Przemek Klosowski wrote:
Radiation bit flips are of course a concern, but I think you
concentrate on a pretty improbable case. Why worry about this
particular bit flip when there are so many other locations that could
be changed (registers, memory, internal state) that would result in
other random application failures. If you are concerned about
radiation flips you have
to implement watchdogs or even periodic resets.
Periodic resets are a good idea (and new for me).
External watchdogs:
- consume current
- why is there an internal if an external is needed anyway?
I think every application which requires an uptime > 1 year have to take
care about bit flipping or the manufacturer of the µC has to take care
that those bit flips are as impossible as possible ( ;-) ).
And this can be done mostly via technology selection. So an application
note from TI would be necessary. Does it exist?
To put some numbers on it and give you an idea on the relative
probability of such events, we had an application in a slightly
elevated radiation field, where a 4MB memory array would get several
bit flips a week. We dealt with it by checksumming the data so that
we knew which bit flipped and flipping it back.
Yes, I know such techniques. But I bet, if you change the manufacturer
of your memory, than you would get completely different numbers! In a
former project where I participate we had problems with RAMs which
immediately turned into latch-up after 'seeing' a proton. In our
application (it was a satellite instrument) this was fatal. Other RAM
types (all others we could get) with the same size/organization did not
show the misbehaviour.
To come back to the original topic: I will write a short summary in a
new (oh no) post and try to clarify it.
Hardy
