Sampling based random number generator for stochastic computing

Abstract

Linear feedback shift register (LFSR) has been widely used to generate stochastic bit streams. Although using LFSR's offers feasibility because of their compatibility with CMOS technology, lack of randomness and related area consumption which is linearly proportional to the number of bits in a stream satisfying a certain probability value, can easily go beyond practical limits. Until now, no distinguished and practical way has been found to compete with LFSR to generate stochastic bit streams. True random number generators (TRNG) are widely used to compensate the poor randomness of LFSR but their complex design which is increased by the sake of acquiring random source, and their uncontrollability to generate random bit stream with desired probability, which is necessary for stochastic applications, make them out of action. Here we propose a novel programmable sampling based stochastic number generator (SBRNG) using CMOS technology. We achieve 100x higher speed, and 640x effective length of stochastic bit streams compared to LFSR based generators. We also claim that the circuit area complexity in terms of the number of effective bits is much better for SBRNG compared to LFSR based generators.