Abstract
<title>Abstract</title> <p>High-speed optical interconnects are increasingly constrained by bandwidth-induced inter-symbol interference, while maximum-likelihood sequence estimation (MLSE), although theoretically optimal for channels with inter-symbol interference, becomes computationally expensive as the channel response length and modulation order increase. Here we demonstrate an Ising-MLSE receiver that reformulates MLSE as an Ising energy-minimization problem and solves the resulting sequence-search task using programmable optoelectronic dynamics. For a finite-memory inter-symbol-interference channel, the channel response and received waveform in MLSE define the sparse spin-coupling matrix and external field in the Ising model. This mapping converts trellis-based sequence detection into a structured low-energy-state search for near-MLSE detection. We implement the mapped problem on an optoelectronic Ising machine, in which digital circuitry constructs the communications problem defined Hamiltonian and an electro-optic nonlinear feedback loop performs noise-assisted physical evolution toward low-energy spin states. In an intensity-modulation/direct-detection link with 55-GHz bandwidth, the Ising-MLSE receiver enables 215 Gb/s PAM-2 and 250 Gb/s PAM-4 transmission over 500m standard single-mode fibre, with bit-error rates below the 20% hard-decision forward-error-correction threshold of 1.5×10-2. The approach reduces the evaluated sequence-search cost compared with digital MLSE, with increasing benefit for longer channel response length. These results establish optoelectronic Ising dynamics as a physical sequence-search mechanism for bandwidth-limited optical interconnects, pointing to receiver architectures in which deterministic digital processing is retained while the most expensive sequence search is offloaded to physical evolution computing.</p>