All-optical silicon microring spiking neuron

Abstract

With the rapid development of artificial intelligence and machine learning, brain-inspired neuromorphic photonics has emerged as an extremely attractive computing paradigm, promising orders-of-magnitude higher computing speed and energy efficiency compared to its electronic counterparts. Tremendous efforts have been devoted to photonic hardware implementations of mimicking the nonlinear neuron-like spiking response and the linear synapse-like weighting functionality. Here, we systematically characterize the spiking dynamics of a passive silicon microring neuron. The research of self-pulsation and excitability reveals that the silicon microring can function as an all-optical class II resonate-and-fire neuron. The typical refractory period has been successfully suppressed by configuring the pump power above the perturbation power, hence allowing the microring neuron to operate with a speed up to roughly sub-gigahertz. Additionally, temporal integration and controllable inhibition regimes are experimentally demonstrated for the first time, to the best of our knowledge. Our experimental verification is obtained with a commercial CMOS platform, hence offering great potential for large-scale neuromorphic photonics integration.