freeman@spar.SPAR.SLB.COM (Jay Freeman) (09/15/87)
References:
I have not been following the current discussion of DRam error rates in
detail, so excuse me if this is old hat.
Several people recently presented calculations of multi-bit DRam error
rates due to radiation based implicitly on the assumption that multi-bit
failures were composed of either (a) independent single-bit failures or (b)
nearly-adjacent non-independent single-bit failures caused by the same
high-energy particle. There is an interesting point here which probably
should not be overlooked: There is in fact a natural physical mechanism
which produces bursts of high-energy particles which can arrive nearly
simultaneously (nanoseconds) over a wide (square meters) area. This
mechanism is the "air shower", in which one whoppingly energetic cosmic ray
interacts with a nucleus in the atmosphere, to produce a handful of
energetic secondaries, each of which produces tertiaries, and so on; so that
when the whole mess reaches ground level you have effectively a "shotgun
blast" of all kinds of elementary particles.
The term "air shower" is a bit misleading, for of course, the initial
collision need not be in the air -- building structure, etc., will do nicely.
I once worked in the Cosmic Ray Lab at Caltech, during two summers when I
was an undergraduate there. We had a bunch of Wilson-type cloud chambers,
each about a half-meter square by ten or fifteen cm thick, stacked like
flapjacks, with ionization chambers and threshholding detectors to expand
the stack only when something interesting (read "high-energy") took place.
At such times, a pair of cameras would take a stereo picture of the
condensation trails for further analysis. A substantial percentage of the
events we saw -- perhaps half, if my memory serves (it may not) -- were air
showers, most of which turned the negative so solidly black that it was
impossible to distinguish individual tracks.
I seem to recall that most DRam soft errors have to do with trace
radioactive elements in the package, so it may not matter much; but
certainly, any attempt to analyze the effect of cosmic rays by describing
them as independent events with some kind of Poisson distribution, is
missing part of the picture.
(It was quite interesting to look at the pictures: I remember one stereo
pair when a *single* particle entered the top of the stack, collided with
something a little way down, produced a dense shower of secondaries, and
continued out through the bottom of the stack on an undeviated trajectory,
as though nothing had happened. The amount of energy that the shower left
behind in our detectors was 2 X 10**12 eV -- more than an erg. And all
evidence suggested that that was only a tiny portion of the incident
particle's total energy ...)
-- Jay Freeman