基于氧化镍薄膜具有类神经突触行为的忆阻器件研究

The human neural network is a complicated system with billions of neurons which are interconnected with each other by the neural synapses. Thanks to the gigantic network, human beings can have memorization and cognitive functions, etc. Realization of synapse-like electronic devices can help to overcome the bottleneck of current computing technology, which sequentially processes and stores data. Therefore the working speed and efficiency of the integrated circuits and chips can be increased significantly. How to design the device structure based on certain smart materials and realize the tunable electrical characteristics is key to realize electronic devices which can emulate neural synaptic behaviors. This proposal is going to fabricate electronic synaptic devices based on reversible resistive switching behaviors between two or among multiple nickel oxide layers with different compositions and characteristics. Magnetron sputtering together with plasma treatment in oxygen ambient will be used to synthesize the nickel oxide thin films. With nickel oxide thin film sandwiched by metal electrodes, two-terminal devices can be formed. Then current-voltage (I-V) measurements with voltage sweeps and pulses will be used to study the electrical characteristics and reveal the charge transport mechanism as well as the resistive switching mechanism. Based on that, the memristive devices can be fabricated with two-layered or multi-layered nickel oxide thin film structure of different composition and microstructure as the active layer and certain metal films as electrodes. By means of suitable pulse stimulation, the memristive characteristics can be investigated. Eventually through the optimization of device structure and performance, the emulation of neural synaptic functions such as analog memory, plasticity, and self-learning ability can be realized.

人类神经系统是一个由数亿神经元及彼此间神经突触联结形成的庞大网络，它使人类具有记忆、认知等能力。实现具有神经突触功能的电子器件，将能克服现有计算机技术采用串行方式处理和存储数据的瓶颈，革命性地提高集成电路及芯片的运行速度和效率。而如何制备智能材料、设计器件结构、实现器件电学性质的可调是研制类神经突触器件的关键所在。本项目基于氧化镍薄膜，利用成分、特性各异的双层或多层氧化镍薄膜间的可逆反应实现忆阻效应，研制能模仿神经突触行为的器件。拟采用磁控溅射法结合氧气氛等离子处理合成氧化镍薄膜，并配合金属电极制备双端器件。通过扫描及脉冲I-V测试研究器件电学特性，揭示电荷传输机制及阻变机理。在此基础上，合成不同成分、结构的氧化镍形成双层或多层结构作为有源层，配合适当的电极制备忆阻器件。采用脉冲激励研究忆阻器电学特性，并调整器件结构、优化性能，以期实现对神经突触行为如模拟记忆、可塑性、自我学习能力的模仿。

人工智能是时下十分热门的话题，也是未来科技发展的新高地，成为了一个新的研究热点。人工智能意味着对人的意识、思维过程进行模拟，需要从根本上改变原本计算机架构中数据串行处理的方式，而从器件层面研发类似人类神经系统中突触的数据处理与存储并行方式的器件，具有十分重大的意义。本项目采用金属氧化物作为有源材料，试制忆阻器件，实现对突触部分功能的模拟。主要的研究内容包括制备导电银、氧化镍、氧化铝和氧化铟镓锌等材料，设计和制备两端式存储器件，测试器件的阻变和忆阻特性并研究分析机理。重要结果包括：1) 合成了适用于喷墨打印的多种材料墨水并制备成薄膜；2) 基于两端式结构，分别制备了基于氧化镍、氧化铝、氧化铟镓锌和氧化铝/氧化镍复合膜的阻变存储器件，并获得了较好的性能；3) 氧化铝/氧化镍复合薄膜器件相较于氧化铝薄膜器件的稳定性有显著改善，器件高低阻态的阻值和擦写电压分布范围更小，对电路应用十分有利；4）利用氧化镍薄膜制备了一写多读存储器件，获得了优秀的性能，适用于数据的长期存储；5）利用氧化铟镓锌器件中电压激励下的阻值渐变来模仿神经突触的增强和抑制，具备了对神经突触可塑性初步的模仿能力。通过项目的开展，我们掌握了以溶液法（含旋涂和喷墨打印）为主要工艺手段来制备导电银和金属氧化物材料、阻变和忆阻存储器件的方法，为电子器件实现模拟存储及神经突触功能打下了一定的研究基础。