This I got from computer historian, Simon Lavington.
The (Manchester) Ferranti Mark I had a hardware random number generator.
This was specified by Alan Turing - (A copy of his original
Internal Report, dated 1949 I believe, still exists.) The random
number instruction was based on electron noise in a thermionic
diode. Turing's report even gives a possible circuit diagram.
The same strategy is used today in the UK Dept. of National
Savings' ERNIE Premium Bond machine.
Another curiosity of the Mark I's instruction set was a sideways
add ('population count'), also specified by Turing. I've always
assumed that the two instructions could be useful for cryptography -
eg for the generation of one-time coding pads and the testing of
decryption procedures.
