[Cryptech Tech] Avalanche noise fascination

[Bernd Paysan]

Am Samstag, 8. November 2014, 12:08:09 schrieb Fredrik Thulin:
> Compared to my earlier use of an msp430, this indicates excellent assumed 
> accuracy. I'm saying assumed accuracy since the timing delta plot (attached
> in  two sizes, plus plot data as text) are very smooth looking to my human
> eye, but I don't have a theoretical model to match this against so who
> knows if the irregular "fallout" on the right is to be expected, is within
> probabilistic limits etc.?

Yes, the rule of statistics is "big numbers" - all events are non-
deterministic, so yo need big numbers for stable results.  On the right end, 
you have 5-15 events, and that few events are way too few to be 
representative.  It is smooth on the top part, because you have big numbers.  
The theory is that you have an equal likelyhood for a decay per unit of time, 
which results in an exponential dropoff of the lifetime (just like radioactive 
decay).

The real-world. non-ideal statistics is that after a avalance discharge, you 
have a dead time until the next decay can happen.  The amount of that (non-
deterministic) dead time depends on the current you are feeding in, the 
higher, the faster (but on the down-side, high current has its own set of 
problems).

So I think this looks pretty much perfect to me.

-- 
Bernd Paysan
"If you want it done right, you have to do it yourself"
http://bernd-paysan.de/