基于阻变存储的神经网络研究

The innovation of comuputer has great effect on the science, society, and modern civilaztion. As a super mechine, it hopes that comupters can reach hunman intelligence and work as human brains.The new hope is brought by the field of neuromorphic engineering, which tries to achieve artificial intelligence by emulating neurobiological systems. Research communities in the field have invented electronic neurons by virtue of very-large-scale-integration (VLSI) technology. However, manufacturing neuromorphic circuits still needs the other basic device – electronic synapse. Fortunately, resistive switching appearing in recent years will maybe provide practical electronic synapses for us. We have grown high quality thin films of metal oxides. We also investigate the materials by advanced equipments and methods to find the electric properties,optical properties, and so on. In this research, the physic mechanism behind the resistive switching phenomenon will be studied. Based on the similarity between the Resistive switchings and synapses, we will try to design excellent electronic synapses and artificial net work models, and find the way to optmizing device performance are . Our long-time research experience and abundant knowledge on resistive switching material and devices will help us establish systematic theory of synapses based on resistive switching phenomenon.

现代计算机的发明深刻地改变了科技、社会和现代文明。几乎从计算机问世那天起，人类就希望有一天计算机能像人脑一样工作。模仿生物神经的神经网络系统承接了此项任务。研究人员已经能够利用超大规模集成电路技术制备出人造电子神经元。但是建立完善的神经网络体系还需要另外一种基本元件-电子突触。近年来出现的电阻开关技术有望成为实现电子突触的关键技术，已有的研究表明阻变存储器件具有突触的特性。但是如何利用阻变存储构建神经网络还存在许多亟待解决的问题。我们已经制备出具有阻变特性的金属氧化物，研究了电阻开关器件的性能和影响器件的因素。考虑阻变特性与突触的共同点，设计基于阻变存储的电子突触，构建神经网络体系，寻求制备人造电子突触的系统理论。

本项目围绕着过渡族而原金属氧化物的阻变存储的存储机理、性能优化、衍生功能等方面的问题，主要从阻变存储器件性能、阻变存储的产生机理两个方面展开了研究，并取得了一定的成果。综合考虑信息技术发展的趋势，从阻变存储的特点出发,研究了阻变存储的类突触特性，为人工智能的实现提供有效方式。