served as the Defense Threat Reduction Agency’s Program Area Reviewer for Single Event Effects, and chaired the 10th and 11th Single Event Effects Symposia. Since 1998, Dr. Marshall has worked for the NASA Goddard Space Flight Center, evaluating radiation effects in emerging technologies and providing flight program support. She has participated in the IEEE NSREC on several committees and chaired sessions (including the Poster Session in 1998). She is a member of the IEEE and NPSS with over 80 published papers.
William F. Heidergott Motorola, Inc., Satellite Communications Group William F. Heidergott received his B. S. in Electrical Engineering from the University of Arizona in 1974. Since joining Motorola, Inc., he has worked in the design and development of CMOS custom devices, application specific integrated circuits and standard products, and subsystem design and system development for numerous DoD, NASA, and commercial space programs. Mr. Heidergott’s recent assignments include program engineering support in design for radiation environment compatibility and management of technology for commercial space programs. In 1998 he chaired the inaugural NSREC session on radiation effects in commercial electronics and space systems.
I-5
BASIC MECHANISMS FOR SINGLE-EVENT EFFECTS Paul E. Dodd Sandia National Laboratories Albuquerque, New Mexico 1.0
Introduction
2.0
Brief Overview of Environments 2.1 Trapped Particles 2.1.1 Protons 2.1.2 Heavy Ions 2.2 Transient Particles 2.2.1 Solar Event Protons and Heavy Ions 2.2.2 Galactic Cosmic Rays 2.3 Secondary Particles
3.0
Basic Mechanisms for Non-Destructive Single-Event Effects 3.1 Charge Deposition 3.1.1 Direct Ionization 3.1.2 Nuclear Reaction Effects 3.2 Charge Collection 3.2.1 Basic Physics of Charge Transport 3.2.2 New Charge-Collection Mechanisms in Submicron Devices 3.3 Single-Event Upset Mechanisms in DRAMs 3.3.1 Storage Cell Errors 3.3.2 Bit-Line Errors 3.3.3 Combined Cell and Bit-Line Errors 3.4 Single-Event Upset Mechanisms in SRAMs 3.5 Single-Event Upset in Other Circuit Types 3.6 Single-Event Multiple-Bit Upset 3.7 Particle Energy Effects 3.8 Mitigation Techniques 3.8.1 Technology Hardening 3.8.2 Circuit- and System-Level Hardening
4.0
Basic Mechanisms for Destructive Single-Event Effects 4.1 Single-Event Latchup 4.1.1 Single-Event Latchup Mechanism 4.1.2 Mitigation Techniques 4.2 Single-Event Gate Rupture 4.2.1 Single-Event Gate Rupture Mechanism 4.2.2 Single-Event Gate Rupture in Power MOSFETs 4.2.3 Single-Event Gate Rupture in Thin Gate Oxides 4.3 Single-Event Burnout
II-1
5.0
Modeling and Simulation of Single-Event Mechanisms 5.1 Interaction Models 5.1.1 Track Structure Models 5.2 Physics-Based Device Models 5.2.1 3D Device and Mixed-Level Simulations 5.2.2 Recent Enhancements
6.0
Future Trends 6.1 Technology Drivers Impacting Single-Event Effects 6.2 Hardening Strategies 6.3 Terrestrial and High-Altitude Single-Event Effects 6.3.1 The Atmospheric Radiation Environment 6.3.2 Historical Perspective and Recent Studies 6.3.3 Mitigation Techniques
7.0
Summary and Conclusions
8.0
Acknowledgments
9.0
References
II-2
1.0
INTRODUCTION
Single-event effects (SEE) in microelectronics are caused when highly energetic particles present in the natural space environment (e.g., protons, neutrons, alpha particles, or other heavy ions) strike sensitive regions of a microelectronic circuit. Depending on several factors, the particle strike may cause no observable effect, a temporary disruption of circuit operation, or even permanent damage to the device or integrated circuit (IC). Single-event effects may be broadly characterized as either non-destructive or destructive SEE. Examples of non-destructive SEE include single-event upset in logic or memory circuits (SEU, or equivalently, soft errors) and single-event current transients (SET) in photodetectors. Destructive SEE include such phenomena as single-event latchup (SEL, which can be either destructive or non-destructive depending on circuit design), single-event burnout (SEB), and single-event gate rupture (SEGR). Any of these effects can cause unacceptable system performance in space applications, and possibly jeopardize mission success. In this short course segment, we will examine the basic physical mechanisms causing singleevent effects in microelectronics for spaceborne applications. We start with a brief overview of the particle environment encountered in space in order to get to know the enemy. We will then discuss the mechanisms and characteristics of many of the non-destructive and destructive SEE mentioned above in detail, including techniques for mitigation. Next we review modeling and simulation methods that have proved useful for gaining physical insight and predicting SEE in microelectronics. We conclude with a look into technology trends that may affect future device susceptibility to SEE and areas of emerging concern. Reflecting their relative importance in the commercial marketplace, most of this short course segment will focus on silicon MOS devices and circuits, but where appropriate we will examine other technologies of interest.
II-3
2.0
BRIEF OVERVIEW OF ENVIRONMENTS
Previous NSREC Short Course segments have given excellent detailed descriptions of the space radiation environment [1,2]. Several review articles on the subject were also published as part of the recent Special Issue on Single Event Effects and the Space Radiation Environment in the IEEE Transactions on Nuclear Science [3-8]. The intent is not to repeat a great deal of this information here, but rather to give a brief overview of some of the more important points that may be of use later as SEE mechanisms are discussed. The reader is directed to the above references for more details on the space radiation environment. 2.1
Trapped Particles
Charged particles that come into contact with the Earth’s magnetic field can become trapped in the near-Earth environment. These particles include electrons, protons, and heavy ions. Electrons are very important components of the space environment because they inflict damage on spacecraft through total ionizing radiation dose and spacecraft charging effects [9-11]. Because electrons do not usually contribute to single-event effects, we will limit our discussions to trapped proton and heavy ion populations. The trapped particle belts (Van Allen belts) consist of two regions of trapped particles: an inner belt centered at about 1.5 Earth radii, and an outer belt of particles at about 5 Earth radii, separated by a region of reduced (but non-zero) particle flux (the so-called “slot” region). References [1] and [2] contain helpful illustrations of the belt structure around the Earth. Although the origin of trapped particles in the near-Earth environment is not completely understood [1], sources include the solar wind and transient solar events, cosmic ray particles from interplanetary space, and reaction products from cosmic ray collisions with the Earth’s atmosphere. It was recently discovered that transient solar events can actually produce new trapped particle belts of surprisingly long duration [4,6]. 2.1.1
Protons
Regardless of origin, energetic protons do exist in the near-Earth environment and are one of the most prominent sources of SEE. They range in energy from tens of keV to hundreds of MeV, with fluxes as high as 105 protons/cm2/sec for protons with energy > 30 MeV [1]. Protons with these energies are easily able to penetrate shielding and impinge on electronics within spacecraft. The altitude at which proton flux peaks depends on the proton energy, with high energy (>30 MeV) protons being cut off by around 3.5 Earth radii, but lower energy protons existing throughout the slot region. Probably the most important region for proton SEE is the South Atlantic Anomaly (SAA), a region off the east coast of South America with greatly increased proton flux at altitudes less than 1000-2000 km. The SAA exists because of the difference between the Earth’s geographic spin axis and its magnetic axis, which causes a localized region of lower magnetic field off the Argentine coast [1]. During passes through the SAA, the flux of energetic (>30 MeV) protons can be more than 104 times as intense than at equivalent altitudes over other regions of the Earth [12]. The SAA is illustrated in Figure 2.1, which shows flux contours for protons with energy > 30 MeV as a function of latitude and longitude at altitudes of 500 km (Fig. 2.1a), 1000 km
II-4
(Fig. 2.1b), and 3000 km (Fig. 2.1c) [1]. At low altitudes, the SAA is highly localized, and as altitude increases, the SAA becomes less distinct until at 3000 km (Fig. 2.1c) the normal Van Allen belt structure re-emerges. 2.1.2
Heavy Ions
The Van Allen belts are predominantly composed of trapped electrons and protons, but it is now well accepted that heavy ions are also trapped by the Earth’s magnetic field. The origin of these particles is thought to be anomalous cosmic rays, which are neutral interstellar particles that drift into the solar system, become ionized by the solar wind and accelerated to 10’s of MeV/nucleon, and are subsequently trapped by the magnetosphere [1]. Trapped heavy ions (e.g., He, C, N, O, and Ne) have been measured by the SAMPEX spacecraft, and an example geographic distribution is shown in Figure 2.2 for oxygen ions [13]. Similar results have been
a)
b)
c) Figure 2.1
Integral proton flux contours as a function of latitude and longitude after Barth [1]. Altitudes are a) 500 km, b) 1000 km, and c) 3000 km. Note the South Atlantic Anomaly, which is visible at lower altitudes but disappears by 3000 km. II-5
obtained by the Japanese satellite MIDORI [14]. The oxygen ions shown in the polar regions in Fig. 2.2 are not trapped heavy ions, but are due to galactic cosmic rays, as discussed below. Trapped oxygen ions are evident in the region near the SAA for the same reason that proton fluxes are greatest there, i.e., the lower magnetic field in the SAA allows the ion flux to dip down to lower altitudes. The peak in trapped heavy ion fluxes is at altitudes just above the inner proton belt (i.e., 1.8 to 2 Earth radii). Because the trapped heavy ions have relatively low energies (10’s of MeV/nucleon), these particles may not penetrate through spacecraft shielding and therefore are not expected to be a major concern for SEE [1]. 2.2
Transient Particles
In this section, the classification of “transient” particles encompasses all particles in the nearEarth space environment that are not stably trapped in the magnetosphere. This includes particles introduced into the environment by solar events such as flares and coronal mass ejections (CMEs), as well as energetic ions incident from interstellar space. 2.2.1
Solar Event Protons and Heavy Ions
The activity level of the Sun is never constant, but follows a cyclical variation of active years followed by quiet years. The period of recent solar cycles has varied between 9 and 13 years, with an average of about 11 years. Solar cycle activity is frequently gauged by the observed number of sunspots, but many solar processes show the same variation. Importantly for the
Figure 2.2
Geographic distribution of trapped oxygen ions during solar quiet time as measured by the Mass Spectrometer Telescope (MAST) on SAMPEX [13]. II-6
present case, this includes the incidence of energetic solar events, with maximum numbers of solar flares and CMEs occurring during active years. Solar events still occur during solar quiet times, but they occur less frequently. We have long associated solar flares with an increased flux of energetic particles, but the evidence appears to support that CMEs may be more important [1]. Solar events can be broadly characterized as being either gradual or impulsive. The gradual events produce a raised particle flux that decays slowly over several hours or even days, and have been correlated to CMEs. These events are proton-rich and can produce high-energy (> 30 MeV) proton fluences higher than 109 protons/cm2 accumulated over a few days. Gradual events are responsible for the majority of large proton fluence events, and occur at a frequency of about 10 per year during solar maximum conditions. Impulsive events are by definition of much shorter duration (hours at most), and are marked by increased fluences of heavy ions and low energy electrons. Impulsive events produce heavy ion fluences that can be orders of magnitude above the galactic cosmic ray background. These heavy ions have energies ranging from tens of MeV/nucleon to hundreds of GeV/nucleon, but at the upper end of this range the flux falls below the galactic cosmic ray background. Impulsive events may be associated with solar flare activity and are responsible for about 1000 small solar particle events per year at solar maximum [1,3]. Figure 2.3 shows solar event proton fluences for the last three solar cycles, superimposed over a plot of the sunspot number [1]. The cyclical variation of the sunspot number is readily apparent, as is the fact that most (but not all) high-fluence proton events occur during solar active years. While solar particle events can be broadly classified as gradual or impulsive, individual events have their own very unique properties in terms of duration, particle fluence, energy spectrum, etc. Stassinopoulos et al. have presented a classification system where events are classified from small to extremely large, and solar cycles are classified from extremely mild to Event Fluences For Cycles 20 - 22 1011
Cycle 21
Cycle 22
> 10 MeV; > 108 p/cm2 > 30 MeV; > 107 p/cm2 Zurich Smoothed Sunspot Number
200 180 160
Protons/cm2
1010 140 120 109
100 80 60
108 40 20 10
Zurich Smoothed Sunspot Number
Cycle 20
0
7
1965
1970
1975
1980
1985
1990
1995
Year
Figure 2.3
Correlation of proton solar event fluence to sunspot number for solar cycles 20-22 [1]. Sunspot number is shown by the solid line plot, and proton solar event fluences by the bars.
II-7
extremely severe [5]. It can be seen in Figure 2.3 that solar cycle 20 was in general extremely quiet with the exception of an extremely large solar event in August 1972; this solar cycle was classified as very mild. Solar cycle 21 was extremely mild, with no individual events of highenergy proton fluence > 109 protons/cm2. By contrast, solar cycle 22 had 8 events with highenergy proton fluence > 109 protons/cm2; this cycle is considered extremely severe [5]. Some events (such as the August 1972 event) are actually a series of closely-spaced events, where proton fluxes have not yet decayed to their original level before the next event occurs. The importance of individual large solar events cannot be overestimated – the August 1972 event by itself accounted for 84% of the total high-energy proton fluence received by spacecraft during the entire 20th solar cycle. Accurate models to predict worst-case particle fluence from solar events are crucial to appropriate parts selection and survivable spacecraft design. 2.2.2
Galactic Cosmic Rays
6-Hour-Averaged Flux (cm2-s-sr)-1
Solar event particles are true transient particles in the sense that elevated fluxes of particles are observed only for a short time following an event (although recall that following a large event increased levels of trapped particles are observed and in some cases can produce new trapped particle belts). In contrast, galactic cosmic rays (GCR) form a background component of radiation that shows a slow cyclical variation with solar activity. GCRs are composed of very highly energetic protons and heavy ions that come from outside the solar system. These particles must fight against the solar wind to enter the solar system and are therefore at their maximum intensity at solar minimum and drop off a factor of 2 to 10 at solar maximum. The cyclical variation of GCRs is shown in Figure 2.4, which is a plot of 25-95 MeV/nucleon He flux over a 20-year period as measured by the IMP-8 satellite [15]. The spikes on this plot are due to increased heavy ion fluences from solar events. Note that these spikes are more likely to occur at solar maximum, when the baseline GCR heavy ion flux is lowest.
Figure 2.4
Solar cycle variation of heavy ions (in this case He nuclei) as measured by the Cosmic Ray Telescope (CRT) aboard the IMP-8 satellite [15]. Note the spikes in He flux caused by solar events. II-8
The particle composition of GCRs is shown in Figure 2.5 [16]. Protons comprise about 83% of the GCR flux, He nuclei (alpha particles) account for 13%, 3% are electrons, and the remaining 1% are heavier nuclei. Even though they are not very abundant, heavy ions are very important to SEE because they deposit the most energy per unit pathlength, as discussed in later sections. Note that beyond Fe, the heavy ion flux drops dramatically. This turns out to be important, because the energy deposited by an ion per unit pathlength depends on its atomic number. Ions heavier than Fe are more ionizing, but are much less abundant. Therefore, radiation hardening microelectronics so that they don’t experience SEE from ions up to Fe can result in low SEE rates. This is sometimes referred to as the “iron threshold.” GCRs that come into contact with the near-Earth environment encounter Earth’s geomagnetic field. Because they are so energetic (tens of MeV/nucleon to hundreds of GeV/nucleon), they do not become trapped and are not significantly attenuated by spacecraft shielding. GCRs that hit the atmosphere form a cascade of secondary particles, as mentioned in the next section. GCRs with polar trajectories can penetrate to low altitudes because of the reduced geomagnetic rigidity in the polar regions. Spacecraft with high inclination angles are therefore at greatest risk of encountering SEE due to GCRs. 2.3
Secondary Particles
Secondary particles are produced when GCRs strike the Earth’s atmosphere and produce a shower of particles in the atmospheric environment. Because this component of the radiation environment is important only for microelectronics operating in the Earth’s atmosphere, a description of the secondary particle environment will be delayed until Section 6.3, Terrestrial and High-Altitude Single-Event Effects.
Figure 2.5
Particle composition of galactic cosmic rays [16]. Note that Hydrogen and Helium nuclei (i.e., protons and alpha-particles) account for the vast majority of GCR flux, while heavy ions comprise only about 1%. II-9
3.0
BASIC MECHANISMS FOR NON-DESTRUCTIVE SINGLE-EVENT EFFECTS
All non-destructive single-event effects are caused by the same fundamental mechanism: collection of charge at a sensitive region of a microcircuit following the passage of an energetic particle through the device. In this section we look at the release of mobile carriers along the path of an incident particle, the collection of these carriers, and how the resulting transient currents interact with the circuit to generate a single-event upset. 3.1
Charge Deposition
By definition, as ionizing radiation passes through a target material electrons and holes are released along the path of ionizing particles. There are two primary methods by which carriers are released: direct ionization by the incident particle and ionization by secondary particles created by nuclear reactions between the incident particle and the target material. Direct ionization can cause SEU if the incident particle (such as a heavy ion) is ionizing enough to free a very high density of carriers. For lighter particles (e.g., protons), direct ionization may produce an insufficient amount of charge to cause upset directly and SEU may instead be due to ionization produced by secondary particles. 3.1.1
Direct Ionization
As we have said, when an energetic particle passes through a semiconductor material it frees charged carriers along its path as it loses energy. When all of its energy is lost, the particle comes to rest in the semiconductor, having traveled a total path length referred to as the particle’s range. We frequently use the terms linear energy transfer (LET) or dE/dx to describe the energy loss per unit path length of a particle as it passes through a material. LET has the seemingly odd units of MeV-cm2/mg, but this is simply because the energy loss per unit path length (in MeV/cm) is normalized by the density of the target material (in mg/cm3). We can easily relate the LET of a particle to its charge deposition per unit path length, because for a given material it takes a certain amount of energy to release an electron-hole pair. For example, in silicon one electronhole pair is produced for every 3.6 eV of energy lost, and silicon has a density of 2328 mg/cm3 [17]. Using these values it is easy to show that an LET of 97 MeV-cm2/mg corresponds to a charge deposition of 1 pC/µm. This conversion factor of about 100 is handy to keep in mind if you need to go back and forth between energy loss (LET) and charge deposition. A curve of particular interest for understanding the interaction of a given energetic particle with matter is the LET of the particle versus depth as it travels through the target material. Figure 3.1 shows such a curve for a 210-MeV chlorine ion traveling through silicon (a common heavy ion beam used for SEU testing at the Tandem Van de Graaff at Brookhaven National Laboratory). Such curves are readily obtained using computer codes derived from the work of Ziegler, et al (e.g., the TRIM and SRIM family of codes, [18]). This figure shows the basic characteristics of ion-induced charge deposition as a function of depth. A peak in the charge deposition occurs as the particle nears its range, and then a precipitous drop in deposition as the particle reaches its range and comes to rest. The peak in charge deposition is referred to as the Bragg peak, and in general occurs as the particle reaches an energy near 1 MeV/nucleon. A more rigorous discussion of the Bragg curve and the Bragg peak is found in reference [19].
II-10
30
2
LET (MeV-cm /mg)
Bragg Peak
20
10
0 0
20
40
60
80
Depth (µm) Figure 3.1
Linear energy transfer (LET) vs. depth curve for 210-MeV chlorine ions in silicon.
Whether or not the charge deposited through direct ionization is sufficient to cause an upset of course depends on the individual device and circuit that has been struck as well as the strike location and trajectory. Direct ionization is the primary charge deposition mechanism for upsets caused by heavy ions, where we rather loosely define a heavy ion as any ion with atomic number Z ≥ 2 (i.e., He and above, or put another way, particles other than protons, electrons, neutrons, or pions). Lighter particles such as protons do not usually produce enough charge by direct ionization to cause upsets in memory circuits, but recent research has suggested that as devices become ever more susceptible, upsets due to direct ionization by protons may occur [20,21]. This effect has not been experimentally confirmed to date. As discussed in references [21] and [22], these upsets will likely be difficult to observe and of limited importance to the overall upset rate because the upset rates will continue to be dominated by proton-silicon reactions for typical orbits and spacecraft configurations. In contrast to memories, upsets in photodiodes used in optocoupler applications have been observed and correlated to direct ionization by protons [2326]. Single-event current transients (SET) resulting from proton direct ionization are capable of causing upsets in these photodiodes because they are by design very large and operate at very high data rates [27]. A recent analysis suggests that a combination of direct ionization and recoils are responsible for the anomalous angular dependence of proton upsets in optocouplers [28]. Charge-coupled devices (CCDs) can also be sensitive to direct ionization by protons because of their large collection depths [29]. For more information on these devices and their susceptibility to protons the reader is directed to the third segment of this short course [30]. 3.1.2
Nuclear Reaction Effects
As mentioned above, direct ionization by light particles usually does not produce a high enough charge density to cause upsets. Unfortunately, this does not mean that we can ignore
II-11
these lighter particles. Protons and neutrons can both produce significant upset rates due to indirect mechanisms. As a high-energy proton or neutron enters the semiconductor lattice it may undergo an inelastic collision with a target nucleus. This may result in the emission of alpha (α) or gamma (γ) particles and the recoil of a daughter nucleus (e.g., Si emits α-particle and a recoiling Mg nucleus), or a spallation reaction, in which the target nucleus is broken into two fragments (e.g., Si breaks into C and O ions), each of which can recoil. Any of these reaction products can now deposit energy along their paths by direct ionization. Because these particles are much heavier than the original proton or neutron, they can deposit higher charge densities as they travel and therefore may be capable of causing an SEU. These inelastic collision products typically have fairly low energies and do not travel far from the particle impact site. They also tend to be forward-scattered in the direction of the original particle; this has consequences for the SEU sensitivity as a function of the angle of incidence [31]. Historically neglected, recent experiments and simulations have shown that elastic collisions may become important for very sensitive devices [32,33]. Low-energy secondary particles (primarily protons and neutrons) generated within packaging or shielding may be an even more significant source of SEU in these sensitive devices [33]. Once a nuclear reaction has occurred, the charge deposition is not greatly different in character from a directly ionizing heavy ion strike. Therefore, once deposited, it is subject to the same fields and concentration gradients and is collected in a similar manner. 3.2
Charge Collection
The basic properties of charge collection following a particle strike have been studied using a variety of experimental and theoretical methods. Broadbeam charge collection spectroscopy measurements have been used to determine SEU-sensitive volumes in SRAMs [34-36], and ion microbeams and lasers have been used with high-speed sampling oscilloscopes to measure charge-collection transients in Si and GaAs devices [37-46]. Ion microbeams and lasers have also been used to map integrated charge collection as a function of position in ICs [45,47,48], and more recently as a function of both time and position [49]. The physics of charge collection have also been intently studied through the use of two- and three-dimensional numerical simulation [50,51]. It is beyond the scope of this short course segment to comprehensively review the massive literature on charge collection; we seek to touch the main highlights and recent developments only. 3.2.1
Basic Physics of Charge Transport
There are basically only three mechanisms that act on the charge deposited by an energetic particle strike: 1) carriers can move by drift in response to applied or built-in fields in the device, 2) carriers can move by diffusion under the influence of carrier concentration gradients within the device, or 3) carriers can be annihilated by recombination through direct or non-direct processes. These three mechanisms are of course not unique to the particle strike problem and are in fact the governing processes of charge transport in semiconductors under most operating conditions [52]. When a particle strikes a microelectronic device, the most sensitive regions are reversebiased p/n junctions, as illustrated in Figure 3.2. The high field present in a reverse-biased junction depletion region can very efficiently collect the particle-induced charge through drift processes, leading to a transient current at the junction contact. The situation is not too dissimilar II-12
from the operation of a solar cell, where large-area p/n junctions are reverse-biased to collect charge liberated in the semiconductor by sunlight. Strikes near a depletion region can also result in a significant transient current as carriers diffuse into the vicinity of the depletion region field where they can be efficiently collected. Note that even for direct strikes, diffusion plays a role as carriers generated beyond the depletion region can diffuse back toward the junction. Shortly following the discovery of SEU, researchers at IBM used numerical device simulators to compute the response of reverse-biased p/n junctions to alpha-particle strikes [50,53,54]. An important insight gained from these early charge-collection simulations was the existence of a transient disturbance in the junction electrostatic potential, which was termed the “field funnel,” as shown in Figure 3.3 [53]. Charge generated along the particle track can locally collapse the junction electric field due to the highly conductive nature of the charge track and separation of charge by the depletion region field. This funneling effect can increase charge collection at the struck node by extending the junction electric field away from the junction and deep into the substrate, such that charge deposited some distance from the junction can be collected through the efficient drift process. The funnel effect has been investigated in further detail by later researchers [55-59], including the influence of epitaxial substrates on the transient charge-collection characteristics [60-62]. Several important additional insights have been gained from these studies. The reader is referred to [59-62] for more comprehensive discussions of funneling, but here are a few key points to keep in mind:
Ion Pa th
1) The term “funnel length” is a misnomer (as is funneling itself to some extent). There is no well-defined length that can be associated with funneling from a physical standpoint [58,61,62]. Funnel length is thus only a useful concept inasmuch as it can be used as a quasi-physical “fudge factor” in error rate calculations. In this case it can be used to represent any charge collection from outside the defined sensitive volume, including the effects of funneling.
Funnel
–+ +– – + Drift +– –+ + – Diffusion ++ –+
Depletion Region
Recombination Figure 3.2. Illustration of an ion strike on a p/n junction showing drift, diffusion, recombination, and funneling. II-13
2) Expecting funnel-assisted drift charge collection to be “prompt” can be misleading. In cases where the substrate is lightly-doped, funneling can take a long time (nanoseconds) to develop and to collect charge [54,60]. 3) Funneling does not require a direct strike on a depletion region. Near misses can also cause funneling if a high enough carrier density diffuses into the depletion region to collapse it [54,61]. 4) Funneling in epitaxial diodes is limited by the heavily-doped substrate underneath the epi layer. Once sufficient minority carriers are removed from the epi layer, the depletion region is reformed, funneling stops, and charge collection continues at a much slower rate. This leads to a characteristic knee in the charge-collection characteristics of n/p epitaxial diodes, as shown by the dashed curve in Fig. 3.4 [60,61]. 5) Because of differences in the hole and electron mobility, funneling occurs in reversebiased n/p diodes, but is much weaker or nonexistent in equivalent p/n diodes. This is indicated in Fig 3.4 by the lack of a knee in the p/n charge-collection characteristics (solid curve). [59,61]. 6) The total charge collected in an epitaxial diode can be approximated by the charge liberated in the epitaxial layer plus the charge liberated within a diffusion length of the epi/substrate boundary [61]. Funneling affects the rate at which the charge deposited in the epi layer is collected, but does not appear to change the total charge collection, since this charge would usually be collected even in the absence of funneling. While in some cases important to charge collection in isolated p/n junctions with constant applied bias, the role of the funnel is less significant in the case of static circuits such as SRAMs, where reverse-biased transistor junctions are connected to active external circuitry. In this scenario, the applied voltage at the struck junction is not constant, and in fact very often the
Figure 3.3. Illustration of funneling in an n+/p silicon junction following an ion strike: a) electrostatic potential, b) electron concentration. Note that contours of electrostatic potential are distorted along the path of the ion [53]. II-14
Figure 3.4. Charge-collection transients in n+/p and p+/n epitaxial diodes following ion strikes [61]. Note the lack of a knee in the p+/n diode curve, which indicates that funneling is very weak for this case. struck node may switch from being reverse-biased to zero-biased (see Section 3.4). This loss of bias at the struck node tends to lessen the importance of drift collection (and hence the funnel) as the single-event transient proceeds [63]. In such cases, funneling may play a role in the early-time response of the circuit by helping initially flip the node voltage, but it is late-time collection by diffusion that ensures the bit stays flipped (see Section 3.4 for further details about the upset process in SRAM circuits). 3.2.2
New Charge-Collection Mechanisms in Submicron Devices
The charge-collection response of a single p/n junction is generally presumed to accurately depict the response of the sensitive junction of a transistor, typically a reverse-biased drain region. Studies have indicated that a new charge-collection mechanism may exist for submicron MOS transistors which requires considering the entire transistor [64,65]. Termed the alphaparticle source-drain penetration effect (ALPEN), this charge-collection mechanism results from a disturbance in the channel potential that the authors referred to as a funneling effect. The effect is illustrated in Figure 3.5a and is triggered by a particle strike that passes through both the source and the drain at near-grazing incidence. Immediately following the strike, the electrostatic potential in the channel region is perturbed to the extent that i) there is no longer a potential barrier between the source and channel, and ii) there is a substantial potential gradient between the source and the drain. These two conditions together can lead to a significant (but short-lived) source-drain conduction current which mimics the “on” state of the transistor. This mechanism was revealed by 3D alpha-particle simulations and has been experimentally verified. The experiments indicate that source charge injection due to the ALPEN mechanism increases rapidly II-15
a)
b)
Figure 3.5. a) Illustration of the alpha-particle induced source-drain penetration (ALPEN) effect [64]. b) Charge injected by the source due to the ALPEN mechanism as a function of gate length. For gate lengths below 0.5 µm, this mechanism may become critical. for effective gate lengths below about 0.5 µm, as shown in Fig. 3.5b. Later work predicted the same direct channel conduction mechanism can occur in 0.3-µm gate length MOSFETs even for normal incidence strikes, and can lead to charge multiplication [66]. This mechanism may forebode a serious vulnerability to SEU for deep submicron MOSFETs. A somewhat similar, but distinct mechanism exists when electrons or holes released by a particle strike are confined to a well or body region in which a transistor is located. For example, for an n-channel MOS transistor located in a p-well, electrons induced by a particle strike can be collected at either the drain/well junction or the well/substrate junction. However, holes left in the well raise the well potential and lower the source/well potential barrier, and the source injects electrons into the channel, as illustrated in Figure 3.6 [67-70]. These electrons can be collected at the drain, where they add to the original particle-induced current and can cause an increased SEU sensitivity. Because the electrons are injected over the source/well barrier, this is referred to as a bipolar transistor effect, where the source acts as the emitter, the channel as the base region, and the drain as the collector. Reducing the channel length effectively decreases the base width, and the effect becomes more pronounced [69]. Essentially the same effect occurs in floating body silicon-on-insulator (SOI) devices, where excess carriers in the body turn on the parasitic bipolar transistor [69,71,72]. This bipolar effect severely impacts the intrinsic SEU hardness of floatingbody SOI transistors.
II-16
Source
Gate
Drain
P-Well
N-Substrate
Figure 3.6. Electron concentration contours inside an n-channel MOS transistor following a heavy ion strike [70]. The bipolar effect is evidenced by the contours emanating from the source, showing that the source is injecting electrons into the p-well, where they may be collected at the substrate or at the drain.
3.3
Single-Event Upset Mechanisms in DRAMs
We’ve seen how an ion strike releases charge along its path through a semiconductor and how this charge can be collected by p/n junctions, but what really matters is determining whether the event actually causes an error in circuit operation. In the following subsections we’ll study how charge collection interacts with the circuit type and design to create a single-event upset. The focus will primarily be on memory circuits, but we’ll briefly address SEU in other circuit types at the end of the section. SEUs in terrestrial electronics were first observed in DRAMs [73,74]. DRAMs have historically been quite susceptible to soft errors because they rely on passive storage of charge to represent information. Their charge state is readily modified by funnel-assisted drift or diffusion following an energetic particle strike. DRAMs have therefore received less use in space systems as engineers have preferred SRAM technologies [75]. As the need for very large amounts of onboard memory is increasing, the use of DRAM technologies in space systems is becoming more common [76-81]. DRAMs are prone to SEU due to three primary mechanisms: storage cell errors, bit-line errors, and a combination of the two. The reader is directed to the review article by Massengill [75] for an excellent in-depth summary of SEU effects in DRAMs. 3.3.1
Storage Cell Errors
Figure 3.7 illustrates the mechanism for storage cell errors in a field plate capacitor DRAM [75]. In this kind of DRAM a stored “0” is represented by electrons occupying a potential well under the field plate, while a stored “1” corresponds to electrons being depleted under the plate. Following a particle strike, electrons can be collected at the reverse-biased field plate. In the case of a stored “0”, this just reinforces the original state, but a stored “1” can look like a stored “0” after electron collection. DRAM storage cell errors therefore can characteristically show a very II-17
large pattern dependence. Note also that data patterns may be inverted internally depending on the memory, so that “0” to “1” errors are observed externally [73,74]. Two other effects can lead to storage cell errors in DRAMs. The ALPEN effect described in Section 3.2.2 can lead to channel conduction that shunts charge onto the storage cell, disturbing the cell information [64]. In DRAMs with closely spaced trench storage cells, charge can be transferred between two storage nodes by a particle track shunt effect [82]. In this case, the particle track intersects neighboring storage nodes in opposite memory states, and the resulting charge transfer between nodes can upset one or both storage nodes. 3.3.2
Bit-Line Errors
Bit-line errors can result if sufficient charge is collected during a read cycle to create a charge imbalance on the precharged bit lines [75]. Because they can only occur during a read cycle, bitline errors have a direct dependency on the read access frequency, with an increasing error rate as the access frequency increases [83,84]. Bit-line errors can be caused by strikes to the access transistor drains along the floating bit-line or strikes to the sense amplifier circuitry itself [75,85]. Because of their inverse dependency on cycle time, bit-line errors become a significant reliability concern as memory speeds increase. 3.3.3
Combined Cell and Bit-Line Errors
A new failure mode for DRAMs was demonstrated ten years ago when it was found that charge collection at both the storage cell and bit line that was insufficient to individually cause an upset could cause an error in combination [86]. This new failure mode, dubbed the combined cell-bit line (CCB) error, was shown to dominate the storage cell and bit-line error rates for very short cycle times. The three components of soft errors in a 512K DRAM are shown in Figure 3.8 as a function of the cycle time. Note the independence of storage-cell errors on cycle time, and the domination of CCB errors for short cycle times [86].
Ion Path Bit Line
Poly Word Line Poly
Field Plate Poly +
+ -+
+
n+
n+
Stored Information Charge
Figure 3.7. Illustration of storage cell SEU in a field-plate DRAM after Massengill [75]. Collection of electrons at the reverse-biased field plate reinforces a stored “0,” but can lead to an upset of a stored “1.” II-18
104 Source: AM241
Soft Error Rate (arbitrary units)
103
102
10
1
CCB Component Bit Line Component Cell Component
10-1 10
102
103
104
105
Cycle Time (ns)
Figure 3.8. Components of soft-error rate in a DRAM [86]. The storage cell component is not dependent on the cycle time, while soft errors involving the bit lines increase dramatically as the cycle time decreases. 3.4
Single-Event Upset Mechanisms in SRAMs
The upset process in SRAMs is quite different from DRAMs, due to the active feedback in the cross-coupled inverter pair that forms a typical SRAM memory cell. When an energetic particle strikes a sensitive location in a SRAM (typically the reverse-biased drain junction of a transistor biased in the “off” state [70,87], for example the “off” n-channel transistor shown in Figure 3.9), charge collected by the junction results in a transient current in the struck transistor. As this current flows through the struck transistor, the restoring transistor (“on” p-channel transistor in Fig. 3.9) sources current in an attempt to balance the particle-induced current. Unfortunately, the restoring transistor has a finite amount of current drive, and equally importantly, a finite channel conductance. Current flow therefore induces a voltage drop at the drain of the restoring transistor. This voltage transient in response to the single-event current transient is actually the mechanism that can cause upset in SRAM cells. The voltage transient is essentially similar to a write pulse, and can cause the wrong memory state to be locked into the memory cell. In SRAM cells, there are four possible sensitive strike locations, namely the four transistor drains interior to the SRAM circuit. An important consideration for charge collection is whether the junction is located inside a well or in the substrate [70,87]. A diagram of mechanisms for each drain strike location in an n-substrate technology is shown in Figure 3.10. Generally speaking, in each case the transient resulting from an ion strike has a quick initial response followed by a slower, sustained current mechanism [70]. For example, in the familiar outsidethe-well “off” strike (Figure 3.10a), there is an initial drift photocurrent (ip ) followed by diffusion II-19
VDD “on” p-channel (restoring transistor)
Rfb
Ion Strike
Feedback
Rfb
“off” n-channel (struck transistor)
Figure 3.9. Competition between the feedback process and the recovery process governs the SEU response of SRAM cells. current (id ). Both currents are directed such that they tend to raise the struck node voltage and cause SEU. This is the most sensitive strike location for most technologies [87]. In the outsidethe-well “on” strike (Figure 3.10b), the initial drift current again raises the node voltage, but this reinforces the stored logic state. As the node rises, the struck junction becomes slightly forwardbiased, and a conduction (diffusion) current flows opposite to the initial drift current, returning the SRAM to its initial state. This strike location does not produce SEU. Inside-the-well strikes are particularly interesting because of shunt and bipolar effects that can occur in multilayer structures [67]. For the inside-the-well “off” strike (Figure 3.10c), the initial drift current pulls down the struck node potential, initiating the upset process. As the transient proceeds, holes deposited in the p-well are collected at the p-well ties, raising the well potential and leading to injection of electrons by the source [67,68,70,88]. This initiates the inside-the-well bipolar effect discussed in Section 3.2.2 and illustrated in Fig. 3.6. Electrons collected by the substrate do not contribute to upset because the substrate is attached to VDD . However, electrons collected by the n-drain constitute a bipolar current in the same direction as the initial photocurrent, and do contribute to the upset process [70]. For small geometries, the inside-the-well “off” strike can become an important mechanism. In the inside-the-well “on” strike (Figure 3.10d), the ion strike bridges two n-type regions that are at an initial potential difference of VDD . The result is a shunt current that quickly raises the struck node voltage and starts the upset process [89]. This mechanism is self-limiting. As the struck node voltage rises, the potential difference across the shunt disappears. Similar to the inside-the-well “off” strike, holes collected at the p-well ties initiate a bipolar effect, but this time the bipolar current tends to restore the node to its original state. For small geometries, the bipolar effect is strong enough to provide a great deal of intrinsic protection from such strikes [70]. Interestingly, incident particles far below the upset threshold are often sufficiently ionizing to induce a momentary voltage “flip” at the struck node of an SRAM. For example, Figure 3.11 shows drain voltage transients in an SRAM for a particle strike with LET well below the upset II-20
threshold, just below the upset threshold, and just above the upset threshold. Even the particle with LET well below the upset threshold causes a significant voltage transient at the struck drain. Whether an observable SEU occurs depends on which happens faster: the feedback of the voltage transient through the opposite inverter, or recovery of the struck node voltage as the single-event current dies out [63,89,90]. Note that in terms of the fundamental charge-collection characteristics discussed in Section 3.2, drift (including funneling effects) is responsible for the rapid initial flip of the cell, while long-term charge collection by diffusion prolongs the recovery process; both mechanisms are critical to the upset process. The recovery time of an SRAM cell to a particle strike depends on many factors, such as the particle LET, the strike location, etc. From a technology standpoint, the recovery time depends on the restoring transistor current drive and minority carrier lifetimes in the substrate [89,90]. A higher restoring current leads to a faster recovery time, as do decreased minority carrier lifetimes. a) Outside-the-well OFF Strike
b) Outside-the-well ON Strike
VDD
VDD VDD
n-epi p+
VDD
p+
ip
n-epi p+
p+
id
ip
id
VDD➟VDD+
0➟VDD
ip = drift photocurrent id = diffusion current ib = bipolar current
c) Inside-the-well OFF Strike VDD
is = shunt current
VDD
VDD
n+ p-well n-epi
VDD
VDD➟0
0➟VDD➟0
ip
is
ib
n+
n+ p-well n-epi
VDD
Figure 3.10.
d) Inside-the-well ON Strike
ib
n+
VDD
Illustration of possible SRAM upset strike locations and mechanisms [70]. II-21
5.0
Struck Drain Voltage (V)
Threshold LET ≈ 42 MeV-cm2/mg
4.0 2
LET = 9 MeV-cm /mg
3.0 2.0
2
LET = 40.5 MeV-cm /mg
1.0
2
LET = 45 MeV-cm /mg
0.0 10-14
10-13
10-12
10-11
10-10
10-9
10-8
10-7
10-6
Time (s) Figure 3.11. SRAM struck drain voltage transients for ion strikes with LET well below, just below, and just above the SEU threshold. Even the ion strike with LET well below the actual SEU threshold is sufficiently ionizing to momentarily flip the struck node voltage. This is because a higher restoring current is more quickly able to re-establish the struck node voltage, while decreased substrate minority carrier lifetimes reduce the diffusion current at the struck node. The cell feedback time is simply the time required for the disturbed node voltage to feed back through the cross-coupled inverters and latch the struck device in its disturbed state. This time is related to the cell write time and in its simplest form can be thought of as the RC delay in the inverter pair. This RC time constant is thus a critical parameter for determining SEU sensitivity in SRAMs – the smaller the RC delay, the faster the cell can respond to voltage transients (including write pulses) and the more susceptible the SRAM is to SEU. Obviously this has implications for the sensitivity of future, higher speed technologies, as discussed in Section 6. 3.5
Single-Event Upset in Other Circuit Types
Although we have concentrated on SEU in memory circuits, the reader should be aware that SEUs can occur in almost any integrated circuit. For example, SEU is not constrained to digital circuits, but occurs in analog circuits as well. We have already mentioned the problem of SET in optocouplers, which can lead to errors in data transmission [23,28]. Errors are observed in many analog circuit types, including operational amplifiers [91-93], comparators [93,94], and analogto-digital converters (ADCs) [95-97]. Upsets in ADCs are interesting in that analog errors are observed as corruptions in digital output codes [98]. Figure 3.12 shows an example of analog SEU in a 12-bit ADC, where the SEUs appear as a distribution of erroneous codes around the expected output (4 codes centered at 1428 that were removed in this plot for clarity [95]). An excellent summary of SEU in analog ICs is found in [98]. II-22
Figure 3.12. SEU in a 12-bit analog-to-digital converter [95]. Errors are indicated by deviations from the expected digital output codes (4 codes centered at 1428 that were removed in this plot for clarity). SEUs also occur in digital circuits other than memories, prime examples being microprocessors [99-103] and digital signal processors [103,104]. These digital logic circuits are very difficult to test for SEU due to their complexity, especially for advanced high-speed technologies [105]. Errors in logic circuits are also very sensitive to critical timing windows and logic paths, and may never propagate to the output pins [106-109]. As circuit operating speeds continue to increase, the probability of a momentary glitch in a line voltage from a single-event transient (SET) propagating through a logic path to become an observable error rises. Given the ever-growing demand for higher functionality, faster speeds, and greater integration in tomorrow’s space systems, SEU testing increasingly complex logic ICs will be one of the biggest challenges for the future. 3.6
Single-Event Multiple-Bit Upset
Single-event multiple-bit upsets (MBUs) occur when a single particle strike causes more than one error in an IC. For example, diffusion of charge to closely-spaced junctions can upset more than one bit in both SRAM and DRAM cells [60,110]. For a particle striking an IC at a grazing angle of incidence, the charge track may intersect several sensitive regions and cause multiple upsets [111]. The effects of MBU are typically alleviated by a combination of error-correcting codes that work on a word-by-word basis (see Section 3.8.2) and layout rules that prevent physically-adjacent bits from belonging to the same word of memory. Still, single-word multiplebit upsets (SMUs) can occur and pose a substantial threat to system integrity [112]. MBUs have been observed in on-orbit spacecraft data [113,114].
II-23
Multiple-bit upset in DRAMs has been studied fairly extensively using experiments and 2D cylindrically-symmetric simulations [110,115,116]. Figure 3.13 shows an example error map obtained from heavy-ion exposure of a 256Kb DRAM. MBU error locations are shown as solid capacitor areas, surrounded by the outlines of adjacent capacitors where no errors were observed [110]. Simulations showed that for normal-incidence strikes the charge-collection mechanism responsible for MBU in DRAMs was lateral diffusion of the ion-induced charge to multiple junctions. Funneling can increase charge collected at the struck node, reducing the charge available for diffusive transport to adjacent nodes [115]. Zoutendyk et al. found that for the small capacitances typically associated with individual storage elements, funnel-assisted drift collection for direct hits on a storage node is very limited. For direct bit-line hits with larger nodal capacitance, however, substantially reduced charge collection at adjacent nodes was observed [110]. Ion tracks that do not hit a junction therefore allow the greatest amount of lateral charge transport and constitute the worst-case strike location for MBU [110]. The issue of MBUs in SRAMs has also been studied using simulations [60,117]. In [60], two adjacent SRAM cells were simulated, and the strike location was varied to determine the region of sensitivity to multiple-bit upset from normally-incident heavy ion strikes. A simulated error map was produced showing the region of vulnerability to MBU, as shown in Figure 3.14. Only strikes incident between the two cells were capable of producing MBU for the ion strikes simulated in this work. Strikes directly incident on the drain of one cell resulted in upset of the struck cell, but insufficient charge was collected at the adjacent cell to cause upset. The mechanism for multiple-bit upset in the simulated SRAMs thus was found to be diffusion from between-cell strikes, similar to the DRAM results [60]. For grazing angles of incidence and very closely-spaced cells, however, drift collection and funneling can be expected to play greater roles.
Figure 3.13. Physical map of a cluster of errors due to single-event multiple-bit upset in a 256K DRAM [110]. Error locations are shown as solid black capacitor regions, while outlined capacitor regions indicate bits without errors. II-24
4 µm D ra in
xx
x
xx
D ra in
x x x
T1
G a te
x x x x x x
G a te
T2
S o urc e
S o urc e
S trik e h e re u p s e ts bo th c e lls S trike h e re up s e ts a s in g le ce ll
Figure 3.14. Simulated multiple-bit upset map for two adjacent SRAM cells [60]. For this SRAM, diffusion of charge from ion strikes between cells is responsible for MBU.
3.7
Particle Energy Effects
Heavy-ion tests at accelerator facilities are frequently performed to study mechanisms of SEU, estimate on-orbit error rates, and qualify parts for use in space-based systems. In most facilities used for such SEU testing, the energy of the particles is on the order of a few (1-10) MeV per nucleon (or equivalently, MeV/amu). However, in the actual space environment, the energy of particles reaches hundreds of GeV/amu, with a peak flux at a few hundred MeV/amu [1]. Figure 3.15 shows the galactic cosmic ray energy spectrum for Fe ions in a geostationary orbit at solar minimum conditions behind 100 mil of Al shielding [118]. The range of ion energies available at typical low-energy accelerators such as the Tandem Van de Graaff at Brookhaven National Laboratory (BNL) is marked by the light gray band. Higher energy ions (10-100 MeV/amu) are available at a few facilities, for example the GANIL cyclotron in France and the National Superconducting Cyclotron Laboratory (NSCL) at Michigan State University. The range of ions available at these facilities is marked by the darker gray band in Figure 3.15. Note that even the “high” energy facilities produce ions with energies well below the maximum encountered in space. The linear energy transfer (LET) of the Fe ions as a function of energy is denoted in Figure 3.15 by the shading of the symbols. The usual low-energy test facilities produce ions with energies where they have maximum LET (greater than 20 MeV-cm2/mg for the case of Fe) but lower fluences in space. Ions with intermediate energies (10-100 MeV/amu) are much more abundant and can still possess large enough LETs (between 5 and 20 MeVcm2/mg for Fe) to be of concern, especially for sensitive commercial technologies. Note also that spacecraft shielding becomes ineffective with increased ion energy [1]. Extremely high-energy ions usually have low LETs (less than 0.2 MeV-cm2/mg for Fe) and therefore do not pose a serious SEU threat except for very sensitive devices. At a given LET, a high-energy cosmic ion track will have a much larger radius and much longer range than a low-energy accelerator ion track. This might lead to different chargeII-25
Ion Fluence (#/cm2/day/MeV/amu)
100 2
Ion LET (MeV-cm /mg) 0.2 5
10-1
20
10-2 10-3 10-4
"High" Low Energy Energy Test Test Range Range (GANIL (BNL) & MSU)
10-5 10-6 10-1
100
101
102
103
104
105
106
Energy (MeV/amu) Figure 3.15. Galactic cosmic ray iron spectrum vs. energy with LET denoted by symbol shading [118]. Ion energies available at typical accelerator facilities range from approximately 1-100 MeV/amu. collection properties, and hence, different SEU response. Recent SEU tests at high-energy accelerator facilities have indeed suggested a difference in IC response to particles with different energies. For example, differences between low- and high-energy ion tests have been seen to date in SEU thresholds and saturation cross-sections [119,120], single-event multiple-bit upset (MBU) occurrence [111,121,122], and unexplained inconsistencies in single-event burnout data [123]. Yet, in some cases, no appreciable difference has been observed between low- and highenergy heavy ion data [124-126]. The results that show differences are worrisome in that they raise concerns about the fidelity of accelerator-based tests for simulating the response of parts to the real high-energy ion environment found in space [127,128]. There are certainly cases where low-energy and high-energy ions have unambiguously given different test results. In these cases, however, it is the greatly increased range of high-energy ions that has led to differences rather than radial track structure effects. For example, in grazing angle irradiations, the greatly increased track-length of high-energy ions leads to multiple-bit upsets of many more cells along the track than would be seen with the shorter range of low-energy ions [111,121]. There may be other cases where the larger range of high-energy ions leads to different results, such as cases where the ions must pass through a great deal of material to reach the sensitive region [129]. Devices with vertical structures (e.g., power MOSFETs) may be sensitive to charge collection deep within the substrate and may exhibit interesting energy effects [130]. A recent study using carefully controlled experiments demonstrated that for four CMOS SRAM technologies, no significant differences were observed between low- and high-energy II-26
heavy ion SEU data [118]. For example, Figures 3.16 and 3.17 show low- and high-energy data for two of the tested technologies, a 256Kb SRAM and a 1Mb SRAM. Within experimental uncertainty, the cross-section data measured with low-energy ions (open symbols) and highenergy ions (solid symbols) appear the same. In the few inconclusive cases where there may have been small differences, the low-energy ion data were always conservative compared to the highenergy data. This suggests that standard low-energy ion testing is sufficiently (but not overly) conservative for CMOS SRAM technologies down to 0.5 µm (the smallest technology tested). The results were explained by the fact that the track structures of low- and high-energy ions are not greatly different except in the outer regions where they have low charge density. Due to the large radial extent of the track, high-energy ions might be expected to cause more MBUs than equivalent low-energy ions. The authors of [118] therefore examined their highenergy SEU data for evidence of MBUs. None of the almost 4000 error maps taken at low- and high-energy facilities showed physically adjacent upset locations. Unfortunately, only about 400 of the error maps were from high-energy facilities, but none of these maps gave evidence of an increase in normal-incidence MBUs due to high-energy ions. Of course, at grazing angles of incidence we would expect a big difference between low- and high-energy MBU results due to range effects. In the future it is possible that there will be devices sensitive to the low induced charge densities in the outer regions of high-energy ion tracks, but this does not appear to be the case yet. If such effects do appear, they may manifest themselves as an increase in the cross-section for high-energy ions as devices become more sensitive to indirect strikes that do not hit the junction [118]. These results validate the continued use of standard low-energy heavy ion test facilities such as the BNL Tandem Van de Graaff and the Berkeley 88" Cyclotron for part testing
Cross-section (cm2 )
100 10-1 10-2 Matra 65656 256Kb SRAM All SNs VDD = 4.5 V
10-3 10-4
Brookhaven GANIL MSU
-5
10
10-6 0
10
20
30
40
2
Effective LET (MeV-cm /mg) Figure 3.16. High- vs. low-energy SEU cross-sections for 256Kb SRAMs manufactured in a 0.8-µm radiation-tolerant technology [118]. II-27
Cross-section (cm2 )
100 10-1 10-2 Matra 65608E 1Mb SRAM VDD = 4.5 V
10-3 10-4
Brookhaven GANIL MSU
-5
10
10-6 0
10
20
30
40
2
Effective LET (MeV-cm /mg) Figure 3.17. High- vs. low-energy SEU cross-sections for 1Mb SRAMs manufactured in a 0.5-µm radiation-tolerant technology [118]. and qualification. This is important because the low-energy facilities are in general much better developed as user facilities, more accessible, and considerably cheaper to use. There are cases (e.g., grazing angle of incidence studies) where the use of high-energy ions is essential because of their increased range, and it is possible that in the future energy effects will become more significant. It is also possible that other device types or materials will prove to be more sensitive to energy effects than the Si CMOS SRAMs tested here. Based on the similarity in low-energy and high-energy ion track structures this seems unlikely, but cannot be positively ruled out. For the present, it appears that in most cases standard low-energy heavy-ion test methods are adequate to qualify similar technologies for current spacecraft use. 3.8
Mitigation Techniques
SEU mitigation techniques can be roughly classified into three distinct categories. Systemlevel techniques deal with SEU at the system architecture level. Circuit-level techniques rely on changes in the circuit design to reduce SEU sensitivity. Technology- or device-level hardening requires fundamental changes to the underlying fabrication technology used to manufacture ICs. Because they flow directly from the basic mechanisms of SEU, we will concentrate on technology hardening methods and give only a brief description of circuit- and system-level hardening. A review of several hardening techniques is also found in [131]. 3.8.1
Technology Hardening
The most fundamental method for hardening against SEU is to reduce charge collection at sensitive nodes. This can be accomplished in DRAMs and SRAMs by introducing extra doping layers to limit substrate charge collection [132]. In advanced SRAMs, triple-well [133] and even II-28
quadruple-well [134] structures have been proposed to decrease SEU sensitivity. Retrograde wells and buried layers can also be used to provide an internal electric field that opposes substrate charge collection [135,136]. Even the simple use of an epitaxial substrate instead of a bulk substrate affords some level of reduced charge collection [60,83]. In GaAs-based heterojunction insulated gate FETs (HIGFETs), the use of GaAs buffer layers grown at low temperatures (LT GaAs) has proven a very effective means of limiting charge collection [44,137,138]. The LT GaAs buffer layer prevents charge-enhancement mechanisms and substantially reduces charge collection (by as much as a factor of 100 [44]) because of its subpicosecond minority carrier lifetimes [139]. Figure 3.18 shows measured drain charge-collection transients for 3-MeV α-particle strikes, illustrating the dramatic reduction in charge collection for the LT GaAs HIGFET [44]. An effective technique for reducing charge collection in silicon devices is the use of SOI substrates [140]. In this case the collection volume is reduced by the fact that the active device is fabricated in a thin silicon layer that is dielectrically isolated from the substrate. A diagram of a thin-film partially-depleted SOI transistor is shown in Figure 3.19. In a typical thin-film SOI device, the source and drain penetrate all the way to the buried isolation oxide (BOX). This substantially reduces the SEU-sensitive area, because the reverse-biased drain junction area is limited to the depletion region between the drain and the body of the transistor, as shown in Fig. 3.19. Charge deposited in the silicon substrate underneath the BOX cannot be collected at the drain due to the dielectric isolation. Unfortunately, as briefly mentioned in Section 3.2.2, charge deposited in the body region (for example, by a particle strike to the gate region) can trigger a bipolar mechanism that limits the SEU hardness of SOI circuits [71,140]. Following a
Figure 3.18. Measured charge-collection transients in LT GaAs and conventional GaAs HIGFETs following a 3-MeV alpha-particle strike [44]. Charge collection is greatly reduced by the incorporation of the LT GaAs buffer layer. II-29
Gate
STI
n+ Source
n+ Drain
p Body Buried Oxide
~2000 A STI ~2000 A
Substrate
Figure 3.19. Diagram of a thin-film partially-depleted n-channel SOI transistor. The chargecollection volume is greatly reduced because the top silicon active layer is very thin. Unfortunately, ion strikes to the body region can initiate a bipolar conduction mechanism that limits the SEU hardness of SOI circuits. particle strike to the body of an n-channel SOI transistor, electrons can be collected at the source and drain electrodes. Holes can only escape through the body tie contact, if there is one, or slowly through recombination if there is no body tie. Residual holes left in the body raise the body potential and trigger the lateral parasitic bipolar transistor inherent to the SOI transistor, with the source serving as the emitter, the body as the base, and the drain as the collector. This bipolar current considerably lowers the SEU hardness of SOI and can even lead to a single-event induced latch-up condition in SOI transistors without body ties if the bipolar conduction is sustained [72]. Body ties are sometimes used commercially to reduce floating-body effects under DC operation, and careful attention to body tie design is crucial to maintaining good SEU performance [71,141146]. Even in body-tied SOI designs, manufacturers have found it necessary to incorporate other hardening methods for applications where very high upset thresholds are desired [147,148]. Fully-depleted SOI transistors exhibit reduced floating-body effects, and have shown excellent SEU performance [149]. Given an understanding of the upset mechanism in SRAMs, we can immediately understand another typical technology-level technique used to harden SRAMs against SEU. We have discussed (Section 3.4) how the SEU process in an SRAM is essentially a race condition between the feedback and recovery processes. To harden an SRAM, we need to either slow the feedback process or decrease the recovery time. The feedback process can be slowed by adding either resistance or capacitance to the feedback loop [150]. Cross-coupled feedback resistors (see Fig. 3.9) are the classical method of increasing the cell feedback time by increasing the RC delay in the feedback loop. Unfortunately, because the SEU process in SRAMs looks just like the write process, this same RC delay directly impacts the write pulse width of the SRAM. The resistive decoupling technique is very effective, as illustrated in Figure 3.20. This figure shows the measured SEU cross-section (area of the chip sensitive to SEU) as a function of LET for a microprocessor before and after hardening by resistive decoupling [101]. The predicted upset rate in a geosynchronous orbit for the unhardened microprocessor is about once a day, while in the resistively-hardened part it is about once per century. Unfortunately, the effectiveness of resistive hardening does not come without a price. We have already mentioned the speed penalty incurred by adding feedback resistors. Probably even
II-30
Figure 3.20. Measured SEU cross-section of a 16-bit commercial microprocessor showing the effectiveness of resistive hardening for mitigating SEU [101]. worse than this penalty is the increased process complexity introduced by adding feedback resistors. These resistors are typically implemented in the cell layout as lightly-doped polysilicon regions. Because the resistivity of polysilicon is very sensitive to the doping concentration and numerous other factors, it is very difficult to control the resistor value [151]. Rockett reports that SEU feedback resistor variance may be as much as half the mean value [152]. To add to this problem, polysilicon resistivity has a negative temperature dependence. This leads to a temperature dependence of both the write speed and the SEU response. As low temperature, the feedback resistance increases, providing good SEU performance, but slowing the maximum operating frequency. At higher temperature, the part operates faster, but the SEU threshold drops. These characteristics make it challenging to optimize a resistively-hardened technology so that it will operate within specifications across a wide temperature range. A variation on the resistive hardening method is the use of gated resistors (essentially thinfilm polysilicon transistors) as a feedback element [152]. The advantage of this approach is that during a write pulse the feedback resistance can be lowered to reduce the impact of the feedback element on operating speed. A diagram of the gated resistor-hardened SRAM cell is shown in Figure 3.21. During a write operation, the resistor gates are clocked into a low-resistance condition by the wordline pulse, greatly improving the write speed over a standard resistivelyhardened SRAM. At all other times the resistors are in their “off” (high-resistance) state and protect against SEU in the usual manner. The gated-resistor SRAM is therefore faster than a standard resistively-hardened SRAM, has reduced temperature dependence, and may require no additional area to implement [152]. It does, however, require two levels of polysilicon, which although fairly common, are not present in all technologies. II-31
Figure 3.21. Diagram of a gated-resistor hardened SRAM [152]. Clocking the gated resistor with the wordline reduces the feedback resistance during write pulses, thereby minimizing the timing penalties usually associated with feedback resistors. A similar, but distinct, approach has been used to harden high-density SRAMs in SOI [153]. In this approach an extra transistor is fabricated in the silicon device layer in parallel with a standard cross-coupled polysilicon feedback resistor. Under normal operation, this transistor is off, and the feedback signal must pass through the resistor in the usual manner. During a write pulse, the transistor is turned on, effectively shorting out the feedback resistor and maintaining the write performance of the memory cell. Other decoupling techniques have been proposed that place resistors or diodes in different locations [154-157], usually for the purpose of reducing the impact of the resistors on timing parameters or increasing manufacturability. For the most part these techniques have not been widely used (if at all) and have their own associated tradeoffs. Capacitors have been successfully used as a feedback element in SOI SRAMs [140,147,148], and very recently as a means to improve the soft-error performance of deep-submicron CMOS SRAMs for terrestrial applications [158]. While adding capacitance still degrades timing parameters, one advantage is reduced temperature-dependence compared to resistive hardening. 3.8.2
Circuit- and System-Level Hardening
Because of the invasive nature of device-level hardening (i.e., the requirement for fundamental changes in the manufacturing process), easier to implement methods have been a frequent topic of research. The goal of circuit-level hardening techniques is to design circuits that are inherently SEU-hardened and can therefore be manufactured using any commercial fabrication process. Also sometimes referred to as design-level hardening, this can be an important technique in cases when parts from a radiation-hardened technology are not available II-32
[159]. Several design-hardened SRAM and latch circuits have been proposed and fabricated [160-165]. These memory cells typically rely on redundant circuit elements (usually 12-16 transistors per memory cell as opposed to 6 in a standard unhardened cell) to protect against SEU. Because of the large number of transistors per cell, these designs consume considerably more area (and frequently more power) than 6 transistor cells. While these cells can be appropriate for protecting a limited number of critical data paths, they are not usually suitable for very highly-integrated circuits. A design-hardened SEU tolerant DRAM cell that is comparable in size to conventional single transistor DRAMs has also been proposed [166]. System-level hardening approaches include the use of error detection and correction (EDAC) circuitry to monitor and correct errors as they occur [167-169]. This approach requires that extra bits of information be stored with the data to reconstruct the original data in the event of an upset. System overhead can be non-negligible, but this is sometimes the only method available if relatively susceptible parts must be used. Cruder methods such as watchdog timers, redundancy, lockstep operation, majority voting, etc., are commonly used to detect control system errors [170]. While simple in concept, overhead can again be significant, but in many cases this is the only option available to the spacecraft engineer. Last year’s NSREC Short Course segment by Kinnison and this year’s segment by Heidergott contain additional information on system-level hardening methodology [170,171].
II-33
4.0
BASIC MECHANISMS FOR DESTRUCTIVE SINGLE-EVENT EFFECTS
In this section we will briefly discuss mechanisms for destructive SEE, namely single-event latchup (SEL), single-event gate rupture (SEGR), and single-event burnout (SEB). This is not intended to be a complete list of destructive SEE failure modes, but are some of the more commonly encountered effects. For a more comprehensive discussion of destructive SEE, the reader is directed to the recent NSREC Short Course segment by Johnson and Galloway [172]. 4.1
Single-Event Latchup
It is well-documented that in semiconductors with a p/n/p/n structure, a high-current destructive failure mode known as latchup can occur [173]. Unfortunately, just such a structure occurs in all CMOS technologies when an n-channel transistor is located next to a p-channel transistor, such as in a CMOS inverter used in an SRAM. The p/n/p/n structure can be thought of as a connection of two parasitic bipolar junction transistors (BJTs), as illustrated in Figure 4.1 [174]. For the CMOS structure shown in this figure, a parasitic lateral npn transistor is formed by the n-channel source (emitter), p-substrate (base), and n-well (collector). Similarly, a parasitic vertical pnp transistor is formed by the p-channel source (emitter), n-well (base), and p-substrate (collector). Note that the collector of each parasitic BJT is the base region of its counterpart. This leads to a positive feedback loop between the two transistors, although in normal operation both transistors are in an “off” condition. Depending on interdevice spacing, junction depths, doping densities, etc., the parasitic BJTs can be turned “on” and the positive feedback loop can be activated by a device overvoltage, improper sequencing of power supplies, or the presence of a large substrate or n-well current [173]. Once triggered, the device goes into a sustained highcurrent mode that can destroy the device due to thermal runaway or failure of metallization [174].
Figure 4.1. Illustration of the parasitic latchup structure inherent to bulk CMOS technologies [174]. II-34
Usually it is necessary to reduce or remove power from a device in a latched state to return it to normal operation before destructive failure. 4.1.1
Single-Event Latchup Mechanism
The astute reader will note that one of the triggering mechanisms for latchup listed above was well or substrate currents. This is, of course, exactly what one has in the case of an energetic particle strike, so it should not be surprising that latchup can indeed be initiated by a single-event strike. Single-event latchup (SEL) was observed in ground tests as early as 1979 [175], and at first was limited to heavy ions and bulk technologies. As technology parameters evolved, it was found that some devices fabricated on epitaxial substrates could exhibit latchup [176], and worse still, that protons could induce latchup in sensitive technologies [177-179]. It is beyond our purpose here to give a comprehensive discussion of all that is known on SEL. The reader is referred to many excellent reviews and analyses of the subject for further information [172,174-178,180-186]. As mentioned above, the latchup condition is triggered when the parasitic BJTs inherent in CMOS turn “on”, establishing a positive feedback loop. For SEL, the triggering involves four distinct steps, as described by Johnston, et al. [174]: 1) The particle strike induces a transient current in the well-substrate junction, which in turn results in a voltage drop within the well. The magnitude of the voltage drop depends on the resistivity of the well, well depth, and distance of the strike from the well tie contact. 2) If the voltage drop in the well is large enough, the vertical parasitic BJT becomes forward biased and magnifies current flow into the substrate due to bipolar amplification. 3) Current flowing in the substrate from the vertical BJT produces a voltage drop in the substrate, which in turn forward biases the lateral BJT. Collector current from the lateral BJT provides base current to the vertical BJT, initiating the positive feedback process. 4) Once the regenerative feedback process begins, current increases rapidly and the device can remain latched even after the subsidence of the original particle-induced current. References [174,180,186] give further details of the charge-collection process and initiation of SEL. 4.1.2
Mitigation Techniques
Like almost any single-event effect, techniques have been developed to mitigate the occurrence of SEL. One difference with SEL is that since latchup can also be initiated by electrical signals in normal terrestrial ICs, it is not unique to the space environment. This means that high-volume commercial IC manufacturers also have to deal with latchup on their own accord. Unfortunately, it does not follow that just because an IC is latchup immune in normal environments it will also be latchup free in space [174]. However, the same general techniques that improve latchup characteristics on the ground can usually be expected to improve the SEL response in space environments. The keys to decreasing latchup susceptibility are to decrease the ability of the parasitic BJTs to become forward biased, and to decrease the current gain of the BJTs should they become forward biased. Factors which make it harder to forward bias the BJTs include increasing the substrate and well doping levels (heavier doping decreases series resistance and hence limits II-35
voltage drops) and increasing the well thickness (thicker decreases resistance and again limits voltage drops). Using an epitaxial substrate can decrease latchup sensitivity both by limiting the charge-collection volume, and by decreasing the substrate series resistance. Increasing the spacing between devices lowers the current gain of the lateral BJT (basically by increasing the base thickness), and increasing the well thickness can do the same for the vertical BJT. While at first it seemed that fabrication on epitaxial substrates was enough to eliminate latchup, this can no longer be considered to be the case. Figure 4.2 shows a compilation of data where the latchup threshold LET for several devices and technologies is plotted against the epitaxial layer thickness. The data were compiled from [186,187] after the method of [174]. Because latchup depends on a diverse set of process and layout parameters, there is no clear trend, and in fact one of the most SEL-sensitive parts ever tested is an AMD K-5 microprocessor (latchup threshold = 0.4 MeV-cm2/mg) fabricated on a very thin (2.5 µm) epitaxial substrate [186]. This unexpected result indicates that SEL could become a serious concern for advanced devices that may be designed with very little margin to the latchup point. Other technology changes which decrease latchup susceptibility include the use of retrograde wells and trench isolation [188,189]. The effectiveness of trench isolation for decreasing latchup sensitivity depends heavily on the trench depth, with deeper trenches preventing latchup better than shallow trenches. In technologies such as CMOS/SOI or CMOS/SOS, the latchup path can be completely eliminated due to total dielectric isolation between adjacent devices [190]. Guard bands have been used to lower latchup sensitivity [180,191,192], as have neutron-irradiated substrates to kill parasitic BJT gain [193].
AMD K-5 (0.35 mm)
Figure 4.2. Latchup threshold as a function of epitaxial layer thickness for several technologies [174]. The AMD K-5 is one of the most single-event latchup-sensitive parts that has been tested, even though it is fabricated on a thin epitaxial substrate. II-36
4.2
Single-Event Gate Rupture
Dielectric breakdown can occur when the electric field across an insulating material exceeds some threshold value. When initiated by an energetic particle strike to the gate region of an MOS device, this phenomenon is referred to as a single-event gate rupture (SEGR). In this section we will present the basic mechanism for SEGR, and discuss recent results obtained in power MOSFETs and devices with thin gate oxides. 4.2.1
Single-Event Gate Rupture Mechanism
Single-event gate rupture has been studied most extensively for power devices such as double-diffused power MOSFETs (DMOS), so we will use this device for describing the SEGR mechanism. As shown in Figure 4.3, current flow in the DMOS structure is vertical rather than lateral as in a standard MOSFET [194,195]. Application of a positive bias to the gate in this nchannel DMOSFET inverts the p-body region to form a channel between the n-source at the top of the structure and the drain (substrate) contact at the bottom of the structure. To handle large currents, the full structure usually contains hundreds or thousands of these cells connected in parallel. The thick lightly-doped epitaxial region allows the power MOSFET to sustain high voltages without breakdown. When an ion strikes the neck region through the gate oxide, SEGR can occur as charge is transported near the Si/SiO2 interface. As charge from the ion strike accumulates underneath the gate region (and depending on the gate bias), the electric field in the gate insulator can temporarily increase to above the critical field to breakdown, causing a localized dielectric failure (i.e., an SEGR). The SEGR response in vertical power MOSFETs has two components [196]. The “capacitor response” describes the interaction of the ion directly with the gate dielectric, inducing an oxide breakdown at a lower field than would occur in the absence of the ion strike. If a drain bias is applied when the ion strike occurs, part of the drain voltage Io n P a th neck
G a te
p+
S o u rc e
n+ p
n+ p
p+
ho le s ele ctron s n ep ila ye r n + s u bs tra te
D ra in
Figure 4.3. Structure of a vertical power MOSFET and current flow paths following a heavy ion strike [195]. II-37
may be transferred through the epitaxial layer to the gate interface [194]. This part of the response is referred to as the “substrate response.” Increasing the gate voltage increases susceptibility to SEGR through the capacitor response by increasing the pre-existing electric field in the oxide. Increasing the drain voltage also increases the susceptibility to SEGR because part of this voltage can be coupled to the interface through the substrate response. 4.2.2
Single-Event Gate Rupture in Power MOSFETs
Typical SEGR data taken for vertical power MOSFETs include the critical field to dielectric breakdown as a function of LET [197]. Note that because SEGR is a destructive event, obtaining such data requires that many devices be destroyed during the experiment. Semi-empirical expressions that relate the voltage threshold for SEGR as a function of bias conditions, oxide thickness, and incident particle LET have been developed for specific devices and fit measured SEGR data very well [197,198]. As implied above, the sensitive region for SEGR in vertical power MOSFETs is the neck region under the gate contact, since this is where high fields exist in an oxide. Much of this gate area is not necessary to device operation, and recent studies have proposed eliminating parts of the gate polysilicon (e.g., the cross-hatched region in Fig. 4.3) to decrease the area sensitive to SEGR [199]. Two interesting recent papers have studied the effects of ion energy on SEGR in vertical power MOSFETs [196,200]. The first of these showed that the substrate response had a clear dependence on ion energy [200]. As ion energy increased, the maximum gate voltage that could be applied without producing SEGR decreased until the ion range was sufficient to penetrate the entire epitaxial layer. At this point the device was most sensitive to SEGR. The results were explained by the fact that this condition resulted in the maximum amount of charge being deposited into the epitaxial layer [200]. The authors noted that at zero drain bias (i.e., no substrate response component), there appeared to be no energy dependence of the SEGR gate bias threshold. This observation spurred further research to explicitly measure the capacitor response as a function of ion energy [196]. In this study, no apparent difference was found in the SEGR gate bias threshold as a function of ion, corroborating the earlier data. Data from this study are shown in Figure 4.4. Note the surprising result that the gate bias at the onset of SEGR does not track with the LET value, appearing to be a constant value regardless of the ion energy. Although the results are still not fully understood, the authors presented a new model for the critical gate field to cause SEGR in terms of the atomic number of the incident ion rather than its LET. Further study is necessary in this field to develop physical explanations for this result. 4.2.3
Single-Event Gate Rupture in Thin Gate Oxides
SEGR effects have been studied for some time in power devices, but a topic that has recently received a considerable amount interest is SEGR in logic and memory ICs. As gate oxide thicknesses decrease, SEGR could become a problem in ICs because they will likely be operated at somewhat higher electric fields. Hard errors have occasionally been observed during heavy ion tests of SRAMs, and have been attributed to local total ionizing dose deposition (“microdose”) [201-203]. In 1994, Swift et al. noted the occurrence of a new kind of hard error in 4 Mbit DRAMs that was clearly inconsistent with the microdose phenomenon [204]. For example, these errors did not disappear with high-temperature annealing and showed different retention-time characteristics than microdose hard errors. Other characteristics of the mechanism seemed
II-38
Figure 4.4. Energy dependence of SEGR failure threshold in vertical power MOSFETs for niobium ions [196]. The failure threshold is roughly constant even though the ion LET varies considerably over the energy range. consistent with SEGR failures, but the mechanism was not conclusively identified. Of particular concern was the fact that 4 Mbit DRAMs from the same manufacturer that had undergone a die shrink (0.8 µm to 0.6 µm) were even more sensitive to SEGR, showing hard errors at lower LET and with a much greater cross-section. The electric fields at which ruptures were observed were less than 3 MV/cm. These results led Johnston et al. to conclude that SEGR could become a major failure mechanism for scaled microelectronics in space [205]. Below a feature size of about 0.25 µm, it was predicted that the threshold LET for SEGR would fall below that of Fe, a significant component of the galactic cosmic ray flux. If this were to happen, the SEGR error rate in space would increase several orders of magnitude, making the usage of deep submicron devices in space problematical. This prediction has inspired recent research into scaling trends and mechanisms for SEGR in thin gate oxides [206-209]. The first of these studies investigated the impact of oxide thickness on the critical electric field to rupture in gate oxide capacitors and CMOS ICs [206]. The authors found much higher critical fields to rupture than in [204]. Their results for gate oxide capacitors with an area of ~10-3 cm2 are shown in Figure 4.5 as a function of particle LET and gate oxide thickness. The critical field for SEGR is higher than 5 MV/cm in all cases, and decreasing the oxide thickness increases the critical field for SEGR at any given LET. Similar trends were observed for 16 Kbit and 256 Kbit SRAMs and 12-stage delay chains, although the critical fields to rupture were somewhat lower [206]. The thinnest oxides in Figure 4.5 have critical fields approaching 8 MV/cm, suggesting that advanced technologies may in fact show improved resistance to SEGR. This improvement in critical field to rupture II-39
12
ECR (MV/cm)
10 8
6.5 nm 6.0 nm
6
12 nm 18 nm
4 2 0 20
30
40
50
60
70
80
90
100
2
LET (MeV-cm /mg) Figure 4.5. Critical field to SEGR in thin oxide capacitors as a function of ion LET [206]. correlated well to improved breakdown field strengths prior to irradiation, due to reduced defect generation by hot carriers in thin oxides. It was cautioned, however, that future voltage scaling is highly uncertain and if fields do exceed 5 MV/cm, SEGR could still be a significant concern. Another issue is that not all gate oxides can be expected to behave the same. Oxides from different processes have in fact exhibited different critical fields to rupture. In [207], for example, capacitors with 4.5-nm and 7.5-nm gate oxides from a different process showed much lower critical fields to SEGR compared to the data of [206], although the qualitative trend was similar. As noted by the authors, these differences due to processing make it difficult to conclusively establish whether in the future SEGR will be a problem for a particular technology operating at a specific oxide thickness and power supply voltage. It does appear, however, that the preirradiation dielectric breakdown field strengths are a reliable indicator of relative SEGR vulnerability. Regardless of processing technology, the published data all confirm that for a given oxide thickness, the higher the pre-irradiation breakdown field, the higher the critical field to SEGR [206,208,209]. Clearly there are some technologies where the critical field to SEGR is in the range of 5-6 MeV/cm, at least for high LET ions. Whether SEGR will in fact be a problem in the future depends on many factors, including what voltage scaling trends are actually adopted and if alternative dielectric materials are used [208]. Recent work has probed not only scaling trends, but also the underlying mechanisms for SEGR in thin gate oxides [208,209]. Based on a limited amount of data for varying ion fluences, earlier work suggested that multiple ion hits might be necessary to initiate SEGR [207]. More extensive data gathered since then conclusively show that there is at most a small effect of fluence per irradiation step on the SEGR threshold. Figure 4.6 shows the critical voltage to rupture measured in capacitors with a 7-nm oxide as a function of the number of ions received per irradiation step [209]. There is only a very weak dependence: 15% difference in critical II-40
voltage to SEGR over more than three orders of magnitude of fluence per step. Two pre-stress points are also marked on the plot. For these two points, capacitors were pre-stressed at zero bias with a fluence of 2×108 ions/cm2. Even after this enormous pre-stress†, the voltage to rupture remained unchanged. These data strongly support a true single-ion model for gate rupture [209]. Although SEGR appears to be a true single-ion effect, there is no question that an accumulated fluence of particles does introduce damage to the oxide. This damage (termed precursor ion damage in [209]) does not affect the SEGR threshold, but does increase capacitor leakage currents. Figure 4.7 shows measured I-V curves for a 7-nm capacitor after exposure to successive fluences of Au ions at increasing bias voltages [209]. The fluence step between each curve is 106 ions/cm2. At irradiation bias voltages below 3.7 V, increasing leakage currents beyond 2 V are observed with accumulated particle fluence. This leakage current is associated with precursor ion damage. After 3.7 V was applied during the heavy ion exposure the capacitor’s I-V curve changed dramatically, with the leakage current increasing by four orders of magnitude. This pronounced change in the I-V curve is the characteristic signature of SEGR. The critical voltage to SEGR in this 7-nm oxide capacitor for 360-MeV Au ions is thus 3.7 V. Accumulated fluence damage effects such as those shown before rupture in Fig. 4.7 are an interesting topic to study from a basic mechanisms standpoint. However, the data show that precursor ion damage and SEGR in thin oxides are largely unrelated effects. Also, because of the very low flux of high-LET particles in the near-Earth environment (on the order of 7 Br
6
VCR (V)
5 Au
4 3 2
After prestress of 2E8 ions/cm2 w/ no bias
1
Nominal Fluence
0 1
10
100
1000
10000
Average Ions/Step Figure 4.6. Critical voltage to rupture as a function of heavy ion fluence per step [209]. Only a weak dependence of the rupture voltage is observed over more than three orders of magnitude of particle fluence per step. †
For reference, it would take roughly the present age of the universe to accumulate a fluence of high-LET particles this large in the near-Earth environment.
II-41
10-4
Gate Current (A)
10-5
Area = 0.0025 cm2
Irradiation Bias Pre 3.0 V 3.2 V 3.3 V 3.4 V 3.5 V 3.6 V 3.7 V
10-6 10-7 10-8 10-9 10-10 10-11 10-12 0
1
2
3
4
5
Gate Voltage (V) Figure 4.7. Thin oxide capacitor current-voltage characteristics after successive 106 ions/cm2 irradiations with increasing gate bias [209]. Precursor ion damage is evidenced by the increasing gate current above 2 V as ion fluence was accumulated. The sudden increased current at all voltages for the 3.7-V irradiation is the characteristic signature of SEGR; this capacitor had a SEGR failure threshold of 3.7 V. 0.4 ions/cm2/year with LET ≥ 40 MeV-cm2/mg), precursor ion damage is not expected to have any practical impact on reliability in space. For further information on particle-induced damage effects in thermal and nitrided oxides, the reader is directed to [209]. SEGR has been investigated in oxides thinner than 5 nm [208,209]. These oxides can behave fundamentally differently than thicker oxides. Specifically, they often exhibit a soft breakdown characteristic rather than the hard breakdown shown in Fig. 4.7. For example, 4.5-nm oxide capacitors tested in [208] showed soft breakdown leakage currents on the order of tens to hundreds of µA, compared to 1-10 mA rupture currents observed in identical capacitors with 7.5-nm oxides. Similarly, 5-nm oxide capacitors tested in [209] typically showed only a gradual increase in leakage current similar to precursor ion damage buildup, even at fields as high as 12 MV/cm. For a few capacitors clear rupture signatures were obtained, but these were rare. In cases such as these, the ability to sweep full I-V curves at the irradiation site is invaluable, as simple current monitoring is not very instructive. Interestingly, reliability studies of very thin oxides often show similar soft breakdown characteristics [210]. Another recent result of interest is the angular dependence of SEGR in thin oxides. Traditionally, normal incidence is considered to be the worst-case ion trajectory for SEGR, with the critical voltage to rupture showing a somewhat less than cos(θ) dependence [194]. In early work on SEGR, Wrobel predicted that the angular dependence would disappear when the oxide thickness approached the diameter of the ion track [211]. In [209], the authors investigated the angular dependence of the critical field to rupture as a function of oxide thickness. Their results II-42
are shown in Figure 4.8, where the critical field has been normalized to the value at normal incidence. The results clearly show that the angular dependence of SEGR does indeed disappear for thin oxides. The fact that off-normal ion strikes are effective at producing SEGR increases the threat environment in the isotropic heavy ion environment of space, and this must be accounted for when estimating SEGR susceptibility for thin oxides [209]. The angular data have been explained in terms of ion track structure, and an analytical model for the angular dependence based on a conductive pipe mechanism has been proposed [209]. 4.3
Single-Event Burnout
Single-event burnout (SEB) due to heavy ions, neutrons, and protons has been observed in both vertical power MOSFETs and bipolar transistors [194,195,212-216]. Similar to single-event latchup, SEB is a destructive failure mechanism that comes about due to a parasitic bipolar transistor structure inherent to some devices. Looking again at the power MOSFET structure in Fig. 4.3, a parasitic bipolar transistor is formed by the n-source (emitter), p-body (base), and nepitaxial (collector) regions. Following an ion strike, currents flowing in the p-body can forward bias the emitter-base junction of the parasitic BJT due to the finite conductivity of the p-body region. The parasitic BJT is now operating in the forward active regime, and if the drain-tosource voltage is higher than the breakdown voltage (BVCEO) of the parasitic BJT, avalanche multiplication of the BJT collector current can occur. If this positive feedback (regenerative) current is not limited, it can lead to junction heating and the eventual burnout of the device [212].
2.5 Normalized ECR (MV/cm)
283-MeV Br LET ~ 37 MeV-cm2/mg 1/cos(θ)
2.0
19.2 nm Wrobel '87
1.5
18 nm 12 nm
1.0
7 nm
1
2
3
1/cos(θ) Figure 4.8. Angular dependence of the critical field to rupture for oxides of varying thickness [209]. The angular dependence disappears as the oxide thickness approaches the diameter of the 283-MeV bromine ion track. II-43
One fortunate difference between SEB and SEGR is that non-destructive test procedures exist for SEB. In the simpler of these techniques, a load resistor attached to the drain of the MOSFET provides current limiting and prevents destructive SEB [217]. Current pulses at the source can be counted as SEB events and the SEB cross-section can be non-destructively obtained as a function of either the drain voltage or the particle LET. A newer technique, called Energetic ParticleInduced Charge Spectroscopy (EPICS), measures the collected charge from SEB events rather than sensing current pulses [218]. This technique was developed to give more information on the triggering mechanisms responsible for SEB, and has offered experimental confirmation of the mechanisms described above. An example of results using EPICS for power MOSFETs is shown in Figure 4.9 [218]. At low drain voltages (Fig. 4.9a), a single peak at a collected charge of about 10 pC was observed and corresponded to charge collection in the depletion layer. At moderate drain voltages (Fig. 4.9b), a second peak appears and indicates that the parasitic BJT has been turned on and the source is injecting current into the device. Note that the original peak also moves to the right (higher charge collection) somewhat. This movement was postulated to be due to avalanche multiplication of the primary ion-induced current [218]. At still higher drain voltages (Fig. 4.9c), the regenerative feedback mechanism is established and SEB occurs. This results in a new peak with 3-4 orders of magnitude more collected charge; this peak is the characteristic signature of SEB in EPICS measurements. A new variation on the EPICS technique was recently reported by Kuboyama, et al. [216]. In this study, the authors connected an EPICS system to both the base and collector electrodes of the bipolar transistors under test. Using the “2-dimensional” EPICS spectra, the authors were able to distinguish between SEB events due to ion strikes to different areas of the emitter.
Figure 4.9. Illustration of energetic particle induced charge spectroscopy (EPICS) technique for measuring single-event burnout [218]. II-44
Many other studies of SEB have been performed; not all of them can be summarized here. These include studies of the position-dependence of SEB [219], temperature-dependence of SEB [220], and hardening strategies [221-223]. The interested reader is directed to references [212223] and references therein for further information.
II-45
5.0
MODELING AND SIMULATION OF SINGLE-EVENT MECHANISMS
From the earliest history of numerical device modeling, the radiation effects community has recognized the value of computational modeling for providing insight into the effects of ionizing radiation on microelectronic devices. In fact, pioneering work on one-dimensional drift-diffusion numerical modeling was presented at the Nuclear and Space Radiation Effects Conference as early as 1967 [224,225], winning the Best Paper award for that year [224]. This early work focused on transient radiation response, as single-event upset (SEU) would not be observed experimentally until almost 10 years later [226]. Given the consequences of SEU (potential loss of mission for space applications, a possible show-stopper for increased integration density in terrestrial memory cells), the rapid development of models to explain and predict radiation effects was essential. The development and use of numerical models for radiation effects has proceeded on many levels: the interaction of ionizing particles with matter, physical device simulators that predict the response of devices to incident radiation, circuit simulators that model circuit response to a single event, and codes that predict the error rate that will be observed for a specific part flying in particular orbit. In this short course segment we will only discuss the models that most directly pertain to the actual upset process itself, namely models for simulating ion track structure in semiconductors, and device/circuit models that predict the response of devices to the ion track. Models for predicting error rates have been covered elsewhere [19,22,227]. 5.1
Interaction Models
As discussed earlier, energetic particles can interact with an IC through either direct ionization or nuclear interactions. Computer codes have been developed that simulate spallation reactions and/or upset rates for protons, neutrons, and other cosmic ray components [36,228232]. Since spallation reactions eventually cause upset through the direct ionization process also, we will limit our discussion here to models that predict the direct ionization-induced ion track structure in the semiconductor. The reader is directed to the above references if they desire further information on the simulation of nuclear reactions. 5.1.1
Track Structure Models
An aspect of SEU and charge-collection simulation that has received considerable attention recently is that of the ion-strike track structure used as input to physical device simulation. Most work until the last few years has employed a simple cylinder of uniform charge generation to represent the ion strike. Detailed calculations of ion strike track structure have been performed using Monte Carlo methods [128,233]. As mentioned in Section 3.7, differences in the track structure between typical particles used in SEU experiments and high-energy particles which might be encountered in a real space environment have raised concerns over the fidelity of the operating environment simulated by accelerator tests [118,128]. Charge production around the path of an incident particle is accomplished by the release of energetic electrons (also referred to as delta rays) along the track, which subsequently travel away from the path and produce further electron-hole pairs. The higher the energy of the incident particle, the higher the energy of the delta rays and the larger the radial extent of the induced charge distribution. As it passes through the silicon device, the particle loses energy and hence II-46
the delta rays also become less energetic, releasing charge nearer the center of the path as the particle nears the end of its range. Incident particles therefore generate a characteristic coneshaped charge plasma in the silicon IC. Analytic models for ion track structure have been developed [234-236] and introduced to the radiation effects community [118,237,238]. These models are easy to implement as computer programs, and due to their analytic nature run very quickly. Analytic methods of computing track structure divide the target material into thin slabs perpendicular to the ion path. The incident ion and energy are specified, and the delta ray distribution and radial deposition of charge are calculated analytically in the first slab [237]. The particle energy is updated based on the energy deposition in the slab and the calculation proceeds to the next slab. In some codes, the LET in the slab is calculated by a call to a TRIM subroutine [18], and the energy is instead updated based on this LET and the thickness of the slab [118]. This method may be more accurate because the radial energy deposition tends to somewhat underestimate the energy loss in the slab, especially near the end of the ion path [237]. Analytical track structure codes have been extensively validated against available Monte Carlo results in the literature. Figure 5.1 compares the analytically-computed radial track structures of high- and low-energy ions that have the same incident LET of 11.4 MeV-cm2/mg: 210 MeV Cl and 5.04 GeV Kr [118]. The higher-energy ion strike is more nearly representative of a cosmic particle that might be encountered in the space environment, but such particles are difficult to achieve in typical laboratory particle accelerators. The Cl ion has an energy of 6 MeV/amu, and the Kr ion has an energy of 60 MeV/amu (ranges in silicon are 63.1 µm and 1.1 mm, respectively). The generated charge density at the silicon surface is shown, as a function of radius from the center of the ion track. Note that both axes of this figure are plotted on a log scale, and the charge density is very 1023
Charge Density (cm-3)
1022 210 MeV Chlorine
1021 1020 1019 1018 1017
5.04 GeV Krypton
1016 1015 1014 1013
Gaussian Approximation
1012 1011 0.001
0.01
0.1
1
10
100
Radial Distance (µm) Figure 5.1. High- and low-energy ion track radial charge distribution profiles [118]. Both ions have the same surface-incident LET. II-47
highly peaked about the center of the ion track. The maximum delta-ray radius at the surface for the 210 MeV Cl strike is about 1.9 µm, so beyond this point the generated charge density falls to zero. For the 5.04 GeV Kr strike the maximum delta-ray radius is about 100 µm, so a low density of carriers (with respect to the central core of the track) exists out to this point. For the highenergy Kr ion, the charge deposited beyond 1.9 µm amounts to less than 15% of the total. Since the curves have the same integral charge (i.e., the ions have the same surface-incident LET), this means the high-energy ion has 15% less charge in the central region of the track (evidenced in Figure 5.1 by the fact that the Kr curve is slightly underneath the Cl curve at small radii). If there were a difference in the SEU response caused by the two ions, one would expect the low-energy ion to be more upsetting since the track is slightly more concentrated and can deposit a greater amount of charge in a small sensitive volume [118]. For the high-energy ion, more charge is deposited at large distances where it may not be collected by the sensitive node [239]. Still, this is expected to be a small effect, since so little of the high-energy ion’s charge is deposited past the boundary of the low-energy ion’s path. For a given ion strike, there is also the more fundamental issue of variation of charge density along the path (i.e., LET is not constant as a function of depth, as seen in Fig. 3.1). Detailed studies of the importance of faithfully representing the track structure within device simulations have shown that it is important to include this effect [240,241]. Neglecting this effect can change the transient current response by as much as 20%, while various fits to the radial track structure typically produce less than a 5% change in the total charge collected for ion strikes on simple p/n diodes [241]. In any event, since the analytic track structures are relatively easy to compute, simulation track structures including at least the variation of LET along the path and reasonable estimates of radial charge distribution are recommended [240]. 5.2
Physics-Based Device Models
In previous sections, we have mentioned device modeling and in some cases shown device simulation results, but so far we have been deliberately vague about the actual techniques used for device modeling. In the present section we will remedy this situation and go into some detail into the actual device models used, what they tell us, and some interesting new developments in the field. Certainly the most commonly used formalism for device simulation is that of drift-diffusion models. In a drift-diffusion model, the semiconductor device equations are derived from the Boltzmann Transport Equation using numerous approximations. The equations to be solved are the Poisson equation and the current continuity equations [17], & (1) ∇ •ε E = ρ , & ∂n ∇ • Jn = q R − G + , ∂t
(2)
& ∂p ∇ • J p = q G − R − , ∂t
(3)
together with the constitutive relationships for current density (the actual drift-diffusion equations): II-48
& & J n = qnµ n E + qDn ∇n , & & J p = qpµ p E − qD p ∇p .
(4) (5)
& & In these equations, E and J are the electric field and current density vectors, ρ is the net charge density, R and G are carrier recombination and generation rates, and n and p are the electron and hole densities. In addition, q is the electronic charge, and µ and D are carrier mobility and diffusivity, respectively. These equations are discretized and solved on the mesh using finite-difference or finite-element techniques [242,243]. Drift-diffusion models are highly evolved, and relatively speaking, not terribly computationally intensive, except in the case of 3D models. Because of the assumptions they are based on, however, they are ill-suited to treat many effects becoming important in small-geometry devices, such as velocity overshoot, carrier heating, and quasi-ballistic transport [244]. Nevertheless, because of their computational efficiency, they remain the workhorse simulation tool, even for deep submicron devices. The next step up the device-simulation hierarchy is hydrodynamic and energy balance codes. Based on fewer assumptions, these codes begin to treat non-local effects, but are correspondingly more computer-intensive, based on five equations of state rather than the three of the driftdiffusion method. Energy balance options are available in commercial drift-diffusion-based codes [245], but have not been extensively used for SEU calculations. 2D energy transport codes have been used for ion-induced charge-collection simulations in GaAs devices, where offequilibrium effects are particularly important [138,246,247]. The top rung of the device simulation ladder is Monte Carlo simulation, which makes the fewest assumptions and approximations [248]. Rather than being based on approximations to complicated macroscopic equations, Monte Carlo methods describe carrier transport on a fundamental, microscopic scale using classical equations of motion (e.g., Newton’s first law). The motion of individual carriers is followed as they drift in fields and interact with scattering centers until statistical significance is achieved. Few assumptions are involved other than transport is described using classical physics. The penalty is very high computational intensity as the trajectories of many thousands of particles must be tracked to attain meaningful statistics. The utility of the Monte Carlo method for simulating radiation-induced charge collection was realized very quickly [249]. Early simulations computed two-dimensional “collection maps” of alpha-particle generated carriers in regular arrays of cells, while a follow-up study looked at scaling effects on SEU in DRAMs [249,250]. The methodology was also extended to cosmic-ray induced SEUs [251]. To reduce computation time, these simulations treated the third dimension (depth into the substrate) with analytic models. Monte Carlo simulation has also been applied to studying SEU in short-channel CMOS on SOI technologies [252]. Recently, a fully threedimensional Monte Carlo simulator has been described [253], and used in combination with a circuit simulator [254]. Due to high computational requirements, the response during only the first hundred or so picoseconds was calculated, but continuing improvements in computer speed and computational efficiency can be expected to overcome these limitations. In the future, Monte Carlo SEU simulations may become more commonplace, especially as Monte Carlo algorithms are frequently inherently easy to parallelize.
II-49
5.2.1
3D Device and Mixed-Level Simulations
One of the many challenges of device simulation of radiation effects is the need for advanced, three-dimensional modeling tools. The inherently three-dimensional nature of an ion passing through a microelectronic device is difficult to address with the two-dimensional simulation programs that are routinely used in the semiconductor industry for device analysis. The development of full three-dimensional tools has been fairly recent, however, and much insight has been gained in the past through the use of two-dimensional programs [50,53,54,56,67, 87,89,90,138,154,182,255-263]. The fundamental problem with representing a three-dimensional ion strike in fewer dimensions is shown in Figure 5.2. In a two-dimensional simulation, all quantities are assumed to be extruded into the third dimension, and hence either the correct generated charge density or the correct total charge can be simulated, not both. Scaling schemes have been proposed that adjust the Auger recombination rate in an attempt to correct for geometry effects [258]. Another method is to use quasi-three-dimensional versions of the popular PISCES-II code, based on cylindrical symmetry and coordinate transformations [264]. Many charge collection and SEU studies have been performed using these modified two-dimensional codes [57,59,61,110,115,116,240,241, 265-268]. Unfortunately, there are few devices that exhibit circular symmetry, although through clever use of geometrical approximations, cylindrically symmetric simulations have proven revealing and surprisingly accurate in some cases [110,116]. Full 3D device codes are necessary to model the effects of angled ion strikes [68,241,269] Fully-3D device simulators were first reported in the literature in 1980 [270,271], and some of the early work on three-dimensional device simulation was motivated by alpha-particle reliability issues [272-274]. An early comparison of 2D and 3D charge-collection simulations showed that while the transient responses were qualitatively similar, significant quantitative differences existed, both in the magnitude of the current response and the time scale over which
-+ + + + + -
-+
-+
-+
-+
+ + + + -
+ + + + -
Ion Strike
Two-Dim ensional
Quasi-Three-Dimensional
Three-Dimensional
Correct charge density or Correct total charge Incorrect geom etry Com putationally efficient
Correct charge density Correct total charge Incorrect geom etry Com putationally efficient
Correct charge density Correct total charge Correct geom etry Com putationally intensive
Figure 5.2. Illustration of the inherently three-dimensional nature of ion strikes [51].
II-50
collection was observed [275]. The implication of these results is that while 2D simulations may provide basic insight, 3D simulation is necessary if truly predictive results are to be obtained. Throughout the 1980s companies developed internal 3D device simulators [276-280], but most of these were proprietary and optimized for super-computers. Only in the present decade have numerical techniques and microprocessor speeds been sufficiently improved to bring such tools to the desktop workstation. In the last few years fully three-dimensional device simulators have become commercially available [245,281-283]. Optimized for high-end workstations, a fairly large 3D simulation can generally be performed in about a day. Another consideration that arises in modeling charge collection and SEU is the tight coupling of device and circuit response to incident ionizing radiation. In SRAMs, for example, modeling the struck transistor with typical constant boundary conditions (or even including passive, lumped elements) will never result in an upset being observed in the simulation – by construction the device will always return to its pre-strike state. The best that can be done in this situation is to study the charge-collection characteristics of the struck device, and compare the collected charge to some critical charge to upset. However, the usefulness of this approach is extremely limited for SRAMs, since the charge-collection characteristics are greatly influenced by external loading and the feedback mechanism in latches [67,284]. Additionally, the critical charge itself may be ill-defined, once again being dependent on external loading and specific circuit designs [63,88,285]. Nevertheless, unloaded device simulation has been useful for studying the basic physical properties of charge collection, and for studying DRAMs, where loading effects are not as prevalent and critical charge is well-defined by noise margins [75]. For studying the upset process itself in SRAMs, circuit simulation has been a more usable tool. The single-event induced transient current is modeled as a current source at the struck node, and the effects on the SRAM are calculated with a circuit simulator such as SPICE [285-288]. One strength of this approach is the large scale of the circuit in question that can be modeled; another is its computational efficiency. A drawback is the accuracy of the transient current used as the input stimulus. For example, if the current is based on device simulations of a struck, unloaded device [117], then the circuit simulation inherits the inaccuracy of the improperlyloaded device simulation. If the charge collection input is approximated using one-dimensional, analytical models [288], even greater inaccuracy may result, particularly for heavily-ionizing particles. Still, circuit simulations have provided considerable insight into SEU in SRAMs and have resulted in improvements to hardening techniques for a variety of circuits [131,150,152, 155,166,286,289,290]. In the best of all possible worlds, one would desire a simulation methodology for SRAMs that allows either modeling the entire memory cell in the device domain, or allows concurrent solution of device and circuit equations. The former method has been used quite successfully with the simulation code SIFCOD [291], which permits numerical modeling of multiple devices at once. Modeling an SRAM cell consists of four simultaneous device simulations, tied together through contact boundary conditions, as illustrated in Figure 5.3 [67]. These simulations were the first reported that were capable of studying the actual upset process at the device level, and were thus key to forming an understanding of fundamental mechanisms leading to upset [67,89,90]. Because four numerical device simulations must be performed at once, this technique is fairly computer intensive. More importantly, however, the SIFCOD program performed only 2D computations and hence suffered from the geometric limitations that have been discussed above. II-51
N+
N+
P+
N-EPITAXIAL
N+
+ P
VDD
P2
P1
N+
P-WELL
+ P
N+
N-EPITAXIAL
P+
+
N+
N+
N
N+
N-EPITAXIAL
P
+
P+
ION PATH
N1 N
N+
+
N2
P
+
P+
P-WELL
N-EPITAXIAL
N+
VSS
Figure 5.3. Illustration of the use of multiple simultaneous device simulation to compute the SEU characteristics of a CMOS SRAM [67]. Each device’s characteristics are simulated in the multidimensional device domain. Commercial 3D simulators such as Davinci have a similar capability to simultaneously simulate multiple devices [245], but are not generally used in this mode. Recently, the simultaneous solution of device and circuit equations has been increasingly used. This technique, known as mixed-mode or mixed-level simulation, was developed by Rollins at USC/Aerospace in the late 1980s [292]. The term “mixed-level” is probably less confusing and more descriptive than “mixed-mode.” In a mixed-level simulation of SEU, the struck device is modeled in the “device domain” (i.e., using multi-dimensional device simulation), while the rest of the memory cell is represented by SPICE-like compact circuit models, as illustrated in Figure 5.4. The two regimes are tied together by the boundary conditions at contacts, and the solution to both sets of equations is rolled into one matrix solution [292,293]. The advantage is that only the struck device is modeled in multiple dimensions, while the rest of the circuit consists of computationally-efficient SPICE models. This decreases simulation times over multiple-device techniques and greatly increases the complexity of the external circuitry that can be modeled. Mixed-level capability has been incorporated into many of the commerciallyavailable 3D device codes [245,281-283]. These codes were first used to study SEU in CMOS SRAMs in 1991 [294], and since then have received a great deal of continued use for this purpose [51]. II-52
VDD
Circuit Simulation
Particle Strike
Device Simulation
Figure 5.4. Mixed-level simulations solve the device and circuit equations simultaneously, taking advantage of the localization of the ion strike. Only the struck device is modeled in the multidimensional device domain, reducing the computational burden. 5.2.2
Recent Enhancements
A drawback of the mixed-level method is that coupling effects between adjacent transistors have been shown to exist at the device level using 2D simulations [258]. These effects cannot be taken into account when only the struck device is modeled at the device level. To address this difficulty, a recent paper has described simulation of the entire SRAM cell in the 3D device domain [295]. An illustration of the technique is shown in Figure 5.5 (interconnects between device regions are defined in the usual manner for an SRAM cell, but are not shown in the illustration for clarity). The authors compared the results to standard mixed-level simulations and found that in cases where no coupling effects between transistors existed, mixed-level simulations were adequate to reproduce the full SRAM cell results. For some strike locations, however, coupling effects between adjacent transistors were observed [295]. Mixed-level simulations are incapable of predicting such effects. As inter-device spacing decreases with increasing integration levels, coupling effects can be expected to become more important, and simulating entire the SRAM cell in the device domain may become routinely necessary [295]. Techniques like mixed-level simulation are useful for in-depth studies of SEU in specific small-scale circuits and for given ion strikes. A system designer, however, is more likely to be interested in the total error rate for a large circuit containing many transistors, and operating in some particular environment of interest. Because this is a very difficult problem, requiring detailed environmental models, the probability that a given ion strike causes an SEU is usually treated using analytical methods such as the rectangular parallelpided (RPP) model. Typical methods of solution have been covered in a previous short course and review articles [19,22,227]. Two groups have recently reported on large-scale SEU simulation systems which are aimed at predicting system error rates using a more first-principles basis of the interaction of ions with devices.
II-53
The simpler of these two models [296] targeted the terrestrial alpha-particle-induced softerror rate (SER) in SRAMs. In this system, analytic models with fitting parameters based on 3D device simulations were used for the alpha-induced charge collection and noise current. An equivalent circuit was extracted and circuit simulations were used to determine error margin, or the probability of an upset for a given incident alpha-particle energy and angle. In parallel with this, the probability of a given alpha-particle being emitted by the metal wiring (interconnects or on-chip solder bumps) and impinging on the circuit was computed from topography simulations and emission rates of the materials. The SER is then just the running summation of the probability that a given alpha-particle will be emitted multiplied by the probability that such an alpha-particle would cause an error. Comparisons of simulated and experimentally-observed error rates were reasonably close at low supply voltages, but only within a factor of five at higher voltages, with the simulations underestimating the observed SER [296]. The authors later rewrote the model to include neutron-generated reaction products and found improved agreement with accelerated data obtained using a high-flux neutron beam [297-299]. IBM has taken a similar approach in the development of their advanced soft-error modeling system (SEMM) [300-302]. This code models both terrestrial cosmic rays and on-chip alphaparticle sources, and includes a nuclear spallation simulator to model particle-silicon recoil spectra from first principles [230,303]. In addition to a nuclear spallation model, SEMM includes a Monte Carlo model for diffusion of the incident charge based on the earlier work of Sai-Halasz [249-251], fitting to transient device simulation to determine the temporal shape of the charge collected at the various junctions, circuit simulation to determine critical charges, and can also
p-channel transistors
Ion Path
n-channel transistors
Figure 5.5. Illustration of the simulation of an entire SRAM cell in the three-dimensional device simulation domain. II-54
incorporate statistical data from process variations. This is very computationally intensive, so SEMM makes extensive use of history files to reduce the amount of computation necessary for any one run. Good agreement has been obtained between SEMM results and field experiments with no parameter adjustments, indicating the predictive nature of the model. Figure 5.6 shows the simulated components of the soft-error rate in a bipolar chip [304]. Components are computed due to both alpha-particle sources and the cosmic ray spectrum. Alpha-particle sources include three levels of metal, Pb-Sn solder pads, and a ceramic layer. Note that for events with small critical charge, alpha particles are dominant and mostly come from the Pb-Sn solder and ceramic materials. For critical charges above 250 fC, only cosmic ray events are capable of inducing upset in this chip. IBM has marketed a soft-error simulation service to outside customers based on these capabilities [305].
Figure 5.6. Simulations of relative soft-error rate in a bipolar memory cell due to terrestrial cosmic rays and various on-chip and package-related alpha-particle sources [304].
II-55
6.0
FUTURE TRENDS
One of the biggest worries for single-event effects is how technology trends will impact device susceptibility in the future. Effects such as SEU have already been getting worse with technology evolution. In this section, we discuss key technology drivers, how technology trends may affect hardening strategies, and the phenomenon of single-event effects in ground-based and aircraft microelectronics, a topic of growing concern. 6.1
Technology Drivers Impacting Single-Event Effects
Technology parameters that influence the SEU sensitivity include gate length, gate oxide thickness, and power supply voltage. Increased SEU susceptibility as gate length is decreased has been well documented, and may be due to direct channel conduction [66], bipolar coupling mechanisms between the source and drain [70], or decreased gate capacitance as the gate area is reduced [260]. Figure 6.1 shows an example of the decreased threshold for SEU for two strike locations as gate length is decreased and all other parameters are held constant [70]. The strike locations are the center of the drain for the outside-the-well “off” transistor and the inside-thewell “off” transistor (panels (a) and (c) in Figure 3.10). The results shown were obtained using 3D mixed-level simulations of a model p-substrate technology with 150 kΩ feedback resistors. As gate length is decreased, the SEU threshold for outside-the-well “off” strikes decreases slightly, and the threshold for inside-the-well “off” strikes decreases rapidly due to the bipolar effect [70]. As the power supply is decreased and all other parameters are held constant, the SEU threshold again decreases due to reduced current drive in the restoring transistor and less stored
1000 Outside the well OFF Inside the well OFF
100
10 0.0
1.0 Gate Length (µm)
2.0
Figure 6.1. Scaling trend for SEU threshold LET as a function of gate length from simulations of a p-substrate technology with feedback resistors [70]. All other technology parameters were held constant. II-56
charge in the cell. Simulations of this situation are shown in Figure 6.2a. However, reductions in power supply voltage are usually accompanied by reduced gate oxide thickness. Thinner gate oxides result in higher gate capacitance and higher current drives, so SEU thresholds increase with decreasing gate oxide thickness if all other parameters are constant (Figure 6.2b). However, if all parameters are taken into account, the overall trend is still a general increase in SEU susceptibility with each technology generation. This trend is shown in Figure 6.3, which is a plot of the simulated and experimentally-measured SEU threshold for SRAMs without feedback resistors in the three most recent Sandia CMOS technologies [70]. A factor that is at least as important as the fundamental changes to the physics is simply the reduction in total capacitance as technologies shrink. Remember that the feedback time for the SRAM cell is to first order related to the RC delay in the inverter pair. As device areas shrink, the gate and drain capacitance shrinks, making the device faster but consequently much more susceptible to SEU. SEU can therefore be expected to continue as a growing concern for scaled technologies. Another area that is likely to become increasingly important is the propagation of singleevent transients (SETs) in digital logic circuits. The problem here is that as circuit speeds rise, the probability that a momentary glitch will be clocked as valid data and propagated down the line increases. For example, in Figure 3.11 we saw that even a particle well below the upset threshold can cause a momentary flip in the state of an SRAM cell. Consider the case where this memory cell is actually a digital latch circuit in a microprocessor. In the example of Figure 3.11, if this latch value is read 10 picoseconds after an ion strike and the value is clocked down the line, it matters little that the struck latch eventually returns to its original state, because the corrupt value has already been passed on to the next stage of the circuit. These types of errors are likely to become a pervasive problem as clock speeds continue to increase, and will be difficult to protect against, especially in commercial microprocessors where speed is paramount. It has been suggested that for circuits built in technologies below 0.35 µm, propagated SETs will be a 100
50
10
10 2.5
3.0
3.5
4.0
4.5
5 50
5.0
VDD (V)
a)
100 150 200 Oxide Thickness (Angstroms)
250
b)
Figure 6.2. Simulated scaling trends for SEU threshold LET as a function of a) power supply voltage, and b) gate oxide thickness. II-57
Threshold LET (MeV-cm2/mg)
100 Experimental Range
10
1 0.0
1.0
2.0
Feature Size (µm) Figure 6.3. Overall SEU threshold LET technology trend for the last three generations of Sandia CMOS SRAMs without feedback resistors [70]. The points are the results of 3D mixed-level simulations, and the shaded regions are the measured range. primary single-event failure mode [306]. The interested reader is directed to the final segment of this short course for more information on SETs in digital logic circuits and mitigation strategies [171]. For SEL, the trends from technology evolution are more difficult to predict. Technology trends working to increase SEL susceptibility are decreased well thicknesses and interdevice spacing. On the other hand, the decreased power supply voltages demanded in newer low-power technologies may alleviate latchup concerns, while technologies such as trench isolation also tend to improve latchup resistance. If CMOS/SOI technology proves to be manufacturable and becomes a mainstream technology solution, traditional latchup concerns may vanish. 6.2
Hardening Strategies
A considerable concern for SRAMs requiring SEU hardness in the future is whether traditional resistive hardening techniques will remain a viable option. The resistive decoupling technique is fundamentally incompatible with high speed because it relies on slowing the cell down so that it cannot respond quickly to SEU voltage transients. This tends to mean that the write performance of hardened SRAM cells has to be kept at a nearly constant level to maintain a constant level of SEU hardness. Also, because a smaller, faster SRAM cell has lower capacitance, the feedback resistance required to harden the cell has to be raised to compensate. This can lead to a requirement for very large resistors in advanced technologies. This point is illustrated in Figure 6.4, which shows the results of 3D mixed-level simulations of two resistively-hardened SRAM technology generations. The simulated SEU threshold LET is plotted as a function of feedback resistor value for a 0.6-µm, 5-V technology with transistor widths of II-58
2.3 µm and 1.3 µm, and for a 0.35-µm, 3.3-V technology with a transistor width of 0.75 µm. As can be seen from the predictions, achieving a target LET threshold of 40 MeV-cm2/mg would require a 200-kΩ resistor for the 2.3×0.6 µm2 device, a 600-kΩ resistor for the 1.3×0.6 µm2 device, and a whopping 1.1-MΩ resistor for the 0.75×0.35 µm2 device. Such large resistors would be very difficult to controllably manufacture and would show a significant temperature dependence [151]. Because of this problem and stagnant performance levels, it is likely that for applications requiring a high level of SEU hardness, devices below about 0.5-µm will utilize other hardening techniques, such as SOI, active feedback elements, or circuit hardening. 6.3
Terrestrial and High-Altitude Single-Event Effects
To this point we have discussed the natural space radiation environment and how it interacts with microelectronics to cause single-event effects. A radiation environment also exists in the Earth’s atmosphere, and although less harsh than the space environment, it can also give rise to SEE. In the following sections we will discuss the high-altitude and terrestrial radiation environments and their effects on microelectronics in aircraft electronics systems and at ground level. 6.3.1
The Atmospheric Radiation Environment
2.3 x 0.6 µm
80
2
Threshold LET (MeV-cm /mg)
The atmospheric radiation environment comes about as a result of the space radiation environment impinging on Earth’s atmosphere. As very highly energetic cosmic rays enter the upper atmosphere they interact with oxygen and nitrogen in the atmosphere and produce a
60
1.3 x 0.6 µm
40
0.75 x 0.35 µm
20
0 0
500
1000
1500
Feedback Resistance (kΩ) Figure 6.4. Predicted SEU threshold LET vs. feedback resistance from 3D mixed-level simulations of a 0.6-µm, 5-V technology and a 0.35-µm, 3.3-V technology. Note the requirement for extremely large resistors as the fabrication technology scales. II-59
cosmic ray shower of daughter products [1,307-309]. Note that the primary galactic cosmic rays are so energetic that some of the daughter products can reach all the way through the atmosphere to ground level, equivalent to passing through more than 13 feet of concrete [308]. A diagram of a cosmic ray shower is shown in Figure 6.5. The daughter products primarily responsible for causing upsets in high-altitude and terrestrial electronics are neutrons and protons [307]. The fluxes of neutrons and protons have similar characteristics with respect to energy and altitude variation, with both populations extending to energies greater than 1 GeV. Both neutrons and protons show a maximum flux at an altitude of 55,000-60,000 feet (17-18 km), with the sea-level flux being several hundred times lower than at aircraft altitudes [307]. A plot of the altitude variation of the neutron flux is shown in Figure 6.6a. The neutron flux also varies as a function of latitude, as shown in Figure 6.6b. The neutron flux is highest at the poles, because the primary galactic cosmic rays can penetrate furthest into the atmosphere there due to reduced geomagnetic rigidity [1]. Models have been produced that give the variation of atmospheric neutron flux as a function of altitude, latitude, and solar activity [310]. Note that the atmospheric neutron flux varies with solar activity due to its dependence on the incident galactic cosmic ray flux. Further details on the terrestrial and high-altitude radiation environment can be found in [307,309] and references therein.
Figure 6.5. Illustration of the terrestrial cosmic ray shower caused by the interaction of galactic cosmic rays with the Earth’s atmosphere [308]. II-60
a)
b) Figure 6.6. Variation of the atmospheric neutron flux with a) altitude, and b) latitude [307]. 6.3.2
Historical Perspective and Recent Studies
Oddly enough, the first paper to ever deal with the issue of SEU was not a paper on the use of electronics in the space environment, but a paper assessing scaling trends in terrestrial microelectronics [311]. Interestingly, the authors predicted that the minimum volume of semiconductor devices would be limited to about 10 µm on a side due to terrestrial cosmic ray upsets! The actual occurrence of soft errors in terrestrial microelectronics did not manifest itself until some time later, shortly after the first observations of SEU in space [73]. This first paper from authors at Intel found a significant error rate in DRAMs as integration density increased to 16K and 64K, spurring a flurry of SEU-related work in the late 1970’s [312]. The primary cause of soft errors at the ground level was quickly diagnosed as alpha-particle contaminants in packaging materials. For example, according to Ziegler, the Intel problem was traced to a new LSI ceramic packaging plant that had just been built downstream from the tailings of an abandoned uranium mine [308]. Radioactive contaminants in the water used by the factory were contaminating the ceramic packages they manufactured. After considerable early activity, the II-61
terrestrial soft error problem was mostly alleviated by using low-activity materials and on-chip shielding coatings [308,313]. Occasionally, changes in suppliers or procedures have caused semiconductor manufacturers temporary but considerable headaches due to raised radioactive contaminant levels in materials such as nitric and phosphoric acid [308,314]. The march toward higher integration densities has made soft-error concerns a continual design consideration for advanced DRAM and SRAM development in the last decade. A particular area of recent concern is flip-chip packaging technologies that place a source of alpha particles (Pb-Sn solder bumps) right on the die itself, where they cannot be shielded by coating layers [299]. Even in the absence of on-chip sources of radiation, recent studies have conclusively proved that terrestrial cosmic rays (primarily neutrons) are a significant source of soft errors in both DRAMs and SRAMs [315-317]. Upsets have been observed both at ground level and in aircraft and have been convincingly correlated to the altitude and latitude variation of the neutron flux, as shown in Figure 6.7 [307,315,317]. Lage, et al. have shown that even without alpha particles, a baseline of cosmic-ray upsets still exists for high-density SRAMs [316]. This result is shown in Figure 6.8, which plots the correlation between the measured accelerated soft error rate (ASER, measured using an intense alpha-particle source placed over the SRAM) and the system soft error rate (SSER, the real application error rate measured in many SRAMs operating over a long period of time). If the system SEU rate were due entirely to alpha particles, there would be a simple direct correlation between the SSER and the ASER, but instead the data show a baseline SER due to neutron upsets even in the absence of alpha-particle upsets [316]. O’Gorman has shown that neutron upsets disappear for DRAMs placed 200 meters underground in a salt mine, while they increase dramatically for systems operated above 10,000 feet in Leadville, CO [315]. In addition to SEU observed in memories used in large computer systems and aircraft, upsets have been observed in SRAMs used in implantable medical devices such as cardiac defibrillators
Figure 6.7. Excellent correlation is observed between upsets recorded in avionics and the atmospheric neutron flux [307]. II-62
Figure 6.8. Correlation between system SER (SSER, from field testing) and accelerated SER (ASER, from intense alpha-particle source testing) [316]. The SSER exhibits a baseline error rate due to neutron-induced upsets. [318]. Neutron-induced single-event burnout has even caused destructive failures in large-area, high-voltage power diodes used for railroad applications in Europe [319]. 6.3.3
Mitigation Techniques
Neutron-induced ground-level upset rates have been estimated to be 1-2×10-12 upsets/bit-hr across a range of commercial DRAMs and SRAMs [317]. For a computer workstation with 1 Gbit of memory (128 Mbytes), this could lead to as many as 1-2 errors per month, depending on how much of the memory space is in daily use [317]. Revelations such as these have significant implications for manufacturers of commercial memory chips and computer systems, because systems can’t realistically be shielded against incident neutrons. Meeting specified failure rates is expected to be a significant challenge for commercial semiconductor manufacturers. Terrestrial soft error failure rate specifications are usually given in terms of FIT rates, where FIT = Failure in Time = 1 error in 109 device hours. A typical specification is to maintain a FIT rate less than 1000 [316]. A complicating factor is that since FIT rates are specified per device, meeting a constant FIT rate specification actually requires reducing the error rate per bit as the number of bits per device is increased. It has been suggested that because manufacturers of commercial microelectronics for terrestrial applications have had to deal with alpha-particle-induced upsets from packaging materials, commercial parts will by design remain hard to at least the alpha-particle threshold, as illustrated in Figure 6.9 [207]. This figure shows the critical charge to upset as a function of gate
II-63
length, with the horizontal line indicating the maximum charge supplied by a 5-MeV alpha particle. Indeed, there is historical evidence supporting this view as data from more than 10 years of microprocessor evolution show a constant upset threshold just above the threshold for alpha particle upset. However, it is also known that many manufacturers have specific soft-error driven design rules for placement of devices relative to on-chip solder bumps and/or use hardened circuit designs for I/O circuitry that must be in the vicinity of such on-chip alpha sources [299,320]. This clearly implies that many devices being manufactured are in fact already below the alpha-particle threshold for upset. Many of the techniques traditionally used in the radiation effects community to SEU-harden devices are of such a nature that they would never be adopted by commercial manufacturers. They tend to consume more power, reduce manufacturability, and severely impact IC performance. Commercial DRAMs have generally exhibited a fairly constant SEU performance because DRAM manufacturers have intentionally maintained the unit cell capacitance through the use of clever modifications to the storage cell [321]. The nodal capacitance for SRAMs, however, has been steadily shrinking [316]. To counteract increased terrestrial soft error rates, manufacturers may find it necessary to explicitly add capacitance to high-density SRAMs [158,316]. Lage has predicted that this will be necessary for the 4 Mbit generation of SRAMs and beyond [316]. Design-hardened circuits may be useful for critical paths or circuitry, but because of area penalties will not be adopted on a large scale [320]. The use of error-correcting memory architectures is already becoming more common again and this trend will likely continue. Mitigating soft errors in high-speed digital logic circuits will be especially challenging. Fault-
Figure 6.9. Technology trend for critical charge to upset as a function of feature size, suggesting that SEU sensitivity will be kept above the alpha-particle limit by manufacturers [207]. II-64
tolerant systems are routinely used in aircraft mechanical systems and seem a natural choice for preventing neutron-induced SEU in avionics [322]. SOI is a possible solution to the terrestrial SEU problem, although as noted previously SOI is not automatically upset immune. Manufacturers have demonstrated the ability to produce ultra large-scale SOI DRAMs [323], microprocessors [324], and SRAMs [325], although others remain skeptical of the advantages of SOI [326]. In any event, the fact that commercial manufacturers will be studying SEU should prove beneficial to the radiation effects community inasmuch as it brings new resources to bear on the problem.
II-65
7.0
SUMMARY AND CONCLUSIONS
In this short course segment we began with a review of the radiation environment encountered by space telecommunications systems. In general, the environment consists of trapped and transient protons, heavy ions, and electrons. Because electrons are rarely important for SEE, they were not treated in this segment. Protons are trapped by the Earth’s magnetic field in two belts around the planet, an inner belt and an outer belt. Trapped heavy ions also exist but are not in general energetic enough to cause SEE in shielded systems. The South Atlantic Anomaly is a localized region of intense proton flux at low altitudes ( 7 krad/year
Ecco
project
2000 x 2000
0°
81 M rad/year
670 krad/year 50 krad/year
note: w ith overshielding = >
MEO (10000km)/GEO (36000km):GPS, Telecommunication Relays,etc
Table 1.
G eosynchronous any satellite
35790 x 35790 0°
1.1 G rad/year 1.1 M rad/year overshielding = > 1-2 krad/year
G PS-II
operational
22600 x 22600 55°
9.1 G rad/year
(or G LO N ASS
= end of life
19132 x 19132
overshielding = >
64.8° )
1.9 M rad/year 1-10 krad/year
Yearly radiation doses for various types of satellites (overshielding example is given for RadpakTM shield case [SEI]).
in
circular
orbits
DOSE in Krad(Si)
1 000 D o s e v s A lt itude 9 Y ears & 5 5 ° ( 10 m m A l )
100
10
1 100
1 000
10 000
100 000
C IRC U LA R O R B IT A L T IT U D E in km
Figure 3.
Indication of the dose received behind 10 mm Aluminum for 50 to 60° orbit as a function of altitude. Trend for present and new spacecrafts or constellations (Space Station, Celestri, Skybridge, GPS-II, etc) [Calv-99].
This increased regime needs enhanced strategies of procurement, modeling of radiation effects and radiation hardness assurance. 2.3.
THE INTERACTIONS: PERMANENT EFFECTS OF THE CUMULATED DOSE
III-6
2.3.1. INTERACTIONS AND TRACK RADII. UNIFORM OR DISPERSED DOSE? In matter, the interaction of photons, ions, protons or high-energy electrons present in the environment generates secondary electrons that are themselves very energetic with respect to the energies of the valence electrons or even of core electrons of atoms.
Figure 4.
Cascades of electrons resulting from the primary interaction.
Thus, secondary electrons can ionize atoms, generating electron-hole pairs or collective excitations of valence electrons (plasmon modes). Oscillations of plasmon modes can eventually generate electron hole pairs. As long as energies of the generated electrons and holes are greater than the minimal energy required for an electron-hole pair, they can in turn generate supplementary pairs. Consequently, one lone photon of high energy, or electron or proton or ion, can create thousand or million of electron-hole pairs. 2.3.2. EQUIVALENCE BETWEEN FLUXES AND DOSE RATES For uniformly deposited dose, a simple relation ties energy, fluence and absorbed ionizing dose derivatives: & = 1.6 10 −5 D
(1)
dE ρdx
Φ
with & D Φ dx dE ρ In
the derivative of the absorbed dose in rad(material).s-1 the particle flux in cm-2.s-1 the elementary abscissa in cm, projected along the particle track (flux vector) the ionizing energy in MeV transferred per dx mass density of the material in g.cm-3
dE , the Linear Energy Transfer (LET) in MeV.cm2.mg-1 can be easily recognized. ρdx
From the tables of LET versus particles and energies, the following set of curves can be obtained III-7
(fig. 5). For practical use, one has to be careful to observe conditions of validity, i.e. sufficient number of particles and sufficiently far from interfaces between materials so that secondary electrons clouds overlap and be considered as homogeneous. Integrating over time, it gives also the relation between dose and fluence.
Figure 5.
Relation between dose and fluence for homogeneous dose deposition (after [VanL-75], completed).
2.4. RADIATIVE ENVIRONMENTS AND SIMULATION OF THESE ENVIRONMENTS Photon interactions are not a primary concern for satellites in the natural environment. However, photon interactions are important in hardness assurance testing of devices devoted to space equipment, since most laboratories use photon sources (low energy X-rays or high energy gamma rays) to simulate total-dose effects for space applications [Holm-93], [Adam-91]. This is well suited for simulation of military environments and the nuclear power industry [Shar-94], [Coen-95] where gamma photons are encountered [Glas-77], [Boud-93], which both result from nuclear energy. But the concern exists for the reproduction of space ionizing radiation, and for the recent concern of high energy physics equipment [Stev-92], where the environments deal mostly with charged particles whereas tests are made with low energy X-rays or high energy gamma rays sources (this issue of great concern has been treated at length in other references, notably: NSREC Short Courses and [Brow-90], [Flee-95], [1019], [1892], [22900], [Wino-94]). 2.5. HARDENING VS. SHIELDING? TRADE-OFFS BETWEEN WEIGHT AND COST The figure 6 gives the range (in aluminum) of typical particles encountered in the field of radiation hardening. It can be seen that, for typical thickness compatible with reasonable shield III-8
weight (1 to 5 millimeters), most of radiation penetrates, if energy exceeds cutoff of photons above 20 keV, electrons above 1 MeV, and protons above 10 MeV. Hopefully, the disposition of the equipments in satellite provide natural shielding, but nevertheless total dose cannot be avoided, especially for those satellite orbiting deep into van Allen Belts. Extra-weight must be more and more avoided, as the goal for constellation is a drastic gain in launch cost. On the other hand, the use of components from the global commercial market rises questions about hardness assurance methods. This is why these new requirements force to pay more attention to prediction of in-orbit behavior of components, especially so as to take profit of their natural annealing capabilities. To get grounds to these efforts, a method is to go back to basics and step-by-step provide models capable of accurate numerical description of reality based ground-tests, being extrapolated to in-orbit mission. Range in aluminum (mm)
Energy (MeV) Figure 6.
2.6.
Orders of magnitude of particle range in matter [Adams-91].
EXAMPLE AT THE COMPONENT LEVEL: LEO SATELLITES
Table 2 illustrates the range of doses in a typical LEO orbit (800 to 900 km). It shows that the required hardness of a given part depends on the position inside the spacecraft. This results in a very broad spectrum from 1 to 16 kilorad(Si)/year. High-level functions are protected by positioning the component deep inside the spacecraft. In Constellation satellites orbiting higher, the dose can be 10 times as much.
III-9
Category Diodes
Transistors
Linear Integrated Circuits Digital Integrated Circuits
Function Ultra High Frequency Signal Zener UHF Bipolar Power MOS Amplifiers Regulators Bipolar TTL-LS CMOS 4000B HCMOS
Microprocessors Memories Table 2.
Dose (krad(Si)) per Year 12 8 4 16 5 1 3 4 8 4 1.5 1 1
Example of a LEO Satellite [CTTS].
3. ANALYTICAL MODELING OF TRAPPING EFFECTS IN MOS TRANSISTORS The ionizing dose affects MOS technologies by mainly two different phenomena that lead to the presence of trapped charge. This results in an induced electrical field, which is added to the applied field existing in the functioning component and disturbs this field. As an example, one can take the simplest case of signal (low-power) MOS transistor 2N4351 (Ntype) and 3N164 (P-type) irradiated all pins shorted (fig. 7).
Figure 7.
Examples of variations of the threshold voltage for small-signal discrete N and P-MOS transistors (irradiated with Co-60 gamma rays under VGS=VDS=VBS =0 V).
III-10
In this figure are summarized certain typical features of total dose effects on MOS devices: -
linear dependence at low dose
-
sublinear regime at intermediate dose
-
saturation or very slow dependence at high dose
-
variation with manufacturing lots
-
difference between N and P type transistors
The basic frame of understanding is that “something” is building up under the influence of dose. This “something” may takes several forms according to the type of the device, the time scale at which it is considered and the way the device has been manufactured, and therefore a comprehensive modeling is required to master all the aspects. This course will only introduce certain aspects, some others being beyond the scope, or some question being too much a matter of research or debate to be addressed in a course. The similar beginnings at low dose is attributed to a common phenomenon to N and P-type, whereas the difference at larger dose is attributed to another different phenomenon which can produce similar effects in N and P, but with sign opposed. More specifically, the variation of threshold voltage is divided into two components named as follows: (2)
∆VtN = ∆VotN + ∆VitN
Figure 8.
and
∆VtP = ∆VotP − ∆VitP
Interpretation of threshold voltage shifts as the sum of two components.
Figure 8 displays the typical negative threshold voltage variation for PMOS, and the firstly negative then positive variation for NMOS. This last phenomenon is due to the fact that ∆VOT saturates, while ∆VIT presents much weaker (or no) saturation. The position from the point of III-11
turn-around according to laws and parameters, as well as the thickness of the oxide and the electric field (bias) has to be analyzed. In the following, we first address the common component to the two types of transistors. 3.1.
CONSEQUENCES FOR TRANSISTORS AND CIRCUITS
For transistors of N type, consequences of negative threshold voltage values, or positive but close to zero (0) Electric Field
Figure 9.
Illustration of charge transport and trapping in the MOS structure for positive gate bias [Herv-93].
III-12
3.3. CHARGE CONTAINED IN THE INSULATOR AND THRESHOLD VOLTAGE SHIFT OF MOS TRANSISTOR For a standard thermal oxide, the drift of the threshold voltage is related to the density of charge present in the insulator by:
1 t ox x ρ( x ) dx C ox ∫0 t ox
(3)
∆Vt = −
with
0 and tox ρ( x ) = q p( x) − n ( x) C ox =
ε ox t ox
the abscissa of gate and silicon interface, respectively the net volume charge (charge density) present in the oxide the gate oxide capacitance per unit area
The net trapped charge density resulting from the irradiation can have many causes: (i) In volume - free charge just after their creation - self-trapped charge again capable of movement by processes such hopping from trapping site to other trapping site [Mott-77]. (ii) In volume or in interface boundary - charge deeply trapped on defects, or more generally modification of the state of defect charge, in volume or in interface. (iii) At the interface with silicon strictly speaking - interface states. Interface states are, in principle, sites that are able to exchange charge with the silicon in indefinitely reversible manner, in thermodynamic equilibrium with the electronic population of the surface of the silicon. This must be done in a finite time (generally in the order of a microsecond for so-called “rapid states” that are usually the most common). These sites are therefore located practically in the “interfacial layer” and, in the sense of extended wave functions of this interface, they belong both to the silicon and to the oxide. It is important to note that interface states are not charged by themselves, but “porters” (sites) of charge (Bardeen states [Scho-39]). It is therefore natural to think that lattice defects, notably those caused at the interface by secondary effects of irradiation, are the cause of such electrically-active sites (e.g. Pb centers) (c.f. also mechanisms of Revesz-Griscom, reminded hereafter). But one can also note that some states cannot be defects in the crystalline sense of the term, but quantum levels caused by the passage of a system to an other (states of Tamm [Tamm32]). Some authors have imagined that charges trapped close to the interface could create allowed levels in the forbidden band-gap of the interfacial silicon by simple electrostatic influence. III-13
The standard classification is by convention that resulting from the “Deal Commission” [Deal80], approved by the Electrochemical Society and the IEEE (fig. 10).
Figure 10. Classification of charges [Deal-80].
Since then, finer studies have shown that reality is not as clear. It seems likely that there exist sites in the volume close to the interface that exchange charge with the surface of the silicon [Fleet-92], in a reversible manner as for states, but much more slowly (these slow states are called “border traps” or “switching states”). There also exist time-evolving phenomena that are identified by the (irreversible) neutralization of charge in the volume close to the interface. In what follows, we develop simple statements without describing the “border traps” concepts. Restrained to the charges that do not exchange reversibly with silicon, the global formula for threshold voltage shift here becomes: (4)
∆Vot = −
q ε ox
∫
t ox
0
x . [p t ( x ) − n t ( x ) + p( x ) − n ( x )] dx
where Vot is the oxide-trapped charge, and nt and pt are the oxide-trapped electron and hole densities of charge (in units charge.cm-3). The exact trap profile is most of the time unknown. It is therefore convenient to define an integral of this density rather than the profile itself. One can define ∆Qe and ∆Qh (more simply Qe and Qh when no ambiguity occurs), as: (5)
t Q e = ∫ ox x . [n ( x ) + n t ( x )] dx
(6)
t Q h = ∫ ox x . [p( x ) + p t ( x )] dx
0
0
To more easily allow reasoning on trapped charge profile, two useful additional quantities can be III-14
defined to represent the charge distribution by its first moment, i.e. the location of charge centroids in the oxide:
xe =
(7)
xh =
(8)
∫
t ox
0
x . [n ( x ) + n t ( x )] dx Q tot
∫
t ox
0
x . [p( x ) + p t ( x )] dx Q tot
where t Q tot = ∫ ox
(9)
0
[p t ( x) − n t ( x ) + p(x ) − n (x )] dx
is the total charge present (except the interface and border traps), and: (10)
t Q ot = ∫ ox x . [p t ( x ) − n t ( x ) + p( x ) − n ( x )] dx 0
is the projected total charge (note that all the quantities Q are surface densities of charges expressed in unit “number of charges per unit area”). Therefore, one can write a simple expression of the voltage shift, each term with a different signification:
(11)
∆Vot = −
Q ot (Q − Q e ) t =− h = − ox C ox C ox ε ox
xh xe − Q tot t ox t ox
The free carrier density, as will be computed numerically in next sections, is insignificant in long-lasting natural irradiations (n and p are between 100 and 106 carriers.cm-3 for dose rates ranging between 10-3 and 103 rad.s-1). It is of course not the case for trapped charges as they accumulate on trapped sites: densities up to 1018 cm-3 can be reached. In this course, we will develop as possible the models that link: -
The ∆Vot threshold voltage shift and the trapped charge Qe and Qh, projected at the silicon interface.
-
The trapped charge Qe and Qh and the charge density profile nt and pt.
-
The charge profile nt and pt and the density of traps Ntn and Ntp.
-
The density of traps Ntn and Ntp and the material and process characteristics.
3.4. POSITIVE AND NEGATIVE TRAPPED CHARGE: PRESENT UNDERSTANDING III-15
The positive and negative charges trapped in the oxide therefore consist of the fraction of the holes and electrons that has not been lost via recombination, and eventually trapped. In the first decade of radiation effects, it was believed that trapped holes only exist, as the trapped charge always revealed as positive. However, evidences of negative charge appeared in metal-implanted oxides [e.g., Wang-75]. Later, it was recognized that trapped charge could be evidenced, even before irradiation, in pure thermal oxides [Shan-83] by combination of Thermally Stimulated Current technique (TSC) and Capacitance-Voltage measurement (C-V). However, the amount of trapped electrons is generally lower as compared to the quantity of trapped holes, which explains the general observation of a net positive trapped charge.
Table 3.
Type of oxide
Thickness
Range of dose
Thermal hard (ATT) Thermal hard (wet) Thermal hard (wet) Thermal hard Thermal hard (wet) Thermal hard (dry) Thermal hard Thermal soft Thermal soft SIMOX (SOI oxide)
18 nm 23 nm 47 nm 45 nm 100 nm 104 nm 98 nm 350 nm 348 nm 380 nm
5 Mrad 2 Mrad 1 Mrad 0.1-5 Mrad 0.5 Mrad 0.25 Mrad Idem 6 krad 20 krad 1 Mrad
Range of the electric field +1 to 2 MV/cm +1 to 2 MV/cm +1 to 2 MV/cm + 1 MV/cm +1 to 2 MV/cm +1 to 2 MV/cm + 2 MV/cm + 1 MV/cm +1 to 2 MV/cm -0.75 MV/cm
Qe/Qh and method used 0.48 TSC 0.083 TSC 0.47 TSC 0.55 TSC 0.14 TSC 0.43 TSC 0.46 TSC 0.15 TSC 0.19 TSC 1.3 (∆Vot bi-ex. fit)
Ratio of trapped electrons Qe to trapped holes Qh as measured under various circumstances (TSC from [Flee-92], bi-exponential fit (sect. 3.15.2 for SIMOX from [Lera-88, Pail95c]).
As recalled in table 3, the ratio of trapped electrons to trapped holes ranges from some percent to one half [Fleet-92], table 2, adapted). However, it is likely that a coherent modeling should require the two components. These data were interpreted in terms of wet and dry process, as in fig. 11.
Figure 11. Trapped electrons/trapped holes ratio for various processes [Flee-92].
III-16
In some specific cases, such as oxides used as substrate insulator for Silicon-On-Insulator under negative bias during irradiation, this ratio can be superior to 1, meaning that the net charge is negative in this case [Lera-88]. However, we shall consider as a first approach valid a first sight that the net charge is globally positive, as it has been considered for 30 years from the late 50’s In table 4, electron and hole densities of traps are measured instead of trapped charges (according to a method described in [Pail-95a] and [Pail-99] and summarized tn section 3.14). As will be seen later, the two items generally differ because a quantity depicting capacity of traps to be filled (NTN and NTP) is not generally the same as another measuring the amount of traps filled (Qe or Qh). Type of oxide
Thickness
Range of dose
SIMOX
380 nm
1 Mrad
SIMOX
80 nm
1 Mrad
UNIBONDTM
400 nm
1 Mrad
Thermal wet 400 nm annealed 1320°C
1 Mrad
Table 4.
Range of the electric field + and – 1 MV/cm + and – 1 MV/cm + and – 1 MV/cm + and – 1 MV/cm
NTH/NTP and method used 0.2 to 0.8 Vt-Vg 1 Vt-Vg 0.06 Vt-Vg 0.75 Vt-Vg
Reference [Pail-93] [Pail-95a] [Pail-95a] [Pail-93]
Ratio of electrons trap NTN to hole traps NTP densities as measured under various circumstances.
It can be seen that holes, although generally the most numerous, never come alone. In this course, this will be the guideline for further understanding. 3.5. FIRST STEP: CREATION AND SEPARATION OF ELECTRON-HOLE PAIRS IN SIO2 Globally, electron and holes are created in equal quantities at a rate: (12)
R0 =
∂n ∂p = = g 0 D' ∂t ∂t
g0 =
ρ = 7.8 1012 w
with: (13)
cm −3 . rad(SiO 2 ) −1
and ρ=2.27 g.cm-3. w=18 eV is the average energy required to generate one electron and one hole in SiO2 via high-energy interactions (an experimental value). Note that, the experimental uncertainty being of the order of 1 eV, some authors use g0=8.2 1012 coming from w=17 eV. However, SiO2 seems to be an exception as compared to Si, GaAs, C, etc., because a certain amount of electron and holes recombine, and this recombination depends essentially on the electric field applied. This “mutual recombination” of the generated carriers is the first part of a series of processes, well separated in time, whereas charge trapping is the last picture. In this series of successive interactions and physics that takes place from the very early moment of energy deposition, let us come back to this early events when the tracks take form. III-17
Figure 13. Fraction of unrecombined pairs versus the applied electric field for various incident radiations [McGa-80], [Oldh-83].
Figure 12. Separation of electron and holes.
According to the particles that interact with the oxide, the density of ionization is more or less important and two borderline cases are considered: -
The recombination of one particle of the pair undertaken with the parent particle. It is the case for electrons or incident photons that give a weak density of ionization.
-
The recombination undertaken randomly with a not-correlated particle in the ionized column. It is the case for ionization due to heavy particles.
The fraction of pairs that overcome this initial recombination and separate depends therefore on energy and the type of the incident particle and also on the electrical field present in the insulator that favors this separation. This fraction has been characterized by several independent workers working on scaled technologies since the mid 70’s: [Ausm75] initiated with 105 nm oxide and 4 keV electrons, [Srou-77b] compared the yield at 77K and 300K, [Hugh-75] for thermal SiO2 and suprasil glass in pulsed irradiation mode, [Boes-76] with 13 MeV electrons at 80K, [Peas-85] studied thick oxide under very low fields used in bipolar emitter-base spacers. A representation can be found in data recalled on fig. 12. Brown and Dozier made thorough studies of dependence versus the energy of photons and the structure of the ionizing tracks ([Dozi-81], [Brow-81], [Oldh-81/83/85]). [Oldh-82, [Oldh-84] and [Ausm-86] presented a refined analytical or numerical model taking into account the track density. A very simple model helps to picture the reasoning [Brow-81]. First, it is assumed that the cloud of holes and the cloud of electrons are swept apart in time tr by an applied field. Second, it is assumed that while the electron and the hole clouds overlap, electron-hole recombination takes place following bimolecular kinetics, i.e. (14)
dR = −χ R 2 dt
where χ is a recombination coefficient and R is the electron (and the hole) density at time III-18
t. This equation can be integrated simply: (15)
1 1 = + χ tr Rf Ri
with t r = d µ E is the time needed for columns separation, and µ the (ambipolar) mobility. This leads to the simplest formula for the non-recombination or yield function: (16)
Y(E)= R f / R i =
E / Ec 1 + E / Ec
with Ec a critical field defined by µ E c = χ . This simple calculation of “columnar” recombination is not always adequate because it supposes sufficiently dense tracks ([Oldh-84] showed it is valid for 107 electrons and holes generated per cm of track, i.e. for protons below 3 MeV, and electrons below 3 keV). Otherwise, a better model would be the “geminate” recombination, based on similar physics but where carriers are more dilute so that they recombine as isolated pairs [Ausm86]. It leads anyway to similar conclusions. On fig. 13, the yield dependence is presented versus energy of particles, that is to say, track density. One notes the low yield for heavy particles and the influence of the electrical field Eox. For Co-60 and 10 keV X-rays, several updates were made in the late 80’s and in the 90’s, and precise comparison have been made ([Bene-86], [Shan-91]). When one comes to analytical approximations, sigmoid curves can be suggested, such as, for X and γ photons:
(17)
E Y(E) = 1 + c E
−m
where m and Ec are coefficient that depends on the type of radiation. Table 5 gives examples of fitting parameters. Photons From [Flament95] From [Oldh-81/83/85] Table 5.
Co-60 γ rays 1.17-1.33 MeV Ec (MV/cm)=0.65 m = 0.9 0.5 0.7
X rays 10 keV 1.35 0.9 1.35 0.9
Coefficients for the empirical function of non-recombination Y(E).
More specifically to take into account a possible residue Y0 at zero field and temperatures above 150 K (as evoked by [Ausm-86]):
III-19
(18)
(19)
Y(E) = Y0 +
Y(E) =
1 − Y0 E 1 + Ec
( )
E + E0 m E + Ec
m
or (other approximate expression):
and E0 = Y0 . EC
with Y0 is about 0.05 at room temperature [Farm-73], [Lera-89a]. 3.6.
SECOND STEP: MOTION OF THE HOLES
3.6.1. BASIC OBSERVATIONS From the point of view of electron and hole processes, one can consider that, whatever the duration of irradiations, it can be decomposed into an infinite series of small quasiinstantaneous elementary irradiations. This method of the “pulsed response” allows a more didactic statement of the successive processes that follow the creation of electron-hole pairs (fig. 14). The use of «flash X» or linac irradiation machines provides conditions to measure the real impulse response of MOS and to suitably model the successive phenomena [Hugh-75a, Hugh-75b, Curt-77, Lera-85, Lera-89a, Peyr-91].
BEFORE
+V
DURING IRRADIATION
DURING AND AFTER IRRADIATION
+V
+V
gate SiO2
+- +- + - +- +- + - +-+ - + - + - + - + +-+-+ - + - + - + -
+ + + + + + + +
Si IONIZATION AND RECOMBINATION
DRIFT, DIFFUSION AND TRAPPING
Figure 14. Generation and the trapping of charge in the insulator of MOS structures.
Contrarily to semiconductors where the free carrier mobilities of electrons and holes have similar magnitudes, they differ here by several orders of magnitude. For electrons, III-20
mobility µ e of about 10 cm2.V-1.s-1 has been reported [Hugh-75] from photocurrent measurements. On the contrary, the positive charge due to holes is slowly evacuated, allowing one to define an effective mobility derived from the time-of-flight of the moving holes [Boes-75, Curt-77]: (20)
µ peff =
t ox E . t1/ 2
with t1/2 the time needed for decreasing the charge to one-half its original value. This behavior has been first attributed to shallow traps situated to the vicinity of the valence band. Their apparent mobility µ peff varies therefore strongly with the temperature [Srou-76], [Srou-77b], [Boes-75], [Curt-77], [Lera-85]: (21)
µ peff = µ p0 exp(-Ei/kT)
In [Hugh-75b], an example of parameter set is given: µ peff = 20cm2/V.s exp(-0.6eV/kT) for thick thermal SiO2 film (862 nm).
III-21
3.6.2. INTERPRETATION In fact, there is a broad distribution of activation energies, and it has been shown that the shallow traps were intrinsic to SiO2 in the sense that the hole becomes self-trapped due to its own electrostatic potential on oxygen atoms constituting the network (polaron effect, [Mott-77, McLe-77]). This is probably why this behavior is so universal among the various kinds of oxides tried, although a slight dependency of activation energies has been found to depend on process [Lera-85], with high-temperature annealed oxides having smaller t1/2, higher hole mobility and lower activation energies. This might be related to the hopping mechanism between sites, favored by denser oxides [Lera-89a]. The figure 15 summarizes the main transport and trapping mechanisms in SiO2 by showing the importance of the spatial and energy distribution of traps.
Free hole
self-trapping
hopping transport
Figure 15. Simplified illustration of transport and trapping mechanisms of holes in SiO2 [McLe-77].
A global hole movement occurs as determined by this trap-controlled mobility under the electrical field toward electrodes. This movement has been measured after irradiations delivered in pulsed mode with X-ray or electron flash machines, and modeled by describing the transport equations in the presence of traps distributed in volume and in energy [Boes-74], [Boes-77], [Lera-89a], [Peyr-91]. For time scales of the order of irradiation experiments, [Wino-94] suggests to use an activation energy of about 0.4 eV.
III-22
It must be noted that such polaron behavior has been investigated for electrons, but no evidence has been found [Othm-80]. R.C. Hughes made thorough investigations on SiO2 photocurrents, and mesured the dependence of electron drift velocity versus electric field (fig. 16). It is similar with other type of semiconductors, except that the critical field is higher (200 kV/cm). This is attributed to the high value of the optical phonons in this material. Ferry calculated the theoretical value of the saturation velocity (2 107 cm s-1) [Ferr], which is in good agreement with the experimental value of Hughes [Hugh-73, Hugh-78].
Figure 16. Drift velocity of electrons in SiO2 versus electric field for three thicknesses of Suprasil-II glass [Hugh-75]. The slope is the electron mobility of 21± ±2 cm2 V-1 s-1.
3.6.3. MODELING OF THE EARLY REGIME OF MOTION AND THE SELF TRAPPING OF HOLES For “free” holes (called also “dry holes” as opposed to “self-trapped” holes or “polaron”), Hughes introduced and indirectly measured a lifetime τp0 before it becomes self trapped. For this to achieve, he measures the product µ p0τp0 ≈ 1.4 10-12 cm2 V-1 by integrating the current released in thermal silicon dioxide irradiated by 3 nanosecond pulses [Hugh-78]. This value is also found in other type of experiments [Curt-75, Boes-76, Srou-76, Srou-77a]. Hughes derived a value for τp0 by theoretical estimation of µ p0, based on the value of the mass attributed to the free holes. With µ p0=1 cm2 V-1 s-1, it gives τp0 ≈ 1.4 10-12 s which is a value compatible with the self-trapping phenomenology [Hugh-98]. So, if we note p0 and jp0 the “dry” hole density and current:
(22)
∂p 0 ∂jp 0 & Y(E) − p 0 + = go D ∂t ∂x τ p0
In steady state, the first term can be neglected (it requires that the dose rate does not vary much in time comparable with τp0, which might be questionable in case of heavy ion strike, for instance). If we neglect diffusion before the drift term, the current equation becomes simply jp0 = p0 µ p0 E. The differential equation can be solved in x for 1D problems: (23)
& Y(E) τp0 (1 − exp − x µ p 0 τ p 0 E ) p0 = go D
Provided that E < tox/µ p0τp0 = 106 V cm-1 for tox=10 nm, the “dry” hole density is nearly uniform and we can have an order of magnitude of the steady-state free hole density: p0 ≈ 7 carriers cm-3 III-23
per rad/s, which could seem very low. However, these carriers must not be neglected. They are not lost as they appear as inputs in the following equations related to the subsequent step where the deep trapping occurs. 3.7.
THIRD STEP: TOWARDS PERMANENT TRAPPING OF CHARGE
In the vicinity of the semiconductor interface, there exists a disturbed region that is the transition between the crystalline silicon and the amorphous silicon dioxide. Deep hole traps, initially neutral, become converted into positive fixed charge by trapping holes (and eventually relaxing into another configuration state which stabilizes the trapping). The probability of capture for a hole is a function of the density of empty traps, and therefore scales with dose initially, before becoming sublinear and saturating at higher doses. 3.7.1.
MODELING THE TRAPPING
Before modeling the detrapping, we shall follow the method developed in [Pail-99], which is a complete treatment of models having been proposed these 20 last years by [McGa-80] for low dose regime, and extended to saturation regime in [Lera-89a]. All these models are based on trapping of hole and electron participating to the flux on a given trap profile, according to the proposition of [Kran-87]. Here, the fluxes are defined by drift-diffusion equations. Note that an alternative formulation exists, where the “flux” is defined as the product of the thermal velocity by the carrier density. These two alternatives categorized as the “J-Model” and the “V-Model” will be compared in the section 4.2. We first describe the J-Model which is probably easier to picture out, and which has been much more in use during the last decade. 3.7.2.
EQUATIONS RELATIVE TO CARRIERS
As mentioned above, during irradiation two charge trapping mechanisms can take place in the oxide for each type of carrier. The continuity equations (for holes in the valence band and for electrons in the conduction band of silica) can thus be written by taking into account the generation term, the flux gradient and both trapping phenomena. Using a one-dimensional formulation, and noting simply p the hole density at interest for this deep trapping model, we get: (24)
∂n ∂jn (x, t ) & Y(E) − σnt(E)jn(x,t)[NTN(x)−nt(x,t)] −σnr(E)jn(x,t)pt(x,t) + = go D ∂t ∂x
(25)
∂p ∂jp (x, t ) p 0 + = − σpt(E)jp(x,t)[NTP(x)−pt(x,t)] −σpr(E)jp(x,t)nt(x,t) ∂t ∂x τ p0
where: n(x,t) and p(x,t) are respectively the density of free electrons in the conduction band and that of free holes in the valence band, go is the density of electron-hole pairs generated per rad(SiO2), represents the dose rate [rad(SiO2).s-1], Y(E) is the probability of escaping initial recombination (also called the yield function), nt(x,t) and pt(x,t) are respectively the density of trapped electrons and trapped holes, t designates the time evolved since the beginning of irradiation, and q represents the charge of the electron. Note that current densities are defined III-24
here as currents of particles (in cm-2.s-1) and not currents of charges (in A.cm-2). By replacing the value obtained for p0 in the steady state, and provided that E < tox/µ p0τp0 = 106 V cm-1 for tox=10 nm, we have:
(26)
&τ g 0 Y(E) D p0 h0 (1 − exp − x µ p0 τ p0 E )= g 0 Y(E) D& (1 − exp − x µ p0 τ p0 E ) = τh0 τh0
So, under these assumptions only, and for x > µ p0τp0 E, the equation of hole trapping simplifies and becomes apparently similar to the equation for electrons:
(27)
∂p ∂jp (x, t ) & Y(E) − σpt(E)jp(x,t)[NTP(x)−pt(x,t)] −σpr(E)jp(x,t)nt(x,t) + = go D ∂t ∂x
The only difference lies in the equation for currents, in which the mobility (and the diffusivity) is defined by a specific law, strongly depending on temperature as recalled before. 3.7.3.
EQUATIONS RELATIVE TO TRAPPED CARRIERS
To be able to derive equations for trapped carriers, we must first assume that there only exists one trapping level for holes and one level for electrons in the bandgap of the insulator, and that these traps are deep enough so that we can neglect carrier detrapping during irradiation, at the temperature of the experiment, which is certainly an oversimplification. The continuity equations for trapped electrons and trapped holes can then be written:
(28)
∂n t = σnt(E)jn(x,t)[NTN(x)−nt(x,t)] − σpr(E)jp(x,t)nt(x,t) ∂t
(29)
∂p t = σpt(E)jp(x,t)[NTP(x)−pt(x,t)] − σnr(E)jn(x,t)pt(x,t) ∂t
The first term of each equation corresponds to the trapping of free carriers. This mechanism depends both on the number of “candidates” (brought by the current), and on the number of available trapping sites (e.g. NTN - nt). The second term accounts for the recombination of already trapped charges with free carriers of opposite sign. The trapping of a free charge on a trapped carrier of opposite sign does not a priori depend on the direction of its displacement. Therefore, only the absolute values of each current are taken into account. In the presence of electric field (coming from the space-charge + from the applied voltage), one expects a cross-section variation in Eox-n [Ning-76], [Kran-87], [Esco-95a]. 3.7.4.
EQUATION RELATIVE TO THE ELECTRIC FIELD
In order to simplify the equations, the internal electric field in the oxide will be written E, although it should more correctly be written E(x,t) since it depends both on the space coordinate and on time. Indeed, E does evolve with the build-up of the trapped charge, both in time and space. The evolution of the field is obtained from the Poisson equation, which for this oneIII-25
dimensional model yields: (30)
∂E q [pt(x,t)−nt(x,t)+ p(x,t)−n(x,t)]] = ∂x ε ox
where εox represents the dielectric permittivity of silica. This equation links the local concentrations of trapped carriers to the local field value E(x,t). Free carriers are not taken into account because during irradiation, their density rapidly becoming negligible compared with the trapped charge density. 3.7.5. EQUATIONS RELATIVE TO THE CURRENTS The expressions of the electron and hole currents must now be defined. Each current density of particles can be written in its very general form, as the sum of a drift term and of a diffusion term. It follows: (31)
jn(x,t) = − n(x,t) µn E(x,t) − Dn
(32)
jp(x,t) = p(x,t) µp E(x,t) − Dp
∂n ( x , t ) ∂x
∂p( x , t ) ∂x
where Dn and Dp are the diffusion coefficients for electrons and for holes, which verify the Einstein relationship recalled below: (33)
Dn =
kT kT µ n and Dp = µ peff q q
with µ peff = µ p0 exp(-Ei/kT)
(cf. Eqn. 21).
The expressions of currents jn and jp can be inserted into the preceding equations. The four continuity equations for free carriers and trapped charges, coupled with the equation of the internal electric field, constitute the system that allows us to calculate the profiles of the charge trapped in the oxide. The voltage shifts associated with charge trapping in the oxide can then be deduced by integration. Without any further simplifying hypothesis, this system of five coupled equations can only be solved numerically. In the following, we shall make simplifying assumptions to derive usable analytical expressions. 3.8. SOLUTION FOR THE STEADY-STATE CASE OF IRRADIATION: MEAN FREE PATHS FOR HOLES AND ELECTRONS. LIMITING CASES Instead of trying to solve numerically this system of five coupled equations, we can attempt to find some simple analytical solutions corresponding to typical profiles of trapped charges. These analytical resolutions are valid as long as the generation of free carriers remains little disturbed by the trapped charge, i.e. when the space charge effect is not too important. They are therefore limited to low doses of irradiation, and need simplifying hypotheses. III-26
• The first simplifying assumption is that corresponding to the stationary regime. The method [Lera-89a, Lee-91, Kran-87] then basically consists in solving the continuity equations for free carriers, by assuming that ∂n/∂t = ∂p/∂t = 0. We assume that the radiation-induced free carriers do not accumulate in the oxide, and thus are either trapped, or swept away at the electrodes.
Si
2 e
GATE jn
λn >> t ox
jp
λp > 1 so
that fe≈1 -
Other mechanisms not described here, such as diffusion, hot electrons. (Extension could be made to electron injection from silicon, not addressed here). We proposed to lump these k T µn τn terms, into two terms Ldif/tox= representing diffusion and LH/tox representing 2 q t ox possible hot-electron component.
We can therefore solve the evolution equation for charged traps at the Si/SiO2 interface by replacing the currents by their expressions. It becomes:
(65)
+
N TPS (D) =
[
]
N TPS 1 − exp(− σ pt g 0 f h (E) Y(E) t ox [1 + F(E)] D ) 1 + F(E)
where
(66)
F(E)=
σ nr f e (E ) X h σ pt f h (E ) t ox
The voltage shift associated with this charge trapped at the interface can be written [Lera-89a]:
(67)
∆Vot h+ = −
[
]
q N TPS t ox 1 − exp(− σ pt g 0 f h (E) Y(E) t ox [1 + F(E)] D ) ε ox 1 + F(E )
It is possible to separate the behaviors at low and high dose by developing the exponential term in linear and in saturation parts. It yields: III-34
- at low doses, (68)
∆Vot h+ ≈ − α β Ω f h (E) Y(E) t ox D 2
- at high doses,
(69)
∆Vot h+ ≈ − A
N TPS t ox 1 + F(E)
with o q = 4.76 Volts / k A / 1012 charges.cm-2, being introduced as another characteristic ε ox constant of SiO2.
A=
III-35
3.11.
ANALYTICAL MODELING
Eqn. 68 and 69 can easily be used. It gives the following set of representation (fig 19.). The two curves a and b refer to the response at low dose (linear) and at high dose (saturation), respectively [Lera-89a].
Large dose Weak dose
Figure 19. Analytical modeling of ∆Vot at low and high dose. ∆Vot versus EG (MV/cm). Case of a 65 nm oxide and σpt =σpr=10-12 cm2
3.12.
PRACTICAL APPLICATIONS
Due to the complexity of the phenomena occurring during the generation, transport and trapping of charges in an irradiated oxide, simple analytical expressions can only be obtained when broad simplifying assumptions are made. The equations derived from these models are then only usable in the experimental context that satisfies the series of assumptions made. To simplify the interpretation of experimental results even further, we attempt next to schematize the behavior of some typical oxides. This will give us a tool to rapidly identify specific trends by the simple observation of some measured electrical characteristics. 3.12.1. ELECTRICAL MEASUREMENT TECHNIQUES Before beginning to discuss and interpret experimental results, let us briefly recall some of the most common measurement techniques used to electrically characterize defects in MOS devices. In practical applications, the experimental structure being tested is either a MOS capacitor or a MOS transistor. • In the case of a MOS capacitor, the technique consists in recording the capacitance-voltage curve (C-V curve), either using a small AC signal superimposed on a DC voltage ramp (high or low frequency C-V), or using a quasi static C-V technique. Fig. 20 shows typical high frequency (1MHz) C-V curves obtained on a p-substrate MOS capacitor, before and after irradiation at a dose of 10 Mrad(SiO2). The dashed lines in fig. 20 show the position of the flatband (Cfb) and midgap (Cmg) capacitances, defined respectively as the capacitance value for which the silicon surface potential equals 0 and φB. Potential φB is the bulk potential given by:
III-36
(70)
φB =
kT N A ln q ni
where q is the elementary charge of an electron, k is the Boltzmann's constant, T is the absolute temperature, NA is the substrate doping concentration, and ni is the intrinsic carrier concentration. The gate voltage applied to reach Cfb (resp. Cmg) is the flatband voltage Vfb (resp. migdap voltage Vmg). The traps at the Si/SiO2 interface are amphoteric, which means that their charge state can change depending on the value of the surface potential of silicon. The traps thus can be positive, neutral, or negative. Traps in the lower half of the Si bandgap are predominantly donorlike (i.e. they “give off” an electron when the Fermi level at the interface is below the trap energy level), and traps in the upper half of the bandgap are essentially acceptor-like (i.e. they “accept” an electron when the Fermi level at the interface is above the trap energy level). The most widely accepted assumption (well confirmed by recent results [Flee-92]) is that interface traps are approximately charge neutral at midgap [Kim-88, Wino-88, McWh-52]. With this assumption, the midgap voltage shift (∆Vmg) caused by irradiation is only due to the oxide trapped charge (∆Vot), while the flatband voltage shift takes into account both the oxide trapped charge and the charge trapped on interface traps between flatband and midgap. CHF (pF) 65 60
Figure 20. Typical high frequency (1MHz) C-V curves of a MOS capacitor before and after a 10 Mrad(SiO2) irradiation. The structure was biased at VG = 0 V during irradiation.
C fb
55
0 krad 50
10 Mrad
45 40
C mg
35
-40
V mg
V fb
30
-30
-20
-10 VG (V)
0
10
20
• In the case of a MOS transistor, the most usual technique consists in recording the currentvoltage curve (I-V curve), usually the drain current IDS versus gate-to-source bias VGS, at a given drain-to-source bias VDS either in the linear region (small VDS) or in the saturation region (large VDS). The threshold voltage of a transistor is determined basically from the intercept with the voltage axis of the ID vs VGS curve in the linear region, or of the (ID)1/2 vs VGS curve in the saturation region. Another method, which is similar to that used for capacitor studies, is to determine the gate voltage for which the surface potential of the silicon substrate equals 2φB, which corresponds to the onset of strong inversion [McWh-86]. Figures 21 and 22 show typical I-V curves obtained on an n-MOS transistor, before and after irradiation. On fig. 22, the position of the “threshold current” and “midgap current” is defined as those values of IDS for which the surface potential of silicon equals 2φB and φB respectively. The amplitudes of the threshold and midgap voltage shifts caused by irradiation are also shown in fig. 22. III-37
The calculated “midgap current” is usually very low, thus the midgap voltage is obtained by extrapolating the I-V curves to the low current level required, and by determining the voltage corresponding to that “midgap current”. Except for the extrapolation step, this method is very similar to the high-frequency C-V method. Using the same assumption, ∆Vot is equal to the midgap voltage shift (∆Vmg), whereas the threshold voltage shift (∆Vth) is equal to the sum of ∆Vot and ∆Vit. Consequently, the voltage shift due solely to the interface trapped charge, ∆Vit, can be determined by the stretch-out of the I-V curve (i.e. by difference ∆Vth - ∆Vmg). Ids (mA) -3
10
I DS
-5
10 10
VOT
VOT
∆Vth
-7 -9
10
Post-rad
Pre-rad
-11
10
-13
10
MOS P
MOS N
V GS
∆Vmg
-15
10
-3
-2
-1
0
1
2
3
VGS (V)
Figure 21. Drift of MOS transistor characteristics in the case of only the trapping of positive charge in the oxide.
Figure 22. Typical sub-threshold I-V curves of an n-MOS transistor revealing the contribution of interface states.
Other techniques have been developed to determine ∆Vot and ∆Vit, especially the dual-transistor techniques, combining threshold voltage measurements with either mobility measurements or charge pumping techniques. The reader is referred for instance to [Wino-92] and [Autr-99] for more details. 3.12.2. BEHAVIOR IN THE LOW DOSE REGIME After a low dose irradiation of a silicon oxide, a net positive charge is created in the insulator. Depending on the process used to manufacture the oxide, this trapped charge can be distributed in different ways in the material. To determine the type of charge trapping taking place, a simple method consists in measuring the trapped charge for different biases applied during irradiation. We can thus evaluate how the electric field influences charge trapping, which gives information on the position of the centroid of the net trapped charge. In the next paragraphs, several typical cases are presented. 3.12.3. CASE OF CHARGE TRAPPING AT THE SIO2/SI INTERFACE This is the simplest case, and that most often used to interpret charge trapping in thermal oxides. III-38
The traps are assumed to be all located at the SiO2/Si interface. • During an irradiation under a positive bias (VG > 0), the generated holes flow toward the interface and can get trapped there. This trapping gives birth to an important image charge in the semiconductor, and therefore to a significant voltage shift of the irradiated structure. ∆Vt
Si
SiO2
ρ
Gate
E>0
ox
E0 and E0 and E0 and E1 MV/cm during irradiation, and ∆Vt / tox< 1 V/10nm). The consequence of these on hole trapping effects is the negative drift of MOS threshold voltages. In the low dose regime, we generalize the simple formula we have derived from the uniform and the interface distribution of traps: (71)
∆Vt ≈ − α β Ω f h (E) Y(E) t ox D 2
with: α=
σ.Nt trapping coefficient (Nt = surface density)
β=
xh/tox position of charge distribution centroid
Ω=
& 2 / krad(SiO2) or 36 mV/nm2 / Mrad(SiO2) 36 V/ µm 2 / krad(SiO2) or 0.36 V/ k A
fh(E) =
function of charge collection at the trapping location
Y(E) =
function of non-recombination
It can be recalled that this type of analysis has been proposed from 70s by Freeman and HolmesSiedle (coefficients “F=R.A.D” [Free-78]). It has been extended by [McGa-80] and [Benedetto] (introducing coefficients ft, fy). The present formulation is that of the CEA team (coefficients a=σ.Nt, Y and ft functions, F functions cf. [Lera-89a], [Pail-99] and this chapter). All these formulations proceed from the same spirit. Authors have widened this formulation to interface states (cf. hereafter) [Flee-92, Berl-92], but physical basis are not so firm as for charge trapping (coefficients fot, fit). Anyway the major drawback of these schemes is to skip all the physics involved in the kinetics of charge build-up and detrapping. But they are instrumental as a first-order approach. 3.13.2. NUMBER OF TRAPS NT AS A KEY PARAMETER When at sufficiently high dose, all traps are filled, and saturation is reached. Charge density is then in the case of interface trapping: (72)
Qot = Nt III-43
and therefore the saturation value of the threshold voltage shift is (under the simplifying assumptions): (73)
with A =
∆Vot ,sat = − A N t t ox
q ε ox
& / 1012 charges.cm-2 = 4.76 Volts / k A
According to this model, the parameter Nt is present at low doses and large doses (fig. 28). It is therefore this parameter that one must reduce. This generally implies the modification of the manufacturing process.
Figure 28. Separation of Vot threshold voltage component at low and high dose under simplifying assumptions
We note that this simple formula is linear as a function of dose (Volt/kilorad) and quadratic as compared to the oxide thickness. It applies only at low to moderate dose (e.g. some kilorads or hundred kilorad) and depends substantially on manufacturing process and oxide thickmess. Beyond this limit, the response is governed by a more complex exponential relationship or can’t be understood without numerical integration of the set of differential equations, with the correct parameters. 3.14.
EXTRACTION OF TRAP PARAMETERS
Once we have determined the trapping behavior of a particular oxide by means of the techniques presented in the preceding sections, we can try to estimate the characteristic parameters of the oxide traps. The reasoning is made then in terms of net equivalent charge projected at the Si/SiO2 interface, which corresponds to what is measured experimentally. We therefore use the equation of the voltage shift due to a trap distribution at the interface in the regime of strong applied electric field (i.e. with fh(E) ≈ 1): (74)
∆Vot+ = −
[
]
q N TPS t ox 1 − exp(− σ pt g 0 f h (E) Y(E) t ox [1 + F(E)] D ) ε ox 1 + F(E) III-44
in which we further assume F(E)≈0. By differentiating with respect to D, we can write:
(75)
d∆Vot + ln dD
= − σ pt N TPS q g 0 Y(E) t ox 2 − σ pt g 0 Y(E) t ox D ε ox
The plotting of ln(-d∆Vot+/dD) as a function of dose leads to a separate extraction of the capture cross section and of the net equivalent charge at the interface, by determining the slope of the obtained curve and its intercept with the y-axis. For hole traps, this procedure of parameter extraction is only valid when the internal field is not too significantly disturbed by the trapped charge, i.e. in the low dose regime with a large applied electric field. When charge trapping occurs at the Si/SiO2 interface, the extraction of trap parameters is straightforward from the set of Eqns. (75) and (76) after extraction of the Vot component,. It becomes more difficult when charge trapping takes place in the oxide bulk (Eqn. 52 or 57). However this method is limited, because it only gives access to the product of the capture cross section by the trap density. The evaluation of the trap density usually requires to push irradiation in the saturation region, where internal electric field is distorted by the space-charge effect due to the trapped charges itself. Anyway, to assess this self-consistent effect, it is necessary to use a model of trap profile. 3.15.
CASE WHERE TWO TYPE OF TRAPS ARE PRESENT
3.15.1. CASE OF UNIFORM DISTRIBUTION OF TRAPS In the same manner as before, we can calculate the voltage shift associated with a negative charge trapped on uniformly distributed electron traps, for a positive applied bias In the case of uniform trapping, we must go back to differential equations (boundary conditions have to be exchanged) and the analytical resolution gives [Pail-99]:
(76)
+
∆Vot ,e = −
qg 0 ε ox
1 1 1 (1 − exp(− σ pt N TN t ox )) Y(E) t ox 2 D − 1 − 2 σ nt N TN t ox σ nt N TN t ox
The net total shift obtained for a positive applied field is then simply the sum of the contributions 1 due to negative and positive trapped charges. We then obtain (with λn = being the σ nt N TN mean free path for electrons): (77) qg − 0 ε ox
+
∆Vot ,e + h = λ p t ox
t exp − ox λ p
λp 1 − exp − t ox − t λ ox p
λ n λ n t + 1 − exp − ox 1− λ t t ox ox n
III-45
Y(E) t ox 2 D
and symmetrically for negative bias: (78)
−
∆Vot ,e + h =
λ λ t ox λ n t ox λ n t ox p p Y(E) t ox 2 D − 1 − + 1 − exp − 1 − exp − exp − λ n t ox t ox λ p t ox λ n t ox We therefore come to a system of two non-linear equations with two variables, which must be solved to extract the values of λn and λp from experimental data. (In the literature, this type of method has been first applied to photocurrents [Bene-87, Penn-90, Penn-92, Boes-94]. We here consider a similar approach, based on voltage shifts). The main disadvantage of such models is that it only gives access to values of the σtNT products. It is not possible to assess independently the capture cross-section and the trap density. qg − 0 ε ox
For that purpose, it is necessary to use a simpler model, based on the following idea: although trapped charges are distributed in the bulk, what we experimentally measure is the net equivalent charge, projected at the Si/SiO2 interface. It is thus possible to assimilate this equivalent charge to a charge located at the Si/SiO2 interface, and to use a simplified model, presented in the following paragraph. 3.15.2. CASE OF INTERFACIAL DISTRIBUTION OF TRAPS Electron trapping in oxides is only revealed in the case of a high dose irradiation when a large negative bias is applied. Due to large discrepancies in the capture cross sections of holes and electrons, the corresponding net voltage shift can be separated into two exponential components given by (67). Each one can be fitted with its set of trap-related parameters.
III-46
Figure 29. Oxide trapped charge contribution ∆Vot as a function of dose for -1 MV/cm back gate applied field [Pail-95b].
∆Vot
(Volts) 5 0 -5 -10
The symbols refer to different SIMOX oxide variants: Thin oxide, Medium I, Medium II, Standard. The solid lines correspond to the biexponential model fit. i S
iO S 2
-15
Gate
ρ E>0 ox
1µm) substrate (“bulk-like technologies”), no effect of parasitic buried transistor with ionizing dose is observed. In this case the hardness of an SOI technology is identical to bulk technology since buried oxide is moved away from Drain/Source [Davi-85, Plat-88, Flam-93, Flam-94].
-
For technology processed on relatively thin or intermediate-thin silicon (tSi ~ 2000 Å), devices operate in partial depletion, and coupling effects with buried transistor are evolving with ionizing dose. This coupling with the effect of field of charge then trapped in the buried oxide obliges us to take into account the hardening of this type of oxide [Lera-85].
-
Transistors manufactured on very thin silicon (tSi ~ 800 Å or less) work in total depletion regime, and this mode is directly controlled by the electric field created by charges which are present in the buried oxide. Thus, the behavior of the buried oxide is of extreme importance, especially for applications where high ionizing dose level is encountered [Ferl-96].
The total dose hardness of an SOI technology depends primarily on the radiation response of the same kinds of oxides as in bulk technologies: gate oxide and lateral isolation. This lateral isolation can be LOCOS or Trench, as in bulk silicon. Another system of lateral isolation specific to SOI is Mesa, which can be view as a Trench without refill. And finally, the ionizing dose response of MOS/SOI transistors can also depends on Buried Oxide, as is descibed in the following subsection. 10.3.
PARASITICS IN A THIN-FILM SOI TECHNOLOGY
Once a SOI structure is irradiated, charge trapping in the buried oxide gives birth to a parasitic effect, called “back parasitic transistor” as represented in fig. 93. A leakage current is usually observed in N-type MOS transistors, due to an electron conduction channel, which shows that the trapped charge is overall positive.
III-98
Polysilicon gate Gate oxide
N+ diffusion LOCOS Back conduction Buried oxide Silicon substrate
Figure 94. Illustration of a radiation-induced back transistor in a SOI structure (after [Ferl-96, Flam-95c]).
Charge trapping not only generates a back-gate FET effect but also a lateral FET effect and often also a parasitic bipolar transistor effect. The improvement brought about by the SOI technology over bulk CMOS technology is that these parasitic structures are located in the silicon island instead of being distributed in the bulk or at the surface of the silicon chip. On the other hand, as noted above, a new parasitic FET transistor appears due to the buried oxide, the underlying substrate acting as a back gate.
Figure 95. Illustration of parasitic elements in a MOS/SOI or MOS/SOS transistor a) The five regions of transistor b) The corresponding parasitic elements [Lera-88b]
10.4.
MODELING THE TRAPPED CHARGE IN SOI DEVICES
In the following, is presented the application of the modeling equations of chapter 3 to the realistic situation of a NMOS transistor on SOI, with the use of codes computing selfconsistently the charges, currents and the electric field. Two kinds of modeling have been presented up to know. A first kind uses the standard codes, by supposing a trapped charge distribution at the silicon surface. This allows useful insights, because the functioning of the silicon part of the device can be evaluated [Ferl-98]. But the modeling is not complete and is based on charge amount and profile that must be supposed. Major improvement is allowed by the use of self-consistent codes, in which the trapped charge is computed by taking into account fluxes, field and traps in the surrounding oxides. Such codes, necessarily in 2 or 3D for SOI, exist only since recently. For example self-consistent modeling with code B has been presented for a lateral cross-section of the SOI structure [Mila98], realistically computing the back-channel induced MOS as a function of dose and of the body-tied-to-source (BTS) dimensions. In the following, as another application of numerical modeling, we present a study using the code A, on a typical thin-film partially-depleted SOI technology, with gate length of 0.5 µm and III-99
LDD feature. This study can be viewed as “orthogonal” to the [Mila-98], as it operates in another cross-section, orthogonal to the previous one. We aim at describing here the other cause of leakage in NMOS/SOI, i.e. the back-channel leakage current, whereas [Mila-98] described the edge-channel leakage current. In SOI, the “mechanical substrate”, i.e. the bulk silicon underlying the buried oxide, is connected to the chip case, and usually tied to ground. In other terms, back-gate is usually at zero Volt with respect to the source. This is typically a case where electric field is low (determined, for instance, by built-in potential between the two silicon facing each other, e.g. P-type for the N-channel MOS and P- type for the bulk silicon substrate). If drain is itself at zero voltage during irradiation, in effect, trapping is at minimum and usually no back-channel parasitic transistor appears. However, it has been recognized since the early times that the worst-case of irradiation occurs when the NMOS drain bias is at supply voltage (e.g. +5 V). In this situation, the back-MOS threshold voltage shifts significantly and the back-MOS transistor can be revealed and cause a major drain-source leakage current. The following set of figures verifies this hypothesis, by computing the electric field, the induced MOS current in the back-side of the silicon film, the trapped charge in the buried with the selfconsistent code A. These set of experiments reveals the field-collapse effect in the two-dimensional situation of the buried oxide laterally biased the drain voltage. The first set of figures depicts the phenomena under a 4-V drain voltage, all the other electrodes being grounded. The last figure compares the trapped hole charge density for drain bias varying between 0 and 2 V. As it could be anticipated, the charge trapping is not nil at zero-bias, because of built-in fields resulting from doping. But, naturally, the amount of trapped charge is increase with the drain bias. This is why the dose threshold at which the back-gate leakage MOS is triggered is strongly influenced by the drain voltage.
III-100
Figure 96. Chart of the electric field in a 0.5 µm NMOS/SOI transistor in worst case during irradiation (VDS=+4V) using the self-consistent code A.
One clearly sees the strong deformation of the equipotential. At the beginning of the irradiation, we observe the situation of the “drain side-gating” in this thin-film SOI transistor: the electric field lines go directly from the drain to the backside of the transistor. Therefore, holes are directed toward the backside and build-up of the charge begins. In this figure, a 2D field collapse is clearly seen, as expected from [Boes-91] and the modeling in 1D exposed above in chapter 4.4. At the end of the irradiation, the field is nearly horizontal and reinforced, because the space charge has considerably extended.
III-101
Figure 97. Apparition of an electron current at the back-side of the MOS between 90 and 210 krads in the back-side of a SOI transistor due to the turn-on of the back parasitic MOS in a N transistor biased in worst case during irradiation (VDS=+4V) using the self-consistent code A.
III-102
VDS=+4V Figure 98. Trapped charge in the oxides of a NMOS/SOI transistor in worst case during irradiation using the self-consistent code A.
VDS=0
VDS=+1V
VDS=+2V
Figure 99. Charts of the charge build-up for other drain voltages of 0V, 1V and 2V using the self-consistent code A.
III-103
11. CONCLUSIONS: TRENDS, MODELING, AND NEW ISSUES As radiation dose increases for new generation of satellites, it becomes more stringent to correctly use devices closer to their limits. For that, we have to look further on modeling and characterization. In this course, we have focuses on the oxide trapped charge as it is of first concern in discrete parts (Power MOS), most post-irradiation effects (rebound) and circuit leakage currents (build-up and recovery), and explored links with process and various situations as case studies. As gate oxide thickness decreases with the progress device integration, the total dose hardness of MOS transistors is mostly limited by parasitic MOS structures. The gate oxide to field oxide bird's beak forms a non-planar transistor responsible for leakage current at low dose. 1D, 2D and 3D-simulation codes are available or are being developed to calculate charge transport and trapping in non-planar MOS structures. Some codes are introducing detrapping phenomena. The dose induced leakage current is due to the threshold voltage shift of the lateral parasitic transistor mainly in the edge, which is directly related to the slope of the surface of parasitic transistors. The isolation geometry is determinant for the behavior of a technology. The isolation shape, the silicon doping level, the supply voltage and the trap parameters in the insulators are the key parameter tools to optimize a rad-tolerant technology. This appears is the key parameter to evaluate the variation of the trapped charge density in the parasitic MOS structures. As test structures are very unpractical, only numerical simulation can help to figure out what is instrumental. Anyway, to feed the codes with data in proper foer, new methods for acquiring parameters have now to be developed. Numerical modeling of interface traps component build-up and bipolar transistor gain degradation is foreseeable, although effective links with the manufacturing process apparently seem more obscure than for the trapped charge. Dose-rate or time-dependent effects are beginning to be implemented, and thus provide help for better understanding of circuit behavior. These conceptual and numerical tools could provide help for technology assessments, ground testing and in-flight predictions. 12. ACKNOWLEDGMENTS Dan Fleetwood, Lew Cohn for encouragement and careful reading. Gérard Barbottin for help in English correction of a large part of this text. The author sincerely wishes to associate Philippe Paillet in co-authorship of large parts of the work concerning, among other topics, analytical solutions and the modeling of detrapping, and co-workers Jean-Luc Autran, Christian Chabrerie, Dominique Hervé, Olivier Flament, Véronique Ferlet-Cavrois and Alphonse Torrès and also Clément Tavernier and Philippe Calvel for their kind participation to the material prepared and exploited.
III-104
13. APPENDIX 13.1.
SELF-CONSISTENT CODES USED
In the following, we report the codes used here, which all share the essential characteristics to compute self consistently the potential, the current and the trapped charge. Only this type of codes can be credibly used for radiation effects modeling in oxides. Code A B C D E F Table 10.
13.2.
Name RAD-TRAPPOXTM v.5 ATLAS/DGEM/ORCHIDTM TRAPPOXR v.4 ACCESS-TRAPPOXR v.3 ACCESS-TRAPPOXR v.2 ASTEC3-TRAPPOXR v.1 and v.0
Origin ISE-CEA SILVACO [Mila-98] CEA [Lera-99b] CEA-CEM [Pail-98] CEA-CEM [Esco-95b, Bris-96] CEA [Herv-94, Lera-89]
Type 2D & 3D J and V-Models 2D & 3D J-Model 1D J and V-Models 2D+detrapping J-Model 2D J-Model 1D J and V-Models
Tables of the self-consistent codes used to illustrate this course.
PARAMETERS
We report here the main parameters used for simulations in chapter 4 (MOS) and 10 (MOS/SOI): 13.2.1. PARAMETERS USED IN CHAPTER 4 - The 450 nm (representing a Field-Oxide for field-collapse study). CEA Code: TRAPPOXR v.4 [Lera-99b] Parameters used: Model = J or V; oxide thickness tox = 450 nm; Finite difference; Number of nodes = 101; Not=1 1012 cm-2; NTP=4.41016 cm-3 ; σpt=1.10-12 cm2 + field dependence [Ning-76]; σpr=1.10-12 cm2; Y0 = 0.1; m =0.9;Ec = 6.5 105 V/cm; µ n= 10 cm2/V.s + field dependence; µ p= 10-5 cm2/V.s [Srour77b]+field dependence. ISE-CEA Code A: RAD-TRAPPOXTM v.5 Parameters used: Model = V; oxide thickness tox = 450 nm; Finite elements; Not=1. 1012 cm-2; NTP=4.41016 cm-3 ; σpt=1.10-12 cm2 + field dependence [Ning-76]; σpr=1.10-12 cm2; Y0 = 0.1; m =0.9;Ec = 6.5 105 V/cm; µ n= 10 cm2/V.s + field dependence; µ p= 10-5 cm2/V.s [Srour77b]+field dependence. SILVACO Code B Parameters used: Model = J; oxide thickness tox = 450 nm; Finite elements; Not=1. 1012 cm-2; NTP=4.41016 cm-3 ; σpt=1.10-12 cm2; σpr=0.10-12 cm2; Y0 = 0.0; Y=exponential saturation; µ p= 10-5 cm2/V. - The 300 nm (representing another Field-Oxide for gate bias switching). III-105
CEA Code: TRAPPOXR v.4 [Lera-99b] Parameters used: Model = J; oxide thickness tox = 300 nm; Finite difference;Number of nodes = 101; Not=0.5 1012 cm-2; NTP=3.31016 cm-3 ; σpt=1.10-12 cm2 + field dependence [Ning-76]; σpr=1.10-12 cm2; Y0 = 0.1; m =0.9;Ec = 6.5 105 V/cm; µ n= 10 cm2/V.s + field dependence; µ p= 10-5 cm2/V.s [Srour77b]+field dependence. 13.2.2. PARAMETERS USED IN CHAPTER 10 ISE-CEA Code A: RAD-TRAPPOXTM v.5 Parameters used: Model = V; Finite elements; gate length Lg= 0.5 µm; LDD ; gate oxide thickness tox1 = 10 nm; NTP1=3. 1018 cm-3; uniform density of traps; buried oxide thickness tox2 = 300 nm; NTP2=3. 1018 cm-3; uniform density of traps; for the two oxides: σpt=5.10-14 cm2 + field dependence [Ning-76]; σpr=5.10-14 cm2; Y0 = 0.1; m =0.9;Ec = 6.5 105 V/cm; µ n= 10 cm2/V.s + field dependence; µ p= 10-5 cm2/V.s [Srour77b]+field dependence. 14. REFERENCES [1019] [1892] [22900] [Adam-76] [Adam-77] [Adam-91] [Aubu-71] [Ausm-86] [Autr-95]
[Autr-99]
[Babc-95]
[Bart-97] [Bene-86] [Bene-87]
[Boes-75] [Boes-76] [Boes-86]
MIL-Std 883, Method 1019, DoD, USA and Web pages. ASTM F-1892, USA and Wep pages. ESA/SCC Basic Specifications, Method 22900 and Web pages for preview. J.R. Adams, F.N. Coppage, “Field oxide inversion effects in irradiated CMOS devices”, IEEE Trans Nucl. Sci., NS-23, 6, 1604 (1976). J.R. Adams, W.R. Dawes, T.J. Sanders, “A radiation-hardened field oxide”, IEEE Trans. Nucl. Sci., NS-24, 6, 2099 (Dec. 1977). L. Adams and A. Holmes-Siedle, “La survie de l'électronique dans l'espace”, La Recherche, Vol. 22, Oct. 1991, pp. 1182-1189. K.G. Aubuchon, “Radiation hardening of PMOS devices by optimization of thermal SiO2 gate insulator”, IEEE Trans. Nucl. Sci. NS-17, 6, (1971). G.A. Ausman, “Field-dependence of geminate recombination in a dielectric medium”, Harry Diamond Laboratories report n° 2097, Adelphi, MD, U.S.A., (1986). J.L. Autran, P. Paillet, J.L. Leray, R. Devine, “Conduction mechanisms and space-charge effects in typical thin-film insulators (SiO2, Ta2O5, PbZrxTi1-xO3)”, 2nd Intl Conf. On Space Cherge in Solids Dielectrics CSC2, edited in “Le Vide”, n°275, March 1995, pp. 44-53. J.L. Autran, “Charge Pumping Techniques. Their use for diagnosis and interface studies in MOS transistors”, in “Instabilities in Silicon Devices”, Vol. 3, pp. 405-494, G. Barbottin and A. Vapaille, Editors (North-Holland ISBN 0 444 81801 4). J.A.Babcock, S.D. Clark, J.D. Cressler, D.L. Harame, R.C. Jaeger and L.S. Vempati, “Ionizing radiation tolerance of high performance SiGe HBTs grown by UHV/CVD”, IEEE Trans. Nucl. Sci ,NS-42, 6, 1558 (1995). J. Barth, “Modeling Space Radiation Environment”, 1997 IEEE NSREC Short Course. J.M. Benedetto and H.E. Boesch, “The relationship between 60Co and 10-keV x-ray damage in MOS devices”, IEEE Trans. Nucl. Sci, NS-33, No 6., 1318 (1986). J.M. Benedetto, H.E. Boesch, T.R. Oldham and G.A. Brown, “Measurement of low-energy x-ray dose enhancement in MOS devices with metal silicided devices”, IEEE Trans. Nucl. Sci., NS-34, 6, 1540 (1987). H.E. Boesch, F.B. McLean, J.M. McGarrity, J. Ausman, “Hole transport and charge relaxation in irradiated MOS capacitor”, T-NS 22, 6, 2163 (1975). H.E. Boesch and J.M. McGarrity, “Charge yield and dose effects in MOS capacitors at 80K”, IEEE Trans. Nucl. Sci., NS-23, 6, 1520 (1976). H. E. Boesch, F. B. McLean, J. Benedetto and J. McGarrity, “Saturation of threshold voltage shift in All references are unclassified
III-106
[Boes-88] [Boes-91] [Boes-94] [Boud-93] [Bris-96]
[Brow-81] [Brow-83] [Brow-85] [Brow-90] [Calv-99] [Chab-97a]
[Chab-97b]
[Chu-90] [Cirb-95]
[Coli-97] [Conl-97] [Cour-99] [Cris-95] [CTVS] [Curt-75] [Curt-77] [Daly-88] [Davi-85] [Dawe-76] [Deal-80] [Derb-75]
NS-33, 6, 1191 (1986). H.E. Boesch, “Time-dependent interface trap effects in MOS devices”, IEEE Trans. Nucl. Sci. NS35, 6, 1160 (1988). H. E. Boesch, G. A. Brown, “Charge buildup at high dose and low fields in SIMOX buried oxides”, IEEE Trans. Nucl. Sci., NS-38, 6, 1234 (1991). H.E. Boesch and C.A. Pennise, “Measurement and modeling of radiation response of multilayer BESOI buried insulators”, IEEE Trans. Nucl. Sci., NS-41, 6, 2322 (1994). J.C.Boudenot, “Interaction Mechanisms and Environments”, Short Course, RADECS-93 Conference (The RADECS Association, C/O Joseph Pinel, Toulouse, France). C. Brisset, V. Ferlet-Cavrois, O. Flament, O. Musseau, J. L. Leray, J. L. Pelloie, R. Escoffier, A.Michez, C. Cirba, G. Bordure, “Two-dimensional simulation of total dose effects on NMOSFET with lateral parasitic transistor”, IEEE Trans. Nucl. Sci., NS-43, 6, 2651 (1996). C.M. Dozier and D.B. Brown, “Electron-hole recombination from a microdosimetry viewpoint”, IEEE Trans. Nucl. Sci., NS-28, 6, 4142 (1981). D.B. Brown, “Thermal annealing of radiation induced defects: a diffusion-limited process? Trans. Nucl. Sci., NS-30, 6, 4059 (1983). D.B. Brown, “Time dependence of interface states production”, IEEE Trans. Nucl. Sci., NS-32, 6, 3900 (1985). D. B. Brown, “Total-Dose Effects at Dose Rates Typical of Space”, 1990 IEEE NSREC Short Course. Philippe Calvel, Alcatel-Space, Priv. Com. C. Chabrerie, “De l’utilisation des recuits isothermes et isochrones pour la caractérisation de structures CMOS irradiées”, Doctoral Thesis in Physics, Université Paris 7, Paris, France, March 1997 (available in microfiche at ANRT/University of Grenoble-II). C. Chabrerie, J.L. Autran, P. Paillet, O. Flament, J.L. Leray and J.C. Boudenot, “Isothermal and isochronal annealing method to study post-irradiation temperature activated phenomena”, IEEE Trans. Nucl. Sci, NS-44, 6, 2007 (1997). A.X. Chu and W.B. Fowler, “Theory of oxide defects near the Si-SiO2 interface”, Phys. Rev. B, 41, 8, 5061 (1990). C. Cirba, “Simulation numérique du piégeage et du dépiégeage dans les oxydes de composants MOS”, Thèse de doctorat en Sciences Physiques, Université de Montpellier II, France, Dec. 16th , 1996, n°96MON2237 (available in microfiche at ANRT/University of Grenoble-II). J.-P. Colinge, “Silicon-On-Insulator Technology: Material to VLSI”, Kluwer Academic Publishers, Boston, (2nd edition, 1997) ISBN 0-7923-8007-X. J.F. Conley, P.M. Lenahan, B.D. Wallace and P. Cole, “Quantitative model of radiation induced charge trapping in SiO2”, IEEE Trans. Nucl. Sci., NS-44, 6, 1804 (1997). A. Courtot-Descharles, P. Paillet, J.L. Leray, “Theoretical study using density functional theory of defects in amorphous silicon dioxide”, Journ. Of Non-Crystalline Solids, 245, 154 (1990). S. Cristoloveanu and S.S.Li, “Electrical characterization of Silicon-On-Insulator materials and , Kluwer Academic Publishers (1995), ISBN 0-7923-9548-4. “Cours de Technologie des Véhicules Spatiaux”, Vol. 1 et 3. Centre National d’Etude Spatiales CEPADUES-Editions (111 rue Nicolas Vauquelin, Toulouse, France). O.L. Curtis, J.R. Srour and K.Y. Chiu, “Physical mechanisms of radiation hardening of MOS devices NS-22, 6, 2174 (1975). O.L. Curtis, J.R. Srour, “The multiple trapping model and hole transport in a-SiO2”, J. App. Phys. 48, 9, 3819 (1977). E.J. Daly, “Evaluation of the space environment for the ESA projects”, ESA Journal, 88, 12, 229 (1988). G.E Davis, T.G. Blake, C.E. Chen, R. DeMoyer Jr, L.R. Hite, H.W. Lam, “Transient radiation effects NS-42, 6, 4432 (1985). W.R. Dawes, Jr., G.F. Derbenwick, and B.L. Gregory, “Process technology for radiation-hardened CMOS integrated circuits”, IEEE J. Solid State Circuits SC11(4), 459 (1976). B.E. Deal, “Standardized terminology for oxide charges associated with thermally oxidized silicon”, IEEE Trans. Electron Devices, ED-27, 3, 606-607 (1980). G.F. Derbenwick and B. Gregory, “Process optimization of radiation-hardened CMOS integrated circuits”, IEEE Trans. Nucl. Sci, NS-22, 6, 2151 (1975). All references are unclassified
III-107
[Devi-94] [Devi-95] [Dozi-81] [Dres-98] [Dupo-91]
[Dyer-98] [Eern-76] [Emil-96] [Enlo-89] [Esco-95a]
[Esco-95b]
R.A.B. Devine, “The structure of SiO2, its defects and radiation hardness”, IEEE Trans. Nucl. Sci., NS-41, 3, 452 (June 1994) and Proceeding RADECS-93. R.A.B. Devine, W.L. Warren, J.B. Xu, I.H. Wilson, P. Paillet and J.L. Leray, “Oxygen gettering and oxide degradation during annealing of Si/SiO2/Si structures”, J.Appl. Phys., 77 (1), 1, (1995). D.B. Brown and C.M. Dozier, “Effect of photon energy on the response of MOS devices”, IEEE Trans. Nucl. Sci., NS-28, 6, 4137 (1981). Paul V. Dressendorfer, “Basic Mechanisms for the New Millenium”, IEEE 1998 NSREC Short Course. E.Dupont-Nivet, E.Delagnes, J.L.Leray, J.L.Martin, J.Montaron, J.P.Blanc, E.Delevoye, J.Gautier, J. de Poncharra, R.Truche, E.Beuville, M.Dentan and N.Fourches, “A hardened technology on SOI for analog devices”, RADECS-91, IEEE Proceedings 91TH0400-2, 15, p. 211 (1991). C. Dyer, “Space Radiation Environment Dosimetry”, 1998 IEEE NSREC Short Course. E.P.Eernisse and G.F.Derbenwick, “Viscous shear flow model for MOS device radiation sensitivity”, IEEE Trans. Nucl. Sci. NS-23, 6, 1534 (1976). David Emily, “Total-Dose Response of Bipolar Microcircuits”, 1996 IEEE NSREC Short Course. E.W. Enlow, R.L. Pease, W.E. Combs and D.G. Platteter, “Total dose induced hole trapping in trench NS-36, 6, 2415 (1989). R. Escoffier, “Simulation numérique de l’effet des charges induites par l’irradiation dans les oxydes des structures MOS”, Thèse de doctorat en Sciences Physiques, Université de Montpellier II, France, Sept. 29th ,1995, (available in microfiche at ANRT/University of Grenoble-II). R. Escoffier, A. Michez, C. Cirba, G. Bordure, P. Paillet, V. Ferlet-Cavrois and J.L. Leray, “Radiation induced shift study in parasitic MOS structures by 2D numerical simulation”, 3nd European Conference on Radiation and their Effects on Devices and Systems (RADECS-95), IEEE 95TH8147.
[Euro-92]
[Farm-73]
[Ferl-96]
J.L. Leray, Training Course ME-10 “Electromagnetic and radiation Effects on Electronics”, Lecture 07 available in video and hard copy at EUROPACE (EUROPean Advanced Continuing Education), Paris La Défense (92), France. J. Farmer, “X-ray induced currents and space-charge build-up in MOS capacitor”, PhD Thesis, Kansas State University, 74-25/599 (1974) (available in microfiche at Xerox University Microfilms, Ann Arbor, Michigan). V. Ferlet-Cavrois, O. Musseau, J. L. Leray, J. L. Pelloie, C. Raynaud, “Total dose effects on short channel NMOS/SOI transistors and its lateral transistor”, IEEE Trans. Electron Devices., ED-44, 6, 965 (1996) and 3nd European Conference on Radiation and their Effects on Devices and Systems (RADECS95), IEEE 95TH8147, p. 142 (1995).
[Ferl-98]
[Ferr] [Flam-92]
[Flam-94]
[Flam-95a] [Flam-95b]
[Flam-95c]
[Flam-98]
[Flam-99]
V. Ferlet-Cavrois, O. Musseau, O. Flament, J. L. Leray, J. L. Pelloie, C. Raynaud and O. Faynot, “Total dose induced latch in short channel NMOS/SOI transistors”, IEEE Trans. Nucl. Sci., NS-45, 6, 2458 (1992). D.K. Ferry, “Semiconductors”, p. 323, Maxwell MacMillan Engineering Editions. O. Flament, D. Hervé, O. Musseau, Ph. Bonnel, M. Raffaelli, J.L. Leray, J. Margail, B. Giffard, A.J. Auberton-Hervé, “Field dependent charge trapping effects in SIMOX buried oxides at very high ,NS-39, 6, 2132 (1992). O.Flament, J.L.Leray, J.L.Martin, J.Montaron, M.Raffaelli, J.P.Blanc, E.Delevoye, J.Gautier, J.de Poncharra, R.Truche, E.Delagnes, M.Dentan and N.Fourches, “Radiation effects on SOI analog devices parameters”, Trans. Nucl. Sci., NS-41, 3, 565 (1994) and RADECS-93, IEEE Conference Records Catalog Number 93TH0616-3, ISBN 0-7803-1793-9. O. Flament, “X-ray radiation sources for process development and qualification”, L'Onde Electrique, 75, May 3rd (1995), and Proc. of the RADECS-SEE Workshop, Toulouse 23 Sept. 1994. O.Flament, J.L.Leray, F. Martin, E.Orsier, J.L.Pelloie, R.Truche, and R.A.B. Devine, “The effect of rapid thermal annealing on radiation hardening of MOS devices”, IEEE Trans. Nucl. Sci., NS-42, 6, 1667 (1995). O.Flament, J.L.Leray, O.Musseau, “A mixed-mode analog-digital CMOS-NPN-PJFET on insulator technology and applications”, in “ , G.A. Machado Ed., IEE Circuits and Systems Series 8, Laventham Press, Laventham, UK ISBN 0-85296-874-4 (1995). O. Flament, C. Chabrerie, V. Ferlet-Cavrois, J.L. Leray, F. Faccio and P. Jarron, “A methodology to study lateral parasitic transistor in CMOS technologies”, IEEE Trans. Nucl. Sci., NS-45, 3, 1385 (June 1998). O. Flament, P. Paillet, J.L. Leray and D.M. Fleetwood, “Consideration on isochronal technique: from All references are unclassified
III-108
[Flee-90] [Flee-91] [Flee-92a] [Flee-92b] [Flee-92c]
[Flee-95] [Flee-98]
[Free-78] [Gall-84] [Gall-85] [Grav-98]
[Grisc-85] [Grisc-88]
[Grov-66] [Hara-95]
[Herv-93]
[Herv-94] [Holm-93] [Hugh-71] [Hugh-72] [Hugh-73] [Hugh-75a] [Hugh-75b] [Hugh-78] [Hugh-85] [John-96]
measurement to Physics”, 1999 NSREC, this Conference. D.M. Fleetwood, P.S. Winokur and L.C. Riewe, “Field dependence of interface-trap buildup in polysilicon and metal gate MOS devices”, IEEE Trans. Nucl. Sci., NS-37, 6, 1632, (1990). D. M. Fleetwood, R. A. Reber, Jr., and P. S. Winokur, “Effect of bias on thermally stimulated current (TSC) in Irradiated MOS devices”, IEEE Trans. Nucl. Sci., NS-38, 6, 1066, (1991). D.M. Fleetwood, “Long-term annealing study of midgap interface-trap charge neutrality”, Appl. Phys. Lett., 60, 2883 (1992). D.M. Fleetwood, “Border traps in MOS devices”, IEEE Trans. Nucl. Sci., NS-39, 2, 269-71 (Apr. 1992). D. M. Fleetwood, S. L. Miller, R. A. Reber, Jr., P. J. McWhorter, P. S. Winokur, M. R. Shaneyfelt, and J. R. Schwank, “New insights into radiation-induced oxide-trap charge through thermally stimulated current measurement and analysis”, IEEE Trans. Nucl. Sci., NS-39, 6, 2192 (1992). D. M. Fleetwood, “A First-Principle Approach to Total-Dose Hardness Assurance”, 1995 IEEE NSREC Short Course. D.M. Fleetwood, P.S. Winokur, M.R. Shaneyfelt, L.C. Riewe, O. Flament, P. Paillet and J.L. Leray, “Effects of isochronal annealing and irradiation temperature on radiation-induced trapped charge”, IEEE Trans. Nucl. Sci, NS-45, 6, 2366 (1998). R.F.A. Freeman, A.G. Holmes-Siedle, “A simple model for the effects of radiation on MOS devices”, IEEE Trans. Nucl. Sci., NS-25, 6, 1216 (1978). K.G. Galloway, M. Gaitan and T.J. Russell, “A Simple Model for Separating Interface and Oxide Charge Effects in MOS Device Characteristics”, IEEE Trans Nucl. Sci., NS-31, 6, 1497 (1984). K.G. Galloway, C.L. Wilson, L.C. Witte, “Charge-Sheet Model Fitting to Extract Radiation-Included Oxide and Interface Charge”, IEEE Trans Nucl. Sci., NS-32, 6, 4461 (1985). R.J. Graves, C.R. Cirba, R.D. Schrimpf, R.J. Milanowski, A. Michez, D.M. Fleetwood, S.C. Witczak, F. Saigne, “Modeling low-dose-rate effects in irradiated bipolar-base oxides”, IEEE Trans. Nucl. Sci., NS-45, 6, 2352 (1998). D.L. Griscom, “Diffusion of radiolytic molecular hydrogen as a mechanism for the post-irradiation build-up of interface states in SiO2 on Si structures”, J.App.Phys., 58, 7, 2524 (1985). D.L. Griscom, D.B. Brown, N.S. Saks, “Nature of radiation-induced point defects in amorphous SiO2 and their role in SiO2-on-Si structures”, in “The physics and chemistry of SiO2 and the Si-SiO2 interface”, C.R. Helms and B.E. Deal Editors, Plenum Press (1988), p. 287. A.S. Grove and E.H. Snow, Proc. IEEE, 54, 1966, pp. 894-895. D.L. Harame, J.H. Comfort, J.D. Cressler, E.F. Crabbé, J.Y.C. Sun, B.S. Meyerson, and T.Tice, “Si/SiGe epitaxial-base transistors-Part II: Process integration and analog applications”, IEEE Trans. Electron Devices, NS-42, 3, 469 (1995). D. Hervé, “Effets à long terme des rayonnements dans les structures métal-oxyde-semiconducteur. Application aux dispositifs à transfert de charge”, Doctoral Thesis in Physics, n°2711, Université Paris 11, Orsay, France, Jan. 1993 (available in microfiche at ANRT/University of Grenoble-II). D. Hervé, P. Paillet and J.L. Leray, “Space charge effects in SIMOX buried oxides”, IEEE Trans. Nucl. Sci., NS-41, 466 (1994). A. Holmes-Siedle and L. Adams, “Handbook of radiation effects”, Oxford University Press (1993). H.L. Hughes, “A survey of radiation-induced perturbations in Metal-Insulator-Semiconductor th Ann. Proc. Reliability Physics Symposium, 33 (1972). H.L. Hughes, “Dependence of MOS device radiation sensitivity on oxide impurities”, IEEE Trans. Nucl. Sci., NS-19, 6, 256 (1972). R.C. Hughes, “Charge-carrier transport phenomena in amorphous SiO2: direct measurement of the drift mobility and lifetime”, Phys. Rev. Lett., 30, n°26, 1333 (1973). R.C. Hughes, E.P. EerNisse and H.J. Stein, “Hole transport in MOS oxides”, IEEE Trans. Nucl. Sci., NS-22, 6, 2227 (1975). R.C. Hughes, “Hole mobility and transport in thin SiO2 films”, App. Phys. Lett., 26, n°8, 436 (1975). R.C. Hughes, “High field properties of SiO2”, Solid-State Electron., 21, 251-258 (1978). R.C. Hughes, “Theory of response of radiation sensing field effect transistors”, J. Appl. Phys., 58, 1375 (1985). A.H. Johnston, C.I. Lee and B.G. Rax, “Enhanced damage in bipolar devices at low dose rates: effects at very low dose rates”, IEEE Nucl. Sci., NS-43, 6, 3049 (1996). All references are unclassified
III-109
[John-98] [Kako-89] [Kern-92] [Kim-88] [Kooi-91] [Kosi-94]
[Kran-87] [Kran-91] [Lai-83] [Lee-91] [Lena-83] [Lera-85] [Lera-88a] [Lera-88b]
[Lera-88c]
[Lera-89a]
[Lera-89b] [Lera-90]
[Lera-99a]
[Lera-99b]
[Ma-89] [Mahe-94] [McGa-80] [McLe-80]
A.H. Johnston, “Radiation Effects in Advanced Microelectronics technologies”, RADECS-97 Conference and , IEEE Trans. Nucl. Sci., NS-45, 3, 1339 (June 1998). R. Kakoschke and E. Bussmann, “Simulation of temperature effects during rapid thermal processing”, MRS Symposium Proceedings, Vol. 146 (1989). K.G. Kerns and M.S. Gussenhoven, report PL-TR-922201 n°1105 on CRESSRAD experiment, USAir Force Philips Lab. (1992). Y.Y. Kim and P.M. Lenahan, “Electron-Spin-Resonance study of radiation-induced paramagnetic defects in oxides grown on (100) silicon substrates”, J. Appl. Phys., 64, p. 3551, (1988). E. Kooi, “The invention of LOCOS”, IEEE Case studies of achievement in science and technology Vol. 1, NY (1991). S.L. Kosier, W.E. Combs, A. Wei, R.D. Schrimpf, D.M. Fleetwood, M. DeLaus and R.L. Pease, “Bounding the total-dose response of modern bipolar transistors”, IEEE Trans. Nucl. Sci., NS-41, 6, 1864 (1994). R.J. Krantz, L.W. Aukerman, T.C. Zietlow, “Applied field and total dose dependence of trapped charge build-up”, IEEE Trans. Nucl. Sci., NS-34, 6, p. 1196, (1987). R.J. Krantz, J. Scarpulla, J.S. Cable, “Total dose-induced charge build-up in nitrided-oxide MOS devices “, IEEE Trans. Nucl. Sci. NS-38, 6, Dec. 1991, pp. 1746. S.K. Lai, “Interface trap generation in silicon dioxide when electrons are captured by trapped holes”, J.App.Phys., 54, 5, 2540 (1983). Der-Sun Lee and Chung-Yu Chan, “Oxide charge accumulation in Metal-Oxide Semiconductors devices during Irradiation”, J. Appl. Phys., 69 (10), 15 May 1991. P.M. Lenahan and P.V. Dressendorfer, “Micro-structural variations in radiation hard and soft oxides observed through electron-spin-resonance”, IEEE Trans. Nucl. Sci., NS-30, 6, 4602 (1983). J.L. Leray, “Activation energies of oxide charge recovery in SOS or SOI structures after an ionizing NS-32, 6, 3921 (1985). J.L. Leray, “Assessments of SOI technologies for hardening”, Microelectronic Engineering, Vol. 8 , n°3-4, p.187 (Dec. 1988) and 1st European SOI workshop, Meylan, France, 1988. J.L. Leray, E. Dupont-Nivet, P. Lalande, O. Musseau, Y.M. Coïc, A.J. Auberton-Hervé, M. Bruel, C. VLSI : investigations of hardened SIMOX without epitaxy for dose, dose rate and SEU”, IEEE Trans. Nucl. Science, NS-35, 6, 1355 (1988). J.L. Leray, O. Musseau, E. Dupont-Nivet, J.F. Péré and Y.M. Coïc, “A study of the sensitivity to cosmic rays and integrated dose of bit-slice microprocessors in CMOS and CMOS on Insulator technologies”, 6th International Conference on Reliability and Maintainability, Biarritz “λ−µ6”, ADERA edit. (Bordeaux, France), p. 625 (1998). J.L. Leray, “Contribution à l'étude des phénomènes induits par les rayonnements ionisants dans les structures à effet de champ au silicium et à l'arséniure de gallium”, Thèse de doctorat d'Etat es Sciences Physiques, n°3576, Université Paris 11, Orsay, France (Dec. 1989) (available in microfiche at ANRT/University of Grenoble-II). J.L. Leray, “Structures à Effet de Champ en Irradiation. Effets de la Dose de Rayonnement”, Short Course, RADECS-1989 (The RADECS Association, C/O Joseph Pinel, Toulouse, France). J.L. Leray, E. Dupont-Nivet, J.F. Péré, Y.M. Coïc, M. Raffaelli, A.J. Auberton-Hervé, M. Bruel, B. Giffard and J. Margail, “CMOS/SOI hardening above 100 megarad(SiO2): limits and liabilities”, IEEE Trans. Nucl. Science, NS-37, 6 (1990). J.L. Leray, “A review of buried oxides and SOI technologies”, in “Instabilities in Silicon Devices”, Vol. 3, pp. 145-232, G. Barbottin and A. Vapaille, Editors (North-Holland ISBN 0 444 81801 4) (1999). J.L Leray, P. Paillet, O. Flament and A. Torrès, “Oxide charge modeling with Trappox code version 4. Comparison of trapping models on desktop computer”, to be presented at 1999 RADECS Conference, Fontevrault (France). TP. Ma et P. V. Dressendorfer, “Ionizing radiation effects in MOS devices and circuits”, J. Wiley & sons, N.Y. (1989). M. Maher, “Total dose testing of Advanced CMOS Logic at low voltage”, National Semiconductor, Application Note AN-927 (1994) (available in Web site). J.M. McGarrity, “Considerations for hardening MOS devices and circuits for low radiation doses”, IEEE Trans. Nucl. Sci., NS-27, 6, 1739 (Dec. 1980). F.B. McLean “A framework for understanding radiation-induced interface states in MOS structures”, All references are unclassified
III-110
[McLe-87] [McWh-86] [McWh-88] [McWh-90] [Mess-92]
IEEE Nuc.Sci. NS-27, 6, 1651 (1980). F.B. McLean, “Basic Mechanisms of Radiation Effects on Electronic Materials, Devices and Integrated Crcuits”, 1987 IEEE NSREC Short Course. P.J. McWhorter and P.S. Winokur, “Simple technique for separating the effects of interface traps and trapped-oxide charge in Metal-Oxide-Semiconductor transistors”, Appl. Phys. Lett. 48, 133 (1986). P.J. McWhorter, P.S. Winokur, and R.A. Pastorek, “Donor/acceptor nature of radiation-induced interface traps”, IEEE Trans. Nucl. Sci., NS-35, 6, 1154 (1988). P.J. McWhorter, S. L. Miller, and W. M. Miller, “Modeling the anneal of radiation-induced trapped holes in a varying thermal environment”, IEEE Nucl. Sci. NS-37, 6, 1682 (1990). G.C. Messenger and M.S. Ash, “The Effects of Radiations on Electronic Systems”, van Nostrand Rheinhold, ISBN 0-442-2359-1 (2nd Ed., 1992), p. 334, ISBN 0-442-25417-2.
[Mila-98]
R.J. Milanowski, M.P. Pagey, L.W. Massengill, R.D. Schrimpf, M.E. Wood, B.W. Offord, R.J. Graves, K.F. Galloway, C.J. Nicklaw and E.P. Kelley, “TCAD-assisted analysis of back-channel leakage in irradiated Mesa SOI NMOSFETs”, IEEE Trans. Nucl. Sci., NS-45, 6, 2593 (1998).
[Mitc-67]
J.P. Mitchell and D.K. Wilson, “Surface effects of radiation on semiconductor devices”, The Bell System Technical Journal, Vol. XLVI, n°1, pp. 1-80 (Jan. 1967). N.F. Mott, A.M. Stoneham, “The lifetime of electrons, holes and excitons before self-trapping”, J. Pys. Chem. Solid State Physics, 10, 3391 (1977). B.J. Mrstik, R.W. Rendell, “Si-SiO2 interface-state generation during x-ray irradiation and during post-irradiation exposure to a hydrogen ambient”, IEEE Trans. Nucl. Sci., NS-38, 6, 1101 (1990). T.H. Ning, “High-field capture of electrons by coulomb-attraction centers in silicon dioxide”, J. Appl. Phys, 45, 7, 1976. R.N. Nowlin, E.W. Enlow, R.D. Schrimpf, and W.E. Combs, “Trends in the total-dose response of modern bipolar transistors”, IEEE Trans.Nucl.Sci., NS-39, 6, 2026 (1992). R.N. Nowlin, D.M. Fleetwood, R.D. Schrimpf, R.L. Pease, and W.E. Combs, “Hardness-assurance and testing issues for bipolar/BICMOS devices”, IEEE Trans.Nucl.Sci., NS-40, 6, 1686 (1993). T.R. Oldham and J. M. McGarrity, “Ionization of SiO2 by heavy charged particles”, IEEE Trans. Nucl. Sci. NS-28, 6, 3975 (1981). T.R. Oldham, “Charge generation and recombination in silicon dioxide”, Harry Diamond Laboratories report n° 1985, Adelphi, MD, U.S.A., (1986). T.R. Oldham and J.M. McGarrity, “Comparison of 60Co response and 10 keV x-ray response in
[Mott-77] [Mrst-91] [Ning-76] [Nowl-92] [Nowl-93] [Oldh-81] [Oldh-82] [Oldh-83] [Oldh-84] [Oldh-86] [Oldh-87] [Othm-80] [Pail-93]
[Pail-94] [Pail-95a]
[Pail-95b]
[Pail-95c]
[Pail-98]
MOS capacitors”, IEEE Trans. Nucl. Sci., NS-30, 6, 4377 (1983). T.R. Oldham, “Analysis of damage in MOS for several radiation environments”, IEEE Trans. Nucl. Sci,, NS-31, 6, 1236 (1984). T. Oldham, A. Lelis and F.B. McLean, “Spatial dependence of trapped holes determined from tunneling analysis”, IEEE Trans. Nucl. Sci., NS-33, 6, 1203 (1986). T.R. Oldham, A.J. Lelis, H.E. Boesch, J.M. Benedetto, F.B. McLean and J.M. McGarrity, “Postirradiation effects in field-oxide isolation structures”, IEEE Trans. Nucl. Sci., NS-34, 6, 1184 (1987). S. Othmer and J.R. Srour, “Electron transport in SiO2 at low temperatures”, in “The Physics of MOS Insulators”, G. Lukovsky, S.T. Pantelides Editors, Pergamon (1980), p. 49. Ph. Paillet, D. Hervé, J.L. Leray and R. Devine, “Effect of high temperature processing of Si/SiO2/Si structures on their response to X-ray irradiation”, Appl. Phys. Lett., 63 (15), 2088-2090, 11 October 1993. P. Paillet, D. Hervé, J.L. Leray and R.A.B. Devine, “Evidence of Negative Charge Trapping in High Temperature Annealed Thermal Oxide”, IEEE Trans. Nucl. Sci., NS-41, p. 473 (1994). Ph. Paillet,P. Gonon, C. Schwebel and J.L. Leray, “Comparison of X-Ray-Induced Electron and Hole Trapping in Various Materials (YSZ, SIMOX, Thermal SiO2)”, Journal of Non-Crystalline Solids, 187 (1995) and E-MRS Spring Meeting, May 24-27 1994, Strasbourg (France). P. Paillet, J. L. Autran, J. L. Leray, B. Aspar and A. J. Auberton-Hervé, “Trapping-detrapping properties of irradiated ultra-thin SIMOX buried oxides”, IEEE Trans. Nucl. Sci., NS-42, 6, 2108 (1995). P. Paillet, “Effet du procédé de fabrication des isolants sur la charge d’espace créée par rayonnement X: application aux technologies silicium sur isolant”, Thèse de doctorat en Sciences Physiques, n°95MON2042, Université de Montpellier II, France, March 30th, 1995 (available in microfiche at ANRT/University of Grenoble-II). P. Paillet, J.L. Touron, J.L. Leray, C. Cirba, A. Michez, “Simulation of Multi-Level radiationAll references are unclassified
III-111
[Pail-99]
[Peas-74] [Peas-85] [Peas-90] [Penn-90] [Penn-92] [Pers-92]
[Pers-97]
[Peyr-91]
[Plat-88] [Raym-92]
[Reve-77] [Saig-97]
[Saks-89] [Saks-90] [Saks-91] [Schm-95]
[Schm-96]
[Scho-39] [Schr-96]
[Schw-88]
[Schw-96]
Induced Charge Trapping and Thermally Activated Phenomena in SiO2”, RADECS-97 Conference and IEEE Trans. Nucl. Sci., NS-45, 3, 1379 (June 1998). P. Paillet and J.L. Leray, “Defects and radiation-induced charge trapping phenomena in silica”, in , Vol. 3, pp. 723-780, G. Barbottin and A. Vapaille, Editors (NorthHolland ISBN 0 444 81801 4). R.L. Pease, F.N. Coppage and E.D. Graham, “Dependence of ionizing radiation-induced hFE degradation on emitter periphery”, IEEE Trans. Nucl. Sci., NS-21, 2, 1216 (Apr. 1974). R. Pease, D. Emilly, H.E. Boesch, “Total dose induced hole trapping and interface states generation in bipolar recessed field oxides”, IEEE Trans Nucl. Sci., NS-32, 6, 3946 (1985). R. Pease, “Radiation Testing of Semiconductors for Space Electronics”, 1990 IEEE NSREC Short Course. C.A. Pennise and H.E. Boesch, “Determination of the charge-trapping characteristics of buried oxides using a 10-keV X-ray source”, IEEE Trans. Nucl. Sci., NS-37, 6, 1990 (1990). C.A. Pennise and H.E. Boesch, “Photoconduction measurements of the charge trapping and transport characteristics of bond-and-etch-back buried oxides”, IEEE Trans. Nucl. Sci., NS-39, 6, 2139 (1992). V.S. Pershenkov and V.V. Chuikin, “The Effect of junction fringing field on radiation-induced leakage current in oxide isolation structures and nonuniform damage near the channel edges in MOSFETs”, IEEE Trans Nucl. Sci., NS-39, 6, 2044 (1992). V.S. Pershenkov, V.B. Maslov, S.V. Cherepko, I.N. Shvetzov-Shilovsky, V.V. Belyakov, A.V. Sogoyan, V.I. Rusanovsky, V.N. Ulimov, V.V. Emelianov and V.S. Nasibullin, “The effect of emitter junction bias on the low dose-rate radiation response of bipolar devices”, IEEE Trans Nucl. Sci., NS44, 6, 1840 (1997). D. Peyre, “Contribution à l’étude des effets semi-permanents induits par les rayonnements ionisants dans les transistors MOS”, Thèse de doctorat en Sciences Physiques France, 1991 (available in microfiche at ANRT/University of Grenoble-II). D.G. Platteter and T.F. Cheek Jr., “The use of multiple oxygen implants for fabrication of bipolar silicon-on-insulator integrated circuits”, IEEE Trans. Nucl. Sci., NS-35, 6, 1350 (1988). J.P. Raymond, R.A. Gardner and G.E. LaMar, “Characterization of radiation effects on trenchisolated bipolar analog microcircuits technology”, IEEE Trans Nucl.Sci., NS-39, 3, 405 (June 1992) and Proc. RADECS-91. A.G. Revesz, “Chemical and structural aspects of the irradiation behavior of SiO2 films on silicon”, IEEE Trans. Nucl. Sci., NS-24, 6, 2102 (1977). F. Saigné, L. Dusseau, J. Fesquet, J. Gasiot, R. Ecoffet, J.P. David, R.D. Schrimpf and K.F. Galloway, “Experimental validation of an accelerated method of oxide-trap-level characterization for predicting long term thermal effects in metal-oxide semiconductor devices”, IEEE Trans. Nucl. Sci, NS-44, 6, 2001 (1997). N.S. Saks and D.B. Brown, “Interface-states formation via the two-stage process”, IEEE Trans. Nucl. Sci., NS-36, 6, 1848 (1989). N.S. Saks and D.B. Brown, “Observation of the H+ motion during the two-stage process”, IEEE Trans. Nucl. Sci., NS-37, 6, 1624 (1990). N.S. Saks and D.B. Brown, “Effects of switched gate bias on radiation-induced interface trap formation”, IEEE Trans. Nucl. Sci., NS-38, 6, 1130 (1990). D.M. Schmidt, D.M. Fleetwood, R.D. Schrimpf, R.L. Pease, R.J. Graves, G.H. Johnson, K.F. Galloway and W.E. Combs, “Comparison of ionizing radiation-induced gain degradation in lateral, substrate and vertical PNP BJTs”, Trans. Nucl. Sci., NS-42, 6, 1541 (1995). D.M. Schmidt, A. Wu, R.D. Schrimpf, D.M. Fleetwood and R.L. Pease, “Modeling ionizing radiation induced gain degradation of the lateral PNP bipolar junction transistor”, Trans. Nucl. Sci., NS-43, 6, 3032 (1996). W. Schockley, Phys. Rev., 56, 10, 317 (1939). R.D. Schrimpf, “Recent advances in understanding total-dose effects in bipolar transistors”, Proc. RADECS-95 and IEEE Trans. Nucl. Sci., NS-43, 3, 787 (June 1996). J.R. Schwank and D.M. Fleetwood, “Effect of post-oxidation anneal temperature on radiationinduced charge trapping in metal-oxide-semiconductor devices”, App. Phys. Lett. 53, 9, 770 (Aug. 1988). James R. Schwank, “Space and Military Radiation Effects in Silicon-On-Insulator Devices”, 1996 All references are unclassified
III-112
[SEI] [Sext-85] [Shan-83] [Shan-90]
[Shan-91] [Shan-98] [Shar-94] [Shaw-95]
[SIA-98] [Srou-76] [Srou-77a] [Srou-77b]
[Stah-93]
[Sun-80] [Sven-78] [Tamm-32] [Tayl-82] [Terr-89]
[VanL-75] [VanL-80] [VanV-95] [Vasu-94] [Wang-75] [Warr-92] [Warr-94]
[Wino-84]
[Wino-85]
IEEE International SOI Conference Short Course. Space Electronic Inc., San Diego, USA, www.spaceelectronics.com and TID Calculator (1998). F.W. Sexton and J.R. Schwank, “Correlation of radiation effects in transistors and integrated circuits”, IEEE Trans. Nucl. Sci., NS-32, 6, 3975 (1985). Z. Shanfield, “Thermally stimulated current measurements on irradiated MOS capacitors”, IEEE Trans. Nucl. Sci., NS-30, 6, 4064 (1983). M.R. Shaneyfelt, J.R. Schwank, D.M. Fleetwood, P.S. Winokur, K.L. Hughes and F.W. Sexton, “Field dependence of interface-trap buildup in polysilicon and metal gate MOS devices”, IEEE Trans. Nucl. Sci., NS-37, 6, 1632 (1990). M.R. Shaneyfelt, D.M. Fleetwood, J.R. Schwank and K.L. Hughes, “Charge yield for Cobalt-60 and 10-keV x-ray irradiations of MOS devices”, IEEE Trans. Nucl. Sci., NS-38, 6, 1187-1194, (1991). M.R. Shaneyfelt, P.E. Dodd, B.L. Draper and R.S. Flores, “Challenges in hardening technologies using shallow-trench isolation”, IEEE Trans. Nucl. Sci., NS-45, 6, 2584 (1998). R.E.Sharp et D.R.Garlick, “Radiation Effects on Electronic Equipement, a Designers/Users’ Guide for the Nuclear Power Industry”, Radiation Testing Service of AEA Technology (January 1994). D.C. Shaw, G.M. Swift and A.H. Johnston, “Radiation evaluation of an advanced 64 Mb 3.3V DRAM and insight into the effects of scaling on radiation hardness”, IEEE Trans. Nucl. Sci., NS-42, 6, 1674 (1995). “The National Technology Roadmap for Semiconductors. Technology Needs”. SIA Semiconductor Association, San Jose, CA, (1997-1998) and www.semichips.org.. J.R. Srour, S. Othmer, O.L. Curtis and K.Y. Chiu, “Radiation-induced charge transport and charge build-up at low temperature”, IEEE Trans. Nucl. Sci., NS-23, 6, 1513 (1977). J.R. Srour and K.Y. Chiu, “MOS hardening for low-temperature applications”, IEEE Trans. Nucl. Sci., NS-24, 6, 2140 (1977). J.R. Srour, S. Othmer, O.L. Curtis and K.Y. Chiu, “Ionizing radiation effects on silicon-on-sapphire devices and silicon dioxide films”, Harry Diamond Laboratories report n° HDL-CR-77-090-1, Adelphi, MD, U.S.A. (1977). R.E. Stahlbush, A.H. Edwards, D.L. Griscom, B.J. Mrstik, “Post-irradiation cracking of H2 and formation of interface states in irradiated Metal-Oxide-Semiconductor field effect transistors”, J. Appl. Phys., 73, 658 (1993). S.C. Sun and J.D. Plummer, “Electron mobility in inversion and accumulation layers on thermally oxidized silicon surfaces”, IEEE Trans. Electron Dev., ED-27, 1497 (1980). C.M. Svensson, Proc. Intl. Conf. On Physics of SiO2 and its interfaces (Pergamon, NY 1978), p. 328. I. Tamm, Z. Phys., 76, 849, (1932) and 3, 34 (1933) cit. Th. Wolkenstein, (Mir Edit., Moskow, 1973). D.M. and T.P.T. Williams, “The dynamics of space-charge accumulation in irradiated MOS capacitors”, Journal of Physics D: Appl. Phys. 15 , 2483-2493 (1982) (printed in UK). J.M. Terrell, T.R. Oldham, A.J. Lelis, H.E. Boesch and J.M. Benedetto, “Time-dependent annealing of radiation-induced leakage currents in MOS devices”, IEEE Trans. Nucl. Sci., NS-36, 6, 2205 (1989). V.A.J. Van Lint, G. Gigas and J. Barengolt, “Correlation of displacement effects produced by electrons, protons and neutrons”, IEEE Trans. Nucl. Sci., NS-22, 6, 2663 (1975). V.A.J. Van Lint, T.M. Flanagan, R.E. Leadon, J.A. Naber and V.C. Rodgers, “Mechanisms of radiation effects in electronic materials”, Vol 1, John Wiley, NewYork (1980). N. van Vonno, “Advanced Test Methodology”, 1995 IEEE NSREC Short Course. V. Vasudevan and J. Vasi, “A two-dimensional numerical simulation of oxide charge build-up in MOS transistors due to radiation”, IEEE Trans. Electron Devices, ED-41, 3, 383 (1994). C.T Wang, B.S.H. Royce and T.J. Russel, “The effect of ion implantation on charge storage in MOS NS-22, 6, 2168 (1975). W.L. Warren et al., J.ECS, 199, 872 (March 1992). W.L.Warren, D.M.Fleetwood, M.R.Shaneyfelt, J.R.Schwank and P.S.Winokur, “Links between oxide, interface, and border traps in high-temperature annealed Si/SiO2 systems”, Appl. Phys. Lett. 64 (25), (1994). P.S. Winokur, J.R. Schwank, P.J. McWhorter, P.V. Dressendorfer and D.C. Turpin, “Correlating the radiation response of MOS capacitors and transistors”, IEEE Trans. Nucl. Sci., NS-31, 6, 1453 (1984). P.S. Winokur, E.B. Erret, D.M. Fleetwood, P.V. Dressendorfer, and D.C. Turpin, “Optimizing and All references are unclassified
III-113
[Wino-92] [Wino-94]
[Witz-96]
[Witz-98]
[Zupa-93]
controlling the radiation hardness of Si-gate CMOS process”, IEEE Trans. Nucl. Sci. NS-32, 6, 3954 (1985). P.S. Winokur, “Total-Dose Radiation Effects”, 1992 IEEE NSREC Short Course. P.S. Winokur, M.R. Shaneyfelt, T.L. Mesenheimer, D.M. Fleetwood, “Advanced qualification techniques” , RADECS-93 Conference, IEEE catalog number 937H0616-3 p. 289, and IEEE Trans. Nucl. Sci., NS-41, 4, 538 (June 1994) (in English), “Techniques avancées de qualification , L'Onde Electrique, Vol. 75 n°3, 20 (mai 1995) (in French). S. Witzack, R.D. Schrimpf, K.F. Galloway, D.M. Fleetwood, R.L. Pease, J.M. Puhl, D.M. Schmidt, W.E. Combs and J.S. Suehle, “Accelerated tests for simulating low dose rate gain degradation of lateral and substrate PNP bipolar junction transistors”, IEEE Trans. Nucl. Sci. , NS-43, 6, 3151 (1996). S. Witzack, R.C Lacoe, D.C. Mayer, D.M. Fleetwood, R.D. Schrimpf and K.F. Galloway, “Space charge limited degradation of bipolar oxides at low electric fields”, IEEE Trans. Nucl. Sci., NS-45, 6, 2339 (1998). D. Zupac, K.F. Galloway, P. Khosropour, S.R. Anderson, R.D. Schrimpf and P. Calvel, “Separation of effects of oxide-trapped charge, interface trapped charge on mobility in irradiated power MOSFETs”, IEEE Trans. Nucl. Sci. , NS-40, 6, 1307 (1993).
All references are unclassified
III-114
1999 NSREC SHORT COURSE
SECTION IV
PROTON EFFECTS AND TEST ISSUES FOR SATELLITE DESIGNERS
Paul W. Marshall Consultant Cheryl J. Marshall NASA/Goddard Space Flight Center
IV. Proton Effects and Test Issues for Satellite Designers General Introduction This portion of the Short Course is divided into two segments to separately address the two major proton-related effects confronting satellite designers: ionization effects and displacement damage effects. While both of these topics are deeply rooted in “traditional” descriptions of space radiation effects, there are several factors at play to cause renewed concern for satellite systems being designed today. For example, emphasis on Commercial Off-The-Shelf (COTS) technologies in both commercial and government systems increases both Total Ionizing Dose (TID) and Single Event Effect (SEE) concerns. Scaling trends exacerbate the problems, especially with regard to SEEs where protons can dominate soft error rates and even cause destructive failure. In addition, proton-induced displacement damage at fluences encountered in natural space environments can cause degradation in modern bipolar circuitry as well as in many emerging electronic and opto-electronic technologies. A crude, but nevertheless telling, indication of the level of concern for proton effects follows from surveying the themes treated in papers presented at this conference. The table lists themes found in the IEEE Transaction on Nuclear Science (TNS) December issue from the past year and compares them with the December issue’s content a decade earlier. Ten years ago there were nine papers, or about 10% of the total, dealing with the four indicated topics. At that time, single event effects from protons were the primary concern, and these were thought to be possible only when a nuclear reaction initiated energetic recoil atoms. This is shown in the table as the ‘traditional” SEE subject. A decade later, submissions addressing this topic had doubled, while papers devoted to displacement damage studies had increased from one to nine! More importantly, displacement damage effects in the natural space environments have become a concern for degradation in modern devices (other than solar cells), and this was not so ten years earlier. Table: Growth of concern for proton effects over the past decade Topic IEEE TNS, Vol. 35, No. 6, 1988 IEEE TNS, Vol. 45, No. 6, 1998 Environments 1 5 Dosimetry 4 2 Displacement Damage 1 9 “Traditional” SEE 3 6 New Effects 4 Total 9 (~10% of total) 26 (~30% of total) In the recent Transactions, four papers were devoted to effects that were either unknown or considered unimportant a decade earlier. These include soft errors from direct ionization by protons [Mars-98, Reed-98] and from nuclear elastic scattering events [Ingu-97, Savage-98, Johnston-98], along with hard failure SEE mechanisms such as latch-up [Norm98] and dielectric breakdown in power MOS devices [Titu-98]. The aggregate level of
IV-1
concern is obvious with a total of 26 papers, or about one third of the articles, in the 1998 December TNS dealing substantially with proton-related issues. In this Short Course segment we will attempt to survey the important developments that have taken place in the past few years. The material we cover emphasizes the developments affecting design tradeoffs for current satellite systems with the recognition that any given component can potentially be used so long as the risks are identified adequately and mitigated appropriately. We approach this task by citing the studies that identify the various effects that protons can have, and then by indicating and demonstrating the tools available to radiation effects experts and knowledgeable designers to quantify the associated risks. To place the material in the context of the needs of the satellite design engineer, we offer the following list of reasons that might motivate the need for proton testing of a given device or circuit. Reasons to test with protons: 1. Expect proton SEEs and have no satisfactory means of predicting response without proton test data. • Have no SEE data on part type and need to characterize for a proton rich environment • Have heavy ion SEE data and correlation approaches indicate “marginal” performance • Need to gain general idea of heavy ion sensitivity and have package penetration test issues • Suspect a sensitivity to direct ionization induced SEEs from protons (e.g., optocouplers) • Need to assess sensitivity to destructive failure (e.g., latch-up) from protons 2. Expect proton displacement damage and have no satisfactory means of predicting response without test data. • Have no displacement damage data on part type and need to characterize for a proton rich environment • Need to verify response on flight-lot devices where lot-to-lot variations can be large (e.g., COTS) • Have neutron data and correlation approaches indicate “marginal” performance • Need to measure a “damage function” energy dependence to reduce uncertainty associated with predictive tools • Mixed damage and TID from protons in application indicates need for proton tests 3. Need to assess TID response to proton-induced dose with high precision (e.g., calibration of p-MOS dosimeters)
Following the major division indicated in the above list, the subject material for this section of the course is divided into two segments with Part A devoted to ionization effects and Part B to displacement damage effects. The section on ionization effects addresses the environment and satellite configuration considerations to identify scenarios where proton dose can play an important role in TID effects. Proton-induced single event effects occurring from IV-2
either direct ionization or generation of reaction recoils will be included in this section, but the emphasis will be on more recent studies describing new sensitivities to proton-induced ionization events. The reader will be referred to previous Short Course notes and related literature for discussions of mechanisms and rate calculations for “traditional” proton SEE and TID concerns, with the exception of two case studies. These two studies examine special concerns for modern communications satellite constellations that route high-speed signals. The section on displacement damage considers the numerous device types exhibiting sensitivity to displacement effects. The primary tools, like the concept of Non-Ionizing Energy Loss (NIEL), now used to treat proton-induced displacement effects have been mostly developed within the past decade. This section addresses the justifications, methodology, and associated uncertainties in applying these tools to various classes of Si-based devices as well as emerging III-V technologies for electronic and opto-electronic applications. Emphasis will be given to satellite environments, including shielding efficacy and tradeoffs. Our goal is to capture the current understanding of the many proton-related concerns important to the satellite subsystem engineer. At the beginning of the first section on ionizing effects, a top-level treatment of the environments internal to satellites will be provided, along with references to previous Short Course discussions for additional detail. The emphasis of the remaining material throughout both sections is on the effects, tools, and associated information to enable informed tradeoffs of design options.
IV-3
Section IV. Proton Effects and Test Issues for Satellite Designers Part A: Ionization Effects Paul W. Marshall 7655 Hat Creek Road Brookneal, VA 24528
Cheryl J. Marshall NASA Goddard Space Flight Center Code 562 Greenbelt, MD 20771
1.0 Introduction………………………………………………………………………………6 2.0 Proton Environments for Satellites………………………………………..……….…... 6 2.1 Environment Description and Issues……………………………….….………. 6 2.2 Example Proton Environment Description……………………………..……… 9 2.3 Requirements: Proton Specific Issues…………………………….…..………...13 2.3.1 Total Ionizing Dose…………………………………….…………...13 2.3.2
Destructive SEE……………………………………….…………… 13
2.3.3
Nondestructive SEE………………………………………..………. 13
2.3.4
Margin………………………………………………………………14
2.3.5
Nonstandard Parts and Waivers……………………………………14
2.4 Recent Updates to the Proton Environment Models……………………………15 3.0 Total Ionizing Dose and Protons………………………………………………………..16 3.1 Proton-Induced Total Ionizing Dose: Mechanisms and Issues…………………16 3.2 Is a rad always a rad? ………………………………………………………….. 18 3.2.1
Lateral Nonuniformities (LNUs) ………………………………….. 19
3.2.2
Electron-Hole Recombination…………………………………...… 20
IV-4
4.0 Proton-Induced Single Event Effects…………………………………………………...22 4.1 Test Issues and Special Cases………………………………………………….. 23 4.1.1 High Speed Technologies…………………………………………...24 4.1.2
Small Probability Events……………………………………………26
4.1.3
Single Event Transients in Linear Devices………………………… 27
4.1.4
Correlation Between Proton and Heavy Ion SEE Sensitivities……. 28
4.2 Proton Direct Ionization and SEEs…………………………………………….. 29 4.2.1
CCDs ……………………………………………………………….29
4.2.2
Optical Link Photodetectors……………………………………….. 31
4.2.3
Optocouplers and MSM Photodiodes……………………………… 36
4.3 Destructive Failures from Single Proton Interactions…………………………. 39 4.3.1
Latch-Up (and COTS) …………………………………………….. 39
4.3.2
Proton-Induced Single Event Burnout….………………………….. 40
4.3.3
Stuck Bits…………………………………………………………... 41
5.0 Summary……………………………………………………………………………..… 43 6.0 Acknowledgments………………………………………………………………………43 7.0 References for General Introduction and Section IVA…………………………………44
IV-5
SECTION IVA: IONIZATION EFFECTS Paul W. Marshall and Cheryl J. Marshall 1.0
INTRODUCTION
This first segment covers various ways in which proton-induced ionization can affect circuit operation. For our purpose, this includes TID effects as well as both direct and indirect single event phenomena. Before addressing the device and circuit effects, we offer a brief overview of the near-Earth proton environments and issues impacting the environment description internal to the satellite. This section also provides examples and discusses issues concerning the generation of design requirements based on the expected environment. Following the environment section, we offer a brief discussion of proton-specific concerns for TID effects. Next, the fourth section examines soft errors due to protons, again with emphasis on recent developments, and the fifth section looks at hard errors.
2.0
PROTON ENVIRONMENTS FOR SATELLITES
Protons occur in every imaginable orbit with variations in spectral energy composition, arrival rates, and sometimes arrival trajectories. The three sources are trapped protons in the inner Van Allen radiation belt, the proton component of solar particle events, and hydrogen nuclei from intergalactic cosmic rays. Careful discussions of the near-Earth, interplanetary, and other planet proton environment models are available in the Conference Short Course notes from 1997 [Bart-97]. The interested reader should refer to those notes and the cited literature to gain an understanding of the characteristics and shortcomings of the widely used NASA AP-8 [Sawy-76] model for trapped protons, the CREME-96 [Tylk-96] cosmic ray model, and various descriptions of solar proton probability models. These models should be viewed only as working approximations aimed at describing the major features of the external environment with the recognition that both the subtleties and major features of the environments are the concern for numerous space-borne experiments and ongoing modeling efforts.
2.1 Environment Description and Issues With regard to the range of proton environments and the factors affecting them, the basic models cited above are the predictive tools of the environment specialists. The 1997 Short Course by Janet Barth [Bart-97] offers an excellent discussion of these and several other models, their applicability, issues affecting their accuracy, and the proton environments as they change with orbital position and solar cycle period. Detailed treatment is outside the scope of this material, but in addition to those notes, interested readers may wish to locate the 1988 review article “The Natural Radiation Environment Inside a Spacecraft,” [Stass-88] or this conference’s Short Course notes on “Radiation Environments in Space” [Stass-90]. For quick reference, the general character of the trapped proton belts external to the spacecraft is provided here as figure 1.
IV-6
Figure 1. The AP-8 model for solar minimum conditions at 0 degrees inclination indicates the higher energy protons at lower altitudes [after Stass-88]. The orbit altitude in km is related to dipole shell as [(L x 6370 km)-6370 km] where 6370 km is the Earth’s radius.
There is a document in development to supplement the various radiation models with practical considerations for satellite applications and make general trends more accessible to design engineers. The IEEE (Draft) Standard 1156.4 [IEEE-1156.4] document is aimed at establishing generic descriptions of four orbit categories: Low Earth Orbit (LEO) below about 10,000 km, Medium Earth Orbit (MEO) from 10,000 to 20,000 km, Geostationary Orbit (GEO) at 36,000 km, and transfer or Highly Elliptical Orbit (HEO). This document identifies example orbits in each of these categories and illustrates proton and other radiation characteristics of those orbits for the purpose of ionizing (but not displacement) effects in space-borne computers. Be warned though, that these are only examples and there is no justification to generalize from those orbits even to other orbits within the same category. For the designer, detailed understandings of the environment models are fortunately not usually necessary. Instead, the proton and other radiation related requirements are either supplied by the procuring organization or generated “in house” by resident radiation environment experts. Several years ago, it was not uncommon to see radiation design specifications expressed in terms of total ionizing dose (or depth-dose curves) supplemented with either Linear Energy Transfer (LET) spectra or guidance with respect to LET threshold for single event induced hard errors and upset rates from cosmic rays. Proton contributions to IV-7
the depth-dose may have been identified, but often there was no breakout of the expected proton energy spectra and fluxes. In many cases, it was assumed that components that could be upset by protons would be screened out by the requirement for a high threshold LET for cosmic ray effects. With today’s emphasis on high performance systems and the component selections now available, all that has changed. Now a more reasonable assumption would be that proton effects are expected, and part of the designer’s task is to manage the associated risk. Increasingly, it is the responsibility of the design team to assess radiation-related risk, and proton effects are often an important part of this equation. There is still quite a variation in the level of detail called out in the proton environment description provided to the design effort. It usually contains some, but rarely all, of the following elements: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.
trapped proton total ionizing dose contribution to a depth-dose relation trapped proton energy spectra behind typical shield thickness peak trapped proton flux with energy composition (in orbit “hot spot”) average daily trapped proton flux mission fluence for protons > 25 MeV (or some other cutoff) solar particle event (SPE) proton energy spectrum with fluence per mission peak (worst case) proton flux from SPE with energy composition assumed frequency of occurrence of design case SPE cosmic ray LET spectrum including proton contributions variations on each of the preceding to reflect solar cycle related changes variations on each of the preceding to reflect uncertainties and design margins
There are no standardized formats for identifying either the proton environments or the radiation-related requirements for proton effects. The preceding list offers several environment descriptions that are commonly seen in various combinations. Often, the intention is to identify TID levels, a mission-duration proton spectrum for SEE and possibly displacement damage concerns, and a peak flux (or fluxes) for use in assessing peak SEE rates. Item ten in the list touches on the often seen practice of inflating the expected environment to add margin for various reasons, ranging from uncertainties in environment models to part-to-part response non-uniformity and response uncertainty. Margin will be discussed in a later section. In practice, the environments identified in many requirements documents do not always adequately specify the details needed to properly assess proton-related effects. This situation follows, in some part, from a natural lag between the identification of a given important effect (e.g., displacement damage in optocouplers, or proton upset response best described by a two parameter Bendel formalism as discussed later) and the recognition of the need to include detailed proton spectral information rather than just proton-induced rad(Si) and > 25 MeV fluences. Also, the timeline associated with procuring flight hardware may result in periods of years between the definition of the orbit environment description to the detailed design, and new effects, which may emphasize previously unimportant aspects of the environment, are continually identified. From the radiation effects perspective, it’s hard to overdo the level of detail in the environment description called out in a requirement. IV-8
2.2 Example Proton Environment Description As an example of the proton environment description in a program currently under way, the interested reader may examine the document SSP 30512 Rev. C entitled “Space Station Ionizing Radiation Design Environment” [SSP-30512]. This document, released in June of 1994, describes the ionizing radiation environment as calculated for the International Space Station Alpha (ISSA) at an altitude of 500 kilometers and inclination of 51.6 degrees. Five years later, it remains the reference environment description for hardware currently being designed for ISSA, and it has general applicability to the multi-national and multi-agency effort. We include this example because this program is of general interest and also because the proton environment description is unusually thorough. The descriptions of various proton environments for ISSA are contained in table 1. The first item accounts for most of the protons encountered in the low-Earth orbit. Though not indicated explicitly in any of the ten items, most of these protons will be encountered during passes through the South Atlantic Anomaly (SAA). The average daily proton flux predicted in this table is calculated by using the AP-8 model for solar maximum conditions and shown below in integral form as figure 2. In table 1 the second and third listings address the depth-dose relation and are included here since protons account for a portion of the total ionizing dose. Figure 3 shows the relative annual dose contributions for electrons and protons at the center of a solid aluminum sphere. The chart indicates how protons dominate the TID for shield thicknesses greater than about 200 mils Al, and this result is typical of most orbits that encounter the trapped proton belts. Careful inspection of the proton curve illustrates how ineffective shielding is for stopping protons. Note the steep falloff in the electron dose with increasing depth and the relatively flat character of the proton curve. Increasing the shield thickness form 100 to 1000 mils only reduces the resulting dose by about 50%. For this reason, and others discussed later, shielding is often not the best technique for minimizing proton effects. Table 1: List of tables describing proton-related environments for ISSA [SSP-30512] Item Table Description 1 3.1.1-2 AP8MAX differential and integral flux energy spectra for trapped protons 2 3.1.2-1 One year dose at the center of a solid aluminum sphere (rads(Si)) 3 3.1.2-2 One year dose in semi-infinite aluminum medium (rads(Si)) 4 3.2.1.1-1 Daily average internal proton integral flux spectrum 5 3.2.1.1-2 Daily average internal proton differential flux spectrum 6 3.2.1.2-1 SAA pass internal peak proton integral flux spectrum 7 3.2.1.2-2 SAA pass internal peak proton differential flux spectrum 8 3.2.1.4-1 Combined integral flux LET spectra (WI1=4) no solar flare flux 9 3.2.2-1 Maximum solar flare peak proton integral flux spectrum 10 3.2.2-2 Maximum solar flare peak proton differential flux spectrum 1 WI = weather index
IV-9
AP8-MAX Integral Flux for ISSA 1.0E+08
Integral Flux (p+/cm2/day)
1.0E+07 1.0E+06 1.0E+05 1.0E+04 1.0E+03 1.0E+02 1.E-02
1.E-01
1.E+00
1.E+01
1.E+02
1.E+03
Proton Energy (MeV) Figure 2. Integral daily proton flux for the example International Space Station Alpha orbit of 500 km x 51.6 degrees shows the large numbers of low energy protons that are easily shielded. Fluxes in the tens to hundreds of MeV can penetrate to sensitive components resulting in single event effects and both ionizing dose and non-ionizing dose.
Dose (rads (Si/year)
1.E+05 1.E+04
Proton Dose (rad(Si)) Electron Dose (rad(Si))
1.E+03
Total Dose (rad(Si))
1.E+02 1.E+01 1.E+00 1.E-01 10
100
1000
10000
Shield Thickness (mils Al) Figure 3. For the 500 km x 51.6 degree orbit for ISSA, the dose from trapped protons dominates for Al shield thicknesses above 200 mils. With increasing shield thickness, the average proton energy increases. IV-10
The fourth and fifth items in table 1 provide detailed information on spectral energy composition for aluminum shield thicknesses of 0, 50, 500, 1000, 2000, 4000, and 7000 mils. The inclusion of predictions for very thick shielding is appropriate for the manned bays in ISSA, but is atypical for unmanned vehicles. Items six and seven in table 1 address the peak proton flux expected from trapped protons during passages through the South Atlantic Anomaly. Note that the peak flux occurring due to trapped protons is almost always lower than the peak flux due to solar particle events, with the exception being low inclination LEO orbits. However the SAA is encountered on about 50% of ISSA orbits (several times a day) as opposed to the relatively rare solar events. In this case, the peak from SAA trapped protons, integrated over energies > 10 MeV, is 2.04 x 103 p/cm2/s, which is 54 times the orbitaveraged rate. Table 1, items nine and ten describe the modeled peak rates corresponding to solar particle events, including effects of the Earth’s geomagnetic shielding. Data listed in these tables indicate an expected peak (for design purposes) of 3.36 x 105 p/cm2/s. This is 165 times the peak rate during SAA passages. This event would be expected, as described in [SSP-30512], once per 11 year solar cycle, or approximately once per mission for ISSA. Peak proton arrival rates for a given mission cannot be predicted in a deterministic manner. The use of the October 1989 flare as a design environment, shown in figure 4, for ISSA is a somewhat arbitrary choice. Other options exist, including other peak solar particle models incorporated in CREME-96 [Tylk-96, Tylk-96a]. More recently, a probabilistic model for predicting peak flux based on solar cycles 20, 21, and 22 has been proposed [Xaps98], and this allows a more quantitative approach to assessing the risk of exceeding a given proton flux. Please see the discussion of peak SPE proton rates in the 1997 Short Course notes [Bart-97] for more information and discussion on this topic. The eighth table entry (combined integral flux LET spectra (weather index=4) no solar flare flux) describes the cosmic ray design environment for ISSA in terms of the LET spectrum from heavy ions. We acknowledge that for orbits encountering the trapped proton belts there are relatively few cosmic ray protons. However, for devices (e.g., detectors) sensitive to single events from proton ionization, cosmic ray protons may be a concern. Hydrogen ions (protons) are after all the most abundant constituent of the composite cosmic ray spectrum, and they account for 83% of all cosmic rays outside the effects of the Earth’s magnetic field. Figure 5 indicates the composite LET spectrum with two shield thicknesses for ISSA. In terms of LET, protons account for only about 1 in 105 incident particles at the lowest LET value shown in the figure (0.1 MeVcm2/mg). Though not shown, at LET values of 0.01 MeVcm2/mg and 0.001 MeVcm2/mg proton contributions increase to 0.5% and 88% respectively. If lower LET values are important, then the proton component of cosmic rays may dominate. Most cosmic ray protons are very energetic and shielding has little effect. For the ISSA environment definitions, proton spectra are provided with various shield thicknesses. Not all design efforts provide this level of detail, and it is often necessary to modify the provided spectra to evaluate the detailed effects of additional shielding. Several transport code based routines are available to provide spectra and sometimes dose behind additional shielding. The capability and complexity of these tools cover a broad range from transport through relatively simple spherical or slab single thickness to full-up ray tracing and IV-11
Maximum Solar Proton Flux for ISSA
0 MILS AL 50 MILS AL 500 MILS AL
1.E+10
2
Integral Flux (p /cm /day > E)
1.E+11
+
1.E+09
1.E+08
1.E+07
1.E+06 0
100
200
300
400
500
600
700
Proton Energy (MeV)
Figure 4. Peak solar proton flux from the October 1989 event, as propagated into the 500 km x 51.6 degree Space Station orbit, shows that less than a third of the protons reaching that orbit are stopped by 50 mils Al shielding. Also note that some protons above 400 MeV are present, in contrast to the trapped belt models.
Integral Flux (particles/cm2/day > LET)
ISSA Cosmic Ray Environment 1.0E+06 50 MILS AL 500 MILS AL
1.0E+03 1.0E+00 1.0E-03 1.0E-06 1.0E-09 1.E-01
1.E+00
1.E+01
1.E+02
LET (MeVcm2/mg)
Figure 5. Proton contributions to the cosmic ray environment at 500 km and 51.6 degrees are included in the heavy ion LET spectrum. They account for 87% of all particles at an LET of 0.001 MeVcm2/mg, but only .002% at the lowest LET shown in the chart. IV-12
high-density sector calculations. Additional discussion and a list of references are provided in the 1998 Short Course notes from this conference in section V part 3.1 [Kinn-98]. These tools play an important role in determining the dose or associated proton spectra at the box and component level. Usually, after surrounding boxes and satellite structural elements are considered, the resulting exposure at the subsystem and device level is less than that for a 100 mil Al spherical shield. However, the protection offered by structural shielding is of much less benefit than in the case of electron exposure because of the penetrating nature of protons.
2.3 Requirements: Proton Specific Issues Expressions of requirements vary greatly from program to program. Top level requirements usually address performance of a system in a specified environment in terms of system lifetime, availability to perform mission objectives, and accuracy in meeting those objectives. At this level proton effects, along with other radiation effects, are only part of the reliability picture. From this level, system engineers usually arrive at a set of derived requirements that are applied to individual subsystems. These will likely address destructive failures, TID failure levels, and possibly soft error rates. Most often proton effects are lumped together with associated effects from other radiation sources. More recently, with the increasingly important role of displacement effects, proton levels may be specified explicitly for component types known to be susceptible to proton-specific effects, such as optocouplers, Charge Coupled Devices (CCDs), and others. 2.3.1 Total Ionizing Dose: TID requirements for a given mission are usually based on the composite electron-proton dose based on the depth-dose relation for the given mission. TID evaluations are customarily made using Co-60 test facilities with the assumption that there is a linear superposition of dose from protons and electrons (this assumption will be examined in a section IVA.3). 2.3.2 Destructive SEE: Similarly, for the case of single events effects resulting in hard failure, requirements for proton-related failure modes are rarely specified separately from heavy ion induced failure, though they sometimes should be. In practice however, extreme measures are usually taken to avoid hard failure, including the requirement that hard failures (e.g., latch-up) may not occur below the iron cutoff (or some higher LET value) of the heavy ion LET spectrum. If such measures are taken, then proton-induced hard failure will also be avoided, since proton sensitivity to SEEs is not expected where such high LET threshold values apply. Unless such assurances are in place, we suggest that requirements should be written to address the possibility of hard failures due to protons. In LEO applications of commercial power MOSFETs, for example, flight data has demonstrated that burnout is much more probable from protons than from heavy ions, even though the orbit inclination is 700 [Bart-98]. Expression of a requirement in terms of only LET would therefore be inadequate. Associated issues will be addressed in section 4.3.2 on Single Event Burnout (SEB). 2.3.3 Nondestructive SEE: It is more common for requirements to address proton-induced soft errors, at least indirectly, through the expression of requirements for average and peak error rates. Historically, this issue arose due to soft errors in memories, but in recent years the IV-13
literature (and unpublished flight data) provides many cases of microprocessors, ASICs, linear circuits, ADCs, detectors, and other components which exhibit sensitivity to soft errors from protons. Typically, requirement documents call for a level of performance, and it is the job of the box manufacturer to allocate error rates due to various expected sources. For orbits encountering the inner belts, protons often dominate average soft error rates in technologies sensitive to their effects. Where error rates are required to stay below some allowed maximum, the peak trapped proton flux and peak solar particle event fluxes represent the greatest challenges. For orbits outside of the proton belts, such as geostationary at 36,000 km, average proton arrival rates may be extremely low, but proton upset sensitivity can be a design driver with the infrequent solar particle event in mind. 2.3.4 Margin: Design margins arise from a variety of concerns and are expressed in a variety of ways, with some of these concerns specific to protons. Where large uncertainties exist either in the ability to predict the environment with high confidence or in the ability to predict circuit response, margin is applied to mitigate risk. In fact, the degree of margin is often scaled according to the criticality of the function being performed with the idea that survival is most important and some mission objectives are more important than others [e.g., Gate-96]. One of the sources of uncertainty is in the environment models. As an example, for trapped protons, the 23 year old AP-8 model has served to establish average and peak proton fluxes as well as proton dose for the radiation belts. The uncertainty associated with that model is stated to be a factor of two for long term orbit averages and higher for short duration periods (e.g., less than 1 year) [Sawy-76]. Therefore, it is appropriate to either double the predicted trapped particle environment or impose margin on the ability of the system to perform in the predicted environment. System designers frequently employ either approach. New information impacting the uncertainty of the AP-8 model is discussed in section 2.4. Uncertain device response and the inability of imperfect predictive models to accurately describe performance also call for radiation design margin. These factors can be quite large, especially where the key variables governing device response are not well understood. In the portion of this Short Course segment concerning displacement damage we will illustrate this point for the cases of CCD damage and optocoupler degradation. Design margins of over two are warranted in each of these situations. After factoring in more customary arguments for margin (e.g., part-to-part and lot-tolot differences in response) the suggestion of additional factors of two or greater for each of these other proton specific sources usually comes as quite a shock to the design engineer. Resolution of these issues varies from program to program, usually with some element of compromise and the hope that either the components easily meet the requirements with margin or the design can be modified to accommodate the anticipated degradation. It is important that both the radiation effects specialist and the design engineer realize the needs for margin and not fall into the trap of assuming that the factor of 2 applied for environment uncertainty also accommodates uncertainties from other sources such as the response model. 2.3.5 Nonstandard Parts and Waivers: Every flight project design effort tracks parts that do not meet requirements with margin or that require “special” considerations. Often, IV-14
additional testing is required and special considerations are needed to evaluate risk. Nowadays, with the emergence of displacement damage concerns in Light Emitting Diodes (LEDs), optocouplers, and CCDs, many proton radiation effects are dealt with in this forum. As specifications of proton environments and requirements improve, along with predictive models for the environments and device responses, these issues will likely be dealt with on a more routine basis and factors indicating needs for large design margins can be minimized. However, for now, many proton effects are only beginning to work their way into the concerns of the typical design effort. Proton testing is still viewed as an expensive alternative to be used sparingly. In many cases the details of defining proton test approaches and deciding on acceptable margins falls within the scope of “nonstandard parts evaluation” efforts. Resolution of these issues can present a significant set of interesting challenges.
2.4 Recent Updates to the Proton Environment Models The preceding discussions have identified many of the proton environment models now being used, and indicated references for additional information. There are however a few key points to make regarding the dynamic status of the environment models. Janet Barth, in the 1997 Short Course notes [Bart-97], includes a section entitled “Problems with the AP-8 and AE-8 Models,” and she follows this with a section entitled “Dynamic Models, A Beginning.” These notes are well worth reading to see the path toward revisions to the existing NASA models. Last year, at this conference, Houston and Pfitzer presented a paper entitled “A New Model for the Low Altitude Trapped Proton Environment,” [Hous-98]. This “new model” is based on data acquired by instruments on the TIROS/NOAA spacecraft from 1978 through 1995. The key finding, from the satellite designer’s perspective, is that the predicted fluxes are about twice as high as those from the AP-8 model, as indicated in figure 6. This finding seems to hold for proton energies of interest to satellite designers (> 16 MeV). The data cover the altitude range from 250-850 km. In conjunction with the CRESSPRO model [Meff-94], this significantly improves the empirical basis for major revision to the AP-8 model. The probabilistic model for SPE peak fluxes presented at the 1998 NSREC has already been mentioned [Xaps-98]. Environment specialists also rely on a probabilistic modeling tool for predicting solar proton fluences during a mission [Feyn-96]. J. Feynman and co-workers at JPL have performed a statistical analysis including data on solar particle event proton fluences from the past three solar cycles. Their findings show that the largest events, such as the August-72 and October-89 events, belong in the same statistical distribution as other events. They provide a Monte Carlo based tool for assessing probability of exceeding a given fluence level during a specified mission duration. This model is especially important for geostationary and interplanetary missions, and it has rapidly gained acceptance. The key point is that our understanding of even the gross features of the space radiation environment is not precise. Revisions and enhancements to the existing models are ongoing. IV-15
Figure 6. Recent dosimetry data from TIROS/NOAA satellites provide the basis for a revised model for trapped protons. The above figure and discussion in [Hous-98], by S.L. Houston and K.A. Pfitzer, indicate about twice the flux predicted by AP-8 for the three energy bins listed in the legend.
3.0
TOTAL IONIZING DOSE AND PROTONS
In this section we discuss several issues specific to total ionizing dose deposited by protons. Topics will include typical situations where proton-induced TID may be important to satellite systems, a discussion of the equivalence between proton dose and ionizing dose from other sources in the natural space environment, and finally a discussion of microdosimetry issues specific to protons. The emphasis will be on ionizing dose with the recognition that protons also deposit non-ionizing dose, which causes displacement damage. That is treated in the segment IVB of these notes.
3.1 Proton-Induced Total Ionizing Dose: Mechanisms and Issues As protons traverse a solid, their positive charge presents an electrostatic force to the orbital electrons of the surrounding material. Excited electrons are freed from their bound state thereby creating electron-hole pairs. Some of these electrons (called delta rays) are liberated with sufficient energy to interact with other electrons at some distance from the incident proton’s trajectory, thereby leading to an ionization track with some structure. This coulombic scattering process liberates electron-hole pairs at a rate that depends on the proton energy and also on the material it traverses. In Si, for example, the electron-hole pair creation requires (on average) 3.6 eV in energy from the incident proton. This empirically determined value is referred to as the ionization potential, and it depends on a number of factors including the material band-gap for the case of semiconductors. In insulators, ionization potentials are significantly larger, e.g., ~ 17 eV in SiO2. IV-16
LET and Range for Protons in Si 1.E+03 LET (MeV*cm2/mg) Range (cm Si)
1.E+01 1.E-01 1.E-01 1.E-02 1.E-03
1.E-05
1. 00 E 2. -02 80 E 1. -02 00 E 2. -01 80 E 1. -01 00 E 2. +00 80 E 1. +00 00 E 2. +01 80 E 1. +01 00 E 2. +02 80 E 1. +0 00 2 E+ 03
1.E-03
Range (cm Si)
LET (MeVcm 2/mg)
1.E+00
Proton Energy (MeV) Figure 7. Lower energy protons with higher LETs are effectively stopped in satellite structural materials due to their short ranges. Higher energy protons are very penetrating, but fortunately transfer their energy at a lower rate. These relations do not address nuclear reactions occur for higher energy protons (>10 MeV) with a probability of about 10-5 for a pathlength of a few microns in Si.
The rate of energy loss for a proton (or any heavy ion) is termed the Linear Energy Transfer (LET) or stopping power, and the usual units are MeVcm2/mg though it can be converted to energy per unit pathlength by multiplying by the target material density. Again, this is an empirically determined relation, and its dependence on proton energy, known as the specific ionization curve, is shown in figure 7 along with the relation between proton range and energy. Note that the LET tends to decrease with increasing energy in the MeV regime encountered within satellites. Also, as the proton loses energy, the decrease in range is highly nonlinear. These energy loss kinematics are the basis for the situation in which the low energy protons are preferentially stopped by spacecraft materials. When the naturally occurring spectrum encounters satellite materials, protons at all energies lose energy, but the mean proton energy reaching the payload electronics actually increases with increasing shield thickness. Along the trajectory of an individual proton, the path of ionization produces more and more electron-hole pairs per unit length until the proton approaches the end of its range. The LET actually peaks at an energy of about 80 keV in Si. This maximum in energy loss is often (inappropriately) referred to as the Bragg peak. IV-17
The precise details of energy loss are beyond the scope of this discussion, but it should be noted that proton dose is deposited along ionization tracks. On a micro-dosimetry scale the presence of these tracks leads to dose deposition that is therefore highly nonuniform in nature. In the following sections, ionization effects to electronic materials will be discussed and the inherent non-uniformity of proton dose on the microdosimetric scale will be placed into perspective by making comparisons to electron (or Co-60) dose deposition, which is much more uniform in nature. Finally, it should be noted that the coulombic electronic scattering mechanism is by far the dominant mechanism for ionization purposes and for affecting proton energy loss, but other processes do occur. Nuclear elastic and inelastic processes lead to some ionizing dose deposition, but these are orders of magnitude down from electronic scattering. By far their most important role in electronic materials is in imparting non-ionizing energy leading to atomic displacements. These processes are discussed in detail in Part B of this Short Course section. In the previous section on environments, it was noted that protons are encountered in all orbits. However, protons are significant contributors of TID only in certain cases. As indicated in figure 3, from the ISSA example in low-Earth orbit, the relative contribution of proton to electron dose increases with increasing shielding. The exact thickness at which proton dose becomes important varies with orbit. Proton dose may be a concern for low-Earth orbits or highly elliptical orbits that encounter the inner Van Allen belts, for orbits encountering high fluences from Solar Energetic Particle (SEP) events, or for missions reaching proton belts surrounding other planets. For circular orbits between about 1,500 and 5,000 km (the heart of the belts) the multi-year mission doses from protons can easily exceed 100 krad(Si). Whether in an orbit dominated by protons or one with mixed electron/proton exposures, the components most sensitive to TID effects are often buried deep in the spacecraft or protected with spot shielding. Consequently they may receive a substantial fraction of their dose from protons.
3.2 Is a rad always a rad? The widely accepted unit for total ionizing dose from ionizing radiation is the rad (from Radiation Absorbed Dose). The rad is defined as 100 ergs per gram of energy absorbed in the exposed material, and 100 rads is equal to 1 Grey (Gy). For the case of heavy ions (including protons), the exposure in rads is determined from the ion LET and the particle fluence according to equation 1. Note that the LET and corresponding dose for a given particle fluence are material dependent quantities. As an example, a 100 MeV proton has an LET in silicon of 5.93 x 10-3 MeVcm2/mg. For a fluence of 1 proton per square centimeter, the corresponding dose would be 9.5 x 10-8 rad(Si). This relation is depicted versus proton energy as figure 8.
(
)
LET ( matl.) MeV ⋅cm2 ⋅ fluence⋅ 1 ⋅ 1.60⋅10−5⋅mg⋅rad = X ⋅ rad ( matl.) mg MeV cm2
IV-18
[1]
Dose per Unit Fluence for Protons in Si Dose (rad[Si]/proton/cm 2)
1.0E-05
1.0E-06
1.0E-07
1.0E-08 1
10
100
1000
Proton Energy (MeV) Figure 8. On average, each proton deposits dose according to its LET and the relation expressed in equation 1. The dependence in the above figure describes dose deposition from protons in silicon.
The basis for the use, or even the existence, of the rad is that the effects of the ionizing radiation in question, whether it be electrons, photons, protons, or other ions, will be equivalent for a given amount of adsorbed dose, irrespective of the radiation source. The radiation effects community has examined this assumption for several important cases, for example, the equivalence of 10 keV x-rays and Co-60 dose and the role of secondary electronic equilibrium at material interfaces. In the following paragraphs, we consider two microdosimetry issues that are specific to proton dose. 3.2.1 Lateral nonuniformities (LNUs): The term LNU in the context of radiation damage in MOS transistors describes the nonuniform distribution of holes in gate oxides. The initial papers addressing the phenomenon are cited in [Frie-88]. This paper describes a detailed investigation of the causes and effects; especially the false indication of interface state production using the subthreshold method when applied under cryogenic conditions. All of the initial work assessed the role of LNUs in gate oxides that were exposed to either Co-60 gamma rays or 10 keV x-rays. In the paper by Frietag, et al., a statistical formalism for the analysis of microscopic fluctuations in dose was introduced and applied to describe the behavior of the subthreshold current. The following year the microdosimetry formalism from [Frie-88] was invoked and modified to treat the problem of LNUs arising from proton damage using a two component model [Xaps-89]. Conceptually, the problem of nonuniform dose deposition from protons might be suspected to cause significant effects when small geometries are considered such as in thin gate oxides. After all, there is a track structure, at least in the initial dose deposition. While the data and analyses presented in [Xaps-89] do demonstrate measurable effects in subthreshold leakage current stretch-out at 77o K, perhaps the most remarkable result of this body
IV-19
of work lies in the fact that, at least to first order, the microdosimetric fluctuations in ionizing dose deposition from protons are unimportant to satellite electronics. 3.2.2 Electron-Hole Recombination: This issue concerns the fate of the electron-hole pairs produced as the proton loses energy in a material. For traditional TID effects, holes are the more important since either trapped oxide charge or interface state formation lead to parameter shifts in transistor performance. The unique characteristic of electron-hole production along ion tracks is that electron-hole pairs along ion tracks are in relatively close proximity to neighboring charge pairs as compared to those generated by photon or electron radiation sources. This close proximity can affect the charge yield for a given dose deposition through two mechanisms: geminate recombination and columnar recombination. Both recombination processes are well documented in literature extending back to the early 1900’s. For the case of protons incident on Si, [Oldh-84] established what is today recognized as the definitive study on these effects. This reference provides vectors to the relevant literature, including the model developed to assess these effects for ions simulating cosmic ray effects [Oldh-83]. The geminate recombination model applies to the situation in which electron-hole pairs are widely separated from other charge pairs and therefore more likely to interact and possibly recombine with each other. Columnar recombination applies to densely ionized tracks along which charge pairs are so dense, and recombination is as likely to occur with pairs initiated in separate events. Figure 9, reproduced from [Oldh-84], indicates the energy dependence of the two recombination models and shows comparisons with published data acquired on n-type silicon. The yield in figure 9 is fractional, and at high proton energies where the geminate model applies the yield is expected to be the same as with Co-60. Note that even at high proton energies the relative yield does not approach unity. Furthermore, at lower proton energies important to both space environments and to test facilities, the yield decreases dramatically with decreasing proton energy. Recall from figure 7 that the lower energies correspond to increased LET and therefore more densely ionizing track structure. The importance of proton dose in LEO missions notwithstanding, there have been few carefully controlled comparisons of Co-60 versus proton-induced TID response. A discussion of much of the relevant data is found in [Ma-89]. Investigation into p-MOSFET dosimeter response to proton dose has indicated similar behavior to that reported in [Oldh-84], but with indications of correlation between charge yield and electric field strength [Augu-82]. Figure 10, which is reproduced from the later work, indicates the relation between threshold voltage shift and gate bias on 1100 angstrom thick oxides for several different radiations, including 37 MeV protons. In [Stap-85], a clear trend was noted in the threshold voltage shifts of p-MOS transistors for the case of heavy ions versus literature data on Co-60. However, the comparison between Co-60 and 62 MeV protons showed no significant difference at a dose of 10 krad(Si). Recent comparisons of modern p-MOS dosimeters indicate similar behavior with a possible slight reduction is high energy proton response versus that from Co-60 [Peas-99]. Three papers have reported comparisons of photon versus proton-induced dose in devices other than p-MOS dosimeters. A careful experimental and modeling effort into the IV-20
1.0
YIELD
GEMINATE MODEL
0.1 SOFT OXIDES HARD OXIDES
COLUMNAR MODEL
0.01
1
10
100
PROTON ENERGY (MeV) Figure 9. This figure is reproduced from [Oldh-84], and it indicates the energy dependence of the two recombination models. The yield is fractional, and at high proton energies where the geminate model applies the yield is expected to be the same as with Co-60
Figure 10. p-MOSFET dosimeter response to dose from proton and other sources indicates a correlation between charge yield and gate bias which affects electric field strength. The oxide thickness is 1100 angstroms.
IV-21
response of radiochromic dye film dosimeters indicated the importance of the columnar recombination mechanism for protons [Hans-84]. This work also investigated track structure effects on charge yield and modeled these effects for protons from 3 to 16 MeV. [Xaps-90] reported a comparison of proton to Co-60 dose response of hardened gate oxides in terms of interface state buildup as measured by the sub-threshold technique. The data did not show a significant difference in response, and their work noted the high value of electric field as being a possible explanation for apparent dose equivalence. There are three implications for the effects of recombination on the charge yield from proton exposures. First, the reduced effect per unit dose suggests that proton studies to assess TID effects should be approached very cautiously, especially if the intended application is in a mixed proton-electron environment, or worse, an environment dominated by electrons. The proton results could lead to slight underestimation of the environment effects. Secondly, if proton dose dominates in the application and the dose response is based on Co-60 studies, they may be slightly conservative. In the absence of detailed proton and Co-60 response comparisons on the parts being considered for flight, good engineering practice argues against factoring the reduced response to protons into the expected part lifetime. Finally, p-MOS dosimeters flown on board satellites may exhibit a reduced response to protons and therefore slightly underestimate the environment, unless the calibration is performed with consideration of the anticipated environment and the energy (and possibly field) dependence of the device response. In summary, to first order, the concept of dose works well for assessing combined effects of various space radiation sources, including protons. In other words, 100 ergs per gram (1 rad) of energy deposition from one radiation source results in approximately the same device response as the same amount of energy deposited from another source. For TID purposes, LNUs from proton exposure do not appear to be an important effect. Geminate and columnar recombination lead to reduced charge yield, per unit rad, in proton environments. Even so, the assumption of a linear superposition of electron and proton dose, combined with Co-60 or 10 keV x-ray testing for component evaluations leads to reasonable estimates which may be slightly conservative in proton dominated dose environments. In closing, we would like to remind the reader that these comments address the fidelity of response estimated for proton-induced total ionizing dose only. If displacement effects are important, then there is no substitute for carefully planned tests, and x-ray or gamma ray testing is probably not appropriate. More will be made of this point in the section IVB on displacement effects.
4.0 PROTON-INDUCED SINGLE EVENT EFFECTS The basis for the classic approach to single event effects from protons first appeared in [Guen-79]. Here, the upset mechanism description involved the proton (or neutron) initiating an inelastic nuclear reaction leading to high energy recoil atoms and subsequent localized ionization sufficient to cause upset. Proton upset in satellite microelectronics was not a major concern at first because of the larger feature sizes and correspondingly high critical charge required for upset. In time this would change, and within a few years, missions routinely saw upset rate increases corresponding to high proton fluxes. In the 1995 NSREC Short Course
IV-22
segment on “Single Event Effects Qualification,” Bill Stapor includes a list of 30 missions with confirmed proton-induced SEE experiences. Today, that list would be much longer. In 1984 the key paper appeared, “Predicting Single Event Upsets in the Earth’s Proton Belts” [Bend-84]. This semi-empirical model assumes the proton reaction based mechanism and allows the use of SEE test data at a single proton energy or multiple energies for the prediction of upset rates in the proton belts. In the single parameter model, the parameter relates the proton upset sensitivity to the proton energy threshold for upset production. The basic formalism for proton-induced SEEs from nuclear reaction recoil products has not changed appreciably since its introduction in the mid-1980s. In 1989, a paper emphasizing application of the 2-parameter Bendel approach showed improved agreement with test data for modern devices with small feature size [Shim-89]. This was followed the next year by another paper advocating the 2-parameter approach and pointing out the importance of proton testing at higher energies to increase rate prediction accuracy [Stap-90]. In addition to the 1995 Short Course notes, the interested reader should refer to the 1997 NSREC Short Course segment on “Single Event Analysis and Prediction” by Ed Petersen [Pete-97]. This course offers an excellent discussion of proton upset mechanisms, models, and rate prediction tools, as well as practical consideration for proton upset cross section measurement. These course notes also provide detailed descriptions of the expanded range of component types for which proton SEEs are important. This includes not only memories, but also processors, ASICs, linears, and most modern circuit families. Additional related materials are found in Ed Petersen’s review article on, “Approaches to Proton Single Event Rate Calculations,” [Pete-96]. Given the level of discussion on the general topics of proton upset measurement and rate predictions in these two previous Short Courses (and references therein), the following material will be devoted to other proton-related special topics important to spacecraft developers. In the three sections to follow, we first treat two special topics related to the “classic” phenomenology of proton-induced recoil initiated SEEs. Next we treat a relatively recent phenomenon affecting several types of very sensitive devices in which protons can initiate upsets by direct ionization. Finally, the last portion of this section will examine various classes of proton-initiated hard errors.
4.1 Test Issues and Special Cases The “classic” approach to proton SEE testing of memories involves loading a pattern and setting up desired test conditions, exposing the Device Under Test (DUT) to some predetermined fluence, and interrogating the device to determine level of functionality and changes from the initial test conditions. The exposures may take place either in air or vacuum for high energy (>10 MeV) protons, and it is almost always required, for health safety reasons, that test personnel be remote to the exposure area during irradiations. Usually, the test setup involves test equipment (e.g., memory tester, computer with controller card, transient digitizer, etc.) local to the test and the ability to control that equipment remotely from either an extended monitor and keyboard or across a communication link or network connection. IV-23
For devices other than memories, the test instrumentation can become quite complicated. For example, processor tests may require comparison with either a second (ghost) processor or with expected results emulated in software. Such a test usually involves real time monitoring of the DUT during exposure. Often, testing of a given component must be done in situ with supporting flight hardware and software to fully assess the impact of SEEs and their likelihood of propagating through the system. This is especially true for circuit implementations incorporating error detection and correction circuitry. Throughout this section, the examples for discussion and case studies have been selected to illustrate not only some of the effects that protons may have, but also the choices and tradeoffs confronting the test engineer in gathering meaningful data without overcomplicating the test effort. 4.1.1 High Speed Technologies: While there still exists some controversy regarding the ability to accurately correlate proton and heavy ion SEE sensitivities, it is generally agreed that protons are more likely to affect technologies which exhibit lower thresholds for heavy ion SEE effects (e.g., LETth below ~10 MeVcm2/mg). By virtue of their lower nodal capacitance and lower switching energies, as well as the absence of a complementary structure, this tends to include several high speed technologies. Protons are known to cause SEEs in Si bipolar devices (ECL), GaAs MESFET and HIGFET devices, GaAs HBT based devices, and other high speed technologies [McMo-96, and references therein]. In general, if the technology has been developed with high speed in mind, then it is likely to be sensitive to proton SEEs, unless SEE hardening has been explicitly incorporated. For a given component, the effect on the circuit and subsystem can be extremely dependent on how the part functions in the circuit, and how follow-on circuit parameters affect error propagation. Investigations have shown that SEE cross sections in high speed technologies can be very dependent on device clock speeds [Mars-95, Reed-96], and even circuit hardening attempts may exhibit a clock speed dependence [Schn-92]. This is understood in terms of reduced noise margins during switching so that the circuit is more vulnerable to SEEs and that vulnerability occurs more often as the clock speed increases [Reed-96, and references therein]. The combination of inherent SEE softness in high speed logic with the increased sensitivity at high data rates argues for in situ proton testing of high speed circuitry where accurate flight error rate predictions are desired. Such testing is conceptually straight forward, but often challenging to carry out. Complications include the need to provide the DUT with high data rate signals and detect errors “on the fly” without being sensitive to the electrically noisy accelerator environment. The requirement to do this remotely argues for test automation with custom software and hardware. Supporting test hardware must itself be capable of the speeds of the DUT, and should in fact have broadband characteristics with ample bandwidth margin. This often requires the design of test circuits that must be fabricated to operate in the GHz regime. Figure 11 illustrates the test hardware and software environments used in the proton SEE evaluation of a commercial fiber channel transceiver set fabricated in a Si p-ECL process [Cart-97]. The referenced paper describes the DUT and analyzes the test results, but a
IV-24
VXI Mainframe Equipment controller (LABVIEW program) GPIB
GPIB
Digital Input/Output
User Interface (keyboard, monitor) Experimenter facility Irradiation room
GPIB
Programmable Serial Bit Stream Generator
GPIB
Fiber Channel Interface Receiver
20
Fiber Channel Interface Transmitter
1
Bit Error Rate Analyzer
Evaluation board GPIB Signal Generator (53.124 MHz)
Signal Generator (53.124 MHz)
Figure 11. Block diagram for a 1.0625 gigabit per second proton-induced single event effects test of a commercial fiber channel chip set [Cart-97]. Testing required automatic data logging and remote control of both the DUTs and also the bit error rate test equipment.
significant portion of the paper is devoted to the test hardware and methodology with emphasis on the fact that meaningful proton evaluations of high-speed technologies are nontrivial. As the data rates increase into the Gbps regime, availability of state-of-the-art bit error rate test equipment becomes a significant issue. Combined equipment costs can exceed $1M. The test development time and risk associated with transport to accelerator facilities are important concerns. As identified in figure 11, the test relied on a “VME eXtended for Instrumentation” (VXI) mainframe running LABVIEW software to control instruments and capture data. Test conditions were set from the VXI chassis via a digital I/O card interface and General Purpose Interface Bus (GPIB). Software controlling the test flow included interfaces to the commercial 12.5 Gbps bit error rate test equipment as well as the custom hardware evaluation boards. During the test, errors were logged automatically and stored on the VXI controller’s embedded processor module’s hard drive, and simultaneously made available over an Eithernet hub for remote archival storage. The test flow control and software interface were exercised by test personnel using a keyboard and monitor extension, but all high speed test equipment and control equipment had to be placed in the target room. Whenever test equipment must be located near the target it is a good, if not necessary, precaution to be aware of the possibility of scattered protons and neutrons reaching the equipment and place more sensitive units where the exposures will be minimized.
IV-25
The previous example described the evaluation of a commercially available device that was being considered for flight. Often, the interest is in the proton SEE characteristics of a technology, and test circuits may be used resulting in simplification of DUT interfaces. Examples are described in [Mars-95 and Reed-96] along with data showing the importance of going to the trouble of the high speed test approach. Figure 11a illustrates that significant differences in proton upset sensitivity for static versus dynamic testing can result as the data rates enter the hundreds of Mbps regime. The explanation is based on an enhanced sensitivity as data transitions near clock edges, and this is carefully mapped out in [Reed-96]. In general, it is important to test a device with as much fidelity as possible to the intended application, and this is especially important with regard to clock speed where high data rates are required. Design of test hardware and DUT fixtures for high data rates is challenging in itself, and misleading results can follow from bandwidth-limited test configurations [Reed-96]. For this reason, broadband test sets with excess bandwidth are highly recommended.
10
-9
SCFL Static SEU Test SCFL 400 Mbps
2
Error Cross-Section (cm / bit)
Proton Cross-section Speed Dependence
-10
10
-11
10
-12
10
10
30
50
70
Proton Energy (MeV) Figure 11a. Bit error rate testing of GaAs HIGFET shift registers showed that static upset measurements would underpredict the upset rates by large factors if the intended application involves fast clock speeds [Mars-95]. These factors increase as the data rates increase and can approach 100.
4.1.2 Small Probability Events: A recent Short Course [Pete-97] pointed out that the total dose sensitivity of a technology can place practical limitations on the accurate determination of proton SEE cross sections. This can be true for TID hardened parts where assurances of very low cross sections are needed, but it may be much more important for COTS or other unhardened devices for which the TID failure levels may be only a few krad(Si). If the TID failure level of a candidate component is on the order of ~10 krad(Si), then testing on an individual DUT will be limited to that dose. If protons of ~60 MeV are used, this corresponds IV-26
to a device cross section of ~10-11 cm2. Smaller cross sections cannot be measured with a single device, although they may be important, especially for hard failures and disruptive soft error modes. The accurate determination of SEE cross sections for important events can require testing of many devices to TID failure levels to get even poor SEE statistics. Such measures may be required when the SEE may lead to catastrophic failure, when many copies of the same device are present on the same satellite, or when many copies of the same device are used in a constellation of identical satellites. The NASA Hubble Space Telescope offers one such example with its 12 Gbit Solid State Recorder (SSR). The SSR is based on DRAM technology and uses 1440 die with each containing 16 Mbits. Details of the proton SEE response of the individual die are provided in [LaBe-98], along with a description of in-flight anomalies that indicated proton sensitivity. Prior to flight, testing had been performed with both heavy ions and protons and on flight lot die. In these tests, block errors were identified through heavy ion testing, and the LET threshold was measured at about 5 MeVcm2/mg. With such a low threshold, protons might be also be expected to cause block errors, but after 3 flight lot die were tested to proton fluences of ~ 3 x 1011 (~30 krad(Si)), the devices failed from TID without exhibiting the single event induced block error. However, after launch, two block errors were noticed and correlated with the proton environment. To understand the source of the these errors, further analysis pointed to the need for additional proton testing with a larger sample size. After a second round of testing with a sample size of 100 die, it was determined that protons could lead to the block error condition, and 9 such events were noted on the 100 die sample set. Calculations of the error cross section and expected in-flight error rate showed good agreement with the anomaly rate. Fortunately, for the HST case the block errors were easily corrected with robust ReedSolomon EDAC protection. The details of this example and others are found in [LaBe-98]. If the error condition were not easily corrected, or worse, if permanent failure resulted, the condition would not have been predicted with a small test sample set. When large numbers of a given device are flown either on the same spacecraft or across a large number of satellites, correspondingly large samples sizes must be used to assess all possible SEE modes. The exact number will be a function of both the orbit and the TID response of the DUT. If hard failure modes are possible, then large sample sizes may be warranted, even if only one device is to be flown. 4.1.3 Single Event Transients in Linear Devices: Transient effects in linear circuits were first reported by Koga, et al. in [Koga-93]. Their study examined transient signals that propagate to the output of analog circuits as a result of heavy ion irradiation. The following year, Ecoffet and coworkers confirmed the findings of Koga, et al., and extended the study to examine transients in several linear circuits, including LM 108 operational amplifiers, LM 111 voltage comparators, LM 218H operational amplifiers, and LM 211 voltage comparators [Ecof-94]. Their findings demonstrated the impact of the problem in each of these part types, and more importantly, showed that for some cases the heavy ion LET IV-27
threshold for initiating transients can be quite low (e.g. well below 5 MeVcm2/mg for the LM 108 and LM 111 devices). Ecoffet, et al., point out that such a low threshold would be expected to result in sensitivity to high energy protons. Nichols, et al., reported proton-induced transients in LM 111 and LM 139 comparators. Their analysis indicated that the transient duration could exceed 200 ns in some cases. Proton error cross sections were on the order of 10-10 cm2 per device over the range from 30 MeV to 200 MeV. Measurements showed varying sensitivity with input voltage levels and also included heavy ion measurements on the same device types. LaBel, et al., have also noted the sensitivity of linear circuits to proton induced transients [LaBe-95a]. Currently, the risk associated with proton induced transients is being assessed by a number of groups. If anything, it is more complex than with digital circuits since the magnitude and duration of the transient vary greatly with several parameters (e.g., input conditions, output loading, proton energy, etc.), and the effects are highly circuit and application dependent. 4.1.4 Correlation Between Proton and Heavy Ion SEE Sensitivities: Before concluding this discussion of “traditional” proton SEE we briefly examine approaches that have been offered to correlate proton SEE with heavy ion SEE. Such correlation can be useful for estimating proton upset sensitivity when heavy ion data is available. In addition, for present technologies, the correlation approaches can allow the estimation of heavy ion sensitivity when only proton data exist or when packaging issues preclude penetration by heavy ions to the active device regions. These estimates can be useful, but if high confidence predictions are required, these estimates should not be substituted for test results from the flight lot devices in application specific test configurations. PROFIT (for Proton Fit) is an empirical model that allows fitting of heavy ion data or heavy ion data combined with proton test data to extract parameters allowing prediction of proton upset rates [Calv-96]. The approach requires knowledge of the number of sensitive cells. It assumes that sensitive cells have the same spatial dimensions but may have differing critical charge levels. In comparisons with the two-parameter Bendel approach, the referenced paper showed very good agreement for the 18 different device types used in the study. In [O’Ne-98], another approach based on proton reaction kinematics shows how upper bounds on the heavy ion upset rates and failure probabilities can be estimated from 200 MeV proton data. This correlation requires proton data as input, and can be especially useful when heavy ion data are not available. The method does not allow estimation of proton upset sensitivity from heavy ion data. The final correlation technique we will discuss was first reported in 1983 and has been revised several times with the most recent being “The SEU Figure of Merit and Proton Upset Rate Calculations” [Pete-98 and references therein]. This approach is based on the claim that upset sensitivity for a given device can be summarized by a single parameter, the figure of merit (FOM). The referenced paper indicates how the FOM can be calculated based on either the heavy ion upset saturation cross section and threshold or from the proton upset saturation cross section. Once determined from either data set, the same FOM can be used to estimate IV-28
upset rates from either trapped protons or from heavy ions. The aggregate upset rate will then be the combination of the two contributions. The referenced paper shows good agreement for a variety of device types with varying levels of SEE sensitivity. In some instances it is of interest to assess neutron SEE sensitivity, such as in avionics applications. In [Norm-98] the Burst Generation Rate (BGR) technique for assessing proton upset sensitivity is compared with neutron BGR calculations and data. If neutron SEE data were available on a device, this approach could be taken to gain an idea of proton upset sensitivity, but as the reference indicates the correlation is not precise. Though it was not mentioned in the introduction where reasons for proton testing were listed, one possibility would be that the assessment with protons could be used with the BGR correlation to estimate neutron SEE sensitivity for avionics applications.
4.2 Proton Direct Ionization and SEEs The preceding section examined special cases of the conventional indirect proton SEE mechanism, which involves heavily ionizing nuclear reaction recoil products. Until just a few years ago, this was considered to be the only important mechanism for proton-induced single event effects. In several recent studies, SEEs due to direct ionization by protons have been reported. Though these may be “special” cases, their treatment in terms of mechanism identification, test issues, hardening solutions, and rate predictions are unique, and the remainder of this section will address these issues. At the outset, we note that for the indirect mechanism to occur there must be a reaction, and the reaction cross sections are so low such that only about 1 proton in ten thousand undergoes such an event. Most protons traverse the region and leave only an ionization track, which often matters little. However, if the circuit is sensitive to the amount of charge deposited by a single ionization track from a proton traversal, then the event cross sections may be greatly increased, by up to four orders of magnitude over the indirect mechanism. Such devices will therefore be very likely to exhibit SEEs with high rates in proton environments. 4.2.1 CCDs: In order for a device to be sensitive to direct ionization from protons, it is likely designed for an application requiring high sensitivity. In the case of the charge coupled device imaging array, the sensitivity is required to register faint signals from distant objects. For some applications, the signal may literally be only a few electrons integrated into the imager’s depletion volume prior to readout. Not surprisingly, when a proton traverses the same depletion region or nearby material from which the deposited charge can diffuse, the CCD pixel registers a false signal. These false signals from proton SEEs can affect science instruments and star tracker based navigational equipment as well. An excellent reference exists which examines the rates and charge signatures for carefully controlled test conditions and provides orbital predictions for a LEO application in the proton belts [Lomh-90]. There are two techniques to minimize the effects from unwanted proton strikes. Imaging arrays on the NASA HST mission are troubled with these stray signals when in the South Atlantic Anomaly so much that they curtail the science operations when passing IV-29
through this high flux region. When stopping operation is not practical, such as with a star tracker, transient events are usually rejected by using a Kalman filter approach to average over several frames of imagery and reject signals which are not repeated in subsequent frames taken in view of the same region. In figure 12, the four images have been acquired by a 1024 pixel by 1024 pixel CCD incorporated into one of the chronograph instruments on board the Solar and Heliospheric Observarory (SOHO) satellite. SOHO occupies an orbit around the L1 libration point that sits 930,000 miles from the Earth on the Sun-Earth line. The coronagraph instrument filters the bright orb to focus on the details of the coronal structure; hence the dark circles in the center. The four panels depict the development of a coronal mass ejection (CME) on 11/6/97. CMEs and solar flares are the two categories of solar disturbances that can result in solar proton events at satellite positions. The two lower panels show the effects of CME protons reaching the coronagraph’s CCD. Even though the instrument has heavy shielding to protect the CCD, the > 100 MeV protons from the CME penetrated to the focal plane. Note the range of proton
Figure 12. Coronagraphs from the SOHO satellite follow the evolution of a coronal mass ejection. Protons from the event reach the instrument’s CCD and “pepper” the image with transients in the lower two panels. IV-30
transient sizes and path trajectories indicating apparent omnidirectional arrival. Also note that the images are from different frames, and the proton transients are not repeated in the same image locations. For this reason, temporal filtering techniques can minimize the interference from the proton strikes for star trackers and other applications requiring tracking of bright objects against a cluttered background. More can be found on these images at the website, “sohowww.nascom.nasa.gov”. 4.2.2 Optical Link Photodetectors: In recent years several missions have implemented fiber-optic based local area networks for spacecraft telemetry and control busses as well as high data rate payload busses. Data transmission via optical fiber offers advantages in terms of power savings and reduced electromagnetic interference concerns, and these issues become increasingly important at data rates in the Gbps regime. The optical signal level representing a digital “1” may contain very little energy. When received at the link’s terminal and converted back to an electrical signal by an optoelectronic photodetector, the signal level may be only a few hundred or thousand electrons prior to amplification. Several studies have demonstrated how the photodetector, by virtue of its low signal level, can be sensitive to false signals from direct ionization by incident protons. For details and additional information, please see the review article [Mars-96, and references therein]. The sensitivity of the photodetector is perhaps not so surprising in view of the fact that this optoelectronic detector functions to capture digital information at rates into the Gbps regime from optical signals with average powers of only a few µW. This results in valid signals of only a few hundred electrons in some cases. Also, the photodiode must necessarily be large enough to capture the optical signal. For typical multimode fiber, this corresponds to surface areas of thousands of square microns (the device examined in our study has a 75 micron optical aperture with an 80 micron diameter junction). Photodiode physical cross -5 2 sections can easily exceed 10 cm , and due to their extreme sensitivity, the error cross sections can be correspondingly large. Figure 13 depicts the disk-shaped planar photodiode structure under reverse bias conditions and indicates various particle trajectories that deposit charge by direct ionization. The sketch beneath shows resulting current pulses sensed in the receiver circuit which decay with an RC time constant determined by the circuit bandwidth. Also depicted is the received signal provided in a no-return-to-zero (NRZ) protocol containing the digital information. The ratio between the high and low current levels (the “extinction ratio”) is typically about 10. Receiver circuits are almost always designed to accommodate a range of incident average optical powers and automatically adjust the decision level, or threshold, to be midway between the high and low levels. As suggested in the figure, data can be disrupted if ioninduced current exceeding the threshold current is sensed at the critical mid-bit decision when a "0" is being transmitted. Though the photodiode must be large enough to capture the optical signal, it obviously should be no larger. The analysis indicates better SEE characteristics for III-V direct bandgap detectors since a depletion depth of only about 2-3 microns can result in > 80% quantum efficiency. This is in contrast with indirect bandgap detectors, such as Si for 830 nm applications, in which depletion depths are about twenty times larger. Specifically, the thinner InGaAs structure minimizes both the "target" size for ion strikes as well as the ion IV-31
pathlength when hit. Also, the III-V device is characteristic of the design choices being considered for high bandwidth data busses since the thin junction offers minimal capacitance. To take advantage of these benefits, most design efforts use III-V InGaAs detectors for 1300 nm lightwave detection. More recently, 850 nm photodetectors with thin depletion regions and favorable SEE performance have been identified [Mars-98]. 4.2.2.1 Proton SEE Measurements on Fiber Optic Receiver Circuits: Proton testing of operating links can be done in situ using subsystem hardware or on components using a commercial bit error rate test set. For the purpose of understanding the test approach and underlying mechanisms and their effects, we present the latter approach here. The subsystem hardware level effects can in turn be inferred from this material, and the impact on the subsystem will differ according to the particular protocol and architecture. Examples of the relation between hardware level effects from “generic” device testing and the impact on specific subsystems can be found in [Carts-97, Dale-97, Mars-96, and references therein]. The SEE response of all associated circuitry must be considered, but we focus on bit errors in the photodetector receiver since it is primary importance in many cases. Figure 14 illustrates a typical test setup for measuring link bit-error-ratio (BER) performance at the component level. The BER is ratio of bits in error to total bits transferred for a given transmission interval. Full details of the measurement are found in [Mars-94a]. This setup for proton testing is similar to that shown in figure 11 for fiber-channel transceiver hardware, and the need for software controlled data collection and logging applies here too.
Proton Induced Bit Errors L
p+ Proton ionization tracks or reaction recoils generate charge in detectors.
Lmax
“1” is not corrupted
This “0” may be corrupted
This “0” is corrupted
i1 ith i0
Time Decision Points
Figure 13. The reverse biased disk-shaped planar collects charge that is deposited by direct ionization from protons. Resulting current pulses sensed in the receiver circuit decay with an RC time constant determined by the circuit bandwidth. Data can be disrupted if ion-induced current exceeding the threshold current is sensed at the critical mid-bit decision when a "0" is being transmitted. IV-32
P r o to n B E R M e a s u re m e n t P N D a ta S e q u e n c e G e n e r a to r, 2 7 -1 Sequence R e c o v e ry
0 .2 , 0 .4 , o r 1 G b p s
B E R C a lc u la tio n O p tic a l D a ta M o d u la to r 1300 nm Laser
P ro to n B eam O s c illo s c o p e
S h ie ld O p tic a l A tte n u a to r L ig h tw a v e P o w e r M e te r
P h o to d io d e U nder Test
A m p lifie r C lo c k R e c o v e ry
T IA D a ta R e g e n e ra to r
Figure 14. This illustrates a typical test setup for measuring link bit-error-rate performance. Full details of the measurement are found in [Mars-94a]. The automated test set uses commercially available bit error rate test equipment.
The tester was set to generate a serial pseudo random numeric (PN or PRN) sequence of bits in length. Data rates of 200, 400, and 1000 Mbps were established by an external waveform generator. The fiber link included a programmable attenuator so that the desired optical power level could be adjusted over the range of -30 dBm to 0 dBm (or 1 µW to 1 mW). The optical power was monitored by an external light wave meter or coupled onto the surface of the photodiode under test. Light was launched onto the photodiode using a 3axis micro-manipulator stage, and coupling efficiency was maximized by tuning and monitoring the photodiode output on a digital sampling oscilloscope. Signals were amplified by a Trans-Impedance Amplifier (TIA) and evaluated for errors resulting from proton strikes. (27-1)
With protons incident on the photodiode, we monitored the BER and recorded the number of errors. Measurements of BER were typically made with >100 total errors to assure good statistics. This usually covered a time interval of minutes. By logging the percent of error free intervals, we verified that for protons the errors were due to individual events and not contiguous errors from a single strike. Similarly, for higher LET He ions, we determined the average number of errors per strike using this method. These measurements described here and in [Mars-94a] were performed at the Naval Research Laboratory beam-line (beam-line 2) at the Crocker Laboratory, University of California. For in situ measurements of data transmission with bit periods of only a few nanoseconds, one must carefully consider the beam’s temporal structure and its relation to the data stream. We examined the impact of the 22 MHz cyclotron frequency (at 63 MeV) which provides micro-pulses of approximately 1.3 ns duration every 44 ns. Our experiments were IV-33
conducted in a manner to assure this did not influence bit-error cross section measurements. Consideration of the microstructure of the timing of proton arrivals may be an important experimental issue is some situations [LaBe-93], especially when high data rates are involved. Most high energy proton facilities have similar concerns. 4.2.2.2 Analysis and Indication of the Role of Direct Ionization in Photodetectors: Next we consider the example of errors from proton-induced direct ionization in photodetectors to show that they can be quantified with the well-developed tools used in more conventional single event investigations. As is customary with spatially separated arrays of memory elements in Random Access Memories (RAMs), we define bit error cross sections for temporally separated bits in a data stream as the ratio of failed bits to the particle fluence incident on the device during the interval in which the failures are measured. Our objective is to understand the error cross section dependence on environmental factors such particle flux and also the particle energy and angle of incidence, which impact the effective linear energy transfer (LET). Also, for a given receiver design, we measure the cross section dependence on the data link characteristics including the data rate and the optical power incident on the photodiode. The result is a data set that can be readily analyzed with existing descriptions of the expected environment to produce estimates of link performance in orbit. In [Mars-94a] the case is made for treating link bit errors as arising from direct as opposed to the indirect upset mechanism. Several indications point to this interpretation including the angular dependence of measured cross section data, the relation between ionization induced charge and electrical signal size, the relation between device physical size and the magnitude of the error cross section, and the particle and LET dependence of the measured cross section. Realizing that direct ionization causes upsets has two important implications for rate predictions for proton-induced errors. First, the traditional Bendel approach does not apply, and second, the proper approach should more closely follow the approach developed for heavy ion induced upset based on LET. Figure 15, and the discussion found in [Mars-94a], show that the proton-induced error data can in fact be usefully viewed as dependent on the effective particle LET, even though the errors are due to protons. The solid lines in the curve follow the customary Weibull form (equation 2), where a, b and c are fitted parameters and σsat is the saturation cross section of the cross section versus LET relation. The family of Weibull curves corresponds to different levels of optical power used in the operating link. It is important to note that the data of figure 15 correspond to a particular data rate and the LET is for the InGaAs detector material.
σ ( LET )
− = 1 − e
LET − a c b
σ
[2] sat
To assess proton effects on the link at other data rates, we could analyze other data sets measured at other rates as we have for the 400 Mbps data set. However more general results can be obtained by inspecting figure 16, which plots cross section dependence on IV-34
optical power for 200, 400, and 1000 Mbps. The two data sets shown represent the high and low LET particles, namely He ions and high energy protons. Note that across the full optical power range and for these two extremes in particle LET, the cross section exhibits, to first order, a direct proportionality to data rate. This trend was noted in all of the detector BER data, and it is consistent with the arguments pertaining to clock rate dependence made earlier in this section. According to equation 3, a cross section that is proportional to data rate results in a BER which is independent of data rate. The cross section is represented by σ, and φ is the particle flux. The on-orbit error rate in terms of errors per day, however, would be expected to scale linearly with error cross section. It would be up to the application as to whether the BER or the error rate is the more important metric. BER =
σ ⋅φ # errors = Bits Transmitted Data Rate
[3]
By using the Weibull approximation we can describe the LET dependence of the proton-induced error cross section and then combine this response with the LET spectrum arising from direct ionization by protons in the detector material system. Then it is possible to exercise the conventional tools for heavy ion upset rate predictions to assess link BER performance in proton rich orbits. This general approach has been validated against flight data with excellent agreement [LaBe-97a, Mars-96, and references therein].
-3
ETX75 InGaAs p-i-n Diode at 400 Mbps
2
Error Cross-Section (cm )
10
10
-5
Optical Power
10
10
-7
-25 -23 -21 -19 -17 -15
-9
10
100
dBm dBm dBm dBm dBm dBm
1000 2
LET in InGaAs (MeV cm /g) Figure 15. Proton-induced bit error data depends on the effective particle LET, even though the errors are due to direct ionization from protons [Mars-94a],. The solid lines in the curve follow the customary Weibull form, and the family of Weibull curves corresponds to different levels of optical power used in the operating link.
IV-35
BER is # bits lost per bit transmitted
BER =
σ φ Data Rate
2
σ ∝ Data Rate
Error Cross-Section (cm )
Data Rate Effects on Cross-Section 10
10
10
-4
ETX75 18 MeV He Ions at 70°
-6
1.0 GHz 0.4 GHz 0.2 GHz -8
+
63 MeV p at 50° -10
10
-20
-15
-10
Optical Power (dBm)
Figure 16. The Bit Error Ratio (BER) cross section for proton strikes on a photodetector is approximately proportional to data rate irrespective of optical power or LET. The relations indicate that a cross section that is proportional to data rate results in a BER which is independent of data rate.
4.2.3 Optocouplers and Metal-Semiconductor-Metal (MSM) Photodiodes: The introduction and section 4.1 offered discussion and references to the extensive literature describing “conventional” proton effects involving nuclear reactions and the indirect upset mechanism. The section 4.2 then dealt with a newer formalism that applies to devices that exhibit behavior that is dominated by direct ionization from protons. Not surprisingly, some devices show characteristics of both behaviors. This section examines two such examples. Optocouplers have received a great deal of attention for their sensitivity to displacement damage from protons and this will be discussed at length in Section IVB on displacement effects. At the 1997 NSREC it was reported that high bandwidth optocouplers could also be sensitive to proton-induced transient effects [LaBe-97]. That study reported that proton initiated transients exhibited a rapid onset and then dissipated with a time constant governed by the bandwidth of the device. At first this appeared to be another example of the general problem of transient effects in linear devices, caused presumably by proton reaction recoil products. But on closer inspection, the angular dependence seen in figure 17 shows an enhancement in the cross section around the plane of the package. For the classic reaction recoil mechanism, no angular dependence would be expected. In [LaBe-97] the explanation is offered in terms of a combination of mechanisms. Figure 18 depicts the situation in which the device response is dominated by strikes to the optocoupler’s internal photodetector with nuclear reactions causing transients when they occur, regardless of the angle of incidence. However, for protons traversing the plane of the photodetector sufficient signal can follow from direct ionization across the longer pathlengths.
IV-36
-7
2
Device Cross Section (cm )
4x10
3x10
2x10
1x10
-7
-7
-7
0
0
30
60
90
120
Angle of Incidence (degrees) Figure 17. Proton-induced transients in optocouplers exhibit an enhancement in the cross section when the beam is directed in the plane of the package. For the classic reaction recoil mechanism, no angular dependence would be expected. In [LaBe-97] this behavior is explained in terms of a combination of direct ionization through the plane of the photodetector and indirect reaction mechanisms.
Direct Ionization Across Long Pathlengths
+ + - + -- + + + - + +- -+ +- + - -+ ++ - + +- -++- + - - + ++ ++ - + - ++++-- -+- -++++--- -+++--- + ++ +- -- ++ -- + --- ++ - - +++--+-- -++ - + + - ++++ + +-+--+-++- - + + +++-+--- + ++- -- -+++ - --++ + - - + - + + -+ - - + - + + -
Proton
380 µ m + + - + -- + + + - + -- + ++- -++-+-+-+-++- -+- + + - - + -+ +- ++- - - + + - +- -+- -+ +- ++-+ +- - + - ++-
Proton
And Nuclear Reaction Recoils.
Figure 18. This drawing from [LaBe-97] depicts the situation in which the device response is dominated by strikes to the optocoupler’s internal photodetector with nuclear reactions causing transients when they occur, regardless of the angle of incidence. For protons traversing the plane of the photodetector, sufficient signal can follow from direct ionization across the longer pathlengths.
IV-37
Cross-Section (cm2/device)
1.2E-07 0 degrees 85 degrees 87.5 degrees 90 degrees 92.5 degrees 95 degrees
1.0E-07 8.0E-08 6.0E-08 4.0E-08 2.0E-08 0.0E+00 30
80
130
180
230
280
Proton Energy (MeV) Figure 19. Data supporting the role of the direct ionization mechanism for optocoupler transients shows an apparent LET threshold for the effect [Reed-98]. The increased cross section seen when protons traverse the plane of the diode does not occur with protons above a certain LET (corresponding to ~100 MeV protons).
The following year Reed, et al. reported additional data supporting the role of the direct ionization mechanism in terms of an apparent LET threshold for the effect [Reed-98]. Figure 19, reproduced from that paper, reveals that the increased cross section seen when protons traverse the plane of the diode does not occur with protons above a certain LET (corresponding to ~ 100 MeV). Johnston, et al. extended the study to other devices and included heavy ions as well as protons [John-98]. Their findings suggest that at higher LET’s corresponding to cosmic rays the follow-on amplifier circuit may also lead to transient effects, and that for protons the nuclear elastic scattering may be important to the process. Another study has pointed to the important combination of direct and indirect mechanisms for the case of Metal-Semiconductor-Metal (MSM) photodetectors for use in digital data links such as the fiber-based data busses already described [Mars-98]. The MSM detector technology offers the advantage of a nearly planar geometry which minimizes the pathlengths (and ionization signature) when traversed by protons. Enhanced proton transient cross sections in the plane of the detector and the signature of a threshold LET for direct ionization effects were both noted for MSM devices in [Mars-98]. The existence of multiple mechanisms for transient effects has important implications for the methods and accuracy of transient rate predictions. It has been suggested [LaBe-97] that the aggregate transient rate must be calculated as the sum of the rates for indirect effects using the Bendel formalism plus the direct contribution using the modified RPP approach and measured Weibull-type LET dependence as described for photodetectors in the previous section. For devices exhibiting large enhancements in the cross section near the plane of the IV-38
device as in figure 18, both mechanisms are obviously important, and the dominant one would depend on the details of the application and the environment. It should be assumed that the rates would depend on how such a device is operated in terms of expected signal levels and detection sensitivity.
4.3 Destructive Failures The concerns for TID failure and soft errors treated in the last two sections are serious issues for designers, but the possibility of an unrecoverable catastrophic failure from a single particle event ranks among the highest concerns. Where cosmic ray heavy ions are present, proton-induced hard failures may not be the dominant threat from the natural environment; however, these failure modes should not be overlooked. In the three sections below on Single Event Latch-up (SEL), Single Event Burnout (SEB), and stuck bits, we examine the various ways in which proton-induced single events can render a circuit unusable. 4.3.1 Latch-Up (and COTS): We first consider SEL with the recognition that it is not solely a COTS issue, but it is a major concern when using COTS CMOS parts in space. Latch-up in CMOS devices is well understood in terms of a particle-induced triggering of an SCR action in a parasitic p-n-p-n path. Details of the mechanism, modeling tools, hardening approaches, and references to related topics are in covered in the NSREC Short Course Notes from 1996 [John-96]. It should be noted that not all latch-up modes lead to destructive failure, and in some cases power cycling may be used to restore nominal operating conditions. For many years, SEL was considered to be a CMOS phenomenon only in cosmic ray environments, but at the 1992 NSREC two papers reported first the laboratory confirmation of proton-induced latch-up [Nich-92] and the observation of a proton-induced latch-up event in space [Adam-92]. In these studies, the affected CMOS devices were either on a bulk or a thick epitaxial substrate, and the corresponding heavy ion latch-up threshold was fairly low (below 10 MeVcm2/mg). Even so, the proton energy threshold below 50 MeV for both cases suggests that proton-induced latch-up may be a higher risk than cosmic ray induced latch-up in mid-latitude LEO applications. Since 1992 there have been many examples of proton-induced latch-up, both in the laboratory and in-flight. Table 2 has been reproduced from [Norm-98] where it was compiled to show several examples from the literature as the basis for evaluating his formulation of the Burst Generation Rate (BGR) model for predicting latch-up sensitivity in microprocessors. Note the K-5 processor results as fabricated on the 2 micron thick epitaxial material. This is one of the more unexpected results, and it is noted, though not explained, in the original reference [John-97]. Both of these papers, and references therein, note the general correlation between susceptibility to heavy ion induced latch-up with low LET threshold, and the sensitivity to proton-induced latch-up. The lack of a more quantitative correlation is blamed on the differences between charge deposition by heavy ions and proton-induced recoils and the associated charge collection processes [John-97]. For crude estimates, it is probably reasonable to assume that devices which exhibit a low heavy ion latch-up threshold (below 5 MeVcm2/mg) will also be sensitive to protonIV-39
induced latch-up and even devices with LET thresholds of twice that may be suspect. The K5 example also illustrates that fabrication on a thin epitaxial starting material does not necessarily guard against latch-up, though systematic study of varying epi-layer thickness in one process has shown this to be an important step toward hardening against latch-up [LaBe-95]. Conversely, it is probably a safe assumption that if a device has been demonstrated to be hard or immune to latch-up by heavy ions then it will show similar hardness with respect to protons. Table 2: Measured and Calculated Proton SEL Cross Sections Device
Measr’d Proton SEL X-Section, cm²
BGR Calct’d Proton SEL X-Section, cm²
HM65162 NEC-4464 K-5 K-5 LSI-64811 LCA200K XC96002 XC96002 IDT3081
1.4E-10 1.8E-10 6.6E-9 6.6E-9 1.7E-11 1.4-4.1E-11 4.5E-9 4.5E-9 3E-11
1.4E-10 1.5E-10 2.2E-11 4E-9 6E-12 7.6E-11 2.6E-10 8.8E-9 0.9E-11
Remarks
t=6µm t=2µm
t=6µm t=4µm
Ref. for Data 31 32 32 32 32 8 33 33 34
More detailed discussion of the present understanding of the mechanisms and modeling for proton-induced SEL can be found in the references provided. In general, even if knowledge of a thin epitaxial material suggests latch-up immunity, latch-up testing should be performed prior to consideration for flight application when the process under question is not well known. Heavy ion screening would be a first step with proton latch-up testing advised only if the heavy ion threshold were low and there was a need to quantify the risk in a proton rich environment. In practice, most missions would avoid the use any part susceptible to failure by proton-induced latch-up for a critical application based on the risk to heavy ion induced failure alone. 4.3.2 Proton-induced Single Event Burnout (SEB) in Power MOSFETs: Single event burnout occurs when an ion or proton-induced recoil atom strikes a power MOSFET in its “off” state and triggers a parasitic bipolar junction transistor. This “on” transistor creates a conduction path between source and drain, and the resulting regenerative feedback leads to a high current state causing second breakdown and burnout. Single event burnout in power MOSFETs has received considerable attention as a hard failure mode from heavy ion effects [Titu-96, Alle-96, Ober-96, Adol-96, and references therein]. Recently, as was the case with latch-up, both experimental investigations and in-flight experience have pointed to protons as a possible cause for burnout. Two 1996 NSREC papers addressed this issue, one with laboratory measurements [Ober-96] and the other with flight data [Adol-96]. Oberg and co-authors evaluated the response of power MOSFETs to both high-energy proton and to high energy neutron irradiation [Ober-96]. Their evidence indicated some correlation between proton SEB cross sections and those for neutrons and heavy ions, with the IV-40
more energetic protons being more likely to cause burnout. The flight data from the CRUX experiment [Adol-96] both confirmed the effect for orbital protons and showed it to be more likely on the higher voltage (200 V) device, as expected. Figure 20 shows additional flight data and correlation of SEB rate with the applied voltage on the CRUX experiment that appear in [Bart-98]. The bias dependence is expected based on electric field dependence of the problem as described in [Titu-96, and references therein]. If proton-induced hard failure is possible, then heavy ion induced hard failure would also be possible. The determination of which failure mode would be more likely depends on the particle environment internal to the satellite and on the relative sensitivities of the device in question. [Bart-98] compares two sets of power MOSFETS flown on the CRUX experiment and shows that on a given satellite, proton-induced burnout may dominate for one device while heavy ion burnout dominates for another. In practice, power MOSFET applications typically de-rate the devices to improve reliability, avoid gate rupture, and prevent burnout. With proper de-rating, the threat of proton-induced burnout can be avoided. 4.3.3 Stuck Bits: The key paper introducing this topic was presented by Oldham, et al., in 1993 with the title “Total Dose Failures in Advanced Electronic from Single Ions,” [Oldh-93]. Their work describes the ability of a single ion to deposit sufficient energy along its path to result in localized increases in trapped oxide charge and interface state generation. In fact, for small feature sizes, an entire transistor gate can be affected and undergo failure as a result of localized dose deposition and threshold voltage shift. The initial work in DRAMs resulted in bits which could not be rewritten, hence the term “stuck bits”. This effect can occur in any device type, and is not just a problem for memories. Their paper also points out the expected increasing importance of this problem with decreasing feature size, and this has since been observed. The following year, Poivey, et al. expanded the study in terms of device types, ion species, and analysis. In that paper, the more formal term “Single Hard Error (SHE)” was introduced [Poiv-94]. At present, proton-induced stuck bits are not considered to be a significant problem, though scaling trends suggest that this may soon change. Proton-induced stuck bits may be either temporary or permanent as reported in [Sore-95]. Figure 21, from the CRUX experiment, indicates that stuck bits have occurred in all device types included in the experiment. Though they have been correlated with solar particle events, [Bart-98] points out that most events have been outside of the SAA and therefore are more likely due to heavy ions. From a test perspective, it is common to see stuck bits during proton SEE testing as the TID limit of the technology is approached. That, along with increased leakage currents, is an indication of the need to resume the test with a fresh device.
IV-41
SEBs As Function of Voltage Drain-to-Source For L < 3 (August 11, 1994 to May 17, 1996) 200
Number of SEBs
175 150 125 100 75 50 25 Results For 200 Volt Boards 0 160
166
172
178
184
190
196
Voltage Drain-to-Source (Volts)
Figure 20. The results of the power MOSFET single event burnout experiment flying on CRUX confirm that protons can lead to device failure and showed that the proton rates can exceed heavy ion induced SEB rates [Bart-98]. Stuck Bit Errors on CR UX SRAMs
Dipole Shell Parameter - L
100 MICRON 1M ED I 1M HITACHI/E LMO 1M MICR ON 256K ED I 256K IDT 256K L=3 Solar E ve nt
10
1 0
1 00
200
300
400
500
6 00
7 00
Number of Days from Launch
Figure 21. Stuck bits have occurred in all the modern memory devices flying on the CRUX experiment [Bart-98]. The chart indicates that some (for L < 3) are probably due to protons, but most are thought to be heavy ion related. IV-42
5.0
SUMMARY
Satellite microelectronic and photonic devices subjected to the ionizing effects of protons may exhibit responses either from total ionizing dose or from single event phenomena. Heavily shielded devices will likely receive dose primarily from protons in orbits encountering the inner radiation belts. For total ionizing dose, we have reviewed the literature comparing the equivalence of proton dose versus other sources of dose deposition found in either the space environment or in laboratory test facilities. We conclude that Co-60 and electron dose satisfactorily simulate proton dose for most purposes. Many different single event phenomena arise from protons including soft errors from nuclear inelastic reaction events, nuclear scattering events, and even direct ionization in several types of more sensitive devices. Special considerations are needed for SEU testing of high speed devices and for evaluations of devices with low soft error cross sections relative to their TID failure levels. In addition, both destructive and nondestructive hard errors may result from proton-induced reactions. In many cases, the concern for hard errors will be greater for cosmic rays, but in geomagnetically shielded (e.g., low-Earth orbits) the greater risk can be proton-related. Satellite subsystem design efforts benefit from proper expression of the anticipated proton environments in thorough requirements aimed at describing realistic typical and worst case proton flux and fluence levels. We have discussed many aspects of the environment models and their associated uncertainties as they affect the requirement definition.
6.0
ACKNOWLEDGMENTS
We gratefully acknowledge partial support, technical suggestions, and complete encouragement from our colleagues and friends in the Radiation Effects and Analysis Group and Radiation Physics Office at NASA Goddard Space Flight Center. We also appreciate the helpful interactions with the Radiation Effects Branch members at the Naval Research Laboratory and throughout the radiation effects community.
IV-43
7.0
REFERENCES FOR INTRODUCTION AND PART A
(All references are unclassified.) [Adam-92]
L. Adams, E.J. Daly, R. Harboe-Sorensen, R. Nickson, J. Haines, W. Schafer, M. Conrad, H. Griech, J. Merkel, T. Schwall, and R. Henneck, “A Verified Proton-Induced Latch-up in Space,” IEEE Trans. Nucl. Sci., Vol. 39, No. 6, pp. 1804-1808, 1992.
[Adol-96]
John W. Adolphsen, Janet L. Barth, and George B. Gee, “First Observation of ProtonInduced Power MOSFET Burnout in Space: The CRUX Experiment on APEX,” IEEE Trans. Nucl. Sci., NS-43, No. 6, pp. 2921-2926, 1996.
[Alle-96]
M. Allenspach, C. Dachs, G.H. Johnson, R.D. Schrimpf, E. Lorfevre, J.M. Palau, J.R. Brews, K.F. Galloway, J.L. Titus, and C.F. Wheatley, “SEGR and SEB in N-Channel Power MOSFETS,” IEEE Trans. Nucl. Sci., NS-43, No. 6, pp. 2927-2937, 1996.
[Augu-82]
L.S. August, “Estimating and Reducing Errors in MOS Dosimeters Caused by Exposure to Different Radiations,” IEEE Trans. Nucl. Sci., Vol. 29, No. 6, pp. 2000-2003, 1982.
[Bart-97]
Janet Barth, “Modeling Space Radiation Environments,” Notes from the 1997 IEEE Nuclear and Space Radiation Effects Conference Short Course, 1997.
[Bart-98]
Janet L. Barth, John W. Adolphsen, and George B. Gee, “Single Event Effects on Commercial SRAMS and Power MOSFETS: Final Results of the CRUX Flight Experiment on APEX,” 1998 IEEE Radiation Effects Data Workshop Record, pp. 1-10.
[Bend-84]
W.L Bendel and E.L. Petersen, “Predicting Single Event Upsets in the Earth’s Proton Belts,” IEEE Trans. Nucl. Sci., Vol. 31, No. 6, pp. 1201-1207, 1984.
[Calv-96]
Philippe Calvel, Catherine Barillot, and Pierre Lamonthe, “An Empirical Model for Predicting Proton Induced Upset,” IEEE Trans. Nucl. Sci., NS-43, No. 6, pp. 28272832, 1996.
[Cart-97]
M.A. Carts, P.W. Marshall, C.J. Marshall, K.A. LaBel, M. Flanegan, and J. Bretthauer, “Single Event Test Methodology and Test Results of Commercial Gigabit per Second Fiber Channel Hardware,” IEEE Trans. Nucl. Sci., NS-44, No. 6, pp. 1878-1884, 1997.
[Ecof-94]
R. Ecoffet, S. Duzellier, P. Tastet, C. Aicardi, and M. Labrunee, “Observation of Heavy Ion Induced Transients in Linear Circuits,” IEEE NSREC Radiation Effects Data Workshop Record, pp. 72-77, 1994.
[Feyn-96]
J. Feynman and S.B. Gabriel, “High Energy Charged Particles in Space at One Astronomical Unit,” IEEE Trans. Nucl. Sci, Vol. 43, No. 2, pp. 344-352, 1996.
[Frie-88]
R.K. Frietag, E.A. Burke, C.M. Dozier, and D.B. Brown, “The Development of NonUniform Deposition of Holes in Gate Oxides,” IEEE Trans. Nucl. Sci., NS-35, No. 6, pp. 1203-1207, 1988.
IV-44
(All references are unclassified.) [Gate-96]
Michele Gates, Kenneth A. LaBel, Janet Barth, Allan Johnston, and Paul W. Marshall, “Single Event Effects Criticality Analysis,” NASA Report, See NASA GSFC Radiation Effects & Analysis Home Page, http://flick.gsfc.nasa.gov/radhome.htm, 1996.
[Guen-79]
C.S. Guenzer, E.A. Wolicki, and R.G. Allas, “Single Event Upset in Dynamic RAMs by Neutrons and Protons,” IEEE Trans. Nucl. Sci., NS-26, No. 6, pp. 5048-5055, 1979.
[Hous-98]
S.L. Houston and K.A. Pfitzer, “A New Model for the Low Altitude Trapped Proton Environment,” IEEE Trans. Nucl. Sci., Vol. 45, No. 6, pp. 2972-2978, 1998.
[IEEE-1156] IEEE P1156.4, “Standard for Environmental Specifications for Spaceborne Computer Modules,” 1995. [Ingu-97]
C. Inguimbert, S. Duzellier, R. Ecoffet, and J. Bourrieau, “Proton Upset Rate Simulation by a Monte Carlo Method: Importance of the Elastic Scattering Mechanism,” IEEE Trans. Nucl. Sci., Vol. 44, No. 6, pp. 2243-2249, 1997.
[John-96]
Gregory H. Johnson and Kenneth F. Galloway, “Catastrophic Single Event Effects in the Natural Radiation Environment,” Section IV from the 1996 IEEE Nuclear and Space Radiation Effects Conference Short Course Notes.
[John-97]
A.H. Johnston, G.M. Swift, and L.D. Edmonds, “Latchup in Integrated Circuits from Energetic Protons,” IEEE Trans. Nucl. Sci., Vol. 44, No. 6, pp. 2367-2377, 1997.
[John-98]
A.H Johnston, G.M. Swift, T. Miyahira, S. Guertin, and L.D. Edmonds, “Single Event Upset Effects in Optocouplers,” IEEE Trans. Nucl. Sci., Vol. 45, No. 6, pp. 2867-2875, 1998.
[Kinn-98]
James D. Kinnison, “Achieving Reliable, Affordable Systems,” Section V from the 1998 IEEE Nuclear and Space Radiation Effects Conference Short Course Notes.
[Koga-93]
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 SEUs in Analog Microcircuits,” IEEE Trans. Nucl. Sci., NS-40, No. 6, pp. 1838-1844, 1993.
[LaBe-93]
Kenneth A. LaBel, Paul Marshall, Cheryl Dale, Christina Crabtree, E.G. Stassinopolous, Jay T. Miller, and Michele M. Gates, “SEDS MIL-STD-1773 Fiber Optic Data Bus: Proton Irradiation Test Results and Spaceflight SEU Data,” IEEE Trans. Nucl. Sci., Vol. 40, No. 6, pp. 1638-1645, 1993.
[LaBe-95]
K.A. LaBel, Donald K. Hawkins, J.A. Kinnison, W.P. Stapor, P.W. Marshall, “Single Event Effect Characteristics of CMOS Devices Employing Various epi-Layer Thicknesses,” IEEE Proc. of RADECS, 95TH8147, pp. 258-262, Sep 1995.
[LaBe-95a]
K.A. LaBel, A.K. Moran, D.K. Hawkins, A.B. Sanders, E.G. Stassinopoulos, R.K. Barry, C.M. Seidlick, H.S. Kim, J. Forney, P.Marshall, and C. Dale, “Single Event Effect Proton and Heavy Ion Test Results in Support of Candidate NASA Programs,” NSREC Radiation Effects Data Workshop, pp. 16-32, 1995.
IV-45
(All references are unclassified.) [LaBe-97]
Kenneth A. LaBel, Paul 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. Stassinopolous, “Proton Induced Transients in Optocouplers: In-flight Anomalies, Ground Irradiation Test, Mitigation and Implications,” IEEE Trans. Nucl. Sci., Vol. 44, No. 6, pp. 1885-1892, 1997.
[LaBe-97a]
K.A. LaBel, P.W. Marshall, C.J. Marshall, J. Barth, H. Leidecker, R. Reed, and C.M. Seidlick, “Comparison of MIL-STD-1773 Fiber Optic Data Bus Terminals: Single Event Proton Test Irradiation, In Flight Performance, and Prediction Techniques,” Proceedings of RADECS 97, pp. 332-338, 1997.
[LaBe-98]
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 Anomaly: Investigation of Proton Induced SEE Test Results for Stacked IBM DRAMs,” IEEE Trans. Nucl. Sci., Vol. 45, No. 6, pp. 2898-2903, 1998.
[Lomh-90]
T.S. Lomheim, R.M. Shima, J.R. Angione, W.F. Woodward, D.J. Asman, R.A. Keller, and L.W. Schumann, “Imaging Charged-Coupled Device (CCD) Transient Response to 17 and 50 MeV Proton and Heavy Ion Irradiation,” IEEE Trans. Nucl. Sci., Vol. 37, No. 6, pp. 1876-1885, 1990.
[Ma-89]
Ionizing Radiation Effects in MOS Devices and Circuits, T.P. Ma and Paul V. Dressendorfer, John Wiley and Sons, New York: 1989.
[Mars-94]
Paul W. Marshall, Cheryl J. Dale, E. Joseph Friebele, and Kenneth LaBel, "Survivable Fiber-Based Data Links for Satellite Radiation Environments," SPIE Critical Review CR-50 on Fiber Optic Reliability and Testing, 1994.
[Mars-94a]
Paul W. Marshall, Cheryl J. Dale, Martin A. Carts, and Kenneth A. LaBel, “Particle Induced Bit Erors in High Performance Fiber Optic Data Links for Satellite Data Management,” IEEE Trans. Nucl. Sci., NS-41, No. 6, pp. 1958-1965, 1994.
[Mars-95]
Paul W. Marshall, Cheryl J. Dale, Todd R. Weatherford, Michael La Macchia, and Kenneth A. LaBel, “Particle Indiced Mitigation of SEU Sensitivity in High Data Rate GaAs HIGFET Technologies,” IEEE Trans. Nucl. Sci., NS-42, No. 6, pp. 1844-1854, 1995.
[Mars-95a]
Paul W. Marshall, Cheryl J. Dale, Martin E. Fritz, Michael de La Chapelle, Martin A. Carts, and Kenneth A. LaBel, “Total ionizing dose and single particle effects in a 200 Mbps star-coupled fiber optic data bus,” Proc. of SPIE Conference on Photonics for Space Environments III, Proc. # 2482, 1995.
[Mars-96]
P.W. Marshall, C.J. Dale, and Kenneth A. LaBel, “Space Radiation Effects in High Performance Fiber Optic Data Links for Satellite Data Management,” IEEE Trans. Nucl. Sci. NS-43, Vol. 2, p. 645 (1996).
IV-46
(All references are unclassified.) [Mars-98]
C.J. Marshall, P.W. Marshall, M.A. Carts, R.A. Reed, and K.A. LaBel, “Proton Induced Effects in a Metal-Semiconductor-Metal (MSM) Photodetector for Optical Based Data Transfer,” IEEE Trans. Nucl. Sci., NS-45, No. 6, pp. 2842-2848, 1998.
[McMo-96]
D. McMorrow, T.R. Weatherford, S. Buchner. A.R. Knudson, J.S. Melinger, L.H Tran, and A.B. Campbell, “Single Event Effects in GaAs Devices and Circuits,” IEEE Trans. Nucl. Sci., NS-43, No. 2, pp. 628-624, 1996.
[Meff-94]
J.D. Meffert and M.S. Gussenhoven, CRESSPRO Documentation, PL-TR-94-2218, Phillips Laboratory, Hanscom AFB, Mass., 1994.
[Nich-92]
Donald K. Nichols, James R. Coss, R. Kevin Watson, Harvey R. Schwartz, and Ronald L. Pease, “An Observation of Proton Induced Latch-up,” IEEE Trans. Nucl. Sci., Vol. 39, No. 6, pp. 1654-1657, 1992.
[Nich-96]
D.K. Nichols, James R. Coss, Tetsuo F. Miyahira, and Harvey R. Schwartz, “Heavy Ion and Proton Induced Single Event Transients in Comparitors,” IEEE Trans. Nucl. Sci., NS-43, No. 6., pp. 2960-2967, 1996.
[Norm-98]
Eugene Normand, “Extensions of the Burst Generation Rate Method for Wider Applications to Proton/Neutron Induced Single Event Effects,” IEEE Trans. Nucl. Sci., NS-45, No. 6, pp. 2904-2914, 1998.
[Ober-96]
D.L. Oberg, J.L. Wert, E. Normand, P.P. Majewski, and S.A. Wender, “First Observations of Power MOSFET Burnout with High Energy Neutrons,” IEEE Trans. Nucl. Sci., NS-43, No. 6, pp. 2913-2920, 1996.
[Oldh-83]
T.R. Oldham and F.B. McLean, “Charge Collection Measurements for Heavy Ions Incident on n- and p-Type Silicon,” IEEE Trans. Nucl. Sci., NS-30, No. 6, pp. 44934500, 1983.
[Oldh-84]
T.R. Oldham, “Analysis of Damage in MOS Devices in Several Radiation Environments,” IEEE Trans. Nucl. Sci., NS-31, No. 6, pp. 1236-1241, 1984.
[Oldh-93]
T.R. Oldham, K.W. Bennett, J. Beaucour, T. Carriere, C. Poivey, and P Garnier, “Total Dose Failure in Advanced Electronics from Single Ions,” IEEE Trans. Nucl. Sci., NS41, No. 6, pp. 1820-1830, 1993.
[O’Ne-98]
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,” IEEE Trans. Nucl. Sci., NS-45, No. 6, pp. 2467-2474, 1998.
[Peas-99]
Ronald L. Pease, private communication.
[Pete-96]
E.L. Petersen, “Approaches to Proton Single Event Rate Calculations,” IEEE Trans. Nucl. Sci., NS-43, No. 2, pp. 496-505, 1996.
[Pete-97]
E.L. Petersen, “Single Event Analysis and Prediction,” Section III, 1997 NSREC Short Course Notes. IV-47
(All references are unclassified.) [Pete-98]
E.L. Petersen, “The SEU Figure of Merit and Proton Upset Rate Calculations,” IEEE Trans. Nucl. Sci., NS-45, No. 6, pp. 2550-2562, 1998.
[Poiv-94]
C. Poivey, T. Carriere, J. Beaucour, and T.R. Oldham, “Characterization of Single Hard Errors (SHE) in 1 M-bit SRAMs from Single Ion,” IEEE Trans. Nucl. Sci., NS-42, No. 6, pp. 2235-2239, 1994.
[Reed-96]
R.A. Reed, M.A. Carts, P.W. Marshall, C.J. Marshall, S. Buchner, M. LaMacchia, B. Mathes, and D. McMorrow, “Single Event Upset Cross Sections at Various Data Rates,” IEEE Trans. Nucl. Sci., NS-43, No. 6, pp. 2862-2867, 1996.
[Reed-98]
R.A. Reed, P.W. Marshall, A.H. Johnston, J.L. Barth, C.J. Marshall, K.A. LaBel, M. D’Ordine, H.S. Kim, and M.A. Carts, “Emerging Optocoupler Issues with Energetic Particle Induced Transients and Permanent Radiation Degradation,” IEEE Trans. Nucl. Sci., NS-45, No. 6, pp. 2833-2841, 1998.
[Sawy-76]
Donald M. Sawyer and James I. Vette, “AP-8 Trapped Proton Environment for Solar Maximum and Solar Minimum,” National Science Data Center Report NSSDC/WDCA-R&S 76-06, 1976.
[Schn-92]
R. Schneiderwind, D. Krening, S. Buchner, K. Kang, and T.R. Weatherford, “Laser Confirmation of SEU Experiments in GaAs MESFET Combinational Logic,” IEEE Trans. Nucl. Sci., NS-39, No. 6, pp. 1665-1670, 1992.
[Selt-80]
Stephen Seltzer, “SHIELDOSE: A Computer Code for Space-Shielding Radiation Dose Calculations,” National Bureau of Standards (NBS) Technical Note 1116, May, 1980.
[Shim-89]
Y. Shimano, T. Goka, S. Kuboyama, K. Kawachi, T Kanai, and Y. Takami, “The Measurement and Prediction of Proton Upset,” IEEE Trans. Nucl. Sci., Vol. 36, No. 6, pp. 2344, 1989.
[SSP-30512] SSP 30512 Rev. C, “Space Station Ionizing Radiation Design Environment,” June, 1994. [Sore-95]
R. Harboe-Sorensen, R. Muller, and S. Frenkel, “Heavy Ion, Proton and Co-60 Radiation Evaluation of 16 Mbit DRAM Memories for Space Application, 1995 IEEE Radiation Effects Data Workshop, pp. 42-49.
[Stap-85]
W.J. Stapor, L.S. August, D.H. Wilson, T.R. Oldham, and K.M. Murray, “Proton and Heavy Ion Radiation Damage Studies in MOS Transistors,” IEEE Trans. Nucl. Sci., Vol. 32, No. 6, pp. 4399-4404, 1985.
[Stap-90]
W.J. Stapor, J.P. Meyers, J.B. Langworthy, and E.L. Petersen, “Two Parameter Bendel Model Calculations for Predicting Proton Induced Upset,” IEEE Trans. Nucl. Sci., Vol. 37, No. 6, pp. 1966-1972, 1985.
[Stap-95]
William J. Stapor, “Single Event Effects Qualification,” Section II, 1995 IEEE NSREC Short Course Notes. IV-48
(All references are unclassified.) [Stass-88]
E.G. Stassinopoulos and J.P Raymond, “The Space Radiation Environment for Electronics,” Proc. IEEE, vol. 76, pp. 1423-1442, 1988.
[Stass-90]
E.G. Stassinopoulos, “Radiation Environments in Space,” in Notes for the 1990 IEEE Nuclear and Space Radiation Effects Conference Short Course, 1990.
[Titu-96]
J.L. Titus and C.F. Wheatley, “Experimental Studies of Single Event Gate Rupture and Burnout in Vertical Power MOSFETs,” IEEE Trans. Nucl. Sci., Vol. 43, No. 2, pp. 533545, 1996.
[Titu-98]
J.L. Titus and C.F. Wheatley, “Proton Induced Dielectric Breakdown in Power MOSFETs,” IEEE Trans. Nucl. Sci., NS-45, No. 6, pp. 2891-2897, 1998.
[Tylk-96]
Allan J. Tylka, James H Adams, Jr., P.R. Boberg, Buddy Brownstein, William F. Dietrich, Erwin O. Flueckiger, Edward L. Petersen, Margaret A. Shea, Don F. Smart, and Edward C. Smith, “CREME96 A Revision of the Cosmic Ray Effects on MicroElectronics Code,” IEEE Trans. Nucl. Sci., Vol. 43, No. 6, pp. 2150-2160, 1996.
[Tylk-96a]
A.J. Tylka, W.F. Dietrich, P.R. Boberg, E.C. Smith, and J.H. Adams, Jr., “Single Event Upsets Caused by Solar Energetic Heavy Ions,” IEEE Trans. Nucl. Sci., Vol. 43, No. 6, pp. 2758-2766, 1996.
[Xaps-89]
M.A. Xapsos, R.K. Frietag, E.A. Burke, C.M. Dozier, D.B. Brown, and G.P. Summers, “The Random Nature of Energy Deposition in Gate Oxides,” IEEE Trans. Nucl. Sci., Vol. 36, No. 6, pp. 1896-1903, 1989.
[Xaps-90]
M.A. Xapsos, private communication.
[Xaps-98]
M.A. Xapsos, G.P. Summers, and E.A. Burke, “Probability Model for Peak Fluxes of Solar Proton Events,” IEEE Trans. Nucl. Sci., Vol. 45, No. 6, pp. 2948-2953, 1998.
IV-49
1999 NSREC SHORT COURSE
SECTION IVB
PROTON EFFECTS AND TEST ISSUES FOR SATELLITE DESIGNERS: DISPLACEMENT EFFECTS
Cheryl J. Marshall NASA/Goddard Space Flight Center Paul W. Marshall Consultant
IV. Proton Effects and Test Issues for Satellite Designers Part B: Displacement Effects Cheryl J. Marshall NASA/Goddard Space Flight Center Electrical Systems Center / Code 562 Greenbelt, Maryland 20771 Paul W. Marshall Consultant 7655 Hat Creek Road Brookneal, VA 24528 1.0 Introduction………………………………………………………………...……...51 2.0 Proton Induced Displacement Damage Mechanisms and Tools………………..52 2.1 Displacement Damage Mechanisms and Defect Formation…………….…. 53 2.2 Displacement Damage Effects in Materials and Devices………………….. 56 2.3 Non-Ionizing Energy Loss Rate (NIEL) Concept…………………………. 59 2.3.1 The Correlation of NIEL to Device Behavior…………………...61 2.3.2 Limitations in the Use of NIEL………………………………….64 2.3.3 Calculation of Displacement Damage Equivalent Fluences……. 68 2.3.4 Concept of “Displacement Damage Dose”……………………... 69 2.4 On-Orbit Performance Predictions……………………………...………….70 3.0 Proton Displacement Damage Case Studies…………………………………….. 74 3.1 Introduction………………………………………………………………… 74 3.2 Laboratory Radiation Test Issues…………………………………………...74 3.3 Case Studies………………………………………………………………... 77 3.3.1 Bipolar Transistors……………………………………………… 77 3.3.2 Charge Transfer Devices………………………………………...79 3.3.3 Photodetectors…………………………………………………... 87 3.3.4 Lasers and Light Emitting Diodes……………………………….90 3.3.5 Optocouplers……………………………………………………. 92 3.3.6 Solar Cells………………………………………………………. 96 4.0 Summary…………………………………………………………………………...99 5.0 Acknowledgments………………………………………………………………....100 6.0 References………………………………………………………………………...100 IV-50
1.0 INTRODUCTION Microelectronic and photonic systems in the natural space environment are bombarded by a variety of charged particles including electrons, trapped protons, cosmic rays, and solar particles (protons and other heavy ions). These incident particles cause both ionizing and non-ionizing effects when traversing a device, and the effects can be either transient or permanent. The vast majority of the kinetic energy of an incident proton is lost to ionization, creating the single event effects (SEEs) and total ionizing dose (TID) effects described in section IVA. However, the small portion of energy lost in non-ionizing processes causes atoms to be removed from their lattice sites and form permanent electrically active defects in semiconductor materials. These defects, i.e., “displacement damage,” can significantly degrade device performance. In general, most of the displacement damage effects in the natural space environment can be attributed to protons since they are plentiful and extremely energetic (and therefore not readily shielded against). For this reason, we consider only proton induced displacement damage in this course. (Nevertheless, we identify solar cells as an important example of a case where both electron and proton damage can be important since only very light shielding is feasible.) The interested reader is encouraged to explore the three previous NSREC and RADECS short courses [Srou88a, Summ92, Hopk97] which also treat displacement damage issues for satellite applications. Part A of this segment of the short course introduces the space environment, proton shielding issues, and requirements specifications for proton-rich environments. In order to exercise the displacement damage analysis tools for on-orbit performance predictions, the requirements document must provide the relevant proton spectra in addition to the usual total ionizing dose-depth curves. Ion-solid interactions and the nature of the displacement damage they generate have been studied extensively for over half a century, yet they still remain a subject of investigation. In this section, a description of the mechanisms by which displacement damage is produced will be followed by a summary of the major consequences for device performance in a space environment. Often the degradation of a device parameter can be characterized by a damage factor (measured in a laboratory using monoenergetic protons) that is simply the change in a particular electrical or optical parameter per unit proton fluence. In addition, we will describe the concept of a non-ionizing energy loss rate (NIEL) which quantifies that portion of the energy lost by an incident ion that goes into displacements. It has been calculated as a function of proton energy, and is analogous to (and has the same units as) the linear energy transfer (LET) for ionizing energy. We will discover that, to first order, the calculated NIEL describes the energy dependence of the measured device damage factors. This observation provides the basis for predicting proton induced device degradation in a space environment based on both the calculated NIEL and relatively few laboratory test measurements. The methodology of such on-orbit device performance predictions will be described, as well as the limitations. Several classes of devices for which displacement damage is a significant (if not the dominant) mode of radiation induced degradation will be presented. The examples IV-51
will illustrate various aspects of displacement damage in more detail. We will see, over and over, that the impact of a particular level of damage on device performance is very application-dependent. It will also become clear that uncertainties in the on-orbit prediction for devices sensitive to displacement damage may require significantly increased radiation design margins. All too often, the design engineer is more familiar with basic total ionizing dose (TID) and traditional SEE effects, and may find it difficult to accept the need for proton testing, and especially, any increased radiation design margin associated with uncertainties in displacement damage analyses. There is an increasing demand to employ displacement damage sensitive devices (e.g., charge coupled devices (CCDs), photodetectors, light emitting diodes (LEDs), optocouplers, solar cells, and high precision linear devices) in harsh proton environments (and/or on longer missions). This has led to a renewed interest in hardness assurance techniques for such devices [LaBe98]. It is hoped that this course will provide the understanding necessary for a radiation effects engineer to identify technologies requiring evaluation for possible displacement effects, use the current literature to make first order estimates of device performance, and help ensure that appropriate laboratory radiation testing and analyses are performed. For those readers interested in surveying proton induced device effects (as opposed to performing displacement damage analyses), we recommend reading section 2.2 on displacement effects in devices followed by section 3.3 which includes case studies of those technologies most affected by displacement damage.
2.0 PROTON INDUCED DISPLACEMENT DAMAGE MECHANISMS AND TOOLS In this section, we describe proton displacement effects, on-orbit prediction tools for device performance and laboratory radiation test issues. We begin with a general description of the underlying physical processes that generate displacement damage. The initial production of defects in the semiconductor by incident protons, and the subsequent evolution of this damage to its final stable defect configuration is then described in section 2.1. We discuss the processes by which these defects electrically alter the semiconductor material, and thereby impact device performance in section 2.2. Section 2.3 contains a description of the non-ionizing energy loss rate (NIEL), after which we present the first order correlation between NIEL and device degradation that is experimentally observed. We identify the implications of this correlation in terms of the basic damage mechanisms described in section 2.1, and provide the basis for understanding the limitations of the correlation in section 2.3.2. The NIEL concept enables comparison of the displacement damage produced by protons of different energies (or a spectrum of proton energies) via the calculation of displacement damage equivalent fluences (section 2.3.3), or the “displacement damage dose” (section 2.3.4). This is analogous to the calculation of total ionizing dose based on the proton fluence and LET [see section IVA, equation 1]. Using these tools we establish a methodology for on-orbit device performance predictions in section 2.4. Figure 1 summarizes the method used to predict the on-orbit device (or circuit) response to
IV-52
displacement damage. Note that some devices may have significant concurrent total ionizing dose effects that must also be considered.
Irradiation of Device with Monoenergetic Protons
Calculation of Incident Proton Spectrum for Given Mission
Measurement of Parametric Degradation
Calculation of Spectrum at Device Location Behind Shielding
Calculation of Device Degradation versus DD Dose*
NIEL
Calculation of DD Dose* for Given Mission
Prediction of On-Orbit Device Performance * DD Dose is displacement damage dose. Alternatively one may substitute the displacement damage equivalent fluence for a selected proton energy.
Figure 1 Block diagram of the generic methodology for performing a on-orbit predictions of device performance when device degradation is dominated by displacement damage effects.
2.1 Displacement Damage Mechanisms and Defect Formation As indicated above, the interaction between a charged particle (such as a proton) and a solid cause both ionizing and non-ionizing effects. Most of the kinetic energy of an incident proton is lost in interactions with atoms in the semiconductor that transfer energy to the electron clouds causing excitation or ionization. However, a very small fraction (< 0.1%) of the energy loss causes the atoms to be displaced from their equilibrium sites, and can lead to lattice disorder. An incident proton may collide with a semiconductor nucleus and displace it from its site producing a primary knock-on atom (PKA). If sufficiently energetic, the PKA displaces more atoms, and the collision cascade proceeds until the magnitude of energy transferred becomes less than the threshold required for displacements. At a given incident proton energy, the recoil atoms can vary in kinetic energy from near zero up to some maximum determined by collision mechanisms. Both
IV-53
the average recoil energy and the shape of the recoil spectrum depend on the energy, mass, and charge of the incident particle and the mass of the target. EXITING PARTICLE
INCIDENT PARTICLE
STABLE DEFECT
Interstitial Vacancy Dopant of Impurity Atom
Figure 2a Cartoon showing the displacement of an atom from its lattice site by an incoming proton, thereby forming a vacancy-interstitial (Frenkel) pair. Surviving vacancies migrate through the lattice and often form electrically active stable defects in conjunction with an impurity or dopant atom.
Regardless of whether an atom is displaced as a part of a damage cascade or as an isolated lower energy PKA, most of the initial vacancy-interstitial pairs recombine and no permanent damage results. The interstitial Si atoms do not form electrically active defects. However, the vacancies that escape recombination migrate through the lattice and ultimately form relatively long-lived and immobile defects. Figure 2a is a cartoon illustrating how the initial formation of a Frenkel pair, which is unstable, ultimately results in the formation of a stable defect. These defects have energy levels within the bandgap of the semiconductor. For example, in Si, two vacancies may combine to form a divacancy that is stable up to about 300 °C, or a vacancy and a phosphorous (or oxygen) atom may form an E center (or A center) which is stable up to about 150 °C (or 350 °C), respectively [e.g., Watk64, Walk73 and Kime79]. The vacancy itself is mobile even at liquid nitrogen temperatures, so it is not practical to attempt to prevent the formation of these defects. The process during which the initial vacancy-interstitial pairs evolve into stable room temperature defects results in the so-called “short term annealing effects” in Si devices, and is usually complete within about a second [Srou70, Hein83, Gove84, Mess86]. Figure 2b is a qualitative pictorial showing the time evolution of the number of surviving defects. Note that the stable damage produced in a space environment is very dilute. Longer term room temperature annealing is often observed over a period of IV-54
days or weeks, but it is generally a small effect. For this reason, displacement damage is considered to be a “permanent effect.” Vacancy-Interstitial Recombination
Number of Defects
Vacancy Migration
Stable Defect Formation
Time
Figure 2b Illustration the time evolution of the initial vacancy-interstitial pairs to the formation of stable defects. The annealing of Frenkel defects occurs in less than 1 millisecond and stable defects are formed on the time scale of seconds.
Displacement Damage Processes in Si PROTON ENERGY
6-10 MeV
> 20 MeV
Log N FREE DEFECTS, Coulomb
SINGLE CASCADE, Nuclear Elastic
MANY SUB CASCADES, Nuclear Reactions RECOIL ENERGY
1-2 keV
12-20 keV
Figure 3 Pictorial relating the initial defect configuration to the primary knock-on atom (PKA) energy in Si material. Note from the plot of the number of interactions (N) versus incident proton energy that most interactions are Coulomb events producing isolated defects. For recoil energies IV-55
above a couple of keV, the overall damage structure is relatively unchanged due to the formation of cascades and subcascades. After [Wood81].
The final configuration of electrically active defects formed by particle irradiation has been a topic of much research, but is still not well understood. As we will see this issue is at the heart of understanding the use and limitations of calculated non-ionizing energy loss rate (NIEL) damage functions to predict the displacement damage response of a device in a proton environment. Figure 3 is a pictorial of the spatial distribution of the initial vacancy-interstitial pairs in Si investigated using the Monte Carlo code MARLOWE [More82]. As can be seen from the plot of the log of the number of interactions (Log N) versus the incident proton energy, most events are Coulomb interactions which produce PKAs with Ethreshold < E < ~2 keV, and result in isolated defects. Although there are many fewer of the nuclear elastic and inelastic reaction events that produce cascades, these events are far more damaging, and can contribute a significant fraction of the total displacement damage at higher proton energies. As indicated in the figure, recoils with energies between about 2-10 keV produce single subcascades, whereas those with energies in excess of 12-20 keV form a tree-like structure with branches containing multiple subcascades. Similar results were obtained for Si by Mueller et al. who also investigated the defect structure near the end of the recoil track. The term “terminal cluster” has been used to describe the damaged region where the recoil ion loses the last 5-10 keV of energy and has the highest elastic scattering cross section [Muel82]. They found that a single cascade is likely to have 2-3 terminal clusters with a characteristic dimension of 5 nm, connected to each other by a string of dilute displacements. (Note that this size is an upper limit since the calculation does not include the initial vacancy-interstitial recombination.) This result is consistent with transmission electron microscopy measurements [Lars78, Nara81] of 1 MeV, 14 MeV and fission neutron-irradiated Si that have found an average size of 4 nm for the damage. It is clear that the early terminal cluster models based on heavily damaged regions extending for 200 nm [VanL80, and references therein] are not supported by more recent work. Unfortunately, the early cluster models derived support from electron microscopy [Bert68] work that later was shown to be compromised by faulty etching techniques [Nara88]. We also note that electrical measurements on irradiated devices performed in the last decade or so are also inconsistent with the early cluster models. The interested reader may refer to the literature for details [e.g., Summ87, Peas87, Dale88].
2.2 Displacement Damage Effects in Materials and Devices The net electrical activity of a given defect with an energy level (Et) in the bandgap is ultimately produced by five basic processes as illustrated in figure 4: (1) the generation of electron-hole pairs, (2) the recombination of electron-hole pairs, (3) carrier trapping, (4) the compensation of donors or acceptors, and (5) the tunneling of carriers.
IV-56
C ond uctio n Ban d
E DO N O R
EC
Et
EV
Tu nnelin g
Vale nce B and Generation R ecom b inatio n Trapping C om p ensation
Figure 4 Schematic of the electrical effects that may occur due to the presence of radiation induced defect levels in the band gap of a semiconductor. After [Gove84].
Physically, electron-hole pair generation occurs by the thermal excitation of an electron from the valance band to the defect level followed by its emission to the conduction band. Midgap energy levels in a depletion region are most effective at generating dark current in a device via this process. Recombination occurs when a carrier of one sign is captured at a defect, and not re-emitted before a carrier of the opposite sign is also captured. The energy may be released in the form of light (radiative recombination), or in the form of phonons (i.e., lattice vibrations) which is termed non-radiative recombination. The minority carrier lifetime, which is a key parameter in device performance, is determined by the recombination rate [e.g., Schr82]. Carrier trapping refers to the process whereby a carrier is captured at a defect and then released to its original band. In the case of CCDs, signal charge may be trapped only to be released after the signal packet has already passed causing the charge transfer efficiency of the device to degrade [Mohs74]. Filled traps with a net charge are more effective scattering centers thereby reducing carrier mobility. Carrier removal results when a majority carrier is trapped. Compensation is also responsible for carrier removal. As seen in the figure (for n-type material), the free electrons provided by the shallow donor levels are compensated by deep lying acceptor levels thereby reducing the net carrier concentration. For example, the resistance in a lightly doped collector of a bipolar transistor can increase as a result of this type of carrier removal. Finally, defect levels can assist tunneling through a potential barrier in the bandgap. This effect can produce increased current in a reverse biased junction, and is most significant in materials with small bandgaps and high electric fields.
IV-57
The most important material parameters for the practical operation of most semiconductor devices are the minority carrier lifetime, the generation lifetime [Schr82], the majority carrier concentration and the majority carrier mobility. Typically, the semiconductor material quality is high so that there is at least an order of magnitude smaller density of recombination and generation centers as compared to the majority carrier concentration. As a result, the proton induced introduction of defects (i.e., recombination centers) will impact the minority carrier lifetime well before there is a noticeable reduction in carrier concentration. The same is true for defects produced in depletion regions that can act to decrease the generation lifetime. Mobility degradation is not generally an issue except at very high displacement damage levels. Hence, devices whose primary characteristics depend on minority carrier or generation lifetimes will be most sensitive to displacement damage. Radiation induced degradation in the carrier lifetime, carrier concentration and mobility in turn impact device characteristics such as transistor gain, transconductance and saturation voltage, dark current, detector responsivity, etc. As just described, the reduction of the minority carrier lifetime is a principal cause of degradation in a number of device types. Examples include gain reduction in bipolar transistors and silicon controlled rectifiers (SCRs), reduced responsivity in photodiodes and Schottky-barrier diodes, decreased solar cell efficiency, etc. Devices with lightly doped active regions are most susceptible to degradation caused by carrier removal. Semiconductor light sources such as lasers and LEDs are generally relatively radiation hard since the carrier lifetimes in the active device regions are very short. However, amphoterically-doped LEDs, employed in some optocouplers, are a notable exception and are quite sensitive to proton induced displacement damage for reasons that are not completely understood. Displacement damage effects do not limit the performance of most MOS devices, which depend on majority carrier transport. Exceptions include optoelectronic device types such as the charge injection device (CID) and charge coupled device (CCD), which are extremely sensitive to displacement damage. CCDs are subject to dark current increases resulting from decreased generation lifetime, and from charge transfer efficiency (CTE) degradation due to carrier trapping. JFET and MESFET technologies, being majority carrier devices, are generally very robust to displacement damage [e.g., Hash94] although their transconductance may be degraded by carrier removal at high proton exposure levels. Table 1 summarizes the relative importance (primary or secondary) of displacement damage in many common device technologies [after Srou88a]. Radiation effects experience over the last 20 years has led to a general understanding of device type sensitivities and degradation modes in response to displacement damage. Summaries of these efforts may be found in general radiation effects texts [e.g., Mess86, Holm93] and in a number of summary papers [e.g., Gove84, Srou88b, Raym87]. The case studies to be considered in this course will also provide brief descriptions of displacement damage effects in selected device types.
IV-58
Table 1. Displacement Damage Mechanisms for Various Technologies1 Component
Lifetime Degradation
Carrier Removal
Trapping
Si MOS Transistors & ICs
Mobility Degradation S
Charged Coupled Devices
P
Si Bipolar Transistors & Linear ICs
P
Photodetectors
P
LEDs & Laser Diodes
P
pn Junctions
P
P S
P
P
JFETs
P
P
GaAs Transistors & ICs
P
S S
P = Primary; S = Secondary 1
After [Srou88a]. Note that TID and SEEs also can be primary radiation concerns for these technologies.
2.3Non-Ionizing Energy Loss Rate (NIEL) Concept As we will see in the next section, it has been shown that the radiation response of many devices can be predicted reasonably well based on calculations of the amount of displacement damage energy imparted to the primary knock-on atoms. The non-ionizing energy loss rate (NIEL) can be calculated analytically from first principles based on differential cross sections and interaction kinematics. NIEL is that part of the energy introduced via both Coulomb (elastic), nuclear elastic, and nuclear inelastic interactions, which produces the initial vacancy-interstitial pairs and phonons (e.g., vibrational energy). NIEL can be calculated using the following analytic expression that sums the elastic and inelastic contributions as: NIEL = (N/A) [σeTe + σiTi].
(1)
The σ’s are total cross sections, the T’s are effective average recoil energies corrected for ionization loss using the Lindhard theory [Lind63], N is Avogadro’s number, and A is the gram atomic weight of the target material. In the case of compounds, the total NIEL is derived as a superposition (weighted by mole fraction) of the contributions for each atomic component [Zeig84]. Notice that the units of NIEL, (keVcm2/g), are the same as IV-59
those for stopping power (or LET) describing energy transfer by ionization and excitation per unit length. Burke has calculated NIEL in silicon for protons and other ions over a broad energy range [Burk86]. More recent calculations by Burke have incorporated improvements in the treatment of the nuclear elastic and inelastic reactions, and the Lindhard correction has been applied to the differential recoil spectrum instead of to the average recoil energy of the target atoms. The more accurate calculation is given by NIEL = N A ∫ L[T ( Θ)]T ( Θ) [dσ dΩ]dΩ
(2)
where dσ/dΩ is the differential cross section for a recoil in direction Θ, T(Θ) is the recoil energy, and L[T(Θ)] is the fraction of the recoil energy that goes into displacements [Lind63]. In the case of Si, the maximum amount of displacement damage energy is about 300 keV, regardless of the energy of the recoiling atom. The maximum damage energy increases with atomic number, and is about 2 MeV for GaAs. Figure 5 shows both the LET and NIEL for Si as a function of incident proton energy. Burke has calculated the proton NIEL for a variety of other materials. The most recent published NIEL calculations can be found in the December IEEE Transactions of Nuclear Science cited as follows: InGaAs [Mars92], GaAs and InP [Summ93], and Si [Dale94].
2
ENERGY LOSS RATE (MeVcm /g)
1000
100
Ionizing 10
1
0.1
Non-Ionizing 0.01
0.001 1
10
100
1000
PROTON ENERGY (MeV) Figure 5 Comparison of the energy loss rate through ionization and excitation of the Si lattice (LET), and through atomic displacements (NIEL) over a wide range of proton energies. The LET was calculated as in [Zeig85], and NIEL as in [Dale94].
IV-60
The nature of displacement damage as a function of proton energy is governed by the interaction cross sections, and the non-ionizing energy of the PKAs as governed by the Lindhard function. For proton energies below about 10 MeV, Coulomb elastic scattering is by far dominant in Si, and produces atomic recoils with non-ionizing energies in the hundreds of eV. At higher energies, the bend in the curve occurs because nuclear elastic scattering becomes more important resulting in recoils with non-ionizing energies in the tenths of MeV range. As the incident proton energy increases the elastic cross section decreases athough it is still larger than the inelastic cross section. By about 100 MeV half of the non-ionizing energy imparted to the Si lattice is from nuclear inelastic reactions with a mean PKA non-ionizing energy that is still about 0.1 MeV (due to the Lindhard partition). NIEL has also been calculated by other means including Monte Carlo programs such as HETC [Alur91], CUPID [McNu81, McNu94] and TRIM [Zeig84]. A comparison between the most recent Burke and CUPID calculations of Si NIEL is discussed in [Dale94]. Although HETC, CUPID and Burke’s calculations of the recoil distributions as a function of incident proton energy show similar trends, they differ in details [Dale94]. The TRIM program only includes the Coulombic interactions, so it is not appropriate to use it directly for damage calculations for proton energies above about 8 MeV or so, depending on the target material. Note that all of the above calculations include a “fudge factor” that accounts for the fact the most of the initially produced vacancy-interstitial pairs recombine and therefore do not produce electrically active defects. For example TRIM is often executed assuming a displacement energy threshold of 25 eV, which is considerably higher than the actual value. This practice helps to account for the efficiency of the initial recombination of the vacancy-interstitial pairs. In other Monte Carlo codes such as MARLOWE, one also has the option to define a radius around each collision point for which all the vacancy-interstitial pairs recombine. In essence, all current NIEL calculations must be scaled to fit the experimental damage factors, unless damage factor ratios are compared. As we shall see, it is the calculation of the energy dependence that is relevant, not the absolute values of NIEL. 2.3.1 The Correlation of NIEL to Device Behavior Device degradation in a radiation environment is often characterized by defining a damage constant, or a damage factor. Damage constants describe the change in basic material parameters such as minority carrier lifetime or diffusion lengths, produced by a given fluence of protons of a specific energy. (Fluence is defined as the number of incident particles per unit target area, and has units of cm-2.) Damage factors are similar except they characterize the observed radiation induced degradation of device or system parameters that may not be readily reduced to basic material parameters because a detailed device model is not available.
IV-61
The following well known equation describes the degradation in minority carrier devices that results from the reduction in the diffusion length that accompanies the introduction of radiation induced defect recombination centers: 1/L2=1/Lo2 + KΦ.
(3)
The initial and post-radiation diffusion length is given by L0 and L, respectively, K is the damage constant, and Φ is the proton fluence. (Sometimes this equation appears in terms of the minority carrier lifetime, τ, using the relation, L = (Dτ)0.5, where D is the diffusion coefficient.) Usually the satellite designer or test engineer is interested in a particular device parameter, and defines a relevant damage factor. In the case studies to follow, we will see examples of other useful device damage factors such as the CCD CTE damage factor, the dark current damage factor, the solar cell efficiency damage factor, and so on. In each case, the device parameter in question changes linearly with fluence, or else is defined in the linear region. Note that parameters such as inverse bipolar transistor gain, detector responsivity, and CCD dark current, which have a well-defined regime with a linear response, may also exhibit a nonlinear response at very low or very high proton fluences. (We will also see examples where a device parameter of interest such as LED light output or optocoupler current transfer ratio does not behave linearly in the proton fluence regime of interest.) Bipolar transistor gain measurements for a variety of incident particles (as a function of energy) have been performed in order to determine whether the NIEL function can be used both to predict the energy dependence of the device damage factor and to correlate the degradation due to different particles [Summ87]. In principle, such a correlation also provides the basis for on-orbit performance predictions based on the measurement of a damage factor at a single proton energy. Likewise, if neutron data already exists, the correlation can be used to predict the device response to protons. In this work, the well-known Messenger-Spratt equation is used to describe the radiation response of the common emitter DC gain, hFE, of a bipolar transistor: 1/hFE = 1/hFEO + K(E)Φ
(4)
where 1/hFEO is the initial reciprocal gain, K(E) is the particle and energy dependent displacement damage factor, and Φ is the incident particle fluence. The transistor gain (given by the ratio of the collector to base currents) decreases with increasing proton fluence primarily as a result of the decreased minority carrier lifetime in the base region. A more detailed description may be found in [Mess86]. The damage factor is determined experimentally by performing device gain measurements (for a particular set of device operating conditions) after incremental exposures at a given proton energy. Figure 6 shows the measured damage factors for protons, deuterons and helium ions normalized to the 1 MeV-equivalent (Si) neutron damage factors as a function of ion energy for a variety of Si bipolar transistors. (We will discuss the meaning of MeV equivalence in a later section, and neutron damage equivalence is explored in [Luer87]. For the present purposes we note that by comparing ratios of measured damage factor to IV-62
the calculated NIEL ratios, no scaling parameter is needed to match data with theory.) The importance of this result is the proportionality between the measured damage factors and calculated NIEL that provides the basis of the on-orbit predictions of device degradation produced by displacements.
Figure 6 The transistor damage factor ratios for a variety of particles with fission neutrons are shown together with the corresponding calculations of the NIEL ratios. Note that both ordinates are identical (with no fitted parameters), which indicates a direct proportionality between NIEL and the damage factors over a wide energy range. After [Summ87].
Research performed in the last dozen years has shown that, to first order, the linear relationship between the device degradation from particle-induced displacement damage and NIEL holds for a variety of electrical parameters, incident particles, and device materials [Summ87, Peas87, Dale88, Mars89a, Walt91, Mars92, Ohya96, etc.]. This is a surprising result when we consider that NIEL calculations describe the energy deposited into the formation of Frenkel pairs (over 90% of which recombine), and do not consider the process by which the stable electrically active defects are formed. Since NIEL is a direct measure of the initial number of vacancy-interstitial pairs created, the implications of the NIEL correlation with device degradation are that: (1) the percentage of initial vacancy-interstitial pairs that survive recombination is independent of the PKA energy, and (2) the resulting stable defects have the same device effect regardless of whether they evolved from a vacancy-interstitial pair originating in a subcascade or as a well-separated pair [Dale88]. In addition, given that various stable defects have quite IV-63
different electrical properties, the correlation also implies that the defect inventory produced is independent of PKA spectrum. Nevertheless, the degree to which the NIEL correlation holds is qualitatively consistent with the Monte Carlo calculations described earlier. These simulations show that a higher energy PKA will produce more overall damage, but that the microscopic nature of the damage is not drastically different. The branching process simply creates more and more subcascades, each separated by a string of relatively isolated defects. It is still important to keep in mind that, although defects produced from isolated vacancy-interstitial pairs (such as those produced by gamma rays and 1 MeV electrons) may have similar electrical characteristics to those produced by heavier particles such as protons and neutrons, there are important differences. These differences are not restricted to short term annealing effects, and also manifest themselves in the long term behavior of a device. For example, E-centers (vacancy-phosphorus defects) produced by 1 MeV electrons anneal at a significantly lower temperature than those produced by protons [Walk73, Kime79], a relevant (and unfortunate) fact for charge coupled device (CCD) engineers who have considered on-orbit warm-ups to mitigate charge transfer efficiency degradation in CCDs [Holl91a]. Differences in the operation of SiGe transistors [Rold98] and AlGaAs/GaAs solar cells [Barn84] have been attributed to differences in the defects produced by neutrons versus protons. Very well controlled deep level transient spectroscopy studies [Eise92, Mind76] have unequivocally demonstrated that, although 1 MeV electrons and protons produce some of the same defects in n-GaAs, there are also different defects produced by each particle. The bottom line for the satellite designer working a mission in a proton environment is that devices that are highly sensitive to displacement damage should be radiation tested with protons. We will see other reasons for this recommendation later in the short course. 2.3.2
Limitations in Usage of NIEL
The NIEL calculation is a useful tool to approximate the expected proton induced radiation response in a space environment, but it is necessary to appreciate the underlying assumptions and limitations in order to use it effectively. Deviations at very low proton energies (approaching the displacement energy thresholds) are expected [Dale 88, Summ93], but they are not generally of concern for proton applications in space because they contribute little to the total displacement damage behind typical shielding, as will be shown in section 2.4. However, indications of other systematic deviations from the NIEL correlation have been observed in Si device measurements (e.g., for several CCDs, a CID, a 2N2907 bipolar transistor [Dale88]), and also in GaAs measurements (e.g., an LED [Barr95], a laser diode [Zhao97], solar cells [Walt99], and a JFET [Summ88]). Depending on how the damage factor measurements were normalized to NIEL, the deviations have been reported either as the damage factors being over-estimated by NIEL at higher energies, or equivalently, being underestimated by NIEL at the lower energies. The choice of a damage function (i.e., the energy dependence given by the calculated NIEL or experimental damage factors) has been shown to be significant. For example, one study found a factor of two difference in the on-orbit predictions of the degradation in Si CCD performance depending on which damage function is employed [Dale91]. IV-64
Figure 7 Transistor damage factors and dark current damage factors for protons (normalized to fission neutron damage factors) versus NIEL. The lower line (with a slope of one) indicates a linear relationship between the damage factor ratios and NIEL. The deviations from linearity are indicated with the upper line. [After Dale89b]. A similar figure in [Dale88] also shows deviations for transistor damage factors measured for electrons.
Deviations from the linear dependence of Si displacement damage factors with the NIEL energy dependence are shown in figure 7, which shows the proton to neutron damage factor ratios for several devices plotted as a function of NIEL [Dale88]. The damage factors represent changes in the minority carrier lifetime in the case of the transistor data, and the generation lifetime in the case of the CID and CCD dark current damage factors. A slope of one on the log-log plot indicates a linear relationship, and the observed deviation from linearity is noted by the top curve. Dale et al. defined a “damage enhancement factor” as the ratio of observed damage factor ratio (upper line) to that expected based on the linearity with NIEL (lower line). In this work, the PKA spectrum produced in Si by the various incoming particles was calculated. Note that the PKA spectrum varies significantly over the range of proton energies of interest in space. It may come as a surprise that the PKA spectrum of a 60 MeV electron is more like that of a 10 MeV proton, than a 10 MeV proton is like a 60 MeV proton. As seen in figure 8, the damage enhancement factor is found to correlate with the fraction of the total NIEL due to PKAs with energies less than 1 keV. It is notable that the result held for the wide range of PKA spectra produced by 4.1 MeV electrons, all the way to 1 MeV-equivalent neutrons that produce very high energy recoils. The observed deviations from linearity would be expected if there were less recombination of initial vacancy-interstitial pairs IV-65
1.0
25 eV 1 keV 1 keV 2 keV
< < <