Measurement-Informed Multi-spin Photonic Ising Relaxation in a Single Spiral TCSEL
Physical Ising machines offer a route to solving combinatorial optimization problems through the natural relaxation of interacting states, but many photonic implementations require replicated optical units or external coupling networks. Here, we introduce SPIRA, a multi-spin photonic Ising relaxation architecture based on a single spiral transverse-cavity surface-emitting laser (TCSEL). The model is informed by measured multimode spectra, current-dependent optical output, and temperature-dependent near-field distributions of the fabricated device. These measurements provide the modal amplitudes and spatial information used to construct a 16-spin effective Hamiltonian, in which spin interactions arise from geometry-defined modal coupling and amplitude weighting, while local fields are represented through gain–loss imbalance. The TCSEL-mapped Hamiltonian achieves 87.5% spin agreement with the target ground state and a relative energy error of 6.42%. In application benchmarks, SPIRA reaches 96.90% of the optimal Max-Cut value, exactly recovers the optimum number partition, and obtains 87.50% agreement with the optimal QUBO solution, with an 18.26% objective gap. Monte Carlo analysis further shows robust near-ground-state recovery under coupling mismatch and device-level perturbations. These results establish measurement-informed multimode TCSEL dynamics as a compact platform for device-level photonic Ising relaxation.
Authors
- Abdolah Amirany (ORCID: https://orcid.org/0000-0003-0298-6945)
- Hamed Dalir (ORCID: https://orcid.org/0000-0002-9998-3830)
- Salem Altaleb
- Pritha Khurana
- Jim Guenter
- Luke Graham
- Mohammad Jooq
- Jim Tatum
- Eric Du
Institutions
- University of Florida (US)
- Quantum Devices (United States) (US)
- Leeds Beckett University (GB)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-20
- DOI
- https://doi.org/10.21203/rs.3.rs-10562696/v1
- Primary Topic
- Neural Networks and Reservoir Computing
- Type
- preprint