Modeling coupled active resonators as an integrated photonic spiking neuron

Le résumé fourni par la source

Photonic spiking neural networks hold great potential for energy-efficient and high-throughput computation. While many photonic integrated circuits can implement linear synaptic interconnections, the complex dynamics of spiking neurons require careful design considerations under energy consumption, footprint, and integration constraints. Here, we propose and model an integrated photonic spiking neuron based on two coupled active resonators, leveraging the dynamics of a saturable absorber and a gain section. Using coupled-mode theory and laser rate equations, we simulate the proposed system, which, under specific conditions, resembles the excitable Yamada model and exhibits passive Q-switching. Numerical simulations confirm two fundamental spiking neuron behaviors: controllable self-pulsation from 0.1 to 0.6 GSpikes/s and optical excitability below the lasing threshold. External optical perturbations with >30 µ W peak power can excite the neuron and result in ∼300 µ W output spikes, enabling fanout to multiple neurons. Furthermore, the neuron exhibits bio-inspired characteristics: leaky integrate-and-fire dynamics and refractoriness. We discuss the necessary design rules, suitable fabrication platforms, and alternative design choices in detail to pave the way for the realization of the proposed integrated photonic spiking neuron.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.