Dammann-grating-enabled parallel-detection-based spatial photonic Ising machine for efficient combinatorial optimization

Photonic Ising machines are specialized computing architectures with potential applications in combinatorial optimization. However, spin scale and detection accuracy remain important challenges for spatial photonic Ising machines. To address the optical readout stability issue, a parallel-detection-based spatial photonic Ising machine (PD-SPIM) is proposed. The system encodes binary spins into the phase pixels of a spatial light modulator and uses a 3×3 Dammann grating together with a nine-channel fiber-coupled photodetector array to acquire replicated far-field intensity readouts. Multi-channel averaging can suppress the measured intensity fluctuations and improve the optical feedback stability. The multi-channel detection configuration can achieve an optical signal-to-noise ratio of 50.47 dB, which is 13.6 dB higher than the single-channel configuration under the same measurement condition. An adaptive cluster annealing algorithm is also adopted to accelerate convergence in the tested cases, reducing the number of iterations by 66.7% compared with the single-spin annealing case. The system is further demonstrated on a 512×512 active encoded spin-pattern array for the number partitioning problem, yielding a relative partition error of 7.61%. These results demonstrate the potential of Dammann-grating-enabled parallel detection for improving readout stability in large-scale spatial photonic Ising machines.