Delay-embedded optical injection for enhanced photonic reservoir computing

Next generation reservoir computing (NGRC) replaces recurrent network evolution with delay coordinate features and nonlinear expansion, but these operations are usually constructed digitally. Here, we investigate delay-embedded optical injection for enhanced photonic reservoir computing (DEOI-PRC), in which the input signal modulated on a master laser is divided into multiple optical paths with different propagation delays and injected into a single reservoir laser. In this way, the reservoir laser is driven simultaneously by the current input and delayed versions of the same input, while the nonlinear response of its output optical field provides the reservoir states used for readout. This architecture uses distinct injection delays to physically supply the current and historical input states required for NGRC inspired processing. On the Santa Fe chaotic time series prediction task, DEOI-PRC improves prediction accuracy relative to conventional time-delay reservoir computing (TDRC) after parameter optimization. The optimized distribution gives the largest injection share to the shortest delay path, which carries the most recent input information, while the longer delay paths introduce historical components. DEOI-PRC broadens the low error parameter region and reduces the number of virtual nodes required for accurate prediction, supporting a compact physical implementation of photonic NGRC.