A New Technique for Stochastic Division in Unipolar Format
Nikos Temenos, Paul Peter Sotiriadis · 2019
Stochastic Computing (SC) is an alternative designing technique where signals are processed non-deterministically. Apart from low-area occupation and greatly reduced power dissipation, SC is inherently fault tolerant due to its probabilistic nature. Considering the aforementioned, SC has regained attention in system design where traditional Digital Signal Processing (DSP) cores require demanding number of resources to perform operations or are sensitive to soft-errors. However, specific operations are considered challenging for implementation due to the fact that processing elements are incapable of constructing non-linear functions such as the division. In this work we propose a new architecture that performs stochastic division in unipolar format that produces direct stochastic output and lowers the hardware requirements. Simulation results of Normalised Mean Root Squared Errors (NRMSE) are provided in order to demonstrate the accuracy of the proposed architecture as well as the simplicity for implementation.