Single-Event Effects Ground Testing and On-Orbit Rate Prediction Methods: The Past, Present and Future

18 Pages·2003·1.5 MB·English

by  Reed, Robert A.

Single-Event Effects Ground Testing and On-Orbit Rate
Prediction Methods: The Past, Present and Future
Robert A. Reed, Member, IEEE, Jim Kinnison, Member, IEEE, Jim Pickel, Fellow, IEEE,
Stephen Buchner, Member, IEEE, Paul W.M arshall, Member, IEEE,
Scott Kniffin, Member, IEEE, Kenneth A. LaBel, Member, IEEE
Abstract Over the past 27 years, or so, increased concern (sometimes call the “single event alphabet soup”). The
over single event effects in spacecraft systems has resulted most studied are Single-Event Upset (SEU), Single
im research, development and engineering activities Event Latchup (SEL), Single Event Gate Rupture
centered around a better understanding of the space
(SECiR), and Single-Event Transients (SET). Another
radiation environment, single event effects predictive
paper in this Special Issue by Paul Dodd [1 J provided a
methods, ground test protocols, and test facility
review of the physical mechanisms for SEES and gave
dlevelopments. This research has led to fairly well
the definition of most SEEs that occur in modern
dleveloped methods for assessing the impact of the space
radiation environment on systems that contain SEE technologies, including those listed above.
sensitive devices and the development of mitigation In April of 1996 several authors published an IEEE
strategies either at the system or device level. Transactions on Nuclear Science Special Issue on
“Single-Event Effects and the Space Radiation
1. INTRODUCTION Environment”. That Special Issue covered topics that
As an ion passes through a semiconductor it generates includes SEE rate predictions approaches, test facilities
dectron-hole pairs, this process is known as direct available at the time, test issues for various
ionization. The charges either recombine or propagate technologies, as well as the components of the space
through a semiconductor via drift or diffusion. Because radiation environment that must be considered when
the function of active microelectronic or photonic evaluating SEEs in a device.
component is governed by the controlled injection of In this paper we summarize the concerns and issues
charge into the depletion layers of pnj unctions, the for modern devices by providing an historical account
uncontrolled charge injection resulting from ionization of the early days of SEE testing and space observation,
can produce an array of effects on the device operation. an overview of the traditional assumptions used to
’These effects are known as Single-Event Effects develop SEE test approaches, a listing of the SEE test
(SEES). facilities available today, a review of SEE rate
SEE ground-based testing is done to characterize how prediction approaches, and finally a listing of some of
;I microelectronic device responds to a single particle the observed phenomena that sever as a reminder that
r:neutron, proton or other heavy ion) interaction with the the traditional methods may not be applicable to all
atonis that makeup the semiconductor. On-orbit rate modern day technologies.
.?redictions methods have been developed that use the The works referenced in this paper-and many others
ground test characterization along with the space that have been published in the IEEE Transactions on
radiation environment definition to estimate the Nuclear Science (TNS), the Proceedings of the
frequency of occurrence for a specific SEE. Radiations Effects in Components and Systems
There is a long list of various types of SEEs (RADECS) Conference, the IEEE Nuclear and Space
Radiation Effects Conference’s Radiation Effects Data
Manuscript received March 27, 2003. ‘This work was
Workshop record, and RADECS Radiation Effects
supported in part by the NASA NEPPERC Project and the
Defense Threat Reduction Agency. Workshoppresent testing methodologies and rate
R.A. Reed, K.A. LaBel are with the NASA Goddard Space prediction techniques that deal with the issues raised
Flight Center, Greenbelt, MD 20771.
hem and that are successful in providing the data needed
J. Kinnison is with JHU/Applied Physcs Lab, Laurel, MD
20723. to develop event rate estimates for space application
J. Pickel is with PR&T, Inc., Fallbrook, CA 92028. design.
S. Buchner is a NASA/GSFC support contractor with QSS II. SEEs CIRCAI 975-I980
Inc. Seabrook, MD 20706
R.esearch on SEES in microcircuits began as most
P.W. Marshall, is a NASA/GSFC Consultant to QSS. Inc.
Seabrook, MD 20706 radiation effects research does; in 1975 an anomaly
S. Kniffin is a NASAlGSFC support contractor with Raytheon occiirred on a earth orbiting spacecraft that could not
ITSS, Greenbelt, MD 20770
readily be explained by from known phenomenon. (The
possibility of cosmic ray induced SEU in microcircuits ionization by reaction products from a nuclear collision.
was predicted by Wallmark et. a1 [2] in 1962.) The first Typically, effects are dominated by direct ionization for
work detailing this new phenomenon was published in a ions with Z > 1 (know in the radiation effects
paper in 1975 by Binder, Smith and Holman of the community as “heavy ions”). For neutrons and protons
Hughes Corporation [3]. We quoted them here: the effects are typically dominated by indirect
ionization. Because of the different mechanism
“Anomalies in communication satellite operation involved, the methods used to determine rate
have been caused by unexpected triggering of predictions-and thereby different test methods-are
digital circuits.. . The purpose of this paper is to very different.
investigate interactions with galactic cosmic rays SEUs were not the only topic of discussion in the
as an additional mechanism.” early days of SEE. The discoveiy of heavy ion-induced
SEL was first published in 1979 by Kolasinski, Blake,
The authors were able to show that the anomaly was Anthony, Price and Smith 191. The authors reported on
due to Single Event Upsets (SEUs) in a digital flipflop SEL induced in SRAMs by heavy ion nuclei.
circuit. They developed a rate prediction approach based By 1980 the combined impact of all of these papers
on device and transistor parameters, charge collection was significant enough to motivate researchers, project
efficiency, and the solid angle and energy spectrum of managers, and design engineers to pay attention to this
impinging cosmic rays. The calculation was within a new phenomenon of radiation-induced effects. Shortly
factor of 2 of the observed rate. While this work after these early papers were published a sequence of
marked the start of a new era of radiation effects, the symposium devoted to SEEs was held. This meeting-
radiation effects community largely overlooked it. The !jingle Event Effects Symposium, today a biennial
A second paper, motivated by yet another spacecraft event-was (and still is) critical to the development of
anomaly, by Pickel and Blandford [4], was published in the current understanding of SEE.
1978 that details the development of a heavy ion-induce It is interesting to note that the early work detailing
rate prediction model that utilizes the concept of the analysis and rate prediction approaches have proven
describing the space radiation environment as a Linear to be veiy robust even when applied to most modem
Einergy Transfer (LET) distribution, or Heinrich day technologies. However there are certain cases
ciistribution [5]. They compared their calculation of the where these analyses fail to predict the device response.
SEU rate for a NMOS dynamic RAM to on-orbit results
and found agreement to within a factor of two. The 111. TRADrrlONAL ASSUMPTIOUNSSE D TO DEVELOSPE E
observed error rate for the system was near one error per TEST APPROACHES
day, significant enough result to catch the attention of
A. Introduction
many of the radiation effects experts of that time.
In this section we discuss some of the assumptions
Another paper published in 1979 by May and Woods
and inethods used to perform SEE testing (Section VI
(61 detailed the first reported alpha particle induced
describes cases where these fundamental assumptions
SEUs. The alpha particles were emitted from trace
are shown to be inadequate.) First we summarize the
amounts of uranium and thorium present in the
space environment, then we discuss proton and heavy
packaging materials.
ion SEE ground testing, and finally we discuss system
Also in 1979, two independent research groups
level testing implications.
uncovered the fact that the recoil products from a
proton-induced nuclear spallation reaction could have B. SEE Space Environment
sufficient LET to cause an SEU. Wyatt, McNulty,
The space charged particle environment responsible
’Toumbas, Rothwell, and Filz reported on ground test
for single event effects is dominated in particle count by
results on two types of 4k DRAMs [7]. They observed
energetic protons, with smaller contribution from
upsets occurring for protons energies ranging from 18 to
heavier ions (Z>1). However, various sources generate
130 MeV. At the same time Guenzer, Wolicki and
these particles, and the characteristics of the
Allas [8] studied and reported on proton and neutron- environment vary in distinct regions of space. The
induced effects. They observed SEUs in 16k DRAMs
environment is traditionally divided into three parts-
for neutron energies that ranged from 6.5 to 14 MeV galactic cosmic rays, particles from solar events, and
and for 35 MeV protons.
particles trapped in planetary magnetospheres.
There are veiy distinct differences in SEE testing and
Galactic cosmic rays (GCRs) are highly energetic
rate prediction approaches between SEEs induced by
ions (Z?l) that arise from sources outside the solar
direct ionization from the primary particle and indirect
system, and are generally considered to form a spatially
constant background modulated by the changes in the parameters that are used to determine the on-orbit event
solar magnetic field during the solar cycle and by the rate.
presence of planetary magnetospheres. The fundamental assumption associated with heavy
Solar events produce a range of energetic ions, but ion SEE testing is that the cross-section only depends on
the maximum energy of these particles is much lower the “effective LET” of the incident particle, that is, the
than for GCRs. Correspondingly, the solar ion nominal LET of the particle divided by the cosine of the
environment is significantly modified by spacecraft incident angle-where the angle is that from the normal
shielding, by planetary magnetic fields, and by phasing to the die surface. Division by cosine conies from the
within the solar cycle. Also, solar events are of short fact that the pathlength of the ion through the sensitive
duration, so the solar ion environment consists of a volunie increases with the angle of incidence. This
sequence of impulsive bursts of ions that can increased pathlength gives rise to more charge being
dramatically raise the single event effect rate for a short generated in the sensitive volume.
time. The end result of an SEE test is a measure of the
Charged particles can be trapped in planetary cross-section as a function of effective LET. The cross-
magnetospheres, and spacecraft in these fields will section usually takes the form of a curve with onset of
experience single event effects at rates that strongly SEE at some threshold LET which then rises to an
depend on the details of the orbit. The most important asymptotic value at higher LET. The critical charge is
example of this environment is the Earth’s Van Allen determined fi-om the threshold LET, while the
belts; all low Earth orbiting spacecraft must take into asymptotic cross-section gives the area of the sensitive
account the presence of the trapped proton belt, volume. When combined with the thickness of the
including deviations from a dipole model such as the sensitive volume-typically derived from the
South Atlantic Anomaly (SAA). Another important archkecture of the device-the parameters derived from
example is the environment of ions such as sulfur the cross-section are sufficient to allow calculation of
trapped in the Jovian magentosphere. SEE rates for inany technologies in any given space
The space environment is modified by shielding environment where 2 > 1.
associated with the structure of the spacecraft around a A typical test consists of a series of niono-energetic
device in orbit as well as the packaging of the device exposures for beams over a range of LETs (or effective
itself. While this effect is small for energetic cosmic LETs.). During each exposure, the device is placed
rays, the spectrum of lower energy ions-such as those under bias, either active or passive. Events of interest
produced by solar events-or trapped protons is are counted for a known incident fluence, and the cross-
s:.gnificantly altered by the presence of material around section is given by the ratio of number of events to the
devices, and must be included in rate prediction. Most effective particle fluence. (Where the effective fluence
computer codes used to estimate environments include is the product of the normal incident fluence and the
transport of ions through an assumed thickness of cosine of the angle-this correction is for the reduced
material before calculating the spectrum used for rate effective exposure area of the die surface.) Authors of
estimation. research and test data reports often omitted the word
There has been several Nuclear and Space Radiation “effective”, even when the heavy ion beam is at some
E:ffects Short Courses that give a detailed description of angle relative to the normal to the die.
the charged particle environment [IO], [ 1 I], [ 121, [ 131. Some of the early work was devoted to understanding
Also Janet Barth’s paper in this Special Issue [ 141 gives SEU:; in static RAMS, which are the best example of
a good review of the space radiation environment. many of the assumptions in SEU testing. Each RAM
cell is-to first order-identical, and from an SEE
C. Device Level Testing
perspective, the device is easily seen to be an array of
1.) Heavy Ion SEE Testing
identical sensitive volumes. Tests are usually
The event rate for a given effect in space is
perfcanied by loading a pattern in the menioiy array,
determined by a combination of environment and device
exposing the device to a known fluence of charged
characteristics, which are assumed to be completely
particles at a particular LET. After the exposure, the
tescribed by the geometry of a sensitive volume and a
array is interrogated to count the number of flipped bits,
critical charge associated with the effect in question for
from which the cross-section is calculated. Since each
a given cell within a device. In most cases, a device is
sensitive volume is identical, the per-bit cross-section is
modeled as an array of identical thin right-rectangular
simply the measured cross-section normalized by the
parallelepiped sensitive volumes-we discuss rate number of bits in the memoiy array. A complete
prediction approaches later in this paper. Device level
experiment uses many LET values to fully map the
SEE testing helps to define some of the critical
cross-section of interest. Since the number of different the definition of an error is in question, and devices
tieanis is limited at a test facility, some method of such as field programmable gate arrays [I81 where
changing the LET of the beam-often, non-normal inadvertent rearrangement of &e circuit design while
angles of incidence or degrader foils to change the beam under irradiation causes complications in the operation
e:nergy-is used to provide as many data points as of a device. In each example, methods for determining
needed. Fig. 1 is an example cross-section curve for a cros:;-sections and event rate estimates have been
i’datra 32Kx8 SRAM. developed that allow conservative circuit design in
The cross-section often depends on other factors such spacz applications.
as temperature or electrical bias, or deviates from a Effects other than upset also provide complexity
strict dependence on effective LET. Even in the beyond the basic SEU test. Latchup sensitivity in a
Eimplest cases such as SEU in static RAMS, significant device is a function of LET, but also of operating
deviations from the basic testing assumptions are voltage, temperature and range of the incident particle
observed. For example, if the array of identical, well- in the device. In many devices, care must be taken to
defined sensitive volumes were strictly true, the cross- use beams with sufficient range to deposit charge in
section would be a step function with respect to LET. latchup sensitive volumes deep within the device, which
In reality, the cross-section increases with finite slope in calls into question the concept of effective LET for
the threshold region, followed by a knee region and a latchup.
more gradual approach to the asymptotic cross-section Aka, in devices susceptible to gate rupture or
than seen in a step function. These deviations can be burnout, the goal of a test is not to define the cross-
due to statistical variations in the sensitive volume section as a function of LET, but to measure
geometry or in the critical charge for a volume, and are susceptibility for various device parameter settings.
significant for calculating event rates from cross-section Thes.e data allow the definition of safe operating regions
data (see Section IV). for the device; when an engineer uses the device in the
Another source of deviation in the shape of the cross- ‘‘safe region” the susceptibility to the effect is
siection occurs when more than one sensitive volume is eliminated or greatly reduced [ 191, [20].
found in a cell, or when several different types of cells 2.) Proton SEE Testing
are present-each with their own characteristic sensitive Energetic protons generally do not deposit enough
volumes. energy in a sensitive volume to directly cause SEES.
As a final example, experimenters often find However, approximately one in lo4 to IO6 protons
discrepancies between cross-sections measured at the undergo nuclear reactions with. If produced in or near a
same effective LET with different beams at different sensitive volume, the residual nuclei can deposit enough
incident angles. These discrepancies have been, in part, energy to cause an SEE. The residuals that cause events
attributed to the deviation from the inverse cosine are short-range ions that deposit most or all their energy
relationship between LET and incident angle. ’These within the sensitive volume.
issues have been studied in detail over the last decade, Pxoton testing proceeds much like heavy ion testing
and in each case, methods for dealing with deviations in that the sample is exercised while exposed to a beam
have been developed [ I5 and references therein]. for a given fluence. Events are counted in each
SEU testing can often involve other serious exposure, and the event cross-section is calculated by
complications and deviations to the basic methodology dividing the number of events by the fluence for the
outlined above. For instance, microprocessors contain exposure (recall that the effective fluence is used for
many registers and latches that may or may not be heacy ion testing). This procedure is repeated over a
active at a given time depending on the program range of proton energies to fully characterize the cross-
running on the microprocessor. Therefore, the device section as a function of energy. For the most part,
upset cross-section strongly depends on the software exposures are done in-air.
used during the measurement, and the problem of Three issues, however, make proton testing
actually detecting an error becomes quite coniplex. significantly different from heavy ion testing. First, and
:Early on, Koga, et al, developed several different foremost, samples experience significant total ionizing
methodologies for testing microprocessor devices [ 161. dose damage when exposed to proton beams, and the
These methods are based on comparison between event cross-section can be different as damage
irradiated devices and golden devices or simulated accumulates. Care must be taken to plan experiments so
:golden devices, and are the basis for much of the that the device characteristics are not unduly altered
microprocessor testing today. Other examples include duri:ng the measurements. Second, since the nuclear
,3EU testing of analog-to-digital converters [ 171 where interaction probability does not depend on the beam
incident angle and the reaction products deposit nearly From a modern-day spacecraft developer’s point of
all their energy in the sensitive volume, the cross- view, the overriding principle is that SEEs are a system
section is assumed to only depends on the proton concern and must be understood and managed at the
mergy. As a result, for proton testing, there is no system level. This “management” can be as simple as
equivalent concept to the effective LET used in heavy to select a device that does not exhibit a particular SEE
ion testing. Finally, the physical layout of proton or as complex as error correction and detection schemes
exposure facilities and safety concerns for human that produce impact avoidance or system recovery when
experimenters adds considerable complication to an SEE occurs. Independent of the methods used, the
experiment and equipment design. impzct of SEEs on the system must be understood to
insure reliable spacecraft operation.
D. Spacecraji System Impact on SEE Testing
Single event effects can lead to changes in the
IV. SEE TESTFACILITIES
macroscopic behavior of a spaceflight system. For
instance, changes in microprocessor code in an SRAM A. Introduction
will, at a minimum, cause erroneous execution of the While each single event effect requires special test
program. In the case of SEUs or SETS in a circuit, the considerations, all tests share common components.
s:ysteni event rate is generally not the sum of the The basic SEE test consists of a sample exposed to a
individual device event rates. Instead, the system event series of mono-energetic beams over a range of heavy
rate is deterniined by a complex interplay of the location ion LETs-or over a range of energies for proton
of a given event, the sensitivity of other devices in the tests-while being exercised in some way to look for
circuit, and the timing of the event relative to the the effect of interest. Components of any SEE test are
activity of the system. If the event rate is low, or if the (1) a source of beams with the required characteristics,
interaction between devices is simple, the sum of event (2) a beam monitoring and control system, (3) a
rates is a reasonable approximation to the system event mounting and positioning system to hold the sample in
rate. However, in a few instances, the additive the beam, and (4) a system for biasing the sample and
approximation is inadequate, and the system as a whole measuring its electrical behavior to detect the single
must be tested. event effect. In most facilities of general use in SEE
This testing is generally performed by irradiating testing, the facility provides the beam, beam monitoring
devices individually while operating the system; events and Icontrol, and sample mounting; the user is generally
observed at the outputs of the system are counted as responsible for providing properly prepared samples, an
with a device-level test. In some cases, a device-level electrical system to bias and monitor the sample, any
test cannot be performed outside the context of a interface hardware used to adapt the user’s system to the
system. For example, a microprocessor test is facility equipment, and equipment to provide for special
impractical to perform at the device level. When test needs such as sample thermal control.
embedded in a computer system, however, a
B. Heavy Ion Test Facilities
microprocessor can be readily tested.
There are seven major heavy ion beam facilities-
Devices sensitive to SEE are often used in space
five in the United States and three in Europe-that are
applications. Many mitigation techniques have been
currently available or will be available in the near future
developed over the years-from error detection and
for :SEE testing (others facilities exists, but are not
correction in memories to latchup protection circuits for
considered to be major test facilities at this time). These
individual devices in a system. The combination of
highly capable SEE test facilities are located at
sensitive device and mitigation is a system that must be
laboratories used for basic physics research; generally,
t.ested to verify proper operation-especially when a
basic research will be higher priority at these
destructive effect is mitigated. Mitigation validation is
laboratories than SEE testing. Part of the challenge in
performed by irradiating the sensitive device in a system
providing SEE test facilities will be to negotiate
with mitigation.
sufficient test time to meet the needs of the aerospace
In order to be effective, system-level testing requires
community while not unduly disrupting basic research
:special consideration. The event cross-section strongly
at a laboratory. Each of the facilities described here
depends on the details of the system design and
produces a distinct set of beams, and is a unique set of
function, including any software that might be executed
compromises between cost, ease of use and space
by the system. As a result, the system tests must be as
environment simulation fidelity.
(close to the flight system as practical for reliable test
A major consideration for determining the fidelity of
:results.
a test is the energy of the beam for a given LET; higher
energy beams tend to more accurately reproduce the cockt.liI to extend the LET range of each below I MeV-
effects of the space environment at the expense of cm2/mg. Before passing to the test chamber, beams
higher cost and greater complexity. from the 88-inch Cyclotron are routed through a beam
In 1987, the Single Event Upset Test Facility diagnostic system which is used to collimate and shutter
(SEUTF) was built by a consortium of US government the beam, measure beam characteristics before test runs,
agencies and Brookhaven National Laboratory (BNL) in measlire relevant beam parameters during exposures,
response to increasing demand for SEE data for and allow alignment of samples with the beam. This is
spacecraft hardware design and qualification. Since accornplished with a set of particle detectors and filters,
becoming operational in 1988, the SEUTF has been and a mirror for the alignment laser. All of the elements
available to users including government, academic, and in the: beam diagnostic system except the filter wheel
commercial institutions. The SEUTF consists of a test are mounted on sliding stages and can be inserted or
station attached to the east beam line of the Brookhaven remoded from the beam via software on the SEEF
Tandem Van de Graaff Facility (TVDGF), and is control computer. The main test chamber is a large
maintained and supported by the TVDGF [21]. The vacuum enclosure of about 1 m3 volume surrounding a
TVDGF is a low energy accelerator compared to all 4-axis motion system on which test hardware is
okher test facilities. The maximum energy for the mounted. The Qaxis motion system provides linear
siandard beams is on the order of a few MeV per travel across the beam horizontally and vertically, as
nucleon. Downstream from the beam TVDGF control well as rotation about the beam axis and about the
aid measurement system, a system of five detectors is vertical axis normal to the beam (to change the incident
used to independently monitor the beam just prior to the angle of the beam with respect to the sample surface).
SEUTF test chamber. Four of the detectors are placed The Texas A&M University (TAMU) Cyclotron
evenly around the edge of the beam to measure fluence Institute, jointly hnded by the State of Texas and the
during test runs, while the fifth is mechanically inserted U.S. Department of Energy operates a K500
into the center of the beam between runs. Since the superconducting cyclotron to support research in
TVDGF provides low energy beams, experiments must nuclear physics and chemistry, as well as applied
be performed in a vacuum. The SEUTF chamber is a research in space science, materials science and nuclear
large vacuum chamber attached to the end of the beam- medicine. The Cyclotron Institute has established the
line. The main SEUTF hardware interface is a three- Radizition Effects Facility (REF) as a permanent test
axis goniometer stage driven by absolute-encoded area, and has offered it for use by commercial,
stepper motors. The stage is designed to provide iravel government and educational organizations to study
in all three linear dimensions as well as revolve about single event effects in microelectronic and related
the beamline axis and rotate about the vertical axis of radiation effects research [24]. The Cyclotron Institute
the stage to change the incident angle of the beam is planning a series of upgrades which will link the
relative to the sample surface. The SEUTF is controlled previously-existing 88-inch cyclotron with the K500
through a custom designed software package that cyclotron to expand the availabIe beams and increase
includes the local user beam control and monitoring as the u:jefulness of the facility; one impact of this upgrade
well as control of the sample positioning system and is that more time may be available for SEE testing [25].
data logging for each run. A number of high energy beams have been developed
The 88-inch Cyclotron at Lawrence Berkeley by the Cyclotron Institute as “standard” beams for REF.
National Laboratory (LBNL) has been used for single These beams range in energy from 12.5 - 55 MeV/amu.
event effects testing by experimenters from the Space Many of these beanis can be used without a vacuum
Science Applications Laboratory (SSAL) at the chamber, which greatly simplifies the test equipment
Aerospace Corporation since 1978 - the early days of interface. REF provides beam monitoring and control
tne field of study [223. In 1996, the third generation in a manner similar to the systems used at SEUTF and
Single Event Effects Facility (SEEF) became SEEF. Since the beam LET can be changed with
operational as the latest facility for SEE testing at degrader foils, REF also includes a silicon transmission
LBNL [23]. The cyclotron can develop bearns in detec:tor that can be used to characterize the degraded
cocktails which are mixtures of elements with constant beam. Two systems for hardware interface are available
charge-to-mass ratio. Each element will have the same at REF-a target chamber system for lower energy
energy per nucleon, and so will have different incident beams and an in-air positioning system for higher
LET. Two standard cocktails are available at 4.5 (1 - energy beams. The target chamber is a cylindrical
62 MeV-cni2/mg) and 10 MeV per nucleon (1 - 55 vacuum chamber 76 cni in diameter and 76 cm high.
MeV-cniz/nig). Low LET ions can be added to each Both chambers have autoinated motion control. Target
position verification is performed with a camera co- extensively for SEE testing for more than 15 years [28].
al.igned with the beam axis and a laser that crosses the The facility setup is similar to the BNL SEUTF as is the
bean1 path in the center of the chamber. range of ions and energies. Testing is done in a
The National Superconducting Cyclotron Laboratory vacuum. Dosimetry is similar to the other low energy
(NSCL) at Michigan State University operates a K500 SEE test facilities.
and a K1200 cyclotron for basic nuclear physics The Grand Accelerateur National D'Ions Lourds
research. Over the last five years, beam time has been (GAKIL) is located in Caan, France [29]. A cyclotron
available on a limited basis for single event effects is used to accelerate heavy ions up to 50 MeVIamu.
testing on the K1200, mainly through researchers at SEE testing has been performed there for more than 7
Goddard Space Flight Center. In 1999, construc.tioi1 years.
began on an upgrade project that has significantly The Chalk River Laboratories Tandem Accelerator
upgraded the basic physics research capability of the and Superconducting Cyclotron (TASCC) operated a
NSCL. The upgrade was accomplished by combining superconducting cyclotron that was used for SEE testing
an electron cyclotron resonance source with the K500 until the late 1990's. The facility was closed
arid KI 200 cyclotrons linked in series; and by replacing permanently in August of 1997.
the fragment separator with a higher acceptance
C. Proton Test Facilities
separator [26]. With this upgrade, NSCL can now
produce heavy ion beams for all elements with energy There are six major proton test facilities; four US, one
European and one Canadian (others exists and are used
higher than that available at all other US facilities
except the Alternating Gradient SynchrotrodRelativistic from time to time, but are not considered to be
mainstream facilities at this time). These facilities are
Heavy Ion Collider at Brookhaven National Laboratory.
As part of the NSCL upgrade, NSCL staff and NASA used for SEE, total ionizing dose and displacement
damage studies. Like the heavy ion SEE facilities these
Goddard Space Ffight Center staff are preparing a
test facilities are located at laboratories used for the
facility at NSCL called the Single Event Effects Test
most part to carry out basic physics research or cancer
Facility (SEETF). The facility consists of a beaniline
therapy-again, this work is a higher priority for these
and target area with associated interface, monitoring and
laboratories.
control hardware modeled after the systems found at
These facilities rely on three primary dosimetry
ciirrent test facilities. Beams with energy from 20- 200
systems to determine the flux and uniformity of the
MeV/amu will be available for ions from deuterium to
beam: scintillators (usually plastic/organic), secondary
uranium. The current NSCL plan is to provide 300-600
electron monitors, and Faraday cups. Additionally,
houdyear of beam time to SEE testing. User interface
radiochromic films may be used to determine qualitative
al. the SEETF is under development, and has been
beam uniformity. AI1 of these facilities have test stands
dzsigned to be similar to existing facilities as much as
that allow open-air exposures (not in a vacuum). It is
pxsible.
widely accepted that the dosimetry at these proton
Fig. 2 plots typical heavy ion LET versus range
values in silicon that are attainable at each of the facilities is reasonably accurate, at least within 10%.
The next few paragraphs list, in no particular order, the
facilities listed above. At several of these facilities, the
major facilities and gives some information about the
LET-range values can be changed by tuning the
facility.
accelerator to a specific beam energy. Degraders can
University of California at Davis's Crocker Nuclear
also be used, but the beam energy straggle and
Lab (CNL) [30] has a isochronal cyclotron proton
uniformity must be verified. (The data for SEETF are
accelerator. It can achieve energies in the 1-68 MeV
the expected LET-range values.)
range. The cyclotron is energy tunable. The beam spot
Cyclotron Research Centre at Louvain-la-Neuve,
uniformity across the maximum 6 cm diameter is better
Belgium [27] operates three cyclotrons capable of
than 10%. Beam dosimetry is achieved from calibrated
delivering of heavy ions up to 27.5 MeV/amu. The
secondary electron emission monitors, these are
cyclotron has two different cocktails of heavy ions used
calibrated to a direct faraday cup measurement.
fix SEE testing. The test chamber is a 71x54~76c ni3
Indiana University (IU) Cyclotron Facility [3 I] has a
vacuum chamber with a multi-directional motion
cyclctrodsynchrotron (cyclotron only for SEE). The
controller for moving the device relative to the ion
energy peaks at 230 MeV (typical operation is near 200
beam. The dosimetry system is similar to the other
MeV) and can be tuned. The beam spot can be up to
heavy ion test facilities.
7cm diameter beam spot. A second beam line is
I'Institut de Physique Nucleaire (IPN) in Orsay,
currently being developed by IU and NASA Johnston
France operates a Tandem Van de Graaff that has used
Space Center staff. The dosimetry is obtained via a greate:: than the energy bandgap of the semiconductor.
faraday cup and secondary electron emission monitors. Added flexibility comes from being able to fire the laser
Lawrence Berkeley National Laboratory [23] has an repeatedly without damaging the device, from single
88-inch cyclotron (described above). The proton energy shot to kilohertz rates, and to do so without the need for
range is 1-55 MeV tuned. The beam spot is; 4” a vacuum. The light is typically focused to a -1 pm size
diameter. The dosimetry is done with an ion chamber spot and scanned across the device to obtain the spatial
with rings for uniformity check, and radiochroniic when dependence of SEE sensitivity. The temporal
needed. characteristics of SEEs in dynamic circuits can be
Texas A&M University (TAMU) [25] uses a K500 measured by synchronizing the circuit clock to the laser
superconducting cyclotron (described above). The trigger and adding delay [38]. Although the charge
energy range is 8 - 70 MeV tunable. The beam spot is generation mechanisms for ionizing particles differ
1” diameter. The dosimetry used is foudfive scintillator fundamentally from those for ionizing photons, both
array. experimental and theoretical investigations show that
Tri-University Meson Facility (TRIUMF), located in the resulting voltage transients are, in many cases,
Canada, [32] utilizes a cyclotron to accelerate protons to indistinguishable [39], [40]. The pulsed-laser technique
energies between 65-120 MeV on one beamliiie and does suffer from a significant limitation-the inability
180-500 MeV on another beamline. A two to three inch of the light to penetrate metal layers on the surface of a
square beam spot is available depending on the energy device:. This is the reason why recent reports suggesting
used. The dosimetry on low energy line is ion chambers that a pulsed laser can be used to generate curves of
calibrated against externally calibrated ion chamber. On SEE cross-section versus LET are not promising [41].
the high-energy line dosimetiy is a achieved with a One application of the pulsed laser that has, on
combination of faraday cup, plastic scintillators, and occasion, proved invaluable is to ensure that both the
PIN diodes-all agree very closely. devices selected for testing and the test equipment are
Paul Scherrer Institute, located in Switzerland, Proton functioning properly before being shipped to the
Accelerator Facility [33] can provide protons .with accelerator facility.
energies between 60 and 300 MeV in one facility and A magnetically focused ion microbeam is a powerful
between 6 and 65 MeV in another. Bean1 spot is 34nim tool for studying the basic mechanisms contributing to
(low energy) and 9 cm (high energy) diameter. The SEEs. The ion beam is generated by an accelerator and
dosimetry in the low energy facility is done using ion focused by a set of magnets to produce a beam with a
chambers or CsI(TI) scintillators. In the high-energy diameter of -1 pm. One such facility, located at Sandia
side an ion chambers, PIN diodes and plastic National Laboratory (SNL), generates ions with a
scintillators are used. inaxinium energy of 50 MeV. Of the ions available,
I-Iai-vard University has a cyclotron that has been used those with ranges greater than IO Lni typically have
extensively in the past for SEE testing. However, since LETs less than 15 MeV-cm2/mg. For testing, the
the early 1990’s this facility has decided to focus most devices are mounted in a vacuum chamber and the ion
of its resources on cancer treatment applications and beam is either rastered across areas of interest or is fixed
research. Currently access to Harvard for SEE testing is in one position. Rastering the beam permits the
ve:ry limited. generation of detailed maps of the device response at
specific locations around various sensitive transistors.
D. Other SEE Test Facilities
In this way, images of both SEE sensitive areas and
Although broad-beam accelerators are essential for
charge collection efficiency from specific junctions
SEE characterization, there are other approaches have been measured [42], [43]. Most of the SEE
capable of providing information on SEEs. The
investigations using a microbeam have been reported by
limitations of broad-beam accelerator testing-
two groups-SNL in the USA and Gesellschaft her
including limited availability, high cost and lack of both
Schwerionenforschung (GSI) in Germany. Some of the
detailed spatial and temporal informatiowhave lead to
technique’s limitations are the short range of the ions
the development of alternate approaches for measuring
available, the necessity of using a vacuum, limited
SEE sensitivity. They include pulsed lasers, ion
number of ion LETs, and the damage induced by the ion
microbeams and 252Cf.
beam in the device being tested.
A pulsed laser is a well-established tool for The third approach involves using the decay products
elucidating the spatial and temporal characteristics of
from ,I radioactive source, such as 252Cf to generate
SEES [34], [35], [36], [37]. The basic requirement is SEEs in circuits [44], [45], [46]. The source and the
that the laser generate short (-1 ps) pulses of light, with device: are mounted close to each other in a vacuum
pulse energies greater than a nJ and photon energies
chamber. The decay products fall into two energy given space environment. The rate prediction methods
ranges around average energies of 78 MeV and 102 do a fairly good job of predicting what is actually seen
MeV. Therefore, not only are the ion energies and, on-orbit. Of course the quality of the predictions is a
consequently, the ranges relatively low, the uncertainty function of the quality of the test data and the skill of
in energy leads to an uncertainty in both LET and range. the iiiodeler, taking into account the assumptions and
The maximum available LET at normal incidence is 45 limiiations of the models.
IvleV-cni2/mg. A major issue is the short range (-10
B. Heavy Ion Predictions
pni) of the ions that limits the usefulness of the
To first order, the linear energy deposition rate
technique, because devices with thick passivation layers
(MeV/pm) drives the effects. This allows simplification
or deep junctions cannot be tested reliably. Another
issue is the radiation hazard for which stringent of the prediction problem through use of energy transfer
precautions must be taken to protect personnel. (LET) spectra, as first developed by Meinrich [5]. All
the ion types and distributions of energy in the space
Nevertheless, for quick evaluations prior to doing
environment can be reduced to their LET, and deposited
accelerator testing, or for certain hardness assurance
measurements, radioactive sources do offer a useful ener,gy can be estimated as LET times the chordlength
through the sensitive volume. With this simplification,
approach.
the problem to be solved is to identify the size of the
V. REVIEWOF SEE RATEP REDICTIOTNESC HNIQUES sensitive volume, calculate the rate of ion hits and the
consequent energy depositions, and determine the
A. Introduction subset of total ion hits that cause SEE.
Prediction of SEE rates involves a combination of The SEE rate is the product of the sensitive area on
experimental data, assumptions about the device, and the chip and the flux of ions in the environment that can
knowledge of the energetic particle environment. This cause upset when they hit the sensitive area. The
section discusses how the ground test data, as described threshold for the effect determines the effective flux.
above, can be used to predict rates for SEE due to The problem is complicated by the angular dependence
energetic particles in a space environment. We sinct: the amount of energy deposited in the sensitive
summarize the current rate prediction techniques for volume depends on chordlength, which in turn depends
heavy ions and protons, interspersed with a historical on angle of incidence. The modeling problem can be
glimpse of the early evolution of the concepts and approached from two directions: from a microscopic
approaches. Table 1 lists the key milestones for viewpoint (the chordlength approach) or from a
development of SEE rate prediction techniques. Much macroscopic uewpoint (the effective flux approach).
work has been done to refine the early methods in the Bott approaches give similar results and are effectively
ensuing years, and the reader is referred to the equivalent if the same geometric assumptions are made.
Transactions on Nuclear Science and several sets of The chordlength model determines the minimum charge
Nuclear and Space Radiation Effects Conference required for upset from cross-section versus LET test
(NSREC) Short Course Notes [47], [48], [49] to follow data, considers the distribution of LET in the
that trail. See the Short Course notes by Petersen [47] environment, and puts a criterion on each ion interaction
lor an excellent review of the historical evolution of the with a sensitive volume to select a set of ions (and
development and use of rate prediction concepts. associated flux) that exceed the minimum charge. The
Single event effects are related to charge generation effective flux model transforms the ion flux in the space
along the path of a primary or secondary ionizing envi::onnient to an effective flux (capable of causing
particle, charge collection on circuit nodes, and circuit SEE) based on measured cross-section versus LET test
i'esponse to the charge transient. Both the total collected data for the chip. Several rate prediction methodologies
charge and the rate of charge collection can be and codes are discussed in the literature, but they all fall
important to triggering the effect. SEE rate prediction into one of these two general categories.
models typically use ground test data to extract 1 .;I Chordlength Model
information about the device sensitivity, measured in The original Chordlength model was introduced by
terms ofcross-section (CS) and critical charge (Qc), as a Pickel and Blandford in 1978 [4] and developed into a
hnction of LET and/or proton energy. Testing methods computer code (CRIER) in 1980 [50]. The sensitive
have been devised to generate this information, as volume is modeled as a rectangular parallelepiped
described in the previous sections. Once the CS versus (PI') with lateral dimensions x and y and thickness z.
LET or CS versus proton energy data have been The saturation cross-section per bit (CS,) is given by
experimentally acquired, there are established the product of x and y; conversely x and y are
techniques for using the data to predict SEE rates in a determined by measurement of CS,,, taking into account
the number of bits in the chip. The RPP approxiniates different thresholds and with distributions on their
the depletion region beneath a p-n junction that is parameters. Petersen was the first to address this issue,
determined to be a sensitive volume. The ion is suggesting that the cross-section curve be divided into
assumed to travel in a straight line and the path through several steps in order to more accurately represent it
ti-e RPP is S, determined by thickness, z, and the angle [52]. The common approach is to weight R(E) with the
o F incidence, ?. Ion plasma track structure is ignored. normalized experimental cross-section data
Charge is also allowed to be collected along a funneling
d stance, Sf, that adds to the chordlength S through the R = ?R(E) f(E) dE (4)
depletion region. Epitaxial layer thickness may limit
charge collection by funneling. The energy deposited in when: the integration range is from the measured
the sensitive volume from an ion with LET, L, is threshold, &, to the measured value at saturation, Sa,,
and f(E) is the cross-section versus LET curve
E = (S + Sf) L. (1) converted to a probability density, often described by
the four parameter Weibull distribution. The function
This energy is converted to charge in accordance with R(E) is the rate at which an energy of E or greater is
the ionization energy (3.6 eV/carrier pair for silicon) deposited in the sensitive volume. Moreover, f(E) may
and it is assumed that all charge that is generated within be regarded as the probability density for an event
the charge collection length S + Sf is collected by the caused by deposition of E or greater. Thus the integral
circuit node. It is assumed that there is a sharp is the: expectation of R(E) with the probability f(E).
threshold for upset-ion hits below a threshold LET do This approach is commonly called the Integral RPP
not cause upset, hits above the threshold cause upset. (IRPP) model.
The classic RPP method utilizes an integral LET The integral in Equation 2 is solved numerically. The
distribution and an analytic differential chordlength original implementation was in the CRIER code [50]
distribution function, f(s), and integrates over the and a version of this code is implemented in CREME
chordlengths through the RPP. The rate is expressed as suite of codes [53], and also in commercial codes. The
integral in Equation 4 is solved by dividing the data set
R(Ec) = A, ?? [ Lt(S,Ec)] f(S) dS, (2) into 21 number of bins based on LET. The data set is
Smin = 0 divided into anumber of bins based on LET and the
s,,, = (x2 + y2 + z2)"2 integral in Equation 4 is also solved numerically.
2.) ESfective Flux Model
where the limits on the integral are fioni 0 to the The original Effective Flux model was introduced
imaxinium pathlength through the RPP, 4 is average by Binder in 1988 [54]. The method is based on
projected area of the RPP, ? (L) is integral flux, E, is the consideration of the range of incident angles that can
threshold energy for generating Q,, and L,(S,E,) is the produce an SEU and the ion flux contained in that
iiiinimuni LET which depends on chordlength through range. The model assumes an isotropic flux as a
funct on of LET, F(L), incident on a thin lamina. If the
threshold for upset is J+ and L > ,,.I then all incident
angles produce upset. If L < L,, there is a critical angle,
where the chord length random variable S has been ?c, which produces upset, where
modified to account for charge collection by funneling
and E, is defined by the critical charge. Inputs to the cos(?,) = L / L,. (5)
classic model are x, y, z, Sf and E,.
Alternative formulations that use an integral The ion flux in the environment ? (L) can be
c hordlength and differential LET distribution were trans::"ormed to an effective flux ? e(L) for an assumed
iiitroduced by Petersen and Shapiro in 1982 [51]. The cutoff angle The effective flux is sometimes called
?c.
two approaches are fundamentally equivalent. They redistributed flux. Then the rate is calculated by
differ in how they handle the complexities of integrating
over discontinuities. R = ?? ,(Lo dCS(L) (6)
The classic RPP model assumes a step function for
the cross-section versus LET curve. However, most wherl: CS(L) is the measured cross-section versus LET
devices exhibit a gradual rise from threshold to test data and the limits on the integral are from 0 to the
saturation because chip response generally is the maxiinum LET in the environment.
composite of multiple types of sensitive volumes with