If I need "real" true random numbers, then I use the Johnson noise generator, like discussed here:

http://www.8052.com/forum/read/173175

But this concept is rather slowly (10msec per bit) and needs lots of shielding and filtering.

On the other hand, I have only bad experience with zenerdiodes. The first disadvantage is that you need a high auxiliary voltage to be able to use the noisy ones. Zener diodes running at lower supply voltages need lots of amplification, so that again you have to invest in considerable filtering and shielding.

Another disadvantage is temperature dependency and stability. Some zeners seem to have a threshold temperature, above which they produce lots more of noise than below. This makes it difficult to set an appropriate gain for the amplification stages.

But the main disadvantage is the huge crest factor (peak to average factor) noisy zener diodes demonstrate to have. Individual peaks can become so high, that the amplifier clips even at rather moderate gains. Only at the lowest gains there is no clippling, with the disadvantage, that between the peaks there's not enough noise to produce random numbers. To make things even worse, there is a huge temperature dependency and drift. So, you never actual know, what your zener is doing in the next hour.

Zeners produce lots of noise ,yes, but way too much, more like a demolishing tornado. It's difficult to get a well controlled, well tempered noise signal from a zener diode. So, I tend to use the Johnson noise generator in combination with a stable metal film resistor if I need a stable noise signal.

Kai Klaas