Delay electro-optic dynamics for brain inspired information processing

This work reports on the first experimental photonic demonstration of a neuromorphic computational unit, on the basis of a recently proposed brain-inspired paradigm typically referred as Echo State Network, Liquid State Machine, or also Reservoir Computing in the neuronal computing literature. This paradigm makes use of the computational power offered by high dimensional transient motions developed by complex nonlinear dynamical systems, when the latter are excited by the information to be processed. The originality of the proposed photonic implementation is to exploit the dynamical complexity of delay dynamics, instead of that provided by spatially extended networks of dynamical nodes (as typically proposed in the existing literature). Complex delay dynamics are indeed well known in photonics with many different practical implementations. Our results have been obtained via a hybrid optoelectronic architecture, which has been successfully used in the past in the framework of optical chaos communications. We will report on two practical implementations involving whether wavelength or intensity dynamics subject to a single nonlinear delayed feedback, or even a multiple delayed one with randomly defined weights for each delay. The computational performance is successfully tested on a benchmark test, a spoken digit recognition task, with which state of the art performances are achieved.