[SystemSafety] Difference between software reliability and astrology

[Prof. Dr. Peter Bernard Ladkin]

On 2024-08-13 16:54 , Derek M Jones wrote:
>
> Background radiation (e.g., cosmic rays) is likely to flip
> 1-bit in 4G os memory every 33 hours
> https://shape-of-code.com/2013/12/13/unreliable-cpus-and-memory-the-end-result-of-moores-law/
>
Here is something I missed.

Twenty-five years ago, Boeing shut its Radiation Lab, which was looking at these kinds of things for 
avionics, since cosmic rays are generally stronger and more prevalent at stratospheric altitudes 
than they are on or near the earth's surface. I also worked on the same floor as a renowned group of 
particle physicists. End of the nineties into the millenium. I wrote quite a lot about it on the 
now-defunct Bluecoat mailing list of aviators and aerospace engineers.

So I looked at what was known about cosmic rays and cosmic ray intensities and makeup at various 
altitudes, and talked to my physicist colleagues.

First, it was not actually known what caused bit flips and bit locks, apart from alpha particles. 
There are lots of those in space, so satellite and spacecraft designers have to take that into 
account. The thing with alpha particles, and indeed any charged particles, is you can stop the vast 
majority of them with a thin metal sheet. Boeing Rad Lab workers had posited that most difficulties 
with SRAMs at altitude were caused by neutrons. I followed the paper citation trail back for the 
source of evidence of that (a non-trivial exercise in itself). There was one experiment performed at 
Los Alamos Labs in the neutron accelerator. They put a SRAM half-in and half-out of a beam, and 
noted that the bit in the beam had more bit-flips/bit-locks than the bit outside. However, they were 
significantly asymmetric (90% of them were 0 to 1. Or was it 1 to 0? I forget. I may also have the 
percentage wrong). And this cannot be explained by any known mechanism involving neutrons.

My particle physicist pals were unconvinced. The European Physical Journal C puts out a regular 
Review of Particle Physics, and there is a booklet which accompanies that. I think it's annual. They 
gave me an older version (from 2000). There is really very little reliable knowledge about cosmic 
rays at various altitudes. It is all measured from balloons, and there aren't so many of those. 
Plus, detection is a science unto itself (how does your detector "know" what it has detected?).

The Concorde took some SRAMs on a couple of flights to see what happened. They got statistically 
significant bit-flips/locks as far as I remember, but of course nobody knows why and how, except 
there is more stuff, more energetic, up there.

There was also an incident with a railway locomotive in which a MOSFET suffered a bit lock and they 
put it down to a cosmic ray.

But the state of the knowledge at the beginning of the twenty-first century was that nobody knew how 
many bit-flips/locks to expect at what altitude or latitude, nror in which environment, nor why. I 
doubt things suddenly changed in the decade thereafter (I guess I could ask).

So to say:

> Background radiation (e.g., cosmic rays) is likely to flip
> 1-bit in 4G os memory every 33 hours 

without giving either the altitude, or the latitude, or the source of evidence,or the environment in 
which the subject stands, is highly misleading. For all of those parameters will affect any estimate 
considerably.

Furthermore, the URL doesn't load from where I am.

PBL

Prof. i.R. Dr. Peter Bernard Ladkin, Bielefeld, Germany
www.rvs-bi.de