[Cryptech Tech] ent for rosc entropy in the Xilinx FPGA

[Bernd Paysan]

Am Montag, 8. Dezember 2014, 21:44:33 schrieb Joachim Strömbergson:
> Aloha!
> 
> First results using the ent tool on entropy extracted from the rosc
> entropy source implemented in the Xilinx Spartan-6 device on the Novena.
> this is based on 62 MByte data:
> 
> 
> Entropy = 7.999998 bits per byte.
> 
> Optimum compression would reduce the size
> of this 67121200 byte file by 0 percent.
> 
> Chi square distribution for 67121200 samples is 227.52, and randomly
> would exceed this value 89.14 percent of the times.

That could be higher...

> Arithmetic mean value of data bytes is 127.5040 (127.5 = random).

That's good, and indicates a low bias.

> Monte Carlo value for Pi is 3.141248317 (error 0.01 percent).

Perfect.

> Serial correlation coefficient is 0.000081 (totally uncorrelated = 0.0).

Ok.

> Not too shabby, it seems to be not totally b0rked.
> 
> I'm doing extraktion of more data. And if anybody wants the current raw
> data to do further analysis just tell me.

Sure, where can I get it?  A histogram plot over the bytes is the thing I do 
first to check if there's something odd; from the ent results, there probably 
isn't.

I would also like to have measurements with a high, constant sample rate, e.g. 
every 16 cycles; I only need a few kilobytes of those, so you can push that 
into an on-chip RAM.  This is a measurement of the jitter, so you can see how 
often you can read the roscs.

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