Abstract
Stochastic circuits re-emerge as an alternative solution to implement arithmetic functions for its ability to tolerate errors. In recent researches, there have been many solutions to implement linear functions. For non-linear cases, they transform the non-linear functions to the linear functions in a special form and handle them as the linear cases. Hence, the information of the relation between variables, which is called equivalence classes (ECs) in this paper, is lost. In this paper, we exploit the EC and present a linear programming approach to synthesize non-linear polynomial functions from a global view. A fast decomposition-based approach to extract the information of EC is also proposed. Compared to the previous works, our approach generates much smaller circuits and runs faster.