Advancing optoelectronic reservoir computing: Enhancing performance through ultrafast neuromorphic hardware technologies

Abstract

Reservoir computing is a neuromorphic architecture based on artificial neural networks. It has gathered significant attention due to its simplicity and efficiency in processing complex sequential data for real-world tasks. We propose an advanced optoelectronic reservoir computing system that uses a single nonlinear node comprised of a Mach-Zehnder interferometer, an optical delay line, and several high-bandwidth integrated optoelectronic components. This system shows efficient performance on benchmark tasks such as signal recognition with an accuracy of 100%, nonlinear channel equalization for generating reconstructed signals with symbol error rates of 10−55, and time-series predictions that reach normalized mean square errors in the order of 10−2.