【Device Papers】Energy-Efficient DUV Light Programming Nonvolatile Memory Based on MoS₂/β-Ga₂O₃/MLG Heterostructures for Neuromorphic Computing

      Researchers from the South China Normal University have published a dissertation titled "Energy-Efficient DUV Light Programming Nonvolatile Memory Based on MoS₂/β-Ga₂O₃/MLG Heterostructures for Neuromorphic Computing" in Advanced Functional Materials.

Abstract

      As one of the key components for brain-inspired computing, optoelectrical synapses based on memory devices are capturing growing attention due to their integrated function of sense and memory. However, the high power consumption (large programming voltage, high optical power density) and difficulty avoiding interference from solar radiation are potentially limiting their applications in artificial neural systems. Here, a novel optoelectrical floating gate memory based on molybdenum disulfide (MoS₂)/β-phase Gallium oxide (β-Ga₂O₃)/Multilayer graphene is proposed. Benefitting from the unique photosensitive dielectric properties of β-Ga₂O₃, the device exhibits an excellent current switching ratio of 10⁶ only at a low programming/erasing voltage of ±40 V. Furthermore, the device possesses a strong anti-interference ability which can operate in the solar-blind region (254 nm) only with a low power of 254.4 pJ (0 V). Due to the large single photon energy at the short wavelength, the photon number (3.25 × 10⁸) used per programming operation is much smaller than that of most present works. The neural network model constructed with the memory device achieves an accuracy of 91.07% for image recognition. These results suggest the feasibility of constructing energy-efficient, anti-interference optoelectrical memory based on the van der Waals heterostructures for future artificial visual neuromorphic systems.

DOI：

https://doi.org/10.1002/adfm.202508292