Optoelectronic Synapses Based on MoS2 Transistors for Accurate Image Recognition

Abstract Neuromorphic chips show advantages over conventional computing system based on von Neumann architecture when dealing with complex scenes, such as artificial intelligence, automatic driving, image, and speech recognition. However, current neuromorphic circuitry usually suffers from device complexity, low computing speed, and efficiency issues. Here, the fabrication of artificial optoelectronic synapses based on MoS 2 field‐effect transistors (FETs) to integrate the sensing, memorizing, and preprocessing functions together is reported. Light‐tunable and gate‐voltage modulated behaviors enable those devices to implement many opto‐synaptic functions, such as paired‐pulse facilitation (PPF), short‐term memory (STM), long‐term memory (LTM), the transition of STM‐to‐LTM, and interest‐modulate human visual memory. Moreover, background noise filtering for image preprocessing with a high recognition accuracy ≈85.5% is realized in MoS 2 ‐based optoelectronic array devices. These simple but prospective opto‐synaptic devices provide an alternative approach to construct neuromorphic visual systems.