SAN JOSE, Calif. The market for radiation-tolerant FPGAs is heating up as chip vendors roll out new devices intended for military and aerospace applications. But so too is the level of scrutiny they must endure to prove that their parts can work reliably under the harshest conditions.
New and established chip vendors alike are betting that makers of satellite and other radiation-sensitive equipment are ready to swap their hardwired ASICs for programmable logic. To spark interest in FPGAs, these vendors are migrating to new process technologies to boost gate counts, adding more layers of protection to prevent radiation-induced mishaps and bringing in features previously available only for commercial applications.
Recently, Xilinx Inc. (San Jose) announced that it has started shipping a new line of reprogrammable FPGAs under the QPro brand, its designation for radiation-tolerant devices. Based on the company's commercial Virtex-II family, the devices include up to 6 million system gates, embedded block RAM, multipliers, a digital clock manager and on-chip I/O termination circuitry.
Meanwhile, Actel Corp. (Mountain View, Calif.) is preparing to ship its latest family of rad-hard antifuse FPGAs, known as the RTAX-S line. Based on 0.15-micron design rules, these one-time-programmable devices carry up to 2 million system gates and more robust flip-flops to deter upsets due to ion strikes, the company said.
Among the newer entrants are Aeroflex Inc. (Plainview, N.Y.), which has been making radiation-hardened ICs for 20 years. The company says it will deliver a rad-hard FPGA based on QuickLogic Corp.'s FPGA sometime this year. Another is Atmel Corp. (San Jose), which recently announced the availability of a reprogrammable, 50,000-gate FPGA with built-in single-event-upset protection.
The more ambitious see opportunities to sell into sockets once deemed off limits to FPGAs. Among them are satellite payloads where the most sensitive and specialized electronics reside, such as image sensors. Others are content to find a place for FPGAs whenever there's an opportunity to reduce development cost or leverage their inherent programmability.
"When it comes to satellites, you don't build millions or even hundreds of thousands. You build hundreds of satellites and hundreds of systems. Unless you're driven by performance or density, ASICs become less attractive," said Anthony Jordan, director of standard products for Aeroflex.
Military and aerospace applications may not have the huge revenue potential of commercial markets, but the long-term contracts and high selling prices make them attractive. Actel, for one, derives one-third of its revenue from military and aerospace customers. And the devices fetch a premium: The RTAX-S parts from Actel start at $3,980 per unit in quantities of 25. Xilinx, meanwhile, is asking $318 to $2,655 each for the new QPro Virtex-II in 100-unit quantities.
There's also some cachet in being selected for high-profile missions. Both Xilinx and Actel point to participation in NASA's Mars rover mission as an example of their technological prowess. There are 56 Actel parts in the Mars rover, prompting chief executive officer John East to declare in a recent conference call that "we now own the much-sought-after Martian market."
But this is no place for the thin-skinned. Vendors face more scrutiny " at times, in the public domain " as the price of making highly reliable devices, and often must go to great lengths to retain customer confidence. Moreover, competitors show no mercy if any reliability concerns " real or perceived " come to the surface.
Consider Actel. Recently, the reliability of the company's 0.25-micron RTSX-A and SX-S devices have been the subject of investigation at NASA after General Dynamics, Boeing Satellite Systems and NASA's Jet Propulsion Laboratory reported some higher-than-expected incidents of manufacturing defects between 2002 and 2003.
In a March advisory, NASA said 11 SX-A and RTSX-S parts had confirmed programmed antifuse errors during testing, most of them occurring in "clusters." NASA's office of logic design identified three causes for the failures: devices that were subject to out-of-specification electrical environment; devices subjected to unknown electrical environments; and programming equipment that failed calibration tests. Actel requalified more than 3,000 devices and found no failures for those operating within specification, the report said.
The advisory concluded that engineers should follow Actel specs and conservative design practices. NASA later recommended that Actel update the data sheets to include more-prominent warnings, make the devices easier to design with, improve its programming and software reporting and subject the parts to more testing.
Since concern over the failures began to emerge, Actel has sought to reassure customers with regular updates on its investigation. Last month, senior vice president Esmat Hamdy said in a letter that the company now has a programming algorithm to protect against electrical overstress.
Dennis Kish, Actel's vice president of marketing, said he's confident that pending tests will reconfirm that the parts are reliable. "We're not really looking at what can be done to change anything. We're conducting additional experiments to add data to what exists already, and to prove that everything is high quality and high reliability," Kish said.
As it seeks to get past concerns about failures, Actel itself has tried to raise doubts about competing FPGAs based on SRAM technology " such as those from Xilinx and Altera by calling attention to the susceptibility to single-event-upset errors. Working under contract with Actel, iRoc Technologies studied the SEU rates for FPGAs and concluded that SRAM-based devices were more vulnerable to ion strikes.
Xilinx and Altera have countered that susceptibility to SEU errors has been known for years.
"Within the aerospace community, these are well-known and they can affect all integrated circuits," said Rick Padovani, general manager of Xilinx's aerospace and defense division.