Experimental demonstration of pyramidal neuron-like dynamics dominated by dendritic action potentials based on a VCSEL for all-optical XOR classification task

We experimentally and numerically demonstrate an approach to optically reproduce a pyramidal neuron-like dynamics dominated by dendritic Ca2+ action potentials (dCaAPs) based on a vertical-cavity surface-emitting laser (VCSEL) for the first time. The biological pyramidal neural dynamics dominated by dCaAPs indicates that the dendritic electrode evoked somatic spikes with current near threshold but failed to evoke (or evoked less) somatic spikes for higher current intensity. The emulating neuron-like dynamics is performed optically based on the injection locking, spiking dynamics, and damped oscillations in the optically injected VCSEL. In addition, the exclusive OR (XOR) classification task is examined in the VCSEL neuron equipped with the pyramidal neuronlike dynamics dominated by dCaAPs. Furthermore, a single spike or multiple periodic spikes are suggested to express the result of the XOR classification task for enhancing the processing rate or accuracy. The experimental and numerical results show that the XOR classification task is achieved successfully in the VCSEL neuron enabled to mimic the pyramidal neuron-like dynamics dominated by dCaAPs. This work reveals valuable pyramidal neuron-like dynamics in a VCSEL and offers a novel approach to solve XOR classification task with a fast and simple all-optical spiking neural network, and hence shows great potentials for future photonic spiking neural networks and photonic neuromorphic computing. (C) 2021 Chinese Laser Press

Automated summary

This study demonstrates the optical reproduction of pyramidal neuron-like dynamics dominated by dendritic Ca2+ action potentials using VCSEL for the first time, showing the potential for application in XOR classification tasks. The results indicate the successful emulation of complex neural dynamics in VCSEL neurons, highlighting their promising role in future photonic spiking neural networks.