Wavelength-Selective Reconfigurable Photonic Synapse in a VCSOA With Sub-Nanosecond Bio-Realistic Temporal Dynamics

By using photonic approaches to emulate the brain's neural dynamics, neuromorphic computing provides ultra-fast speeds, electromagnetic interference resilience, and wavelength-selective reconfigurability for extending the domain of artificial intelligence. However, as a critical component of neuromorphic computing, photonic synapses face some challenges in replicating bio-realistic temporal dynamics and wavelength-selective reconfigurability at sub-nanosecond operational timescales. Herein, inspired by the gain dynamics of vertical-cavity semiconductor optical amplifier (VCSOA) under optical injection, we propose a scheme to acquire wavelength-selective reconfigurable photonic synapse. By using the Fabry-Pérot (F-P) analysis approach, numerical investigation of the gain dynamics in VCSOA reveals two distinct operational regimes: gain saturation under sub-resonant wavelength injection and gain overshoot under supra-resonant wavelength injection with lower injection power. Dependent on the distinct regimes, we numerically demonstrate that a VCSOA can be reconfigured as a bio-realistic excitatory or an inhibitory synapse by switching between supra-resonant and sub-resonant wavelength injections. Furthermore, capitalizing on the reconfigurable features of the photonic synapse, we successfully simulate four variants of spike-timing-dependent plasticity (STDP) learning rules: asymmetric Hebbian STDP, asymmetric anti-Hebbian STDP, symmetric Hebbian STDP, and symmetric anti-Hebbian STDP. Notably, all bio-realistic temporal dynamics operate at sub-nanosecond timescales, surpassing biological counterparts in speed by over 7 orders of magnitude. By overcoming the limitations of existing photonic synapses in bio-realistic temporal dynamics and wavelength-selective reconfigurability at sub-nanosecond timescales, this work establishes a VCSOA-driven paradigm for ultra-fast spike dynamics processing in photonic neuromorphic computing.