Enhanced photonic reservoir computing using an optically injected VCSEL with random polarized optical feedback.

Area of Science:

• Optoelectronics
• Nonlinear Dynamics
• Machine Learning Hardware

Background:

• Reservoir computing offers a novel approach to machine learning by utilizing complex dynamical systems.
• Photonic implementations, particularly those based on vertical-cavity surface-emitting lasers (VCSELs), are promising for high-speed computation.
• Enhancing parallel task processing capabilities in these systems remains a key research challenge.

Purpose of the Study:

• To propose and numerically demonstrate a photonic time-delay reservoir computing (TDRC) system.
• To investigate the impact of different optical feedback polarization states on parallel task processing performance.
• To analyze the influence of key system parameters on the TDRC's computational capabilities.

Main Methods:

• A VCSEL subjected to optical injection and random distributed optical feedback was employed as the reservoir.
• Two distinct random feedback structures were implemented: orthogonally polarized optical feedback (OPOF) and parallelly polarized optical feedback (PPOF).
• The system's performance was evaluated using benchmark tasks: chaotic time-series prediction and waveform recognition.

Main Results:

• The proposed TDRC system demonstrated enhanced parallel task processing, particularly with the PPOF structure, due to increased nonlinearity from random feedback.
• The OPOF structure showed potential performance deterioration compared to PPOF.
• The effects of injection strength, feedback strength, pump current, and virtual node count on TDRC performance were systematically analyzed.

Conclusions:

• Random distributed optical feedback in a VCSEL-based TDRC significantly enhances parallel task processing capabilities.
• The polarization state of optical feedback critically influences system performance, with PPOF outperforming OPOF.
• This research provides a pathway for improving parallel computation in VCSEL-based TDRC systems through polarization multiplexing.