忆阻器自被提出以来，由于其低功耗、高读写速度和高集成度等优势，相关研究报道络绎不绝，而其在神经模拟和类脑计算方面的应用无疑是皇冠上的明珠。基于此，本文围绕动力学行为丰富的NbOx基忆阻器，设计并研制了两类忆阻结构，为全忆阻神经网络和听觉模拟提供了基础单元，具体情况如下：一方面为了发展全忆阻神经网络，需要合适的忆阻结构实现神经形态的特征功能。我们制备了Nb/HfOx/Pd结构的忆阻器，其具备两种忆阻机理，即Nb/HfOx界面处形成的NbO2纳米晶粒的金属—绝缘态转变以及介质层中氧离子（空位）的迁移。外部偏压的调节可以控制以上机理，实现不同的忆阻特性。正向电压促使氧空位导电通道形成，使得器件具备非易失的忆阻特性，进而模拟突触可塑性；负向电压抑制氧空位导电通道的形成并激活NbO2纳米晶粒发生金属—绝缘态转变，进而模拟动作电位的发放。该体系能够自适应地响应来自同样结构器件产生的动作电位和突触信号。值得注意的是，该器件在混沌边缘微扰下展示出了丰富的神经动力学行为，蕴含着信号的编码与识别。另一方面为了利用NbOx在混沌边缘微扰下的神经动力学行为模拟生物听觉，我们构建了Pd/Nb/NbOx/Nb/Pd结构的忆阻器，因其内部包含大量氧空位，电流受Poole-Frankel非线性电子传导机制控制，展现出了典型的阈值转变特性，并且定量计算后发现其受温度和电压控制。随后用其构建的振荡电路仅在一定电压范围内发放电流尖峰，其边界被定义为混沌边缘，并阐明了其混沌动力学过程。基于此，在混沌边缘处施加周期性的正弦微扰，通过改变频率实现了听觉的频率选择性，即低频下，周期内电流尖峰数量与频率成反比关系，而中、高频下锁相地发放电流尖峰。通过改变微扰的振幅，实现了强度编码，即大振幅下的电流尖峰数量和响应范围都会增大。通过改变与忆阻器串联的电阻大小，实现了音调拓扑，即不同串联电阻下振荡电路表现出的特征频率也不同。进一步地，我们合成输入固定频率，不同时间延迟的双正弦微扰，实现了电流尖峰发放模式随时间延迟的变化，可模拟水平面的双耳延迟定位。通过合成输入固定时间延迟，不同频率的双正弦微扰，得到了对应时间延迟下的特征频谱，可模拟垂直面的耳廓反射延迟定位。最后，混合输入两个不同频率的正弦微扰，其电流尖峰响应通过傅里叶变换可以将两个单组分频率分离出来，并且出现了相应的增频和差频，可模拟听觉的混频。

Since the memristor was proposed, due to its low power consumption, high read/write speed and high integration density, there have been numerous related research reports, and its great potential in neuromorphic emulation and computing is undoubtedly the most attractive. Based on this, we focus on the NbOx-based memristor with rich dynamic behavior, design and fabricate two types of memristive structure, providing basic units for constructing fully memristive neural networks and emulating auditory senses. The main achievements are as follows:On the one hand, in order to find a suitable memristive structure to perform featured neuromorphic functions for the further development of fully memristive neural networks, we proposed a Nb/HfOx/Pd structure memristor. It has two memristive mechanisms, namely the metal-insulator transition of the NbO2 nanocluster formed near the Nb/HfOx interface and the migration of oxygen ions (vacancies) in the dielectric layer. By modulating the external bias voltage, the above mechanism can be controlled and different memristive behaviors can be obtained. The positive voltage induces the formation of oxygen vacancy conductive filament, resulting in a non-volatile behavior that can emulate synaptic plasticity. The negative voltage restrains the formation of conductive filament and activates the NbO2 nanoclusters to undergo metal-insulator-transformation, thereby emulating the action potential. The system can respond adaptively to the action potential and synaptic signals generated by the same structural device. In particular, the memristor exhibits rich neural dynamic behavior under perturbations at the edge of chaos, implying signal coding and recognition.On the other hand, in order to emulate auditory senses by neural dynamic behavior of the NbOx under perturbations at the edge of chaos, we constructed a Pd/Nb/NbOx/Nb/Pd structure memristor. Because it contains a large number of oxygen vacancies, the current is controlled by the Poole-Frankel nonlinear conduction mechanism and shows typical threshold switching behavior. Quantitative calculations show that it is controlled by temperature and voltage. The oscillation circuit constructed by this memristor only fires current spikes within a certain voltage range, the boundary of which is defined as the edge of chaos, and its chaotic dynamics is elucidated. Based on this, periodic sinusoidal perturbations are applied at the edge of chaos, and frequency selectivity of auditory sense is emulated by changing the frequency. At low frequencies, the number of current spikes in one period is inversely proportional to the frequency, while at medium and high frequencies, the current spikes are fired at locked phases. Intensity coding is emulated by changing the amplitude of the perturbation, that is, the number of current spikes and response range both increase at a larger amplitude. By changing the resistance in series with the memristor, the tonotopy is emulated, that is, the characteristic frequency is modulated. Furthermore, we synthesized two sinusoidal perturbations with fixed frequency and different time delays, realizing the change of current spike mode with time delays, which can emulate sound location in horizontal plane by the binaural time delay. By synthesizing two sinusoidal perturbations with fixed time delay and different frequencies, the characteristic spectrum is obtained, which can emulate sound location in vertical plane by the auricle reflection time delay. Finally, when mixing two sinusoidal perturbations of different frequencies, the current spike mode can separate the mixing frequency into individual ones and give the up-conversion and down-conversion frequencies by Fourier transform, emulating the frequency mixing of auditory sense.