面向人工神经网络的新型相变忆阻器的模型研究

The current neuroscience is mainly built in the classic Von Neumann computer architecture,while the mode that Von Neumann digital computer follows to separate data processing from data storage utmostly forbids the current artificial neural network technology to entirely mimic the full function of human brain. However, the presence of memristor provides the best solution to this problem. Nevertheless, there are some severe drawbacks for current memristors on the aspects of performance and cost, which thus handicaps the application of memristor on artificial neural network. In order to overcome the drawbacks of current memristors, a concept using a novel phase-change (Ge2Sb2Te5) based memristor is proposed in this project.The main research contents include (1) model structure and theoretical algorithm of this memristor are proposed by means of parametric designed method and electro-thermal model based on classical nucleation theory; (2) simulate the memristive property of this memristor and evaluate its parameters in order to prove that compared with other memristors, this memristor exhibits several merits, e.g.fast transitional speed, low power consumption,great stability and low cost etc; (3) simulate the associative memory performance of this memristor. ..The expected outcomes will give a solution to overcoming the drawbacks of current memristors, to finding out a new way for realization of artificial neural network using hardware, and thus to pushing forward the development of artificial neural network technology.

目前的人工神经网络科学主要是以冯o诺依曼计算机体系结构为依托，而冯o诺依曼数字计算机所遵循的数据处理和数据存储相分离的模式正是阻碍人工神经网络实现完全模拟人脑功能的最大症结所在。而忆阻器的出现是目前解决这一难题的最佳方案。 但是现已面世的忆阻器存在着严重的性能及成本缺陷，从而阻碍了其在人工神经网络方面的应用。本项目针对现有忆阻器的缺陷,提出了一款基于相变材料(Ge2Sb2Te5)的新型忆阻器概念。主要研究内容包括 (1) 利用参数化设计方法和原子成核电热模型，研究和提出了该忆阻器的模型结构和理论算法；（2）对该忆阻器的忆阻特性进行仿真并评估其参数指标，从而证实该忆阻器与其他忆阻器相比具有转换速率快，能量损耗少，稳定性强和造价低等优点；（3）仿真模拟该忆阻器的联想记忆功能。..预期成果将给出克服现有忆阻器缺陷的方法，为人工神经网络硬件实现方式找到新的出路，从而推动人工神经网络技术的发展。

目前的人工神经网络科学主要是以冯•诺依曼计算机体系结构为依托，而冯•诺依曼数字计算机所遵循的数据处理和数据存储相分离的模式正是阻碍人工神经网络实现完全模拟人脑功能的最大症结所在。而忆阻器的出现是目前解决这一难题的最佳方案。 但是现已面世的忆阻器存在着严重的性能及成本缺陷，从而阻碍了其在人工神经网络方面的应用。本项目针对现有忆阻器的缺陷,提出了一款基于相变材料(Ge2Sb2Te5)的新型忆阻器概念。主要研究内容包括 (1) 利用参数化设计方法和原子成核电热模型，研究和提出了该忆阻器的模型结构和理论算法；（2）对该忆阻器的忆阻特性进行仿真并评估其参数指标，从而证实该忆阻器与其他忆阻器相比具有转换速率快，能量损耗少，稳定性强和造价低等优点；（3）仿真模拟该忆阻器的联想记忆功能。..预期成果将给出克服现有忆阻器缺陷的方法，为人工神经网络硬件实现方式找到新的出路，从而推动人工神经网络技术的发展。