Neuromorphic Transistors Integrating Photo-Sensor, Optical Memory and Visual Synapses for Artificial Vision Application

Abstract

In commercial artificial vision system (AVS), the sensing, storage, and computing units are usually physically separated due to their architecture and performance gaps, which thus increases the volume, complexity, and energy loss. This work develops a neuromorphic transistor integrating these different modules within one single device. Leveraging the gate-tunable out-of-plane electric field, the device achieves the multi-mode integration of photo-sensor, optical memory, and visual synapse. When operating at negative top gate voltage (VTG), a strong photo-gating effect enables highly sensitive photo-response with responsivity of ≈6.515 kA W−1 and detectivity up to ≈3.92 × 1014 Jones. Due to the charge storage effect, it can also act as a non-volatile multi-level (>4 bits) optical memory with a long endurance of over 10 000 s and a high writing/erasing ratio of up to 106. At zero or positive VTG, the transistor switches to visual synapse mode with neuromorphic computing capability, providing a pathway for complex biological learning and flexible synaptic plasticity. By further combining the synaptic plasticity with an artificial neural network (ANN), it achieves precise image recognition and classification with an accuracy of up to 95.26%. This work develops a multi-mode transistor that integrates key components of an AVS, addressing the existing challenges of all-in-one integration and manufacturing complexity.