In-Stream Correlation-Based Division and Bit-Inserting Square Root in Stochastic Computing

Editor’s notes:
This article presents improved stochastic computing primitives for division and square root operations. Both are nonlinear functions that cannot be reduced to additions and multiplications. The authors make use of the very correlations that are usually considered undesirable in stochastic computing; their controlled injection into computations leads to good compromises between convergence time and area requirements. —
Weikang Qian, Shanghai Jiao Tong University