All-Optical Synapse With Directional Coupler Structure Based on Phase Change Material

With the rapid growth of the artificial neural network, the operational efficiency of von Neumann computing architecture is limited by the separation of memory and processor, and the exploration of the efficient hardware mimicking bionic neurons and synapses has become a matter of great urgency. Moreover, the circuit implements for this architecture have potential limitations such as slow switching speed, high computational power consumption, and high interconnection loss. As an alternative, neuromorphic engineering in the photonic domain has recently gained widespread international attention. In this work, we propose an all-optical synaptic device based on a directional coupler structure, which can control the degree of light field distribution variation by changing the state of the phase change material Ge2Sb2Te5 (GST) distributed in discrete islands on one port. The mode distribution and propagation of the field have been carefully analyzed with pulsed light of different power. More importantly, it is possible to flexibly design the weight update of the synaptic devices by varying the size and location of the GST islands. Verified by the Spiking Neural Network, we can improve the recognition accuracy of handwritten figures from the original 50% to 91% by improving the linearity and accuracy of the synapse weights. This may provide a new solution for future low-power non-volatile photonic integrated circuits.

Automated summary

This paper proposes an all-optical synaptic device based on a directional coupler structure, which controls the degree of light field distribution variation by changing the state of a phase change material, improving neural network performance.