基于金属掺杂氧化钨复合薄膜的光控忆阻器构筑及其神经突触仿生研究

The development of brain-like neuromorphic computing system is a research hotspot in the field of artificial intelligence, in which the development of high-precision and brain-inspired intelligent materials and devices is the foundation to realize brain-like learning functions. Inspired by optogenetics, the construction of light-controlled synaptic devices based on photosensitive material system and using light to regulate neural activity will help to improve the tunability and accuracy of synaptic simulation. This project intends to develop the new photoelectric memristive devices based on the photon-electron coupling effects in metal/tungsten oxide composite thin film, so as to simulate the high-order cognitive function modulated by hybrid optical/electrical signal. The interface barrier resistance can be continuously modulated by the electron doping effect induced by surface plasmon resonance excitation of metal nanoparticles in the composite film under illumination. The photoelectric memristive behavior can be controlled by adjusting the intensity and time of light irradiation, metal doping distribution and plasma treatment on the film surface. The photo-induced redox process of metal particles is investigated by in situ transmission electron microscopy, so as to reveal the microscopic mechanism of the optoelectronic memristor. The study will focus on the simulation of high-order cognitive functions such as logic operation, associative learning and pattern recognition modulated by hybrid multiple signals. This project would take the research idea of “material-device-function”, which is the cutting-edge research in the cross field of materials and artificial intelligence.

开发类脑的神经形态计算系统是人工智能领域的研究热点，而发展具有高精度仿生的脑启发智能材料和器件是实现类脑功能的重要基础。受光遗传学的启发，基于光敏材料体系构筑光控神经突触器件，实现光对神经活性的调控，将有助于提升突触模拟的可调性及准确性。本项目拟发展基于金属/氧化钨复合薄膜光子-电子耦合的新型光电忆阻神经突触器件，并以此模拟光、电混合调制的高级认知功能。结合光照下复合薄膜中金属纳米粒子表面等离子共振激发产生的电子掺杂效应，实现界面势垒电阻的连续调制。通过调节光辐照强度和时间、金属掺杂分布和薄膜表面等离子体处理等方法调控光电忆阻行为。采用原位透射电镜技术方法探究金属粒子的光致氧化还原过程，揭示光电忆阻行为的微观机制。重点关注多信号混合调制的逻辑运算、联合式学习和模式识别等高级认知功能模拟。本项目以“材料-器件-功能”为研究思路，是材料和人工智能交叉领域的前沿课题。

新兴的忆阻器具有结构简单、低能耗、与生物突触信号传输特征相似等优点，是构筑神经形态计算系统的理想元件。本项目基于氧化钨、氧化钛、有机物等忆阻材料构筑光控忆阻器件，从高性能忆阻器件研制、光调控忆阻行为探究及类脑突触功能实现等方面开展了系列研究工作。取得主要结果如下:（一）提出了插入离子缓冲层、引入元素掺杂和构筑微纳孔道等解决方法，实现了对导电通道的精细调控，从而获得了具有超低能耗的高性能忆阻器件；（二）开发了光调控有机阻变式存储器，利用可调控的开启电压特性实现了加密的图像存储；提出了光致还原氧化石墨烯掺杂碳量子点新方法，实现了模拟型阻变的全碳忆阻器件；依靠银/二氧化钛复合薄膜中的局域表面等离子体共振效应，在单一器件中获得了可见光/紫外光作用下可逆调控忆阻行为；（三）采用1T1R结构精确调控限制电流和器件电导，模拟了联想学习的获取和退化功能；构建了氧化锌/氧化铪异质结构建了光电突触器件，实现精确度达96.1%的高效图像识别功能；基于WOx忆阻器实现了电压依赖突触可塑性，实现了图像布尔运算（包括交集、差集和并集）功能；本项目研究成果对提高忆阻器件可靠性及光调控忆阻器件构筑方面具有重要的参考价值，同时研究成果中所提出的忆阻人工突触架构及对神经突触功能的仿生模拟，也为发展高效人工神经网络提供了新的研究思路。