Time delay reservoir computing based on mutually coupled add-drop microring resonators
Lili Li, Yiyuan Xie, Jiang Xiao, Su Ye, Yichen Ye, Zelin Li, Yuhan Tang · Chaos Solitons & Fractals · 2025
Add-drop silicon microring resonator (MRR) has notable advantages in low power consumption, light weight and scalability, making it one of the research hotspots in optical reservoir computing (ORC) and optical chip. In this paper, for the first time, we propose a novel nonlinear dynamic system using mutually coupled (MC) add-drop MRRs with clockwise and counter-clockwise optical injection. Utilizing the proposed dynamical model, which is grounded in modified nonlinear dynamic equations incorporating coupled mode theory (CMT), we further construct an ORC system. Different dynamic behaviors and internal physical mechanisms, affected by key parameters, are analyzed in detail through bifurcation diagrams. Based on this, the effects of key parameters including injection strength, pump power, and injection delay time on the performance of ORC are detailedly analyzed in results. Through comprehensive analysis and optimization, the proposed ORC can achieve the normalized mean square error (NMSE) of 0.4% for the prediction task and the symbol error rate (SER) of 0.2% with SNR of 24 dB for nonlinear channel equalization. By analyzing the effects of the system output state, the number of virtual node, and scaling factor on the above tasks, we achieve remarkable recognition accuracies, attaining 99% on the MNIST dataset and 86.8% on the Fashion-MNIST dataset. The results and analysis underscore the importance of mastering the dynamic mechanism of the proposed model to achieve optimal application performance for constructed ORC systems. An in-depth understanding of the proposed model offers valuable insights and inspiration for the subsequent development of integrated topologies.