Professional Documents
Culture Documents
Overview of In-Orbit Radiation Induced Spacecraft Anomalies: R. Ecoffet, Member, IEEE
Overview of In-Orbit Radiation Induced Spacecraft Anomalies: R. Ecoffet, Member, IEEE
Abstract—Spacecraft anomalies due to radiation effects on elec- spacecraft design, such as celestial mechanics or electromag-
tronic devices have been known since the very beginning of the netism, for instance, which pre-existed before the space era,
space era. Today they represent a major part of on-board anom- the knowledge of the space environment was a consequence of
alies, mission outages, and unexpected workload on ground con-
trollers of operational systems. This text will first describe a few space exploration itself.
known cases of cumulative or transient effects in Earth or plane- A few years after the Explorer-I launch, Telstar was launched
tary environments. Then, it will discuss investigation methodology, on 10 July 1962, designed and built by the Bell Telephone Lab-
and issues such as the statistical aspects of probabilistic anomalies oratories with AT&T funds and NASA support. Telstar defini-
in the assessment of cause to effects relationships, and the neces- tively opened the era of satellite telecommunication with a live
sity of multi-field experts in group root cause analysis, from radia-
tion effects, to satellite fault determination, to identification and re- television picture transmission from the USA to France on that
configuration strategy. A focus on some practical cases of anomaly same day. It was the first satellite equipped with transponders,
analyses will be made. and thus amplification, and thus the first active telecommunica-
Index Terms—In-orbit anomalies, radiation effects, space envi-
tion satellite. On 9 July 1962, the day before the Telstar launch,
ronment, spacecraft anomalies. the USA carried out a high altitude nuclear test. The extremely
high radiation levels induced by electrons injected in the radi-
ation belts caused degradation of some electronic components,
I. INTRODUCTION (diodes in the command decoder) and finally, the early loss of
the satellite on 21 February 1963. This was the first spacecraft
loss due to radiation effects.
TABLE I
MAIN SOURCES OF THE NATURAL SPACE RADIATION ENVIRONMENT.
Fig. 8. Overview of various space environment components in the period 2000–2010 on the right, left: risk tendency chart for 2008–2010. Data are from GOES-10,
NPOES-15, ICARE/SAC-C, and ACE/CRIS. Credits NOAA, CNES, NASA. IPSAT plot, and CNES/ONERA [4].
C. Statistical Comparisons
We will limit ourselves here to the results of two studies,
by NASA [8] and Aerospace Corporation [9]. In “Spacecraft
System Failures and Anomalies Attributed to the Natural Space
Environment”, NASA reference publication 1390 [8], MSFC
compares more than 100 anomaly cases attributed to the envi-
ronment of the satellite, and proposes the following classifica- Fig. 9. Spacecraft anomalies due to the space environment, from [4].
tions:
— plasma (charging)
— radiation (TID, SEE, DDD)
— meteoroids and space debris
— atmosphere (neutral thermosphere: satellite drag, effects
on surface materials)
— solar (effects of solar events, may fall in “plasma”, “radi-
ation” or “geomagnetic” categories)
— thermal (thermal conditions in vacuum—may be consid-
ered as a design problem) Fig. 10. Repartition of radiation spacecraft anomalies, from [4].
— geomagnetic (effects on magneto-torquers for example).
The chart in Fig. 9 uses these classifications. For classifying Although the results may differ from one study to another,
anomalies by their physical cause, we break out the “solar” cat- a general conclusion can be drawn: the two major sources of
egory into “plasma”, “radiation” or “geomagnetic” categories. environment related spacecraft anomalies are plasma and radia-
From this study, the two main environment-related causes of tion effects, i.e., effects related to the charged particles from the
spacecraft anomalies are radiation and plasma effects. If we now space environment.
break out the radiation part into the main radiation effects, we
come to the distribution shown in Fig. 10. The “upset” category D. Feedback From CNES Satellite Operations
should be understood in a wide sense, and may include SET ef- A survey was made at CNES in order to assess the weight
fects. of space environment and space radiation in the number of ob-
The set of anomaly cases used in the Aerospace study [9] in- served anomalies:
cludes more recent anomaly cases and has an overlap with the 1) Returns From CNES GEO Satellites: Since beginning of
set used by MSFC. This set is composed of 326 anomaly cases, a life (1992) and for the 4 geosynchronous TC2 satellites (total
large number of which is related to GEO satellites, thus empha- of 50 years in orbit), about 30% of anomalies were due to the
sizing the ESD issue. Table II from [9] gives the classification space charged particle environment (radiation and charging al-
used and the repartition of anomalies. together).
1796 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013
TABLE II
REPARTITION OF CAUSES FOR 326 SPACECRAFT ENVIRONMENT RELATED
ANOMALIES, FROM [9].
TABLE III
SPACECRAFT ANOMALIES ASSOCIATED TO THE INTENSE SOLAR ACTIVITY OF
OCTOBER/NOVEMBER 2003, FROM[10].
TABLE IV
EXAMPLE OF THE EFFECT OF SHIELDING DESCRIPTION ON ESTIMATED
MISSION LEVELS FOR AFEE/DFEE EQUIPMENT ON INTEGRAL—DOSES IN
RAD(SI) [11]
Fig. 12. Background noise in the L-time diagram for the hipparcos star mapper,
from [12]. Color scale from green to white is background noise intensity in
arbitrary units.
TABLE V
HISTORICAL ARTIFICIAL RADIATION BELTS. SOURCE AND CREDIT: WIKIPEDIA.
a number of other satellites were impaired such as OSO-1 (Or-
biting Solar Observatory).
On 9 July 1962, the day before the TELSTAR launch, the US
conducted the Starfish Prime experiment. The TELSTAR orbit,
942/5646 km 44.8 , took the satellite through the inner radiation
belt and parts of the outer belt. The radiation exposure was very
intense, and increased again after Soviet tests in 1962. It is be-
lieved that TELSTAR experienced fluxes 100 times higher than
those expected. Four months after the Starfish blast, some tran-
sistors failed, and finally the loss of the spacecraft (end of life
21 February 1963) was attributed to total ionizing dose degra-
dation of diodes in the command decoder.
ARIEL, the first international science satellite, designed and
built by NASA/GSFC to study the ionosphere and solar radia-
uncertainties related with space environment models and radia- tion, was launched on 26 April 1962 and provided science data
tion qualification tests procedures. until September 1962, after which it degraded rapidly because
of the degradation of its solar panels by the Starfish increased
A. Artificial Radiation Belts radiation. It continued to function erratically and was cut off
As quoted in the introduction, exo-atmospheric nuclear ex- in November 1964. TRANSIT 4B and TRAAC mission losses
periments conducted by the USA and the USSR, before this was were also attributed to solar cell degradation.
banned by international treaties in 1967 (see Table V), created
dramatic consequences not only on the ground (the EMP effect B. HIPPARCOS
was detected as far as Hawaii after the Starfish test), but also The ESA star mapping mission HIPPARCOS was launched
in space through long-term enhancements of the radiation belt on 8 August 1989 for a geostationary position, but
fluxes. its apogee motor failed and the satellite ended in an initial
The most noticeable US experiment, because of the power GTO orbit 498/35889 km, 6.5 . This caused the satellite to be
of the weapon and the altitude at which it was exploded, was exposed to much higher radiation fluxes than it was designed
the Starfish Prime experiment which led to the formation of en- for. Furthermore, a major solar event in March 1991 caused
hanced flux zones in the inner belt between 400 and 1600 km flux enhancements in the radiation belts. This event was one
and beyond. The USSR experiments are not well documented, of the major observations of the US CRRES satellite and has
but it is believed that they contributed to flux enhancements in revived the interest for understanding the dynamic behavior
the outer belt. These modified fluxes finally extended in all the of Van Allen belts. The event was observed by HIPPARCOS
radiation belt regions and were detectable as late as the early itself through background noise increase in its instruments, as
1970s. In some zones of the inner electron belt, fluxes increased depicted in Fig. 12 from [12].
by a factor 100. Electrons from the Starfish blast dominated the The change in HIPPARCOS orbit lead to total dose expo-
inner belt fluxes for five years. These artificial fluxes delayed sure levels 5 to 10 times more severe than those expected for
the issue of representative natural radiation belts models, and the nominal orbit. HIPPARCOS flew 5 gyroscopes, 3 of which
Starfish artifacts might still be present in some zones of the AE8 were active at a time. The five gyroscopes (one of which was al-
and AP8 models in use today. ready degraded) successively failed within a period of 6 months
In the period directly following these tests in 1962, a series of after the mission lasted three years. For four of them, the spin
at least 10 satellites successively failed (7 in 7 months). Among down and final stop were attributed to the total dose degradation
them were TELSTAR, TRANSIT 4B, TRAAC, and ARIEL, and of the access times of the bipolar PROM which digitally stored
ECOFFET: OVERVIEW OF IN-ORBIT RADIATION INDUCED SPACECRAFT ANOMALIES 1799
Fig. 14. Timeline for some Galileo anomalies, compared with TID accumula-
tion, from [13].
Fig. 13. Timeline of Hipparcos gyroscope failures. then, JPL has produced excellent and careful studies of the ef-
fects of Jovian radiation belts on GALILEO systems [13], [14],
the sinusoidal excitation of the fields spinning the wheels. The and [15].
dose received on the PROM was estimated to about 40 krad(Si). GALILEO was almost entirely equipped with rad-hard com-
Before the final spin downs, problems appeared when trying ponents and the radiation issue had been studied as carefully as
to re-start the gyroscopes at cold temperature because of tran- possible by JPL in the design phase. The use of rad-hard com-
sistor degradation in the DC/DC converter. Noise and erratic ponents eliminated the SEL risk and most of the SEU problems.
data were also due to the degradation of the 262 kHz clock of the For this project, SEE, in the “classical” sense, was not an issue,
wheel motor power supplies. The last gyroscope, kept in cold but background noise in sensors had to be taken into consider-
redundancy until the very last days of the mission, also failed ation. Cumulative effects (TID, DDD) were a strong challenge:
a few months after it was turned on. This time, the failure was in its various orbits around Jupiter, Galileo was submitted to
attributed to the radiation degradation of an optocoupler in the doses between 10 to 50 krad(Si) at each periapsis. A character-
thermal regulation system with an estimated received dose of 90 istic of these orbits, having a long period, one to three months,
krad(Si). Finally, communications were lost with HIPPARCOS. was that this exposure was concentrated at periapsis passes, and
The timeline of the successive gyro failures is shown in Fig. 13. the components could anneal during the remaining part of the
During the last period of its life, HIPPARCOS was operated orbit. This characteristic was used by JPL for extending the life-
with only two of the three gyroscopes normally required, and time of some critical systems. This gave also time for the flight
schemes were drawn for operating it without gyroscopes at all. team to work out circumvention solutions before the next pass.
The first gyro-less data were acquired when a communication Various effects (see cases in Fig. 14) were observed during the
failure with the on-board computer put an end to the science GALILEO mission, and some examples will be given from [13],
mission on 24 June 1993. Further attempts to restore operations [14] and [15].
were unsuccessful and the mission terminated on 15 August The star scanner was sensitive to high energy electrons
1993, 4 years after launch. , resulting in a radiation noise that could be
In spite of these problems, HIPPARCOS observed 118274 misidentified for a star. These misidentifications lead to erro-
stars with an extreme precision. HIPPARCOS was designed for neous celestial attitude estimates. At the periapsis of one of
an operational life of 2.5 years in a GEO orbit. It conducted an the most exposed orbits (“E12”), the attitude estimate from
extremely successful mission in the much more aggressive envi- the star scanner was in strong disagreement with the inertial
ronment of a GTO orbit (and even in enhanced flux conditions), attitude estimates from the gyroscopes (themselves affected by
and accomplished all its scientific goals during more than 3.5 radiation, see Figs. 14 and 15 ). The flight software thought
years. that GALILEO’s scan platform was not pointed where it had
The reason for such operational resilience probably lies in been commanded. The fault-monitors commanded swapping
known and unknown margins in the design. As stated before, to redundant systems twice in 8 hours. Another problem was
the few available cases of total dose anomalies all corresponded related to the averaged radiation signal that increased the signal
to conditions far exceeding the original specifications. to noise ratio and caused some stars to fall outside of the
expected brightness range, resulting in “missing stars” in the
scanner’s acquisitions. For passes inside 8 Jupiter radii, only
C. The Galileo Probe at Jupiter the 10 or 20 brightest stars could be distinguished.
The scale factor of the gyroscopes used for pointing instru-
Extreme environments can also be found around other planets ments was also affected. A discrepancy between the expected
of the Solar System. In this respect, the Jupiter environment is a pointing direction and gyro’s position estimates grew with mis-
pandemonium of intense radiation fluxes trapped in the colossal sion time, adding its contribution to the hardware swapping in
magnetic field of the giant planet. This environment has been orbit E12. After this orbital pass, the gyros gave unreliable infor-
first studied by JPL’s PIONEER and VOYAGER probes, whose mation, and the flight software thought that the spacecraft’s an-
results helped designing the JPL GALILEO spacecraft. Since tenna was no longer pointed at Earth. The AOCS fired thrusters,
1800 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013
Fig. 15. Spin detector error compared in%to real spacecraft spin rate, from
[13].
maneuvering the spacecraft to a wrong attitude. Six unexpected Fig. 16. USO frequency shift during the orbit E11 radiation belt pass, from
maneuvers occurred before the flight team succeeded in taking [13]. Note the 0.29 Hz jump.
control. Fortunately, the antenna was still pointed sufficiently
TABLE VI
toward the Earth for receiving telecommands. The gyroscope SUMMARY OF GALILEO IN-FLIGHT RADIATION ISSUES, FROM[13].
problem was traced back [14] to leaking currents from a DG181
switch that resets the slew rate integrator to zero after the mea-
surements are performed.
On one of its last orbits, while recording data from the
Amalthea moon, data collection stopped abruptly just after an
encounter that had brought GALILEO deeper in the radiation
belts than ever before. The cause was a sudden switch of the
spacecraft to safe mode attributed to proton SETs in the four
phase lock loops of the command and data system. The data
from Amalthea had been recorded but could not be played back
to Earth. The problem was traced back to the tape motor drive
electronics, and more specifically, to three GaAs OP133 LEDs
used in the wheel position encoder.
Displacement damage was incriminated as the cause for the
drops in the LED’s light output. Knowing this, JPL derived a
circumvention strategy based on the current enhanced annealing
properties of the LED. They used a special operating mode al-
lowing current to be applied continuously to the LEDs without
trying to move the motor. The recorder began to operate for a
few seconds. After each annealing step, the recorder operating
period was longer, and finally JPL succeeded in retrieving all
the missing data [15].
GALILEO used also an USO (ultra-stable oscillator) for
tuning the telemetry 2.29 GHz download frequency. This
USO frequency was subject to gradual and permanent drifts
due to cumulated radiation. It was also subject to frequency
jumps during the passes through the radiation belts, some of
D. Ultra-Sensitive Systems
which were close to 1 Hz (Fig. 16). These drifts were precisely
measured because operators of the Deep Space Network had to Some systems are so finely tuned and depend so much on
find the signal by manually sweeping through the frequencies. general spacecraft environment stability that dose effects mani-
Sometimes, the spacecraft was already transmitting before the festations can be observed even in moderate radiation environ-
carrier wave was locked-up. ments. This is the case, for example, of the USO (ultra-stable os-
Other effects are described in reference publication [13]. cillator) of the DORIS positioning payload on board JASON-1.
Table VI from [13] gives a summary of the various systems JASON-1 is a joint CNES and JPL satellite devoted to altimetry
affected and how JPL found clever circumvention schemes. and positioning applied to ocean sciences. The satellite moves
These schemes could not have been derived without a detailed on a 1335 km, 66 orbit that makes it pass through moderate flux
comprehension of the radiation phenomena. regions of the proton belt. This orbit is still much more exposed
ECOFFET: OVERVIEW OF IN-ORBIT RADIATION INDUCED SPACECRAFT ANOMALIES 1801
Fig. 17. DORIS on JASON frequency shifts, courtesy J.M. Lemoine, GRGS Fig. 19. PIONEER VENUS shift register upset count versus time [4].
[4].
upsets as well, but GCR anomalies are found in either “old”
satellites, or “new” ones situated in a proton free environment.
One of the very first upset observations was made on the Pi-
oneer 12 (Pioneer Venus) probe, in a 1024 bit PMOS shift reg-
ister. Fig. 19; using data from the NOAA Spacecraft Anomaly
data base [17] shows the cumulated upset counts from 1978 to
1985.
The CNES SPOT-1,-2 and –3 satellites (820 km, 98.7 )
were all equipped with a central processing unit (CPU) whose
memory array was made out of 1000 1-kbit HEF4736 static
RAMs. These memories, procured in 1986, were only sensitive
Fig. 18. MISR on TERRA dark current increases, from [16], NASA image.
to GGR. Their LET threshold [18] was , so
than more conventional LEO Earth observation orbits such as they responded only to heavy ions from the group of Iron or
800 km, 98 . To give an idea of the proportion factor in rough higher Z particles.
numbers, the dose received on JASON behind a spher- During the lifetime of these satellites, single event upsets
ical shield is about 10 times the dose on an Earth observation (SEUs) were regularly recorded in the CPU memory array [18].
satellite. Nevertheless, the order of magnitude of dose exposure About half of these SEUs lead to operational problems of var-
at each pass through the SAA is less than 1 rad(Si). Some very ious levels of importance, including switching the satellite to
sensitive systems such as USOs can be sensitive to dose incre- safe mode. Later generations of SPOT satellites were equipped
ments of this order. with completely different CPUs, not sensitive to these effects.
The effect of radiation on USO systems is to induce frequency As indicated earlier, SEUs are also now associated with
shifts related to the amount of dose received (see also GALILEO particles other than GCR, and associated in-flight anomalies
effects in the previous paragraph). DORIS sensitivity is such are harder to find. Nevertheless, there is a category of effects
that even slight frequency shifts can have an impact on the final which is still related to GCR and is beginning to become a
system performance. In the case of the JASON application for major contribution to operational spacecraft anomalies. This
DORIS, the satellite passes through the SAA which leads to category of effects is made out of single event transients (SETs)
tiny deposits of dose that can clearly be seen in the processed in analog electronics. The typical SET effect is an unexpected
data. The effect of ascending or descending passes through the switch-off or reset, but other manifestations can occur. Such
SAA can even be observed. Fig. 17 plots the frequency shift unexpected switch-offs occurred on SOHO, along with various
amplitudes versus geographical position: the correlation with other phenomena, and were studied by ESA-ESTEC [19].
SAA fluxes is clearly visible. Table VII from [19] summarizes these anomalies. For power
Other very sensitive systems are finely calibrated imaging supply events SETs, PM 139 comparators and UC1707J dual
systems, which may record even a slight modification of their channel power drivers, were suspected. ESTEC proceeded to
pixels’ dark currents. Fig. 18 from [16] maps the dark current in- ground test engineering models and confirmed this cause as the
creases for the MISR imager on NASA’s TERRA satellite when origin of the anomalies.
the instrument shutter is closed. The effect of trapped protons is Many private communications have convinced many that
clearly visible. SETs are a growing concern for geosynchronous telecommu-
nication satellites. Some cases of operational problems were
V. SINGLE EVENT EFFECTS recorded and attributed to SET effects. These cases were not
correlated to the large solar events of the present solar cycle,
A. Galactic Cosmic Rays and seem to be GCR induced. The susceptibility of the world
Galactic cosmic rays (GCR) are the most efficient particles telecommunication fleet in GEO is not surprising simply be-
for inducing single event upsets, but their flux is very weak. cause these satellites constitute the bigger sample of spacecraft
Most of today’s technologies are sensitive to proton induced in a GCR exposed orbit. The geosynchronous orbit is far above
1802 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013
TABLE VII
SUMMARY OF SOHO UNEXPECTED EVENTS, FROM [19].
Fig. 21. TDRS-1 AOCS RAM upsets compared with GOES flux measure-
ments during October 1989 solar flare, from [21]. From top to bottom, GOES-7
X-ray flux, number of anomalies per hour, GOES-7 proton flux at various ener-
gies, Deep River monitor neutron counts.
October 1989, the day after the GALILEO launch, was one of
the most powerful solar flare events ever recorded. The upset
count on TDRS-1 jumped from about 15 to 20 upsets/week, to
249 events in 7 days (19–25 October 1989).
Fortunately, the GALILEO Jupiter probe was still near Earth
and its ion counter instruments could measure the ion compo-
sition of the flare. The NOAA GOES-7 proton data could also
be used for proton flux estimations. Considering the sensitivity
of the 93L422 memories, the investigators concluded that about
30% of the upsets were due to solar protons and 70% were due to
solar ions. This balance depends on component sensitivity and
satellite shielding and should not, of course, be extrapolated to Fig. 24. MYRIADE reaction wheel anomalies ‘[4].
other applications.
3) SOHO SSR Upsets: Another particularly well docu- Other recorded effects of SAA protons include single event
mented correlation between solar events and radiation effects transients (SET) as experienced, for example, on the CNES
[19] is the case of single event upsets in the Solid State Recorder MYRIADE micro-satellite platform, on an on-the-shelf reaction
(SSR) of the ESA/NASA SOHO spacecraft at the L1 Lagrange wheel as shown in Fig. 24.
point. Fig. 22 shows the evolution of the SEU rate between Latch-ups have also been attributed to trapped protons. One
1996 and 2001. The background rate is due to galactic cosmic of the most famous cases, the first to show proton induced
rays. The solar cycle modulation of this background rate is latch-up, was the failure of the non-ESA PRARE altimeter
visible in the plot. The various sudden increases in the SEU on-board the ESA Earth observation satellite ERS-1 (774 km,
rate correspond to solar events. 98.5 ). At the moment of the switch-off event, an increase
in primary power of about 9 W, lasting between 16 and 32
C. Trapped Protons seconds, was recorded by the housekeeping system of the
NASA/GSFC has carefully studied [23] the upsets observed satellite. This event occurred after only 5 days of operation.
on some solid state recorders (SSRs) on SAMPEX The failure location is right in the middle of the SAA zone
and TOMS/METEOR . (Fig. 25). The anomaly was studied by ESA [24], and ground
Figs. 23 and 24[23] very clearly show the correlation between tests on engineering models proved that the origin of the failure
upset occurrence and the SAA. Note that due to the error detec- was due to a latch-up on one of the 64-kbit CMOS SRAMs
tion and correction routines implemented on these applications, used in PRARE. Occurrence estimations from these tests were
these upsets had little impact on system performance. SAMPEX also consistent with the in-flight observation.
was also a technology demonstrator using new up-to-date tech- Another well documented LEO case is the one of the joint
niques that had not previously been tried in space applications. Swedish-German project Freja which expe-
For example, it used a MIL-STD-1773 optical fiber data link rienced a number of single event effects including false com-
[23]. These data links use an LED as an emitter, fiber optics for mands (0.5 to 1 per day, in the SAA), telemetry format er-
transmission, and a photoreceptor as a receiver. When an upset rors, memory SEUs, and finally latch-ups on the CPU boards.
occurs in a data transmission (mainly due to particle transients The latch-ups shown in Fig. 26[25] were detected in the house-
on the photoreceptor), a bit is corrupted, and the system asks keeping data through records of switches to redundant units.
for a retry. Fig. 23 shows again a clear correlation between The anomalies were studied by ESA [25]. The latch-ups were
MIL-STD-1773 bus retries and the SAA. reproduced on flight model spares using high energy proton
1804 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013
A. Optocouplers—TOPEX/poseidon
B. Bipolar Devices—GLOBALSTAR-1
Fig. 27. Locations of HX6228 SEUs on IASI—METOP [4]. The GLOBALSTAR-1 cellular phone telecommunication
constellation is composed of 48 operational satellites and 4
spares. The original deployment of the constellation took place
beams. The origin of latch-up events was traced back to CMOS in the period 1998–2000. The radiation design lifetime of
NS32C016 circuits in the CPU board. individual satellites was 9 years. As soon as 2002, some type
Finally, proton induced upsets in HX6228 rad-hard devices of on-board receivers began exhibiting failures [27]. In the
have recently been observed through anomalies on the CNES period 2002-2004, 24 receivers on-board 21 satellites were
IASI instrument on the EUMETSAT METOP polar orbiting affected, 3 satellites failed. Later on, a strategy was developed
satellite. About half the upsets are located in the SAA as shown to circumvent this failure type and force signals through the
in Fig. 27. Those upsets may be related to proton or ion induced receivers.
high-Z recoils within the device. The effect becomes observable The time-to-fail distribution over the impacted flight models
at system levels because the satellite uses approximately 130 of is shown in Fig. 28[27]. Note the large spread of this distribution
those devices as a fast memory array. from 2.7 to 6.1 years as shown in Fig. 27.
The in-flight anomaly analysis [28] proved that the failures
VI. DISPLACEMENT DAMAGE originated in the proton displacement damage of LT RH1014
bipolar devices. Proton tests exactly reproduced the anomaly on
Optoelectronic components (LEDs, photodetectors, optocou- spare models.
plers, CCDs, APSs, etc.) tend to be very sensitive to displace- The reasons for those earlier failures are:
ment damage, and so are, to a lesser extent, bipolar components. — At the time of the receiver design , assessing the
The particular case of sensors (CCDs, APSs) will be developed effect of displacement damage on bipolar devices was not
in Section VII. Also, solar cells are well known to be sensitive an engineering practice in satellite design: nobody thought
ECOFFET: OVERVIEW OF IN-ORBIT RADIATION INDUCED SPACECRAFT ANOMALIES 1805
Fig. 31. Blurring of the VIS camera on NASA POLAR satellite during july
2000 solar flare—credits NASA.
Fig. 33. Solar flare proton tracks on SOHO LASCO instrument—credits ESA.
Fig. 34. Close—up on image (A) inside the SAA—proton tracks are clearly
Fig. 32. JPL GALILEO image of Io showing particle transients, image visible [4].
PIA00593 [29]– credits JPL.
TABLE VIII
IASI SEU LOG, MAY 2011.
Fig. 37. Upset count versus time for the IASI HX6228 memories [4]. Fig. 39. Scatter plot (time between consecutive events versus time) for SPOT-2
OBC SEUs [4].
Fig. 38. Scatter plot (time between consecutive events versus time) for SPOT-1
OBC SEUs [4].
Fig. 40. SPOT-1 and-2 OBC SEU count versus time [4].
was due in the first case to a coincidence of a GCR upset fol-
lowed 5 days later by a SAA upset, and in the second case, to a In the case of SPOT OBCs, upsets were recorded either when
SAA upset followed by a GCR upset 12 days later. raising a flag on the satellite, in which case they are precisely
SPOT: Upsets were recorded on Phillips HEF4736 1 kbit dated and located, or through memory dumps. In the latter case,
SRAMs arrays (1440 devices) of SPOT-1 and-2 on board com- there is some uncertainty on the exact date of the upset, so the
puters [18]. Those upsets were due to galactic cosmic ions. scatter plots may suffer from this filtering. Because of this un-
For SPOT-1, on a series running from 1986 to 2003, time certainty, it is difficult to try to reproduce the statistics using a
between two consecutive upsets varied from 0.3 days (i.e., 3 random generator with a small time scale. We can nevertheless
upsets a day) to 432 days (see Fig. 38), where the mean upset try to simulate the process at a larger time scale, i.e., try to sim-
period was 83 days. Sample size was 74 upsets. ulate an upset count per year, for example.
For SPOT-2, on a series running from 1990 to 2007, time The mean upset period was about 90 days (one every 3
between two consecutive upsets varied from 0.5 days (i.e., 2 months). This would correspond to a probability of 1/3 of
upsets a day) to 773.5 days (see Fig. 39), where the mean upset having an upset each month. Using a random generator, we
period was 93 days. Sample size was 66 upsets. simulate, for each month of the series, upset occurrences with
If now we compare the two series of upset counts versus time a probability of 1/3. Table IX shows the total upset count per
for the two satellites, we can observe “ramps” and “gaps” in the year for SPOT1 and SPOT2.
two series as shown in Fig. 40. We will use the SPOT1 series because it has the larger spread
At the beginning (1986–1996), the upset count variations in upset count (1 to 10 per year). Fig. 41 compares the upset
seemed to reproduce the modulation of cosmic ray fluxes, which count versus date for the actual SPOT-1 series (red dots) and 30
are higher in solar minimum and lower in solar maximum. But random series per graph (total of 120 random series for the 4
even in this period, one thing remained unexplained: in 1993, graphs).
SPOT-2 ramped up sharply whereas SPOT-1 remained “quiet”. The SPOT-1 series remains inside the “statistic valley”. Let
The two satellites were exposed to the same environment on us now evaluate the deviation inside the series (SPOT-1:
the same orbit and same orbital plane. In 1997, SPOT-1 SEUs , ). The random series also can reproduce this devi-
ramped up whereas SPOT-2 remained at a moderate rate. By ation, as shown in Fig. 42.
2001–2002 (solar maximum again), both satellites had lower In conclusion: the in-flight SPOT-1 upset series is situated to-
upset rates and appeared to be quiet. wards the most “stretched” zone in yearly upset count variations
ECOFFET: OVERVIEW OF IN-ORBIT RADIATION INDUCED SPACECRAFT ANOMALIES 1809
TABLE IX
SPOT1 AND SPOT2 SEU YEARLY COUNT.
Fig. 44. Latch-up count versus time for CY7C1069 as recorded by the
CARMEN-2 experiment [4]. Fig. 46. Scatter plot of time between consecutive events versus time for the
CARMEN flight data [4].
TABLE X
MEAN, MINIMUM AND MAXIMUM TIME BETWEEN EVENTS IN UNIFORM
PROBABILITY RANDOM SIMULATIONS AND ACTUAL CARMEN IN-FLIGHT DATA.
Fig. 45. Latch-up count versus time for CY7C1069 as recorded by the
CARMEN-2 experiment emphasizing on June-August 2008 period [4].
Fig. 49. Scatter plots (time between events versus time) for the simulated runs
Fig. 47. Comparison between random simulations and actual in-flight
[4].
CARMEN latch-up count. The in-flight series exhibit a 44 day gap [4].
and gaps (low SEE rate periods), and explain larger periods be-
tween events. Using a varying probability function, because the
SEE probability is not the same along the orbit, even with a sim-
plistic model, we reproduce even the smaller periods between
events. To conclude, an observed SEE series exhibiting a mean
period of 1 event/day, with 7 days maximum time between 2
events, and 128 seconds minimum time between 2 events, can
be explained by a pure random process.
Out of 189 events, there were six occasions in which events
were separated by less than 20 min. The average probability of
having 2 SEEs in one orbital pass for a 1 day period phenom-
enon is thus about 3%.
4) Back to SPOT and IASI Cases: If we apply a simple
scaling rule between the CARMEN series and other series using
the ratio of the mean SEE periods, we could predict upset per-
formances for SPOT and IASI. We used the same mean rate for
SPOT and IASI which is conservative for the minimum time
between events for SPOT.
The applicability of this rule has to be assessed, but the
scaling should not be too bad. The SPOT upsets are due to
cosmic rays only (not SAA as for CARMEN) and there are 2
Fig. 48. Examples of simulations of event count using the non-uniform prob- passes over the polar zones per orbit, about 20 min duration
ability function used for modeling SAA passes [4]. for each pass (assuming the “polar” zone starts at 45 north
or south). In this case, the modulation is in fact twice the
events is also well reproduced and now, with the model change SAA modulation (probability spread over 40 min instead of
to a non-uniform probability distribution, the minimum time be- 20 min), so the “averaged peak” probability would be lower
tween events is also reproduced. than for a “one shot zone” as for CARMEN, and thus the
The mean, minimum and maximum time between events for minimum period would be underestimated by the scaling rule.
the sample runs are summarized in Table XI. The IASI upsets are situated half in the SAA and half in the
A scatter plot of time between actual events versus time for polar zones. Here again we remain on the conservative side
the simulated runs is shown in Fig. 48. Fig. 49 shows a side-by- for the minimum time between events (probability spread over
side comparison of the flight data and the simulated runs scatter ).
plots. On IASI, 5 days and 400 days between events have already
Summary of the Case Study: Using a coarse random con- been observed, but with so few events there is more range in the
stant probability process, we reproduce ramps (high SEE rate) mean upset rate. On SPOT-2, 700 days have been observed.
1812 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013
Fig. 50. Comparison between scatter plots (time between events versus time) for the actual flight data (left) and the simulated runs (right) [4].
On the two SPOT satellites, we have a certain number impact assessment. As a proposed methodology, we may for-
of events separated by much less than 3.6 days (see scatter mulate:
plots in VIII.A.2.b., Figs. 37 and 38 ), which behaved out of — Calculate the mean SEE rate.
family with a random event. Those points may be related to — Determine the probability function.
some other effect than SEE. Particularly, one double upset was — Determine min and max periods between events (at confi-
recorded on SPOT-1. This event took place on December 26, dence level to be chosen).
1993. The first upset corresponded to a 0 to 1 transition of bit — Or, which is the same: estimate min and max SEE rates
4 at address 023142, dated 15h 11\min 29s. The other corre- (with confidence levels).
sponded to a 1 to 0 transition of bit 10 at address 132503, dated This would help in determining whether rates are “normal” or
. These addresses refer to physically not and sort out what could be related to mere chance and what
different packages. The satellite was at a position of is abnormal i.e., may be related to environmental variations.
latitude of longitude on both events. These events Determining the “normal” domain would help in operations for
may not be SEEs. evaluating \max/min workload and also for relations with final
5) Considerations on See Test Logs: SEE ground test logs customers when “ramps” of anomalies appear: they simply
also exhibit the same ramps and gaps features that in-flight could be due to “bad luck”. Determining the “normal” domain
records do because the same statistical nature of SEE counting would also help in assessing how strongly environmental
occurs both in space and during ground tests. Fig. 50 depicts variations modify the statistics.
a sample proton latch-up test run for the CYPRESS SRAMs During the project design phase, mean SEE rates may be
flown on the CARMEN payload and used in the case study. accompanied with an evaluation of the statistical max (con-
The same statistical representation could be used to analyze fidence level) risk, especially in the case of destructive phe-
the test logs. In the case of cyclotron beams we also face the nomena when a mean predicted rate is compared with lifetime
issue of a non-uniform rate because cyclotron beams are pulsed or FITs.
beams. Other considerations associated with this statistical behavior
6) Conclusionsmdash;Methodology for See Analysis: The may be: SEE counting is sometimes used as a proxy to assess en-
statistical behavior described above (ramps and gaps) would be vironment conditions. But doing this measure through the proxy
applicable to flight series of different time scales (mean period of discrete, probabilistic phenomena will be subject to the same
1/month, 1/year, etc.). This is also the behavior of SEE count large statistical variations, so detection accuracy should be care-
logs during ground tests, with the same problem of varying fully studied. Also, SEE prediction models should of course be
event probability in case of pulsed beams. For mission opera- as accurate as possible in the determination of the mean in-flight
tion, dependability analysis, or for critical SEE effects, the cal- rate for a given environment, nevertheless the statistic “fog”
culation of a mean SEE rate may not be sufficient for system may introduce some trade-offs in the efforts to achieve this goal.
ECOFFET: OVERVIEW OF IN-ORBIT RADIATION INDUCED SPACECRAFT ANOMALIES 1813
Fig. 51. Sample ground test log of latch-up count versus time for the
CY7C1069 used in the CARMEN experiment [4].
Fig. 52. Comparison between POSEIDON/JASON-1 resets (top) and electron
flux measurements (from top to bottom: NPOES-15 5.35 MeV and 3.35 MeV,
B. Correlations and Cause To Effects Relationships SAC-C ICARE1 MeV and 0.5 MeV). Credits NOAA, CNES. IPSAT plot
CNES/ONERA [4].
Fig. 54. Location of the JASON-2 passenger incident compared with AP8 20
MeV proton flux map [4].
[11] E. Vergnault, R. Ecoffet, R. Millot, S. Duzellier, L. Guibert, J. P. [24] L. Adams, E. J. Daly, R. Harboe-Sorensen, R. Nickson, J. Haines, W.
Chabaud, and F. Cotin, “Management of radiation issues for using Schafer, M. Conrad, H. Griech, J. Merkel, T. Schwall, and R. Henneck,
commercial non-hardened parts on the integral spectrometer project,” “A verified proton induced latch-up in space,” IEEE Trans. Nucl. Sci.,
in Proc. 2000 IEEE Radiation Effects Data Workshop, pp. 68–68. vol. 39, no. 6, p. 1804, Dec. 1992.
[12] E. J. Daly, F. v. Leeuwen, H. D. R. Evans, and M. A. C. Perryman, [25] B. Johlander, R. Harboe-Sorensen, G. Olsson, and L. Bylander,
“Radiation belt and transient solar magnetospheric effects on hipparcos “Ground verification of in-orbit anomalies in the double probe electric
radiation background,” IEEE. Trans. Nucl. Sci., vol. 41, no. 6, p. 2376, field experiment on Freja,” IEEE. Trans. Nucl. Sci., vol. 43, no. 6, p.
Dec. 1994. 2767, Dec. 1996.
[13] P. D. Fieseler, S. M. Ardalan, and A. R. Frederickson, “The radiation [26] B. E. Pritchard, G. M. Swift, and A. H. Johnston, “Radiation effects
effects on Galileo spacecraft systems at Jupiter,” IEEE. Trans. Nucl. predicted, observed, and compared for spacecraft systems,” in Proc.
Sci., vol. 49, no. 6, p. 2739, Dec. 2002. 2002 Radiation Effects Data Workshop, p. 7.
[14] A. R. Frederickson, J. M. Ratliff, and G. M. Swift, “On-orbit measure- [27] B. Smith, Age Distribution of Rx Failures vs On-Orbit Time and Age
ments of JFET leakage current and its annealing as functions of dose of Constellation, Private Communication.
and bias at Jupiter,” IEEE Trans. Nucl. Sci., vol. 49, no. 6, p. 2759, Dec. [28] J.C Azarra and C. Barillot, “In-flight anomaly analysis,” Private Com-
2002. munication.
[15] G. M. Swift, G. C. Levanas, J. M. Ratliff, and A. H. Johnston, “In-flight [29] [Online]. Available: http://photojournal.jpl.nasa.gov/catalog/
annealing of displacement damage in GaAs LEDs: a Galileo story,” PIA00593.
IEEE Trans. Nucl. Sci., vol. 50, no. 6, pp. 1991–1991, Dec. 2003. [30] A. S. Kirankumar, P. N. Babu, and R. Bisht, A Study of On-Orbit
[16] [Online]. Available: http://eosweb.larc.nasa.gov/HPDOCS/misr/ Behavior of InGaAs SWIR Channel Device of IRS-1C/1D LISS-III
misr_html/darkmap.html Camera, Private Communication.
[17] [Online]. Available: http://www.ngdc.noaa.gov/stp/satellite/anomaly/ [31] P. W. Marshall, C. J. Dale, and E. A. Burke, “Proton-induced dis-
doc/anom5j.xls placement damage distributions in silicon microvolumes,” IEEE Trans.
[18] R. Ecoffet, M. Prieur, M. F. DelCastillo, S. Duzellier, and D. Falguère, Nucl. Sci., vol. 37, no. 6, pp. 1776–1783, 1990.
“Influence of the solar cycle on spot-1,-2,-3 upset rates,” IEEE Trans. [32] C. J. Dale, P. W. Marshall, and E. A. Burke, “Particle-induced spatial
Nucl. Sci., vol. 42, no. 6, p. 1983, Dec. 1995. dark current fluctuation in focal plane arrays,” IEEE Trans. Nucl. Sci.,
[19] R. Harboe-Sorensen, E. Daly, F. Teston, H. Schweitzer, R. Nartallo, P. vol. 37, no. 6, pp. 1784–1791, 1990.
Perol, F. Vandenbussche, H. Dzitko, and J. Cretolle, “Observation and [33] G. R. Hopkinson, “Space radiation effects on CCDs,” in Proc. ESA
analysis of single event effects on-board the soho satellite,” in Proc. Electron. Components Conf. ESA SP-313, 1990, p. 301.
RA Conf., Dec. 2001, p. 37. [34] G. R. Hopkinson, “Cobalt60 and proton radiation effects on large
[20] C. Poivey, J. Barth, J. McCabe, and K. LaBel, “A space weather event format, 2D, CCD arrays for an earth imaging application,” IEEE
on the microwave anisotropy probe,” in RA Workshop Proc., Dec. Trans. Nucl. Sci., vol. 39, no. 6, pp. 2018–21025, 1992.
2002, pp. 43–43. [35] G. R. Hopkinson, “Radiation-induced dark current increases in
[21] D. C. Wilkinson, S. C. Daughtridge, J. L. Stone, H. H. Sauer, and P. CCD’s,” Proc. RADECS, pp. 401–408, 1994.
Darling, “TDRS-1 single event upsets and the effect of the space envi- [36] S. Barde, R. Ecoffet, J. Costeraste, A. Meygret, and X. Hugon, “Dis-
ronment,” IEEE Trans. Nucl. Sci., vol. 38, no. 6, pp. 1708–1708, Dec. placement damage effects in InGaAs detectors: experimental results
1991. and semi-empirical model prediction,” IEEE Trans. Nucl. Sci., vol. 47,
[22] D. R. Croley, H. B. Garrett, G. B. Murphy, and T. L. Garrard, “Solar no. 6, pp. 2466–2472, Dec. 2000.
particle induced upsets in the TDRS-1 attitude control system RAM [37] F. Bezerra, E. Lorfevre, R. Ecoffet, D. Peyre, C. Binois, S. Duzellier, D.
during the solar particle events,” IEEE Trans. Nucl. Sci., vol. 42, no. 5, Falguère, T. Nuns, M. Mélotte, P. Calvel, R. Marec, N. Chatry, W. Falo,
p. 1489, Oct. 1995. and C. Deneau, “In flight observation of proton induced destructive
[23] C. M. Seidleck, K. A. LaBel, A. K. Moran, M. M. Gates, J. M. Barth, single event phenomena,” in Proc. RA Conf., Dec. 2009.
E. G. Stassinopoulos, and T. D. Gruner, “Single event effect flight data
analysis of multiple NASA spacecraft and experiments ; implications
to spacecraft electrical design,” in Proc. RA Conf., Dec. 1995, p. 581.