当今人类社会的数据量呈指数型增长，对计算机的低能高效的需求越来越大。然而，基于传统冯诺依曼架构的计算与存储分离导致的高功耗、高延时难以满足需求。受人脑在智能信息处理方面高能效的启发，类脑的神经形态计算系统应运而生。神经形态计算架构模拟生物存储运算一体化的特性，在存储单元中原位地实现信息处理，构成存内计算系统，并且模拟生物以脉冲表达信息的特性，构成仿生脉冲神经网络，从根本上解决冯氏架构存储墙的问题。目前，神经形态计算的硬件实现主要基于传统CMOS器件，仍然面临硬件开销高和功耗高的问题。针对此挑战，诸多研究利用新型材料与器件中独特的物理性质，模拟神经形态计算系统中信息处理单元的功能需求，能显著降低神经形态计算的硬件开销。本文分别对具备神经形态计算应用潜力的二维石墨烯材料、双极性沟道材料的物理特性进行了深入分析和建模研究，并结合新型氧化铪基铁电材料构建了融合存储和运算的神经形态器件，有助于降低神经形态计算硬件实现的硬件开销。
首先，石墨烯材料长的自旋扩散长度与自旋弛豫时间为其纯自旋电流的输运提供了理想条件，是神经形态计算中的一种理想自旋电子器件。本文利用行列式量子蒙特卡洛方法，本文重点研究了三个锯齿边界量子点和量子环，即三角形石墨烯量子环、转角双层三角形石墨烯量子点和转角双层三角形石墨烯量子环。其表现出鲁棒的边缘铁磁行为，与尺寸、单层或双层以及点或环的形状无关。在半填充状态下，边缘磁化率会随着在位相互作用的增加而增加。这对为器件中利用高阻和低阻态进行二值存储等有一定启发。
其次，双极性材料由于其电场调控的载流子浓度与极性，在实现神经形态计算中的线性不可分运算以及仿生突触的兴奋和抑制功能上具有潜力。本文首先探讨了双极性的物理机制，分析了极有潜力的双极氧化物半导体氧化锡的能带结构。其次结合基本电学理论和载流子统计规律，建立了可控双极型场效应晶体管的模型，并分析讨论了其转移曲线、输出曲线，提出了增强双极输运电流的对称性与改变对称中心对应栅压的方法，为双极型沟道材料及器件的神经形态计算应用奠定了基础。
进一步地，铁电材料由于其非易失特性与丰富的极化翻转动态特性，被广泛用于神经形态计算的硬件实现中。本文提出结合铁电存储特性和双极型氧化物沟道的栅控双极电流输运特性，构建了铁电栅双极氧化物沟道场效应晶体管，并对器件电学特性进行建模，仿真结果表明该器件能同时实现了神经形态计算的中权重单元的存储特性和线性不可分的运算功能，显著降低硬件开销，为低硬件开销高集成度的神经形态计算系统奠定了基础。这对利用新材料与器件特性构建人工神经元，并实现存算一体功能有一定的意义与启发作用。

With the exponential growth of the amount of data in today's human society, there is a growing demand for low-energy and high-efficiency computers. However, the separation of computing and storage based on traditional von Neumann architecture leads to high power consumption and high delay. Inspired by the high energy efficiency of human brain in intelligent information processing, a brain-like neuromorphic computing system has emerged. The neuro-morphological computing architecture simulates the characteristics of the integration of biological memory and computation, realizes the information processing in situ in the memory cell, constitutes the in-memory computing system, and simulates the characteristics of biological information expressed by pulses, a Bionic Pulse neural network is constructed to solve the problem of the storage wall in the von Willebrand architecture. At present, the hardware implementation of neuromorphic computing is mainly based on traditional CMOS devices, which still faces the problems of high hardware overhead and high power consumption. In order to solve this challenge, many researches use the unique physical properties of new materials and devices to simulate the functional requirements of information processing units in the neuro-morphological computing system, which can significantly reduce the hardware cost of neuro-morphological computing. In this paper, the physical properties of two-dimensional graphene and bipolar channel materials, which have potential applications in neural morphology computation, are deeply analyzed and modeled, combined with a new hafnium oxide-based ferroelectric material, the fusion memory and operation of the neuromorphic device are constructed, which is helpful to reduce the hardware cost of the neuromorphic hardware implementation.
Firstly, the long spin diffusion length and the long spin relaxation time of graphene provide ideal conditions for the transport of pure spin current, and it is an ideal spintronic device in neural morphology calculation. Using determinant quantum Monte Carlo method, this paper focuses on the study of three zigzag edge quantum dots and quantum rings, that is, triangular graphene quantum ring, twisted double-layer triangular graphene quantum dot and twisted double-layer triangular graphene quantum ring. It exhibits robust edge ferromagnetic behavior independent of size, monolayer or bilayer, and dot or ring shape. In the half-filled state, the edge susceptibility increases with the increase of on-site interaction. This is useful for binary storage of high and low impedance states in devices.
Second, bipolar materials have the potential to achieve linear indivisibility and excitatory and inhibitory functions of bionic synapses due to their electric field-regulated carrier concentration and polarity. In this paper, the physical mechanism of bipolar is discussed, and the energy band structure of the potential bipolar oxide SnO is analyzed. Secondly, based on the basic electrical theory and carrier statistics, the model of the controllable bipolar FET is established, and its transfer curve and output curve are analyzed and discussed. The method of enhancing the symmetry of bipolar transport current and changing the corresponding gate voltage of the center of symmetry is proposed.
Furthermore, ferroelectric materials are widely used in the hardware implementation of neuromorphic computation because of their non-volatile properties and rich dynamic properties of polarization reversal. In this paper, a ferroelectric gate bipolar oxide channel field effect transistor (FET) is constructed by combining the ferroelectric storage characteristics with the gate-controlled bipolar current transport characteristics of the bipolar oxide channel, and the electrical characteristics of the device are modeled, the simulation results show that the device can realize both the memory characteristic of the weight unit and the function of linear indivisibility, and reduce the hardware cost significantly, it lays a foundation for the low hardware cost and high integration of the neural morphology computing system. It is significant and enlightening to construct artificial neurons by using new materials and device characteristics and to realize the function of integration of memory and computing.