Hybrid readout scheme for time delay reservoir computing using a semiconductor laser

Reservoir computing (RC) based on laser dynamics has emerged as a promising computational framework due to its simplicity and low-cost training. However, the majority is constructing virtual nodes based solely on optical intensity (OI) after photodetection, and the performance relies on the node number, which usually results in the laser reservoir with a long feedback loop. This Letter presents and investigates numerically a hybrid readout RC system based on a single semiconductor laser with optical feedback. Unlike previous reports, we here integrate both OI and terminal voltage (TV) signals to improve performance on the Santa Fe time-series prediction task for the first time, to the best of our knowledge. Firstly, simulations show that reservoirs constructed by sampling virtual nodes from TV perform comparably to those conventional approaches based solely on OI. Notably, TV-only sampling bypasses the need for photoelectric conversion, reducing the cost and complexity of the RC system. Secondly, by combining OI and TV signals to construct virtual nodes, the system effectively doubles the number of nodes without altering the physical architecture, significantly enhancing the computational performance. Thirdly, as this novel, to our knowledge, hybrid approach also offers acceptable computing performance similar to those only sampling TV or OI signals in a semiconductor laser with a feedback loop double the length, twice the original processing speed can be expected.