I agree noise diodes, neons, etc (noisy things) are the way to go if you do want this approach, but to make it work you'd probably want to play it safe and operate far below the bandwidth limit. _______________________________________________
So let's say we use that NC104 noise diode and limit the bandwidth to 100 kHz, which gives 1 mV RMS. After sampling the amplified output and processing the bits, it seems reasonable to generate 1 kbyte/s of random numbers. What data rate can be expected from the method that calculates 2*x*x-1 using analog multipliers, and how well does it pass tests for randomness? On the basis of information I've seen, my preferred way to generate hardware random numbers is to wait for the new CPUs that incorporate "Bull Mountain". -- Gene