Novel nondelay-based reservoir computing with a single micromechanical nonlinear resonator for high-efficiency information processing

Jie Sun,Yunfei Liu,Tianyi Zheng,Wuhao Yang,Xingyin Xiong,Zhitian Li,Zheng Wang,Xudong Zou

doi:10.1038/s41378-021-00313-7

Abstract

Reservoir computing is a potential neuromorphic paradigm for promoting future disruptive applications in the era of the Internet of Things, owing to its well-known low training cost and compatibility with hardware. It has been successfully implemented by injecting an input signal into a spatially extended reservoir of nonlinear nodes or a temporally extended reservoir of a delayed feedback system to perform temporal information processing. Here we propose a novel nondelay-based reservoir computer using only a single micromechanical resonator with hybrid nonlinear dynamics that removes the usually required delayed feedback loop. The hybrid nonlinear dynamics of the resonator comprise a transient nonlinear response, and a Duffing nonlinear response is first used for reservoir computing. Due to the richness of this nonlinearity, the usually required delayed feedback loop can be omitted. To further simplify and improve the efficiency of reservoir computing, a self-masking process is utilized in our novel reservoir computer. Specifically, we numerically and experimentally demonstrate its excellent performance, and our system achieves a high recognition accuracy of 93% on a handwritten digit recognition benchmark and a normalized mean square error of 0.051 in a nonlinear autoregressive moving average task, which reveals its memory capacity. Furthermore, it also achieves 97.17 ± 1% accuracy on an actual human motion gesture classification task constructed from a six-axis IMU sensor. These remarkable results verify the feasibility of our system and open up a new pathway for the hardware implementation of reservoir computing.

Highlights

Emerging sensor applications, such as the Internet of Things (IoT)[1] and ubiquitous sensing, require sensors with smaller size and lower power consumption, as well as “edge computing”[2] capabilities, to process a deluge of data locally
Compared with time-delayed feedback RC13, we directly serialize the input stream U(t) and feed it into the reservoir, and different degrees of nonlinear cumulative effects can be obtained by the self-masking process, which simplifies the masking procedures and improves the information processing efficiency
To further illustrate the excellent performance and reliability of the new architecture proposed here, we test the Mixed National Institute of Standards and Technology (MNIST) handwritten digit recognition benchmark by our hybrid nonlinear response (HNL)-reservoir computing (RC) system based on a single resonator, and the classification accuracy is better than that of memristorbased RC, which uses 88 memristors[24]

Summary

Introduction

Emerging sensor applications, such as the Internet of Things (IoT)[1] and ubiquitous sensing, require sensors with smaller size and lower power consumption, as well as “edge computing”[2] capabilities, to process a deluge of data locally These expanding computing requirements have motivated the creation of new and specialized computing paradigms to break through the “von Neumann bottleneck”. RC based on spatially extended nodes provides efficient parallel information processing[11,12] It suffers from the complexity of hardware implementation. Simple RC based on a time-delayed nonlinear system possessing only a single nonlinear node has been proposed It can emulate the spatially extended nodes of RC using virtual nodes temporally extended along with a delayed feedback. An initial report demonstrated the feasibility of RC with a single “delay-coupled” nonlinear microelectromechanical system (MEMS) resonator, and its best classification accuracy was only 78+2% for the TI-46 recognition benchmark

Methods

Results

Discussion

Conclusion