Neuro-MMI: A Hybrid Photonic-Electronic Machine Learning Platform

We demonstrate a hybrid electronic-photonic feedforward neural network which exploits interference patterns in a Multimode Interference coupler (neuro-MMI). The proposed neuro-MMIs accuracy is statistically equivalent to state-of-the-art networks. Neuro-MMI serves as a blueprint for a class of high-performance neuromorphic networks that can solve cognitive tasks.