A Fully Reconfigurable Silicon Photonic Processor for Solving Different NP-Complete Problems

Combinatorial problems-especially nondeterministic polynomial time (NP)-complete problems, poses a formidable challenge for efficient resolution within the constraints of conventional architectures. Recent non-Von Neumann processors, implemented across various platforms such as biological motors and photonics, offer massive parallelism but often suffer from non-reusability and bulkiness. We propose a compact, fully reconfigurable integrated processor for NP-complete problems that leveraging the synergy between algorithm-driven pre-encoding and customized configuration of hardware platform. Our design features a triangular cross-bar network on a silicon photonics platform leveraging microring resonators for node state configuration. By mapping problem features onto the physical parameters of an optical system and controlling light propagation, our architecture addresses NP problems in the optical domain. We demonstrate the effectiveness of this architecture by tackling two representative NP-complete problems: the subset sum problem and the exact cover problem. Two distinct mapping algorithms that encoding the problems and the hardware is configured accordingly. A prototype with 16 layers and 17 output ports produced deterministic solutions that agree with analytical models, the measured signal-to-noise ratio exceed 10 dB. Furthermore, we discuss the potential for extending this solution to diverse platforms and monolithic integration. Our architecture presents a practical alternative for addressing NP-problems by seamlessly integrating hardware and software solutions, offering promise for deployment across various real-world scenarios.