Time Series Processing with VCSEL-Based Reservoir Computer

Abstract

Reservoir computing architectures offer important benefits for the implementation of a neural network in a physical medium, as the weighted interconnections between the internal nodes are random and fixed. Experimental results on a time-delay photonic reservoir computer based on directly modulated Vertical Cavity Surface Emitting Lasers and multi-mode fiber couplers are presented. The neuron is made of photodiode, non-linear amplifier and laser chips. The NARMA10 chaotic time-series task is performed with a configuration having 25 virtual nodes operating at 1 GS/s. Experimental and simulated error ranges are in good agreement, which is promising for an expansion to a more elaborate system. The potential of this scheme for the realization of a photonic reservoir cluster device operating at very high speed with low power and a small footprint with a large number of interacting physical and virtual neurons is discussed.