Title: Actel FPGA (Field Programmable Gate Array) Presentation to NASA and Industry September 22, 2004
1Actel FPGA (Field Programmable Gate Array)
Presentation to NASA and IndustrySeptember 22,
2004
- Aerospace Actel Team
- Team Lead Larry Harzstark
2Problem Statement
- Several contractors experienced device failures
(approximately 36) after successfully programming
the FPGAs - All failures occurred within first 100 hours of
operation - All contractors were operating parts outside of
Actel specification - Multiple post-programming failures raised
potential latent failure and reliability concerns - Aerospace issued first Aerospace Alert/Advisory
in December 2003 - Aerospace hosted three industry meetings - Oct
2003, Feb 2004, Jun 2004 - Industry investigation team formed (headed by
Aerospace) to determine root cause, potential
hardware exoneration screens, long term
reliability.
3Boeing MEC FPGA Test Plan Old-Algorithm
4B125 MHz, 17 I/Os switching, 12.5 I/O toggle
rate -1V undershoot 4B250 MHz, 70 I/Os
switching 50 toggle rate -2V undershoot
Project 4B1
Project 7
550 parts
600 hrs
Project 4B2
2000 hrs
- Project 7 31 failures observed thru 600 hours
- Project 4B1 3 failures observed thru 1000 hours
- Project 4B2 2 failures observed thru 1000 hours
- Projects 4B1 and 4B2 resumed testing 9/20-21
4Boeing Project 4 Test Results
- Failures consistent with continuation of Project
7 Weibull curve - -No evidence that stress (SSU) caused any
additional failures
Test Point Units Tested 4B1 4B2 Identified Failures 4B1 4B2
0 hours 232 232 0 0
24 hours 232 232 2 0
168 hours 230 232 0 1
336 hours 230 231 0 1
500 hours 230 230 1 0
1000 hours 229 230 0 0
5Analysis of Boeing Project 7 Results
- 31 failures observed thru 600 hours of test
- Weibull FIT Rate developed based on Boeing test
results for old algorithm with 90 confidence
limits
6Failure Predictions Based on Boeing Test Results
Hours Parts Screened 1 Year Mission 3 Year Mission 10 Year Mission
100 4.9 6.6 8.8
500 3.6 5.3 7.5
7Tiger-Team MEC Parts Testing
4B250 MHz
4B2 old algo 25 deg C
70 I/Os
switching,
50 I/O
4B2 old algo 85 deg C/ Vcca3
559 parts
toggle rate
-2 V undershoot
4B2 new algo 25 deg C
Electrical test points at 0, 24, 48, 168, 500,
1000 hours
8Tiger Team MEC Parts Testing
- Current Status as of Sept 17, 2004
- Old Algorithm
- Completed 168 hours 9/17
- 500 hours to be complete 10/4
- 1000 hours to be complete 10/31
- Group A at 25 deg C had 4 failures observed thru
168 hours - Group B at 85 deg C and Vcca3.0 V had 11
failures observed thru 168 hours - New Algorithm
- Completed 168 hours 9/14
- 500 hours to be complete 9/30
- 1000 hours to be complete 10/25
- Group A at 25 deg C had 11 failures observed thru
168 hours - 5 F X antifuses
- 5 I, K S antifuses
- 1 unknown-type antifuse
9Tiger Team MEC Test Results
Algo Type No Parts Programmed No Parts Into ATE Test 0 hour 0 1 hour 1 25 hours 25 - 49 hours 49 168 hours 168 500 hours 500 1000 hours
Old Algo Group A 25oC 88 83 2/83 1/81 0/80 0/80 1/80 TBD TBD
Old Algo Group B 85oC 87 83 2/83 4/81 4/77 1/73 0/72 TBD TBD
New Algo 25oC 384 336 4/330 4/326 2/322 1/320 0/319 TBD TBD
2 units kept as controls
10Tiger Team MEC Test Results
- Unexpected failures (5) of new algorithm high
current (F X) antifuses - 2 DPAs completed at Aerospace on F X antifuses
- One device has evidence of residual photoresist
- Potentially affects all other parts in Tiger-Team
tests - Parts tested directly into P4B2 environment at 25
deg C show behavior comparable to Boeing P7 tests - No acceleration apparent
- Parts tested in 85 deg C, 3 V stress environment
does appear to show an increase in the number of
failures - Need additional analysis and investigation
11Additional Testing Planned
- Government working with Actel for formal space
qualification of UMC 0.25 um technology - Includes evaluation of physics of failure,
development of activation energy and acceleration
factors - Aerospace to begin long-term (2-3 year) life test
on UMC parts shortly - 120 plastic A54SX32A parts
- Design configured to detect functional and
timing-shift failures - All programming pulses and waveforms will be
recorded for each part and analyzed - Interim data and results will be published and
provided to industry
12Conclusions
- There is significant risk of future failure with
the MEC old-algorithm parts - Screening does not appear to be feasible
- Tiger-Team tests have not shown sufficient
reduction in failure risk to recommend use of MEC
new-algorithm parts - Wafer processing defects maybe contributing to
failures observed - UMC-parts testing by NASA to begin shortly
- All Actel results to date are promising
- ASIC alternative is appropriate if program
schedules permit