Experimental demonstration of reservoir computing with a silicon resonator and time multiplexing

Reservoir computing (RC) replaces the backbone of deep neural networks with the dynamics of a complex physical system in which only the output synapses are trained. Optical phenomena form a natural substrate for these architectures, while integrated optics can be used to enhance the nonlinear effects. Here, we propose and experimentally validate an all optical RC scheme based on a silicon on insulator microresonator (MR) and time multiplexing. We give proof of concept demonstrations of RC by solving two nontrivial tasks: the delayed XOR and the classification of the Iris flowers dataset. The approach could be scaled up to realize large hybrid spatio-temporal reservoirs of increased computational speed and complexity.

Automated summary

Reservoir computing replaces the backbone of deep neural networks with the dynamics of a complex physical system, training only the output synapses. An all optical RC scheme based on a silicon on insulator microresonator and time multiplexing has been proposed and experimentally validated. This approach can be scaled up to create large hybrid spatio-temporal reservoirs with increased computational speed and complexity.