Automatic Partition for Hybrid Stochastic-Binary-Based Circuits

Stochastic computing (SC) is an emerging computing paradigm with a smaller hardware cost than conventional binary computing (BC). In this paper, we study the effects of circuit partitions that determine which parts are implemented in the SC or BC domain on several applications. We further propose an automatic partition algorithm to find the best partition meeting the accuracy and area requirements. Experimental results demonstrate the trade-off among accuracy, area, and energy using different partitions and an average of 46.53% in energy reduction of the circuits generated by the proposed partition algorithm.