Ising Solver using Weight Profile of Memristor Crossbar Array for Combinatorial Optimization

Abstract

In this work, a weight profile design is presented for efficient Ising solver system based on a Hopfield neural network (HNN) using 32×32 memristor crossbar array. It utilizes device noise in the probabilistic decision process of simulated annealing for binary current inputs. By implementing 0 and 1 weight matrix across various conductance states in the crossbar, we experimentally solve an unweighted max-cut problem. It is confirmed that higher noise levels, concentrated in high resistance states, enable more efficient convergence to the minimum point of the HNN energy function. This approach effectively exploits intrinsic noise, reducing external hardware overhead and demonstrating feasibility for optimization problems.