Dammann grating–enabled spatial-temporal photonic Ising machine for large-scale combinatorial optimization problems
Phys. Rev. Applied 25, 054031 – Published 12 May, 2026
DOI: https://doi.org/10.1103/hc5s-pwcd
Abstract
Photonic Ising machines represent an emerging computational paradigm for solving combinatorial optimization problems, characterized by ultralow power consumption and accelerated processing capabilities. Spatial photonic Ising machines (SPIMs), in particular, have attracted substantial research interest due to their native architectural scalability and parallel processing topology. However, existing spatial photonic Ising machines employing simulated annealing (SA) face challenges in attaining optimal or near-optimal solutions for large-scale problems due to the complexity of the energy landscape for the Ising models. This paper proposes a Dammann grating–enabled spatial-temporal photonic Ising machine for gradient-inspired parallel simulated annealing with tunable spin-cluster size. Using this novel architecture, we experimentally generated a near-optimal solution for the Max-Cut problem on a 1600-node weighted random graph. Compared with an SA-based SPIM, our codesigned photonic framework achieves a 20-fold acceleration in convergence speed and a 2.6-fold improvement in cut value after 400 iterations. Our proposed system effectively unleashes the scalability and parallel processing capabilities of spatial optical computing, enabling accelerated optimization and improved solution quality, thus demonstrating the potential for superior efficiency for large-scale combinatorial optimization problems.
