AI 中文总结
本研究采用微腔CVD法制备MoWS₂合金光电突触器件，实现低功耗神经形态学习，在MNIST数据集上达92.43%识别准确率，还可用于彩色图像滤波，为节能神经形态视觉应用提供新路径。
AI 中文摘要
二维过渡金属二硫化物合金因具有强光-物质相互作用和可控缺陷特性，成为先进光电和神经形态应用的潜在候选材料。然而，这类合金的大面积生长仍具挑战性，其物理特性与神经形态特性之间的关联也大多不明确。本研究中，我们提出一种创新的微腔化学气相沉积（CVD）反应器路径，用于生长均匀且大面积的MoWS$_2$单层及少层合金薄膜，以展示其光电突触功能。经X射线光电子能谱（XPS）、开尔文探针力显微镜（KPFM）和扫描透射电子显微镜（STEM）测量证实，由生长诱导的固有硫空位驱动，我们的光电突触器件（OSD）成功模拟了重要的生物突触特征，如兴奋性突触后电流（EPSC）、双脉冲易化（PPF~170%），以及依赖刺激的短期和长期可塑性（STP与LTP）。该器件每个突触事件的能耗为皮焦耳级，暗电流为纳安级，可实现低功耗神经形态学习，包括巴甫洛夫联想学习的模拟。此外，实验测得的电导权重更新特性使人工神经网络（ANN）模拟在MNIST手写数字数据集上达到92.43%的识别准确率。最后，我们利用该器件的波长选择性光响应特性，执行彩色图像滤波，展示了先进的神经形态视觉处理。这种简单却多功能的器件架构，为节能、光谱选择性的神经形态视觉应用提供了一条有前景的路径。
英文摘要
Two-dimensional transition-metal dichalcogenide alloys are potential candidates for advanced optoelectronic and neuromorphic applications due to their strong light-matter interactions and controllable defect properties. However, large-area growth of such alloys remains challenging, while the correlation between their physical and neuromorphic properties remains largely unclear. In this work, we present an innovative microcavity chemical vapor deposition (CVD) reactor pathway to grow uniform, and large-area MoWS$_2$ mono- and few-layer alloy films for demonstrating optoelectronic synaptic functionalities. Driven by growth-induced intrinsic sulfur vacancies, as confirmed by XPS, KPFM, and STEM measurements, our optoelectronic synaptic device (OSD) successfully emulates essential biological synaptic features, such as excitatory postsynaptic currents (EPSC), paired-pulse facilitation (PPF~170%), and stimulus-dependent short- and long-term plasticities (STP & LTP). With picojoule-order energy consumption per synaptic event and nanoampere-order dark current, the device enables low-power neuromorphic learning, including emulation of Pavlovian associative learning. Furthermore, the experimentally measured conductance weight-update characteristics enabled an artificial neural network (ANN) simulation to achieve 92.43% recognition accuracy on the MNIST handwritten digit dataset. Finally, we demonstrate advanced neuromorphic visual processing by executing color image filtering based on the device's wavelength-selective photoresponse characteristics. This simple, yet multifunctional device architecture provides a promising path toward energy-efficient, spectral-selective neuromorphic vision applications.
Commentsmain manuscript (5 figures, 37 pages) and supporting information (17 figures, 12 pages)
