Programmable Cascaded Optical Memristors Based on Phase-Change Materials for Photonic Neuromorphic Computing

Abstract

Optical memristors are indispensable building blocks for neuromorphic optical computing as they provide both information storage and programmable synaptic weights. Integration of phase-change materials (PCMs) with photonic devices represents one of the most promising approaches toward realizing optical memristors. However, achieving precise and stable optical analogue weight programming remains a major challenge. Here, we demonstrate a precisely programmable optical memristor platform based on structured PCMs for photonic neuromorphic computing. By combining optical excitation methods and structuring the Ge2Sb2Te5 (GST) films into linear memristor element arrays, we achieve independent control of the phase transition in each element, enabling the precise modulation of waveguide transmission. This platform attains high linearity up to 0.997 and an ultrahigh modulation resolution of 0.078 dB/level, which is an order of magnitude better than that previously reported. Leveraging the precisely programmable 3-element optical memristors, we further demonstrate a proof-of-concept MNIST photonic neural network application by cascading it with a wide dynamic range but low-resolution memristor. The cascaded memristor synapse can achieve a 6-bit effective resolution with a significant accuracy improvement of 15% (from 76.79 to 92.28%). The proposed platform offers a promising pathway toward high-order optoelectronic devices with potential applications in large-scale photonic neural networks for high-performance neuromorphic computing.