It's mildly more power efficient, at least in theory, to do a single CPU
with equivalent internal redundancy like the ARM in that kit, and they
further improve things by metal packaging and their proprietary HARDSIL
process, which sounds like design guidelines for geometry and some kind of
lamination process, which reduces the amount of redundancy required to hit
their metrics.
The advantage of full lockstep is that it addresses failures in the voting
and error correction process as well, and can be scaled, so you can get
much longer lifetimes at higher rads than is possible (you'd run out of
space for internal redundancy within a single core/chip). And of course you
can use rad-hardened chips in lockstep themselves, which improves your math
further.
On Fri, Jan 18, 2019 at 11:46 AM adam paul <pauladam316@xxxxxxxxx> wrote:
That's really interesting, thanks!
What is the advantage of a traditional rad-hard CPU versus a lockstep
solution? To the best of my knowledge products like Pumpkin Space's
rad-hard board (
https://www.pumpkinspace.com/store/p199/Vorago_Rad-hard_OBC_%2F_C%26DH_Avionics_Kit.html)
Uses a traditional method. Is it more efficient as commands only need to be
run once?
Thanks,
Adam
On Fri, Jan 18, 2019 at 2:18 PM justin corwin <outlawpoet@xxxxxxxxx>
wrote:
Lots of safety critical applications use multi core or multi chip in
lockstep to prevent bit errors from causing problems, that's not
particularly new. In fact, I'm almost certain a TI Hercules processor was
used in a cubesat, which is by default, a two ARM core lockstep solution.
SpaceX may be trying to save money by lockstepping or voting cheaper CPUs
and trusting their mean time to failure calculations to cut closer to the
bone, but it's an established technique you can apply at multiple levels.
In fact, some "radiation tolerant" controllers are just FPGAs who are
running lockstep IP blocks on a low level with error correction or voting
circuits built in at a low level.
On Fri, Jan 18, 2019 at 11:09 AM adam paul <pauladam316@xxxxxxxxx> wrote:
Hi Everyone,
I was looking at the different ways that spacecraft have approached
radiation hardening for components in the past. It seems like the old way
to do things was to use components specifically designed to be radiation
tolerant, which often resulted in using components that were older and not
state-of-the-art.
However, it seems like SpaceX is using a new method of rad-hardening
Dragon by using multiple CPUs running code in parallel and comparing
results to avoid bit flips. My question is, are there other examples of
spacecraft doing this, and has it ever been tried in a cubesat? If not, why
not?
Thanks everyone,
Adam
--
Justin Corwin
outlawpoet@xxxxxxxxx
http://programmaticconquest.tumblr.com
