Silicon Photonic Modulator Neuron 论文

摘要

Neural networks based on optoelectronics could be more than a million times as fast as electronic implementations, opening uncharted regimes of information processing. Despite advances in laser-based neurons and programmable Si photonic interconnects, a photonic neuron compatible with a photonic network is missing. These challenges could be addressed by photonic-modulator-based neurons that were integrable on the same platform as Si interconnects. The authors fabricate a Si photonic-modulator neuron and observe all essential networking properties of fan-in, cascadability, and high-gain nonlinearity, plus programmable, multi-input, time-resolved, and self-feedback processing behaviors.