3

I'm working on a hardware random number generator based on the below design. It uses two reverse P-N junctions to create avalanche noise.

I was looking at the circuit with my oscilloscope, and I found that the unamplified noise from the two junctions seems HIGHLY correlated. (Enough entropy, however, seems to be added/amplified by the rest of the circuit so the bitstreams appear uncorrelated.)

So, is avalanche noise not the self-contained random process I thought it was, but something highly sensitive to external forces such as minor fluctuations on the rail? Or, have I’ve built a system that is amplifying something other than avalanche noise?

Duel Reverse P-N Junction circuit enter image description here

rob
  • 225
  • 2
  • 7
  • You're sure your +12 V supply is noise free? – The Photon Jan 12 '14 at 17:59
  • One way is to digitise a (large) sample of data then use the [Diehard Test Suite](http://en.wikipedia.org/wiki/Diehard_tests). Note that there are different ways of sampling the raw output from the noise generator to try and remove bias. See [here](http://holdenc.altervista.org/avalanche/) and [here](http://crypto.stackexchange.com/questions/4005/avalanche-noise-rng-for-one-time-pad-use) also. – David Jan 12 '14 at 17:59
  • I've run the combined bitstream against the NIST test suite, and it passes all tests. I'm still troubled, however, by the correlation that I can see visually on the oscilloscope. – rob Jan 12 '14 at 18:07
  • Your circuit diagram doesn't show power supply decoupling caps and possibly series inductors /resistors. Are you absolutely sure your power supply is clean? – jippie Jan 12 '14 at 19:36
  • Sorry, this diagram is incomplete. I do have decoupling caps. I'm not familiar with series inductors/resistors. Even with the caps I can see *some* noise on the line. How can one quantify the noise and at what level can the line be declared clean? – rob Jan 12 '14 at 22:24
  • I see a potential problem with CMOS latchup if microcontroler does not have latchup protection - maybe they all do now. Taking the asynchronous signal to a clocked logic system can cause latchup, which requires a power cycle to cure - provided it doesn't let the magic smoke out. +1 for the topic. I'm going to have to make one for myself. – C. Towne Springer Jan 13 '14 at 07:22

1 Answers1

2

It's your scope or your power supply. Try connecting both scope channels to 0V and see how the signals correlate and, in fact how much residual noise your scope channels have.

Next connect both channels to the 12V and see what happens.

Rest assured that there will be no correlation in the noises from two different semiconductor sources other than that caused by power supply noise (common to both).

Andy aka
  • 434,556
  • 28
  • 351
  • 777