Experimental Evaluation of Performance Improvement by Sparse Operation in Ising Spin Computing

The ability to quickly solve combinatorial optimization problems is essential for improving society and industry. For solving the problems, we present extraction-type majority voting logic (E-MVL) that purposely discards the interaction between the spins by scheduling a parameter, called sparsity. In this paper, the intrinsic computation time of E-MVL is estimated by using step-to-solution (STS) which evaluates the performance independent of implementation. We show that the E-MVL can explore the ground state to the Sherrington-Kirkpatrick model essentially faster than highly optimized simulated annealing (SA). These results indicate that E-MVL is more effective for optimization problems than SA.