Short-term forgetting at ultrafast timescales emulated by the depression function of a VCSOA-based photonic synapse

Abstract

Forgetting, an active optimization mechanism in the human brain, is governed by synaptic plasticity. This insight has motivated the development of artificial synapses for hardware-level emulation of forgetting. Among them, photonic synapses have drawn considerable attention owing to their ultrafast response and low crosstalk. However, short-term forgetting (STF) at ultrafast timescales remains unexplored in photonic synapses. Here, the STF at ultrafast timescales is demonstrated using a photonic synapse based on a vertical-cavity semiconductor optical amplifier (VCSOA). First, STF is realized by replicating bio-realistic paired-spike depression in a VCSOA under a paired optical spike injection, with dynamic control achieved by adjusting injection power and bias current. Subsequently, stimulating the VCSOA with multiple optical spikes produces multiple-spike depression, which can also simulate the STF at ultrafast timescales. To approximate a realistic STF scenario, a letter “T” pattern encoded by different spike numbers and time intervals is injected into the VCSOA-based photonic synapse, and a gradual forgetting trend similar to that of the human brain is clearly observed. Remarkably, all bio-inspired STF occur at nanosecond timescales, providing a tangible artificial synapse for photonic neuromorphic computing at nanosecond timescales.