Ran across an article in the latest EDN (May 12th, 2005) regarding the effect of Cosmic Rays on computer reliability. The problem is called SEE (Single Event Effects). Imagine your system as being a large Geiger counter in which a particle like a neutron matter interaction can temporarily cause a part of a memory or CPU to be in a state other than that caused by the program it is running.
SEEs, suh as soft errors, have since the early 1980s appeared in commercial electronics, but they are now becoming the dominant reliability-failure mechanism in modern CMOS technologies.
It goes on to say that Sun recalled some workstations in the late '90s due to this problem.
On page 50 Table A shows the susceptibility of various chip technologies to SEEs. At 90nm and below SEEs are a problem for both memory and logic circuits.
We have often noted that SW has become more and more prone to random CTDT type problems since SW2000. This parallels the widespread use of
130nm and smaller chip technologies. With all the other possible causes of SW crashing this is one of the more bizarre things that will appear on my list.Apparently it is possible to take steps to minimize the effect of SEEs. ECC memory is a start. There are apparently programming steps that can help also. Encasing a system in a couple inches of lead might also help.