[Cryptography] Entropy of a diode

[Kurt Roeckx]

On Thu, Jul 21, 2016 at 03:36:03AM -0700, Bill Cox wrote:
> I have not seen any papers on this, but I can tell you that if you buy a
> regular zener diode, it will generate a disappointingly low amount of
> noise, many times less than the reverse breakdown Vbe of a cheap
> transistor.  Apparently, zener avalanche noise is highly dependent on the
> process used, and if you tune the process to lessen avalanche noise, you
> can eliminate most it.

They do use a transistor.  I didn't say anything about zener diode.
As far as I understand the transitor or just a normal diode should
have the same effect.

All the schematics and source is available.

>From what I understand, they changed the schematics because the
noise was being distored.

They have a 12 bit ADC, but it seems that because the reference
voltage of the ADC isn't filtered enough they have some bias is
the lower bits.  I think this is currently causing them to loose 1
or 2 bit of entropy.

But running some statistics on it, they still seem to get just over
8 bit of entropy per sample from the raw input data.


Kurt