College of Chemistry and Materials Science, Sichuan Normal University, Chengdu, 610066, China.
Department of Applied Physics, The Hong Kong Polytechnic University, Hong Kong, 999077, China.
Nanoscale. 2023 Jun 15;15(23):10089-10096. doi: 10.1039/d3nr01301d.
Neuromorphic computing inspired by the human brain is highly desirable in the artificial intelligence age. Thus, it is essential to comprehensively investigate the neuromorphic characteristics of artificial synapses and neurons which are the unit cells in an artificial neural network (ANN). Memristors are considered ideal candidates to serve as artificial synapses and neurons in the ANN. Herein, two-terminal memristors based on two-dimensional (2D) MoSe nanosheets are fabricated, demonstrating analog resistive switching (RS) behaviors. Unlike the digital RS behaviors with a sharp transition between the two resistance states, the analog RS provides a series of tunable resistance states, which is more suitable for the realization of synaptic plasticity. Thus, the fabricated memristors successfully implement the synaptic functions, such as paired-pulse facilitation, long-term potentiation and long-term depression. The analog memristors can be utilized to construct the ANN for image recognition, leading to a high recognition accuracy of 92%. In addition, the synaptic memristors can emulate the "learning-forgetting" experience of the human brain. Furthermore, to demonstrate the ability of single neuron learning in our devices, the memristors are studied as artificial nociceptors to recognize noxious stimuli. Our research provides comprehensive investigations on the neuromorphic characteristics of artificial synapses and nociceptors, suggesting promising prospects for applications in neuromorphic computing based on 2D MoSe nanosheets.
受人类大脑启发的神经形态计算在人工智能时代是非常理想的。因此，全面研究人工突触和神经元（神经网络中的单元细胞）的神经形态特征至关重要。忆阻器被认为是作为人工神经网络中的人工突触和神经元的理想候选者。在此，我们制备了基于二维（2D）MoSe 纳米片的两端忆阻器，展示了模拟电阻开关（RS）行为。与在两个电阻状态之间具有尖锐转变的数字 RS 行为不同，模拟 RS 提供了一系列可调节的电阻状态，更适合实现突触可塑性。因此，所制备的忆阻器成功地实现了突触功能，如成对脉冲易化、长时程增强和长时程抑制。模拟忆阻器可用于构建用于图像识别的人工神经网络，从而实现 92%的高识别准确率。此外，突触忆阻器可以模拟人脑的“学习-遗忘”体验。此外，为了展示我们设备中单神经元学习的能力，我们将忆阻器作为人工伤害感受器来识别有害刺激。我们的研究全面研究了人工突触和伤害感受器的神经形态特征，为基于 2D MoSe 纳米片的神经形态计算应用提供了广阔的前景。
