Hybrid reservoir computing based on an external-cavity semiconductor ring laser

We propose and numerically validate a photonic reservoir computing system based on an external-cavity semiconductor ring laser (ECSRL). In contrast to conventional schemes that construct virtual nodes solely from the optical intensity (OI), our system creates hybrid virtual nodes by jointly sampling both the OI of the two counter-propagating modes and the terminal voltage (TV) reflecting carrier dynamics. This approach leverages the intrinsic nonlinear dynamics and dual-mode operation of a single ECSRL to enrich the reservoir state space. Through comprehensive analysis on the prediction task, we demonstrate that the hybrid-node approach based on the ECSRL effectively doubles the number of virtual nodes without altering the hardware structure, thus reducing the normalized mean square error by 34% compared to the counterpart using OI alone. Moreover, for equivalent performance targets, the proposed architecture increases the processing speed by 50% (from 1.11 GSa/s to 1.67 GSa/s). In parallel multi-task processing, while inter-mode coupling degrades performance, the combined configuration of OI and TV exhibits enhanced robustness. These findings establish the multimodal readout and fusion of laser dynamics as a viable pathway for developing high-performance photonic neuromorphic computing systems.