Denver, sunspots, and the hazards of SRAM

Wow, Mike Brogley of Actel took a lot of reader flack at Embedded.com for an intriguing article on sunspot influence on SRAM-based FPGAs. Now, before you think that this has something to do with crystal healing, take a careful look at the article in question. Since Actel is well-known for its rad-tolerant FPGAs in space and military applications, Brogley thought he’d look at the effects of sunspot activity at high latitude and high altitudes, and the potential for flipped bits in SRAM reliability.  Designing systems during the present "quiet sun" period might lead to problems during solar maxima, he concluded.

Readers took him to task for not identifying the competitive architectures that might provide problems (three guesses), and suggested he was shouting “Fire” about a trivial problem. I’ll give this article a reserved thumbs-up, but point out the resemblance it bears to radiation studies of the late 1970s by anti-nuclear activists like John Gofman. The comparison is apt in particular when Brogley suggests that SRAM-based FPGA reliability might be different at different locations on the Earth’s surface.  The effect might be statistically significant in rare cases, but it’s not likely that a real-world user of such FPGAs in Colorado, or even in Tibet, would see it.

We ought to thank Brogley for pointing out the greater neutron flux density encountered during 11-year solar maxima, and the potential impact this could have near the poles and at high altitudes. Because the potential for single-event-upset soft errors in military systems can be large, designers of FPGAs in mil-aero and space-based military systems have to pay close attention to this.

But, as Brogley admits, nonvolatile workarounds already have been offered in some FPGA architectures. Sure, Actel should be proud to tout its own architectural advantages. As feature sizes decline, the potential problems of standard SRAM-based FPGAs will increase.

Readers made good points about a potential “crying wolf” problem, particularly in not listing the time over which failures were encountered, and failure to examine potential failures in FPGAs vs. ASICs, main memory, microprocessors, etc. Do we have a problem, or not?

I liken it to my own choice to live at 7700 feet in altitude. Do I face a minutely increased risk of cancer due to greater ionizing radiation at high altitude? Of course I do. Do I care? I consider it a great tradeoff, given the natural surroundings. Sometimes, when designing mission-critical systems, we have to care about the failure mechanisms described by Brogley. But a commercial system in Denver or Kiruna is hardly likely to experience more failures than those in Miami or Rio, solely due to a sunspot cycle.