[Cryptech Tech] Avalanche noise test boards

[Fredrik Thulin]

On Tuesday, August 19, 2014 10:38:01 AM Peter Gutmann wrote:
> Fredrik Thulin <fredrik at thulin.net> writes:
> >There's also some interesting documentation on the site - like test output
> >showing that they too "fail" to get a good chi square number from 'ent'
> >with just raw avalanche noise data (here:
> >http://moonbaseotago.com/onerng/theory.html).
> 
> I've talked to a few commercial vendors and many of them fail one or more of
> the more rigorous tests, which is why they all postprocess the raw noise.
> That way it doesn't matter if the noise is biased in some way, as long as
> it's unpredictable it doesn't have to be white noise since the
> postprocessing will add that.
> 
> (Think of the canonical perfect RNG, a Geiger counter fed from radioactive
> decay, it's extremely biased since most of the time it's outputting nothing,
> but also extremely unpredictable for the events it outputs).

Yes. We've had lots of discussions on-list and in person bringing everyone 
currently working on this in Cryptech to the same page (I hope =) ) regarding 
the difference between noise, entropy and random data.

For the Cryptech design, it is clear that the building block I'm currently 
working on doesn't have to output good random data, just good entropy.

I think it would be easier to build confidence with others if the entropy 
sources output happens to satisfy the various testing tools too, and it might 
also make the building block usable in other projects as well.

The building block is an avalanche based entropy source with an SPI interface, 
that currently happens to have a Raspberry Pi GPIO compatible pin header on 
it.

/Fredrik