Digital bio-inspired satisfiability solver leveraging fluctuations

Abstract

As the end of Dennard scaling is almost approaching, a number of bio-inspired algorithms have been gaining attention these days to realize new types of computing models and architectures. AmoebaSAT is one of such algorithms that can efficiently solve satisfiability (SAT) problems and optimization problems that are transformed to SAT. It is regarded as a promising algorithm, especially when its hardware architecture is deployed on the Internet-of-Things (IoT) and embedded systems applications due to its inherent parallelism and excellent solution search performance. The key feature of this algorithm is to leverage fluctuations for efficiently searching a solution. However, no prior works sufficiently explored the suitable architecture structure nor the effective fluctuation parameters to realize the hardware solver. This paper intensively studies multiple architectures to implement the AmoebaSAT hardware solver and the effects of fluctuations, which are realized by pseudo random number generators on a digital device. By exhaustively examining the combinations of fluctuation parameters, we provide insightful findings and discussions on the efficient architecture and parameters from the algorithmic perspectives.