Optical Echo State Network Reservoir Computing

TL;DR

The paper tackles the energy and latency bottlenecks of recurrent networks by proposing an all-optical ESN with fixed reservoir dynamics and a trainable readout. It advances a dual-SBS scheme to realize both linear matrix-vector multiplication and a tunable nonlinear activation in-fiber, leveraging dBm-based optical arithmetic and a pump-splitting strategy to tailor the activation function. Through simulations on Sine-Square Classification, NARMA10, Mackey-Glass, and noise scenarios, the optical ESN demonstrates performance approaching that of software ESNs of comparable size, while offering substantial gains in speed and energy efficiency. The work argues for the practicality of on-chip optical reservoir computing and outlines the design choices that balance universality, scalability, and hardware constraints for future photonic implementations.

Abstract

We propose an innovative design for an optical Echo State Network (ESN), an advanced type of reservoir computer known for its universal computational capabilities. Our design enables an optical implementation of arbitrary ESNs, featuring flexibility in optical matrix multiplication and nonlinear activation. Leveraging the nonlinear characteristics of stimulated Brillouin scattering (SBS), the architecture efficiently realizes measurement-free nonlinear activation. The approach significantly reduces computational overhead and energy consumption compared to traditional software-based methods. Comprehensive simulations validate the system's memory capacity, nonlinear processing strength, and polynomial algebra capabilities, showcasing performance comparable to software ESNs across key benchmark tasks. Our design establishes a feasible, scalable, and universally applicable framework for optical reservoir computing, suitable for diverse machine learning applications.

Figures (11)

• Figure 1: (a) SBS operating in large pump (linear) regime. Input-output relationship is shown for various pump powers to demonstrate tunability. (b) SBS operating in self-pumped (nonlinear) regime. The nonlinear shape can be designed by choosing an appropriate $gL$ value. (c) Sigmoid function for comparison.
• Figure 2: Optical ESN Design
• Figure 3: Comparison of the performance of the Sine-Square Classification test. Parameters: $N=10$, $gL=12$, $m=1500$
• Figure 4: NMSE vs activation function parameters for the sine square classification test.
• Figure 5: Comparison of the performance of the NARMA10 test. Parameters: $N=10$, $gL=12$, $m=1500$