LAUSR

Stochastic computing (SC) is an unconventional computing paradigm which enables low power and massive parallelism in various applications. The numerical values in SC are represented as random bit-streams and interpreted as probabilities. The SC design provides the low-cost arithmetic units, while suffering from long latency and accuracy degradation. The division operation is the most challenging in SC. Chen et al. have recently developed a stochastic divider, called correlated divider (CORDIV), by exploiting the conditional probability. It has been proved that the CORDIV has a lower hardware cost than the previous designs and provides good accuracy performance. This brief presents a new stochastic divider, which includes one saturating subtractor and one JK flip flop. The subtractor is designed by taking advantage of correlation among inputs. Simulation results reveal that the proposed divider has the same accuracy performance as CORDIV. However, the new divider has competitive advantages over CORDIV. With applications to the piecewise linear function implementation, the proposed design is superior to the finite state machine (FSM)-based method in terms of accuracy and hardware cost.