High-Speed Compilation of Large-Scale Stochastic Circuits

Stochastic computing (SC) is executed on bitstreams which encode probabilities into the ratio between the number of one bits and the length of the stream. SC is successfully applied, for example, to image processing and beliefpropagation (BP) of low-density parity-check (LDPC) codes. With increased interest into SC, there is a gap with respect to the available scalable design automation tools needed for the fast prototyping of SC circuits. We bridge this gap by starting from the BitSAD.jl library, and implement data structures and methods to achieve orders of magnitude speed-ups for the compilation of SC circuits. We demonstrate the speedups by compiling BP decoders for LDPC codes. Our improvements enable the compilation of circuits which were previously out of reach for the tools. Our methods pave the way towards scalable design automation of SC.