氧化物忆阻材料与人工神经突触器件

Artificial synapse plays key role in in brain-analogous artificial intelligent system. Memristor is one ideal framework element for developing artificial synapse, however it generally suffers from reliability issues (such as variability, retention, endurance, uniformity), which are the crucial obstacle to achieve accurate and diverse cognitive functions for mimicking biological synapse. This project would aim at studying metal oxide materials and adjusting defects and microstructure to optimize memristor material system, and eventually obtaining reliable memristive devices. We plan to reveal the influence on reliability of memristive devices from defect behavior (such as concentration, distribution, migration, relaxation, restructure); we propose several methods to optimize the material system for memristor like insertion of nanoparticles, doping and material amorphization, which are based on the thoughts of localization of electric field, optimization of migration barrier and stabilization of phase structure, respectively. We also plan to establish kinetic model of memristor and set up new framework of heterojunction and nT1R for artificial synapse, and achieve accurate simulation for primary synaptic functions and adaptive evolution functions, and explore advanced synergetic functions based on multiple synapses. This project starts from the fundamental physics question which is the adjustment of defects and micro-structures in memristive materials, and aim at realizing the optimization of device performance and functionalization of artificial synapse. This study can provide technical accumulation for artificial neural network, which is one leading edge topic in overlapping field covering material, electronic information and physics subjects.

神经突触器件在类脑人工智能领域有重要应用。忆阻器是发展人工突触的理想构架，其中忆阻器件的可靠性（包括：参数波动性、电阻保持性、循环耐受性、器件一致性）是制约突触仿生准确性与多功能性的关键难题之一。本项目围绕氧化物材料，从缺陷及微结构调控角度，设计优化忆阻材料体系，构建人工神经突触器件。揭示缺陷行为（如：浓度、分布、迁移、弛豫、重构）对忆阻器件可靠性的影响规律；基于电场局域化、迁移势垒调制、相结构稳定等思想，发展粒子包埋、调制掺杂、材料非晶化等方法，提升忆阻材料与器件的可靠性。建立忆阻器件动力学模型，创建异质结型、nT1R型等人工突触器件新构架，实现对生物突触基本认知功能和自适应演变功能的准确模拟；发展多突触器件协同的高级认知功能的模拟方法。本项目从忆阻材料的缺陷和结构这一微观源头出发，实现器件的性能优化与突触功能化，为构筑人工神经网络积累技术储备，是材料、信息、物理等交叉领域的前沿课题。

本项目聚焦氧化物忆阻器材料与人工神经突触器件，针对忆阻器在类脑功能化与工作可靠性方面的关键科学难题，从阐明忆阻材料微观机制出发，提出了引入界面电场、构筑微纳孔道、光电协同调控等新思路和新方法，开展了忆阻器在存算一体、感存算一体方面的研究，实现了在类脑功能和器件性能上的突破。相关成果在Advanced Materials、Nature Communications等SCI期刊上发表论文90篇；以第一完成人身份获2019年国家自然科学奖二等奖、吉林省科学技术奖一等奖各1项；项目负责人2021年入选中国科学院院士，两名项目成员分别获批国家基金委杰青项目和国家级青年人才项目。本项目从微观机制和性能提升、突触功能与神经网络、光电忆阻及存算一体等多角度开展研究工作，取得重要结果如下：（1）针对氧离子扩散随机的问题，通过构建氧化物异质结型类脑忆阻器件，利用界面内建电场和外场协同作用，促进氧离子的定向迁移和扩散，首次基于单一器件完整模拟了大脑高级认知规律BCM 类脑功能,创制了具有时空模式识别功能的忆阻型人工视觉系统。（2）发展了构筑微纳孔道、离子活化界面等方法引导导电通道生长，获得了热稳定性好、结构有序的纳米尺寸导电通道，研制了具有7.5 fJ 类脑超低能耗、3000 万次超长寿命的高性能忆阻材料与突触器件。（3）基于忆阻器发展了多种可行的存算一体化技术方案，在忆阻器交叉杆阵列中可重构实现了输入与输出均为非易失态的IMP、NOR、OR等多种状态逻辑功能，证实了基于忆阻器阵列的非易失算术逻辑运算单元在功能上的可行性。（4）设计了光电协同激励促进通道可控化的新方案，创制了等离激元型光电忆阻材料与类脑器件，获得了全光学信号调制忆阻特性，实现了准确率达98%的高效图像识别。本项目研究结果对于深刻理解氧化物忆阻材料与人工突触器件具有重要的参考价值，为推进忆阻型类脑智能器件发展提供了科学依据和技术基础。