【作者】 李蕾；

【导师】 闫小兵； 孙永健；

【作者基本信息】 河北大学 ， 电子信息硕士（专业学位）， 2023， 硕士

【摘要】 传统计算系统中存在着冯·诺依曼瓶颈和内存墙的问题,这使得计算机的数据访问及运算效率无法满足未来数据密集型计算的需求。作为一种新型的半导体器件,忆阻器具有结构简单、存储与处理兼容、运算速度快和能耗低的优点,成为创建新型计算架构的有力竞争者。然而,在以往的忆阻器研究中,存在着开关电压大和开关比小的问题,并且在生物仿生功能模拟和高密度电路设计方面存在一些挑战。因此,寻找更可靠的阻变材料和制备高性能的器件以实现多领域应用成为忆阻器研究的重点内容。钙钛矿稀土镍酸盐Sm Ni O₃（SNO）是一种突出的复杂氧化物材料,它具有众多物理性质,很多研究人员将其应用在新能源燃料电池、光调制和传感器件等领域。最近几年的研究中发现,这类材料具有弱化学键和较小的氧空位形成能,易于氧空位的形成和迁移,从而导致阻变现象发生,为其在忆阻器方面的应用前景提供了有力的证明。但是,该种薄膜同时作为人工突触和神经元的行为还缺少进一步的研究。因此,在本文中,利用SNO制备了两种不同结构的忆阻器,分别表现出非易失性存储开关特性和易失性阈值开关特性,具有较大开关比和良好的重复性。并且利用此器件能够实现多种应用,包括人工突触、人工神经网络、布尔逻辑门电路、人工伤害感受器和人工神经元。本论文的主要研究工作如下:一、制备了W/SNO/La_0.67Sr_0.33Mn O₃（LSMO）/SrTiO₃结构的忆阻器。通过常见的半导体表征方法,例如,X射线光电子能谱、X射线衍射、原子力显微镜、透射电子显微镜等分析器件的微观结构。在不改变LSMO薄膜生长条件的情况下,探究了最适宜SNO薄膜生长的条件,为600°C-10 Pa。电学测试显示出器件具有明显的非易失性存储开关特性,开关电压较小,集中在0.72 V和-0.73 V,且重复性良好。器件能够实现双向电导调控和对突触功能的模拟即成对脉冲易化、兴奋性突触后电流、经典双脉冲尖峰时间依赖可塑性、三重态尖峰时间依赖可塑性、短期可塑性到长期可塑性转变、不应期现象和学习遗忘规则等。此外,将此忆阻器应用于人工神经网络中,对三种数据集的识别准确率分别达到78.50%、84.60%和89.10%。重要的是,利用两个突触器件与一个神经元器件成功搭建了逻辑门电路,实现“与”、“或”和“非”功能。以上结果表明此器件具有在由全忆阻器集成的人工神经网络和高密度电路领域的突出前景。二、制备了W/SNO/n-Si结构的忆阻器。利用常见半导体表征方法对器件进行表征。在较小的限制电流(≤1×10^-5 A)下,此器件表现出了明显的易失性阈值开关特性,并且开关比达到10⁷。在更大的限制电流下,器件表现出易失性开关到非易失性开关的转变,这与氧空位导电细丝的形成强度有关。利用此器件成功实现了伤害感受器的模拟,其主要特征有:伤害感受阈值、不适应、松弛和敏化等。同时,利用此器件构建了基于简单RC和传统CMOS的两种神经元电路,实现了人工神经元功能,如阈值驱动放电、强度调制和电导调制频率尖峰响应等。以上结果表明,此器件的成功实现为多功能忆阻器和人工智能领域的发展提供新的可能。更多还原

【Abstract】 There are von Neumann bottleneck and memory walls in traditional computing systems,which make computer data access and computational efficiency unable to meet the needs of future data intensive computing.As a new type of semiconductor device,memristor has the advantages of simple structure,compatible storage and processing,fast operation speed and low energy consumption,and has become a strong competitor to create a new computing architecture.However,in previous research on memristors,there are some problems such as high switching voltage and low switching ratio,and some challenges in biomimetic function simulation and high-density circuit design.Therefore,searching for more reliable resistive materials and preparing high-performance devices for multi-field applications have become the focus of memristor research.Perovskite rare-earth nickelate Sm Ni O₃（SNO）is an outstanding complex oxide material,which have many physical properties.Many researchers have applied them in many fields such as new energy fuel cells,optical modulation,sensor devices,and so on.In recent years,it has been found that these materials have weak chemical bond and small oxygen vacancy formation energy,which are prone to the formation and migration of oxygen vacancies,leading to resistance changes,providing strong evidence for their application prospects in memristors.However,the behavior of this kind of film as both artificial synapses and neurons is still lacking further research.Therefore,in this paper,SNO was used to prepare two kinds of memristors with different structures,respectively showing nonvolatile memory switching characteristics and volatile threshold switching characteristics,with a large switching ratio and good repeatability.And they can be used to implement multiple applications such as artificial synapse,Artificial Neural Network,Boolean logic gate circuits,artificial nociceptor and artificial neuron.The main research works of this paper are as follows:Firstly,the memristor with W/SNO/La_0.67Sr_0.33Mn O₃（LSMO）/SrTiO₃ structure was prepared.The microstructure of the device was analyzed by common semiconductor characterization methods,such as X-ray photoelectron spectroscopy,X-ray diffraction,atomic force microscopy,transmission electron microscope,etc.Without changing the growth conditions of LSMO film,the most suitable conditions for SNO film growth were investigated,which was 600°C-10 Pa.Electrical tests showed that the device has obvious non-volatile memory switching characteristics.The switching voltage is small,concentrated around 0.72 V and-0.73 V,and has good repeatability.The bidirectional conductance regulation and the simulation of synaptic function were achieved,namely,paired-pulse facilitation,excitatory postsynaptic current,classical pair spike time-dependent plasticity,triplet-STDP,transition from short-term plasticity to long-term plasticity,refractory period phenomenon and learning forgetting rule.In addition,when the memristor was applied to an Artificial Neural Network,the recognition accuracy of the three data sets reaches 78.50%,84.60%and 89.10%respectively.Importantly,two synaptic devices and one neural device were used to successfully construct the logic gate circuit,realizing the“AND”,“OR”and“NOT”functions.The above results showed that this device has a prominent prospect in the field of artificial neural networks and high-density circuits integrated by full memristors.Secondly,the memristor with W/SNO/n-Si structure was prepared.The device was characterized by common semiconductor characterization methods.At a small limiting current(≤1×10^-5A),the device showed obvious volatile threshold switching characteristics,and the switching ratio reaches 10⁷.Under a larger limiting current,the device showed a transition from volatile switching to non-volatile switching,which was related to the formation strength of oxygen vacancy conductive filaments.The nociceptor was simulated successfully with this device.Its main characteristics include nociceptor threshold,no adaptation,relaxation and sensitization.At the same time,two types of neuron circuits based on simple RC and traditional CMOS were constructed using this device to realize the artificial neuron functions,such as threshold driven-firing,intensity-modulated and conductance-modulated frequency spike response.The above results showed that the successful implementation of this device provides new possibilities for the development of multifunctional memristors and artificial intelligence.更多还原