This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1 MAPLD2005/1004sakaide
Evaluation of Evaluation of Actel FPGA Products Actel FPGA Products
Japan Aerospace Exploration Agency (JAXA)Japan Aerospace Exploration Agency (JAXA)22
2005 MAPLD International Conference
2 MAPLD2005/1004sakaide
Failures on programmed anti-fuse of Actel FPGA produFailures on programmed anti-fuse of Actel FPGA products which were built in the 0.25 um MEC/Tonami procescts which were built in the 0.25 um MEC/Tonami process have been reported in U.S. since 2003.s have been reported in U.S. since 2003.
While the investigation and evaluation have been perforWhile the investigation and evaluation have been performed by NASA, Industry Tiger Team (ITT) and so forth, tmed by NASA, Industry Tiger Team (ITT) and so forth, the root cause of failure is not clarified and a lot of userhe root cause of failure is not clarified and a lot of users are really concerned about the application of Actel FPs are really concerned about the application of Actel FPGAs (MEC) for flight units under the present condition.GAs (MEC) for flight units under the present condition.
Japan Aerospace Exploration Agency (JAXA) started to Japan Aerospace Exploration Agency (JAXA) started to evaluate Actel FPGA products; A54SX-A (MEC) and RTevaluate Actel FPGA products; A54SX-A (MEC) and RTSX-SU (UMC) in the end of 2004.SX-SU (UMC) in the end of 2004.
BackgroundBackground
3 MAPLD2005/1004sakaide
(1) MEC die devices(1) MEC die devices
- To determine the acceleration factors of the a- To determine the acceleration factors of the a
ntifuse failures by performing long-term life tesntifuse failures by performing long-term life tes
ts at various temperatures.ts at various temperatures.
(2) UMC die devices(2) UMC die devices
- To evaluate the reliability for space applicatio- To evaluate the reliability for space applicatio
ns by performing long-term life tests and radiatns by performing long-term life tests and radiat
ion tests.ion tests.
Test ObjectivesTest Objectives
4 MAPLD2005/1004sakaide
Test SamplesTest Samples
PartNumber
ManufacturerSample
SizeRemark
ProgramAlgorithm
A54SX32A-CQ256M
Actel Corp. 190 MEC dieOld
(ver. 4.42)
A54SX72A-CQ256M
Actel Corp. 320 MEC dieOld
(ver. 4.42)
RTSX32SU-CQ256E
Actel Corp. 110 UMC dieOriginal
(ver. 4.48)
5 MAPLD2005/1004sakaide
Test ItemTest Item ConditionConditionSample sizeSample size
A54A54 SX32ASX32A
A54A54 SX72ASX72A
RTRT SX32SUSX32SU
Operational Life Test
25 deg.C, 1MHz, 1000H 45 77 -
70 deg.C, 1MHz, 1000H 45 77 -
125 deg.C, 1MHz, 1000H 45 77 100
25 deg.C, 33MHz, 1000H 45 77 -
Temperature Cycling Test -65 to +150deg.C,1000 cycles 45 77 90
Test SystemsTest Systems Prevention of EOSPrevention of EOS
- Signals and power supplies within - Signals and power supplies within recommended operating conditions recommended operating conditions described in datasheetdescribed in datasheet
(ex. Power strip with noise filter)(ex. Power strip with noise filter)
9 MAPLD2005/1004sakaide
A54SX32A
Test Results (2): Test Results (2): Initial tInitial tPLHPLH Distribution Distribution
Initial tPLH distribution – A54SX-A (MEC, old programming algorithm)
A54SX72A
tPLH anomalies were observed on initial electrical parameter test for A54SX-A (MEC die) FPGAs
Initial tPLH distribution
0
20
40
60
80
100
120
140
130 132 134 136 138 140 142 144 >145
tPLH [ns]
Fre
quen
cy o
f O
ccur
ence
R3
R2
R1
R0
Initial tPLHdistribution
0
20
40
60
80
100
120
140
0
20
40
60
80
100
120
140
unstable
135 140 145 150
tPLH [ns]
R7
R6
R5
R4
R3
R2
R1
R0
>155
Fre
quen
cy o
f Occ
urre
nce
10 MAPLD2005/1004sakaide
Test Results (3): Test Results (3): Weibull PlotsWeibull Plots
• Weibull plots for 72A samples were successfully obtained and the
failure mode was infant mortality.
• Weibull plots for 32A samples were slightly different and and statistically poor because of small sample size.
Test Results (4):Test Results (4): Failure Rate as a Function of TimeFailure Rate as a Function of Time
• Failure rates were calculated based on the Weibull plots for 72A samples.
• The failure rates are consistent with 32A and 72A data within practical application purpose. It was considered that the difference of the failure rate was caused by lot difference because of the same structure of 32A and 72A.
Comparison of Failure Rate1E+11E+21E+31E+41E+51E+61E+71E+8110100100010000100000Time [Hour]Failure Rate [Fit(32A Equivalent)] 25C(72A)25C(32A)
Failure Rate for 72A
1E+1
1E+2
1E+3
1E+4
1E+5
1E+6
1E+7
1E+8
1 10 100 1000 10000 100000Time [Hour]
Fa
ilure
Ra
te [
Fit
(32
A E
qu
iva
len
t)] 25C
70C
125C
Comparison of Failure Rate
1E+1
1E+2
1E+3
1E+4
1E+5
1E+6
1E+7
1E+8
1 10 100 1000 10000 100000Time [Hour]
Fa
ilure
Ra
te [
Fit
(32
A E
qu
iva
len
t)] 25C(72A)
25C(32A)
12 MAPLD2005/1004sakaide
Acceleration Factor (Ea=0.002eV)
y = 0.9567e-0.0218x
1E-1
1E+0
2 2.5 3 3.5
1000/T [1000/K]
1-b
Test Results (5):Test Results (5): Acceleration FactorAcceleration Factor
Ea=0.002eV
• Temperature acceleration factor was calculated based on the Weibull plots for 72A samples
• Given activation energy was too small to screen out the defective antifuses throughout PPBI (125 deg.C, 240 hours) .
13 MAPLD2005/1004sakaide
Discussion(1):Discussion(1):Failures before life testsFailures before life tests
2/100
18/98 2/79
19/77
# of failure example:
Programming
ATE test
Loading to test board
Waveform check at R.T.
Waveform check at Specified temperature
Start of life test
Rise of Temperature
-
These failures should be included in Weibull Plot
There are several number of operations before start of the life test where defective antifuses can be failed.
It was confirmed that the failures detected in the operations should be included in Weibull plots.
1st2nd
14 MAPLD2005/1004sakaide
1.E+00
1.E+01
1.E+02
1.E+03
1.E+04
1 10 100 1000 10000
t [Hour]
Del
ta T
PD
[ns]
203(R7)
206(R1)
210(R7)
211(R1)
212(R7)
218(R2)
231(R4)
233(R3)
233(R5)
242(R7)
247(R2)
247(R4)
249(R0)
251(R2)
265(R2)
265(R4)
283(R7)
287(R4)
288(R6)
289(R4)
295(R5)
1.E+00
1.E+01
1.E+02
1.E+03
1.E+04
1 10 100 1000 10000t [Hour]
De
lta
TP
D[n
s]
349(R7)
372(R5)
390(R0)
396(R6)
424(R4)
437(R6)
444(R7)
451(R3)
462(R4)
482(R4)
499(R3)
500(R1)
1.E+00
1.E+01
1.E+02
1.E+03
1.E+04
1 10 100 1000 10000
t [Hour]
De
lta
TP
D[n
s]
306(R3)
311(R0)
327(R3)
340(R5)
345(R1)
347(R7)
350(R3)
367(R6)
404(R4)
406(R2)
421(R3)
Discussion (2):Discussion (2): Antifuse Delay Time Trend Antifuse Delay Time Trend
There were several features of the delay time trend observed with defective antifuses.
In most cases, the delay time once increased and then did not drastically changed.
It was observed that the delay time first increased and then returned close to initial value.
SX72A,25deg.C,1MHz,1000H SX72A,70deg.C,1MHz,1000H
SX72A,125deg.C,1MHz,1000H
15 MAPLD2005/1004sakaide
Discussion(3):Discussion(3):Antifuse Delay Time DistributionAntifuse Delay Time Distribution
Discussion(5): Discussion(5): Policy of JAXAPolicy of JAXA
Almost installed FPGA in JAXA satellites and rockets was performed post programmed burn-in (PPBI) . But temperature acceleration factor of this failure mode was too small to screen out the defects by PPBI. On the other hand, any defects were not observed in the evaluation test of UMC die FPGAs in JAXA.Based on these results, it was suggested that MEC die FPGAs shall be replaced UMC ones in almost JAXA projects.
18 MAPLD2005/1004sakaide
ConclusionsConclusions
• Weibull plots for the antifuse failures of A54SX-A (MEC) FPGAs were successfully obtained. The failure mode was infant mortality.
• Given temperature acceleration factor was too small to screen out the defective antifuses throughout PPBI (125deg.C 240hours)
• No defective antifuses were observed for RTSX-SU (UMC) FPGAs.
•Based on the results, the MEC die FPGAs shall be replaced with UMC ones by decision of JAXA projects.
• There was a new finding, i.e. the increased delay time distribution for failed antifuses.
•Temperature cycling tests are being performed. No defective antifuses were observed at 800 cycles for MEC die FPGAs.
•The radiation tests are also being performed for UMC die FPGAs.