Briefing Summary: Independent NASA Test of RTSX-SU FPGAs
A briefing was held on February 16, 2005, to communicate the results of the Independent NASA Test of RTSX-SU FPGAs. First a brief review of the Aerospace Industry Tiger Team results on the MEC-based devices was given by The Aerospace Corporation and then the meeting moved to the primary topic, UMC-based programmed antifuse reliability.
Reliability Results: For the NASA tests, descriptions of the test equipment, vehicles, and protocols were given. It is noted that the test conditions were extreme and the stresses far higher then normal designs and application conditions. 300 RTSX32SU devices are being tested to this regime; On order of 1,000 units of RTSX32SU and RTSX72SU were tested by Actel under the less severe Tiger Team-like conditions. A total of approximately 750,000 device hours of testing has been accumulated, roughly split evenly between Actel and NASA tests. While the Tiger Team tests of MEC devices showed an 8% failure rate of the programmed antifuse, the Actel and NASA tests only showed a single device of concern; it is unknown whether that failure is related to the programmed antifuse and the device will be subjected to a destructive physical analysis to determine the cause of failure. Combining the results of the Actel and NASA tests, and assuming the worst-case result for the one propagation delay anomaly, the failure rate due to the programmed antifuse is estimated at less than 0.1%, a two order of magnitude reduction when compared to the Tiger Team results on the MEC devices with the old programming algorithm.
Test data and analysis of the radiation characterization of the unprogrammed UMC antifuse was presented and discussed. NASA and The Aerospace Corporation both presented their future reliability test plans.
Programming yield, a concern in the industry, was addressed. Statistics taken from approximately 3157 devices showed a programming yield of 95.1%. Based on these results and the testing and analysis performed by Northrop-Grumman, power line conditioning is recommended for during all programming operations.
ESD and device failures. Every anomaly encountered during the NASA test was discussed in detail. This ranged from mishaps related to the test equipment to out of family device performance, even when the device was well within specification. A trend seen by analysis of out of family devices showed a susceptibility of the I/O structures to ESD damage. This is consistent with the ESD tests performed by Actel for the recent RTSX-SU/UMC device qualification. Since the Actel ESD precautions are considered very good by aerospace community standards, it is recommend that users employ strict ESD controls for highly sensitive parts. ESD precautions at Actel are also being upgraded.
Wire bonding in these devices, with gold wires and aluminum pads, was also an industry concern. We had asked a number of organizations (JHU/APL, Lockheed-Martin, JPL, NASA GSFC) to run tests, collect data, and present their findings, in addition to our own OLD accelerated tests. The results did not show any systemic problems. It is recommended, to properly manage the risk of this structure, that lot specific testing and analysis of wire bonds be performed. Note that the next generation RTAX-S technology will employ aluminum wires and bond pads.
International View. Representatives from ESA and JAXA presented their experiences, results, and test plans.
On Wednesday February 16, 2005, a briefing on the Actel RT54SX-S and RTSX-SU devices was given at the NASA Goddard Space Flight Center in Greenbelt, MD. The briefing started at 9:30 am in the Building 8 auditorium and concluded at approximately 8:15 pm. The meeting charts are all available on-line.
This note will be a summary of the briefing, covering the meeting's highlights. It is not a formal, complete, and detailed document such as a NASA Advisory, an application note, or an OLD News release. It is strictly a summary with some opinions.
After a brief Introduction, we had a Summary of Industry Tiger Team Results, presented by Larry Harzstark of The Aerospace Corporation. While there were no fundamentally new results, this served to bring every one up to speed on the results of the RT54SX-S/MEC parts testing and give context to the primary subject of the meeting, the RT54SX-SU tests.
RT54SX-S & RTSX-SU Programming Status was presented by Dan Elftmann of the Actel Corporation. This section covered all of the latest issues with algorithms and programming software along with detailed notes as to the differences and uses of different software versions and hardware models. Note that the algorithm is how an antifuse gets programmed. The software is what controls the device and conducts the tests. First, programming yield issues were covered and the changes in software, which configure the programming adapter and device, were discussed in detail. Next, the status of the MEC "modified new algorithm" were discussed; this algorithm targets improvements in low current dynamic antifuses. Also, the detail of how the modified new algorithm would be tested were discussed, which is briefly summarized below. As can be seen, the tests will yield results comparable to the NASA tests, as much of the work done to establish them is leveraged for this reliability study.
- Test Vehicle: MEC RT54SX32S-CQ208B
- Device Pattern: NASA design as used in KU1, KU2, KM1, and KM2.
- Quantity: 300
- Duration: 2000 hours
- ATE Pull Points: 24,48,168,500,1000,1500 & 2000 hours
- Ambient/Junction Temperature: 125 °C/148 °C
- Bias Voltages (average): VCCA = 2.75V; VCCI = 4.0V
- Clock frequency: I/O = 8 MHz; Array = 32 MHz.
- I/O Undershoot = -1V (average)
NASA Test: Experiment Design
A discussion of the NASA test's experiment design was presented by Rich Katz of NASA's Office of Logic Design, enabling the audience to understand what was tested and how. The discussion started with a detailed look at the Equipment and Facilities being employed. Then, a brief review of the NASA Test Vehicle was given. A more complete description was given at the September 22, 2004 Briefing and at the 2004 MAPLD International Conference in Session A. A detailed discussion of the testing and conditions ensued along with an analysis of the differences between the Industry Tiger Team testing was given, with the Industry Tiger Team's testing being performed mostly at nominal or below nominal stress conditions, and the NASA test being conducted at much higher levels, which are summarized below:
I/O Stress: CLKIO = 8 MHz with 57 SSO's
Array Stress: CLKArray = 32 MHz (100 % toggling); Delay Line = 1 MHz.
Bias Voltages (average): VCCA = 2.75V; VCCI = 4.0V
I/O Undershoot = -1V (average)
Ambient Thermal Environment: T = 125 ºC or -55 ºC
NASA and Actel Tests: UMC Reliability Test Results and Analysis:
Actel Tests: This testing utilized on order of 1,000 units split between RTSX32SU and RTSX72SU and in general followed the test protocol employed by the Industry Tiger Team. Most tests were run at room temperature with some units being exposed to -55 ºC and +85 ºC or temperature cycling. A total of approximately 400,000 device hours have been accumulated with no programmed antifuse failures detected. This can be compared with the Industry Tiger Team data presented earlier by Larry Harzstark and in the next section by Rich Katz, where under the same conditions MEC devices exhibited programmed failures both at initial test and after just 24 hours of test at room temperature, with nominal voltage conditions, moderate speed, and virtually all outputs disabled. The section was concluded with a summary of programming yield, with a 95% yield observed for a population size of 3,157 devices.
NASA Tests: This testing utilized 300 units of RT54SX32SU devices and was tested under more severe conditions as described above in the experiment design section. First, a summary of the testing to date was given, with 9 steps completed, representing approximately 337,500 device hours. Most of the testing was performed at 125 ºC with one 250 hour segment performed at -55 ºC. The test campaign is on-going and this represented a snapshot in time. A discussion of the data products received was given along with the data analysis techniques employed. For review and proper context, results from the Industry Tiger Team were presented, a subset of the slides from the September 22, 2004 Briefing. Then sample data sets from the NASA test were presented, along with the analysis, showing the lack of programmed antifuse damage that was prevalent in the Industry Tiger Team RT54SX-S/MEC tests. A brief overview of all anomalies were given (more details below) and it was noted that parts are identified as anomalous under strict conditions; the criteria is tighter than pass/fail per the device specification. A White Paper on Definitions for and Approach to Anomaly Handling has been written detailing this philosophy. One propagation delay anomaly was detected; the root cause is currently unknown and may or may not be antifuse related. Preparation for physical analysis continues. It is noted that was detected a significant amount of damage to the I/O area of the devices -- where there are no antifuses -- which is a concern, particularly considering the low ESD rating of these devices, which is covered in detail below. Thus, the preliminary recommendation is to both minimize unnecessary handling of the devices and maximize, to the extent practical, the ESD precautions taken in the laboratory. The section included charts showing the observed differences between propagation delay between the MEC and UMC devices utilized in these reliability tests, along with their stability. The results from the NASA tests will be updated and are available on-line.
Dan Elftmann presented a detailed discussion of every anomaly detected during this test campaign. Each device with a detected anomaly is being analyzed or tracked with formal reports being written.
Dennis Dowden of Northrop Grumman Space Technology was charted to provide an independent reliability assessment of the test campaign and results. He posed the question, "What needs to be done to retire risk for space application?" and then presented a spread sheet calculator his group designed as a tool to be used in answering that question. The spreadsheet assumed a constant failure rate and bounds EA from 0.2 – 1.6 and concluded that for an EA of 0.4 and an acceleration factor of 25 for undershoot a useable FIT rate has been achieved. Recommendations are to have strict ESD controls, careful application of the technology, and utilize power conditioning equipment during programming. It is noted that Northrop Grumman's power line conditioning equipment was employed for all NASA tests.
ESD Results and Analysis
As mentioned previously, the ESD rating of these devices is low, approximately 75 volts, and a number of anomalies are likely the victims of ESD. A detailed reporting of ESD test results and analysis was given by Dan Elftmann. The presentation started off with the different ESD models, described how the tests were conducted, the test results, schematics of the damaged areas, and SEM images of damaged locations. The analysis concluded that one circuit area was common to the failures, independent of whether military or JEDEC standards were used. This area involved the transistors responsible for the 5V CMOS input threshold.
Wire Bonding Results
A number of groups presented their results on wire bonding reliability for these devices. Both MEC and UMC FPGAs of this family, the RT54SX-S and RTSX-SU, employ aluminum pads and gold wire. Testing continues and the results presented represent a snapshot. The presentations were as follows:
- Wire Bonding Experiment, Rich Katz, NASA Office of Logic Design
- Gold-Aluminum Intermetallics in Flight Actel RT54SX72S Programmable Gate Arrays, Bruce M. Romenesko, John Hopkins University, Applied Physics Laboratory
- Status of JPL Usage of Actel RTSX-SU/S FPGAs, Douglas Sheldon, Jet Propulsion Laboratory
- Wire Bonding and Recently Examined Actel FPGA Devices, Henning W. Leidecker, NASA Goddard Space Flight Center
In general, the results were considered satisfactory. However, there were some exceptions and it is recommended that appropriate tests be conducted on a per lot basis, when lot-specific data is not available. The test data showed that data from lots "close in time," differing by only a few weeks, can not be extrapolated to other lots.
The International View
There was good international participation in this Briefing, both as attendees and as contributing participants. I personally would like to thank our Japanese and European colleagues for putting in the time and effort to travel to NASA for the briefing. NASA's counterparts in Japan, JAXA, and in Europe, the European Space Agency (ESA), both contributed their results, evaluations, and recommendations.
- ESA’s View & Strategy Regarding Actel RTSX-S Antifuse Reliability Problem, Agustin Fernandez-Leon, European Space Agency
- Evaluation of ACTEL FPGA Products by JAXA, Norio Nemoto, Japan Aerospace Exploration Agency (JAXA)
Radiation results on the RTSX-SU devices were discussed, with the discussion concentrating on unprogrammed antifuse reliability in the heavy ion environment. References to total dose test reports written by Dr. J.J. Wang were given. The results of the heavy ion test trips (April 2004 and November 2004) were discussed along with test methods. The November 2004 test runs objectives were covered in detail:
- Conduct additional voltage margin tests using Bromine
- Test with multiple roll and tilt angles.
- Test with different patterns
- SEE pattern used in April was simple, optimized for SEE evaluation
- Added aggressive application-type pattern that utilized close to 100% of the device
- Test with multiple lots of RTSX-SU
- Test with multiple products
- A54SX-A/UMC (no peak current limiting resistor)
- Add additional runs to increase data base and statistics
Testing to the point of failure was conducted with the variables including angle of incidence (in two degrees of freedom), voltage bias, and ion. The margins observed were presented and discussed with no significant difference observed from the MEC product. The structural differences in the antifuse could be seen with the MEC unprogrammed antifuse being most susceptible to damage at normal incidence and the UMC unprogrammed antifuse being most susceptible at angle. Additional tests will be conducted at the University of Texas.
Device Modification, Other Tests, and Analyses
Revision B of the RTSX-SU products are now being qualified by Actel Corporation. The intent of this die revision is to improve timing margins on the "–1" speed grade devices and to reduce tri-state leakage currents. The structures being changes were discussed along with the qualification test plans and results to date. No changes are planned for the electrical test specification, the datasheet, or the Standard Military Drawing.
The Aerospace Corporation is conducting what they call the "Long Term Life Test" with the following major objectives and methods:
- To assure that there is no underlying low activation energy antifuse failure mechanism
- Tests run with little or no electrical stress
- Record all programming waveforms using high speed digitizer for one-half of the population
- Test conducted at as high a temperature as possible in order to activate any low EA mechanisms (EA ~ 0.3-0.5 eV)
- Test conditions:
- Sample size 240 A54SX72APQ208I/UMC FPGAs (industrial grade, plastic, unscreened)
- 85 ºC ambient
- VCCA = 2.5V
- Test time 2 – 3 years (2.5 x 104 hours)
- Continuous in situ readout of functionality and timing delays
- Note that for the A54SX-A/UMC series parts:
- A54SX-SU/UMC are similar to the spaceflight RTSX-SU devices, but not the same, with respect to antifuse reliability
- A54SX-SU/UMC has a lower programming current then the RTSX-SU
- A54SX-SU/UMC does not contain the current limiting resistors that are present in the RTSX-SU, resulting in higher peak antifuse currents
- These factors lead to a higher programmed antifuse stress level in the A54SX-A/UMC devices than the RTSX-SU spaceflight FPGAs.
This test began on January 5, 2005, on 20 parts, with an additional 20 parts added per week until all 240 parts are integrated into the test. At the time of the briefing (February 16, 2005), 1 device has shown a drift in timing of approximately 50 nsec. The root cause is unknown, it may or may not be an antifuse related problem, and The Aerospace Corporation is continuing to test the device and observe any trends.
Additional NASA tests are planned utilizing both A54SX32A and A54SX72A FPGAs produced at UMC. As is being done in The Aerospace Corporation "Long Term Life Test," the devices will be industrial grade and packaged in plastic. The objective of this test is to test with large numbers of devices for greater statistics significance and with a higher stress on the antifuse, as the commercial, industrial, and military grade products do not employ the current limiting resistors present in the space grade products.The Aerospace Corporation Space Qualification was presented by Larry Harzstark. This program will employ 300 RTSX32SU and 150 RTSX72SU FPGAs.
Henning Leidecker presented a reliability assessment of Actel Devices in three NASA GSFC programs. The analysis considered factors such as number of devices, the amount of testing, reliability fits and models, and redundancy. Henning concluded that to fit using any single distribution (such as a Weibull) suggests that all the devices are subject to the same statistical behavior. His position is that there is some reason to believe that the anomalous devices are a sub-population and should be fit using one distribution, while the remaining devices should be fit using a distinct distribution. A brief review was also given about the magnitudes of changes in propagation delays and user circuit sensitivities to such changes.
Last Revised: February 03, 2010
Web Grunt: Richard Katz