Title:On Accuracy and Area of Digital Filter Circuits Based on Stochastic Computing

Authors: Tatsuyoshi Sugino, Hideyuki Ichihara, Tsuyoshi Iwagaki, Tomoo Inoue

Jounrnals:Functional Integrated Information System

Published Month: 6

Published Year: 2015

Type: techreport

Stochastic computing, which a calculation methodology with probability, draws attention as an alter- native method of (deterministic) digital computing owing to the increase in uncertainty in circuit behavior and in soft errors in the field. In this paper, we propose a design method of digital filters based on stochastic computing; the proposed design method aims at achieving low area and high accuracy of digital filters. First, we consider a design of an efficient data input circuit (DIC), which stores an input sequence and transform it to stochastic numbers. Our design for DIC is based on the relationship between the correlation among stochastic numbers and the accuracy of calculation, so that it can reduce the area size of DICs without losing the accuracy of them. Next, we discuss an optimum design of MUX trees to maximize the calculation accuracy. By formulating the effect of the structures of MUX trees on the accuracy, we present an algorithm to construct the MUX tree that can minimize the errors based on the correlations of stochastic numbers.