Intel Is Investigating Server Errors Caused By Background Radiation
Two employees in Intel's Corporate Quality Network have been running a remarkable chip experiment, 2,150 feet beneath the Chihuahuan Desert in New Mexico.
Norbert Seifert and Matthew Kirsch have been testing how susceptible some of Intel's server devices are to very low-level, low-energy radiation--which can lead to tiny software-correctable errors.
These blips occur at an extraordinarily low rate (the exact figure is confidential) but for mission-critical server parts, Intel wants to keep them as rare as humanly possible.
Their experiment required that they periodically squeeze into a cramped elevator in pitch-black darkness, then descend to the bottom of a salt mine nearly half a vertical mile under the Chihuahuan Desert.
On the earth's surface, it's tough to tease apart the effects of low-level cosmic radiation (mostly neutrons) that streams in from outer space, from another kind of low-level radiation (alpha particles) that regularly emanates out of mundane stuff like your desk, your wristwatch or even a banana.
Or chip 'packaging' materials, i.e the materials that the tiny slices of silicon are placed into so the chips can be attached to a motherboard.
That's why they wanted to check it out in a radiation-free environment. So Intel got access to the US Department of Energy's Waste Isolation Pilot Plant (WIPP), a super-deep salt mine where the US Government plans to store uranium waste -- permanently. With half a mile of solid rock overhead, radiation gets neither in nor out.
Down at the bottom of the WIPP, Seifert and Kirsch hooked up a large array of 45nm SRAM devices and began collecting "soft error" data. They monitored the data from Hillsboro, Oregon for a full year.
It turns out that more than 90 percent of all ionizing particles that cause these rare soft errors in logic chips come from the "outside world"--not from the silicon itself or its own packaging.
In the mission-critical server world, this is important data.