Deep optical reservoir computing based on microring resonators for nonlinear channel equalization and image classification

Add-drop microring resonators (MRRs), with lightweight configurations, low power consumption, and high compatibility with CMOS technology, are becoming the focus of research in the field of optoelectronic integrated circuits. In this paper, an innovative nonlinear dynamic system based on cascading add-drop MRRs with optical feedback in each layer is proposed for the first time. We describe and analyze the constructed nonlinear dynamic system using the modified coupled mode theory (CMT) equations, which further construct a deep optical reservoir computing (ORC) system. By understanding the impact of key parameters on the internal physical mechanisms of the constructed dynamic system through the bifurcation diagrams, we achieve the desired system output and identify the high-performance parameter range of the deep ORC system. Through detailed analysis and discussion of the results, the proposed deep ORC system can achieve the symbol error rate (SER) of 0.06% with signal-to-noise ratio (SNR) of 24 dB for the nonlinear channel equalization task. Besides, the recognition accuracies of 99.3% and 85% are achieved in the image recognition task using the MNIST and Fashion-MNIST datasets, respectively. The results and analysis show that deep ORC system can effectively enhance the performance of related tasks. With its low power consumption and highly integrated design, the proposed deep ORC system offers strong support for subsequent diverse experiments.