Speeding up MUX-FSM based Stochastic Computing for On-device Neural Networks
Jongsung Kanga and Taewhan Kimb
Dept. of Electrical and Computer Engineering Seoul National University, Korea
ajskang@snucad.snu.ac.kr
btkim@snucad.snu.ac.kr
ABSTRACT
We propose an acceleration technique for processing multiplication operations using stochastic computing (SC) in ondevice neural networks. Recently, MUX-FSM based SCs, which employ a MUX controlled by an FSM to generate a bit stream for a multiplication operation, considerably reduces the processing time of MAC operations over the traditional stochastic number generator based SC. Nevertheless, the existing MUX-FSM based SCs still do not meet the multiplication processing time required for a wide adoption of on-device neural networks in practice even though it offers a very economical hardware implementation. In this respect, this work proposes a solution to the problem of speeding up the conventional MUX-FSM based SCs. Precisely, we analyze the bit counting pattern produced by MUX-FSM and replace the counting redundancy by shift operation, resulting in shortening the length of the required bit sequence significantly, together with analytically formulating the amount of computation cycles. Through experiments, it is shown that our enhanced SC technique is able to reduce the processing time by 44.1% on average over the conventional MUX-FSM based SCs.
Keywords: Convolutional Neural Networks, Stochastic Computing.
