Emulating the coherent Ising machine with a mean-field algorithm

TL;DR

This paper demonstrates that the optical component of the coherent Ising machine can be replaced with a classical mean-field algorithm, achieving similar performance but at a significantly faster speed, without relying on quantum effects.

Contribution

It shows that a noisy mean-field annealing algorithm can emulate the coherent Ising machine's performance, challenging the necessity of quantum dynamics in such systems.

Findings

Mean-field algorithm closely matches CIM performance

Classical implementation runs approximately 20 times faster

Quantum effects are not essential for the observed performance

Abstract

The coherent Ising machine is an optical processor that uses coherent laser pulses, but does not employ coherent quantum dynamics in a computational role. Core to its operation is the iterated simulation of all-to-all spin coupling via mean-field calculation in a classical FPGA coprocessor. Although it has been described as "operating at the quantum limit" and a "quantum artificial brain", interaction with the FPGA prevents the coherent Ising machine from exploiting quantum effects in its computations. Thus the question naturally arises: Can the optical portion of the coherent Ising machine be replaced with classical mean-field arithmetic? Here we answer this in the affirmative by showing that a straightforward noisy version of mean-field annealing closely matches CIM performance scaling, while running roughly 20 times faster in absolute terms.