Title:Acceleration of Stochastic Computing by Dynamically Sharing Consecutive Bit Sequences

Authors: Kensuke Takamori, Hideyuki Ichihara, Tsuyoshi Iwagaki, Tomoo Inoue

Jounrnals:IEICE Tech. Repo., vol. 115, no. 449, DC2015-89

Pages: 19-24

Published Month: 2

Published Year: 2016

Type: techreport

Abstract:
Stochastic computing, which is a computational scheme with probabilities, is notable for its applicability to error tolerant applications. In this study, we discuss a dynamic share of consecutive bit sequences, which are called stochastic numbers, in order to accelerate stochastic computing. A basic idea of dynamic share of consecutive stochastic numbers has been presented. We consider the feasibility of this dynamic share of stochastic numbers, and then propose a hardware architecture for implementing this dynamic share. Moreover, based on discussion of area efficiency of the hardware, we propose an algorithm, which is called a selection method, to determine the length of bit sharing. Experimental results show that the proposed method can accelerate stochastic computing with small hardware cost.

Reference: http://www.ieice.org/ken/paper/20160217ibGz/