Performance Enhancement Reservoir Computing System Based on Combination of VCESL Optical Feedback and Mutual Injection Structure

• Published in: IEEE journal of selected topics in quantum electronics Vol. 31; no. 3: AI/ML Integrated Opto-electronics; pp. 1 - 12
• Main Authors: Zhu, Pengjin, Wang, Hongxiang, Ji, Yuefeng
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.05.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Configuration management; Linear polarization; Nonlinear optics; Optical attenuators; Optical feedback; optical feedback and injection (OFAI); Optical polarization; Parallel processing; Performance enhancement; photonic information processing; polarization dynamics; Reservoir computing (RC); Reservoirs; Stimulated emission; Time series analysis; Topology; Vertical cavity surface emission lasers; Vertical cavity surface emitting lasers; vertical-cavity surface-emitting lasers (VCSEL); Waveforms
• ISSN: 1077-260X; 1558-4542
• DOI: 10.1109/JSTQE.2024.3480455

In this paper, a novel performance enhancement reservoir computing (RC) system based on the combination of vertical-cavity surface emitting laser (VCSEL) optical feedback and mutual injection (OFAI) structure is proposed and demonstrated numerically. By simultaneously introducing optical feedback and mutual injection structures into the proposed RC system, the nonlinear and high-dimensional mapping capabilities are significantly improved. The proposed system exhibits the best performance in both single task processing mode and parallel processing mode compared to the other 4 RC systems. Specifically, the minimum NMSE of Santa-Fe time series prediction, waveform classification and NARMA-10 task are 0.0011, 1.058<inline-formula><tex-math notation="LaTeX">\times 10^{-8}</tex-math></inline-formula> and 0.101 respectively. Furthermore, since two linear polarization modes coexist in VCSELs, the parallel-polarized and orthogonal-polarized configuration is considered. Numerical results show that in all benchmark tasks, the performance of the orthogonal-polarized configuration is generally better than the parallel-polarized configuration in single task processing mode, and the conclusion is opposite in parallel processing mode, which is related to the coupling mechanism between the two polarization modes. Finally, the effect of different parameters on the system performance is explored in detail. In summary, the proposed system is interesting and valuable in the field of high-speed and low-power neuromorphic photonics.