忆阻器实现神经元电路的方法研究

发布时间：2018-05-21 18:10

本文选题：人工神经元 + 忆阻器　；参考：《武汉科技大学》2014年硕士论文

【摘要】：人工神经网络是现代信息处理和智能控制领域的一个重要方法。神经网络的全硬件实现技术是神经网络研究的一个重要方面。从神经网络的定义可以看出：神经元是构成整个神经网络的基础。传统的模拟神经元电路都不能连续精准的调整和方便有效的存储神经突触上的权值。因此，设计一种权值精度高、稳定性好、调整便捷的人工神经元电路无疑具有很大的实用价值。忆阻器是继电阻、电容和电感之后的第四种基本电路元件，通过调节忆阻器两端的电压或流经的电流可以改变其阻值的大小，它是一种具有记忆功能的纳米级尺寸的非线性二端无源器件。本文设计了一种基于忆阻器的人工神经元电路，，可以通过一个端口来调整突触权值的方向和大小以及实现信号的输入。由于忆阻器的阻值是连续变化的且具有记忆功能，因此本文所设计的神经元电路在权值调整的精度、存储的稳定性等方面比传统的模拟神经元电路具有很大优势。通过一系列SPICE仿真实验，验证了本文提出的设计方案的合理性和可行性。
[Abstract]:Artificial neural network is an important method in the field of modern information processing and intelligent control. The full hardware implementation of neural network is an important aspect of neural network research. From the definition of neural network, we can see that neuron is the basis of the whole neural network. Traditional analog neuron circuits can not adjust the weights of synapses continuously and accurately. Therefore, it is of great practical value to design an artificial neuron circuit with high weight accuracy, good stability and convenient adjustment. The resistor is the fourth basic circuit element after resistor, capacitance and inductance, which can be changed by adjusting the voltage at the two ends of the resistor or the current flowing through the resistor. It is a kind of non-linear two-terminal passive device with memory function. In this paper, an artificial neuron circuit based on amnesia is designed, which can adjust the direction and size of synaptic weight and realize the input of signal through a port. Because the resistive value of the amnesia is continuous and has memory function, the neuron circuit designed in this paper has a great advantage over the traditional analog neuron circuit in the precision of weight adjustment, the stability of storage and so on. The rationality and feasibility of the proposed design are verified by a series of SPICE simulation experiments.
【学位授予单位】：武汉科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM13

【相似文献】