Photonic reservoir computing for parallel and deep processing based on feedback-free cascaded semiconductor ring lasers

Reservoir computing (RC) is a hardware-friendly neuromorphic computing paradigm whose computational capability can be enhanced along two main routes, namely parallelism and depth, yet these two routes have largely been explored separately and most implementations rely on external feedback loops that hinder photonic integration. A semiconductor ring laser (SRL) naturally supports clockwise (CW) and counter-clockwise (CCW) counter-propagating modes that share the same gain medium but are only weakly coupled, thereby providing two parallel nonlinear channels within a single integration-compatible device. Motivated by this property, we propose a feedback-free cascaded SRL reservoir in which three SRLs are interconnected through unidirectional optical injection. The CW and CCW modes serve as parallel input channels, the cascaded injection provides depth, and the dual-mode intensities of all three layers are jointly read out as a single high-dimensional feature vector. A dual-channel encoding scheme further converts the input-side memory depth into a tunable hyperparameter. On the Santa Fe chaotic time series prediction and the nonlinear channel equalization benchmarks, the proposed architecture consistently outperforms the single-SRL feedback-free baseline, substantially expands the high-performance parameter region, and, through the dual-channel encoding, relaxes the fabrication tolerance of the performance-limiting front-stage inter-laser coupling, offering a viable pathway towards compact, monolithically integrated photonic neuromorphic processors that combine both parallelism and depth.