All-optically modulated PDVT-10/IGZO heterojunction synapses for neuromorphic applications

Abstract

The development of neuromorphic visual systems aims to address the constraints in energy efficiency and stability within machine vision. However, neuromorphic photonic devices mostly encode hybrid optical–electrical signals or adjust the optical response through electrical bias, resulting in limited biological fidelity. Herein, a retina-inspired all-optical PDVT-10/IGZO heterojunction synapse with superior photoresponse tunability is proposed. Leveraging wavelength-dependent programming with 340 nm light for potentiation and 530 nm light for depression, the device functionally emulates bidirectional synaptic plasticity and multiple optical logic operations (i.e., “OR”, “AND”, “NOR”, and “NAND”). This configuration yields an optical conductance tuning ratio of 8.2 and retains stable performance even after 9 months in the atmospheric environment. The light-induced mechanism can be attributed to the ionization and neutralization of oxygen vacancies within the IGZO layer. Such an all-optical synapse is further validated by integration with artificial neural networks, achieving a recognition accuracy of 97.4% in handwritten digit classification and demonstrating effective feature enhancement in image-denoising tasks. This bio-inspired design will endow machine vision systems with high biological fidelity, high energy-efficiency, and fully photonic operation.