Photonic reservoir computing approaches to nanoscale computation

Abstract

This material is based on work in progress. Reservoir computing, originally a training technique for recurrent neural networks, exploits the computation that naturally occurs in physical dynamical systems. Reservoir computing with integrated nanophotonics potentially o ers lowpower, high-bandwidth signal processing for telecommunication applications. We present our recent results for optical signal regeneration. Our simulations show that a smallscale low-power integrated photonic reservoir achieves stateof- the-art performance for regenerating optical signals that have traversed ber lengths of up to 200 km.