Hardware Implementation of Intrinsic Plasticity Learning in Photonic Spiking Neural Networks Based on DFB-SA Laser

Intrinsic plasticity plays a critical role in biological learning and memory. In experiments, we observed that the distributed feedback lasers with a saturable absorber exhibit intrinsic plasticity, allowing modulation of output spike delay by adjusting the gain region bias current. Utilizing this property, we propose a novel delay-current modulation scheme to implement the delay learning remote supervised method (DW-ReSuMe). Applied to the spike sequence learning task and the Iris dataset classification task, this method achieved successful learning of specific spike sequences and classification accuracies of 94.67% and 84.00% in simulations and experiments, respectively. Compared to the previous delay-weight modulation implementation of the DW-ReSuMe algorithm, this new method simplifies the hardware architecture in photonic spiking neural networks and shows significant potential for scaling these networks to larger, more complex systems.