Developments in Secure Random Number Generators

Designs of pseudo-random number generators have been divided into two areas of applications: computer simulation and computer security. The requirement for computer simulation application is long period length and great distributional property whereas generators for computer security need to be secure with unpredictability and a nice distributional property. Most popular secure ciphers have used similar strategies: (a) applying some long key expansion scheme to initialize internal state vectors without using external generators, (b) applying a complex sequence of ARX (Addition, Rotation, XOR) transformations to diffuse the initial state vectors and outputs via additional transformations. We discuss a general framework to design a class of random number generators suitable for both computer simulation and computer security applications. The key idea is to add several good external generators to the existing popular ciphers (with minimal changes) for the purpose of better initialization and continuously updating these internal vectors with external generators.