A VLSI-EFFICIENT TECHNIQUE FOR GENERATING MULTIPLE UNCORRELATED NOISE SOURCES AND ITS APPLICATION TO STOCHASTIC NEURAL NETWORKS

In electronic implementations of parallel stochastic learning neural networks, the technique used to generate noise for the neurons can be crucial. The noise sources must present uncorrelated noise simultaneously to all neurons in the system. A linear feedback shift register (LFSR) generates a pseudorandom bit stream with good noise-like properties, but using a separate LFSR for each neuron (to obtain uncorrelated noise) requires an unacceptable overhead for VLSI implementation. Here we present a method for generating multiple arbitrarily shifted pseudorandom bit streams from a single LFSR. Each bit stream is obtained by tapping the outputs of selected LFSR cells and feeding these tapped cell outputs through a set of exclusive-OR gates. This enables many neurons to share a single LFSR, resulting in an acceptably small overhead for VLSI implementation. This paper presents the concept, its proof, several methods for finding the cell taps, and a description of a VLSI implementation.