Energy-efficient Brainware LSI Based on Stochastic Computation

Abstract

Stochastic computation has recently been studied for soft-error-resilient hardware and approximate computing, such as image processing, machine learning, and deep neural networks. This paper reviews stochastic computation and discusses the advantages and disadvantages along with the recent developments in hardware. In addition, stochastic-computation-based brainware LSIs (BLSIs) for vision information processing are introduced and discussed in terms of energy efficiency.