Tunable optoelectronic response in van der Waals heterojunction transistors for artificial visual recognition

Abstract

Optoelectronic synaptic transistors are advantageous in in-memory light sensing for artificial neural networks. Herein, optoelectronic synaptic junction field-effect transistors (JFETs) based on a Ga2O3/MoS2 heterojunction are fabricated. The devices exhibit robust electrical performances, including a high on/off ratio of 108, a low subthreshold swing of 69 mV dec−1, and a high output current of 3.4 μA μm−1. An inverter and a NAND gate are constructed based on the dual-gated configuration, with the inverter showing a high voltage gain of 28 and the near-ideal noise margin of 90.4%. Additionally, the devices demonstrate outstanding optoelectronic performances benefiting from the strong light–matter interactions of MoS2. Typical synaptic plasticities, including short-term plasticity, long-term plasticity, and spiking-rate-dependent plasticity, are simulated by applying the light pulses. Furthermore, metaplastic excitatory postsynaptic current, metaplastic facilitation of long-term potentiation and transition from potentiation to depression are also readily demonstrated. The artificial neural network, in which neurons are interconnected through our proposed optoelectronic synaptic transistors, achieves a high accuracy of 89.8% in recognizing handwritten digits. This work provides insight into the design of an optoelectronic synapse based on JFETs.