用于神经态计算的基于二维纳米材料的电子突触

This project will be devoted to design, fabricate, characterize, and optimize electronic synapses for neuromorphic computing. The devices will be emulated using memristors, and our strategy to enhance their performance is to use two dimensional materials with advanced properties, so that they could provide several enhanced capabilities to the synapses. The purpose of our work is to understand the main ingredients responsible for the resistive switching phenomenon, as well as finding strategies to control the current on/off ratio, cyclic endurance, retention time, power consumption, switching times, and switching voltages. Furthermore, we will put much effort on improving the variability of these parameters, both cycle-to-cycle and device-to-device, which are the biggest problems of memristive technologies. In particular, we will statistically analyze the switching characteristics between the high resistive state and low resistive state and vice versa, with the aim of elucidating the physical and chemical changes (i.e. atomic rearrangements) that produce the cyclical conductivity changes. Moreover, we will try to supply the devices with additional features, such as flexibility and transparency (among others). The experiments will be used combining device level devices (fabricated using electron beam lithography and photolithography) and nanoscale electronic measurements using conductive atomic force microscopy and in situ electrical characterization via transmission electron microscopy. We will also fabricate devices using ionic liquid to emulate exactly the functioning of synapses in the human brain. The experimental data collected will be matched with computational calculations.

本项目致力于实现神经形态计算中电子突触的设计，制造，表征和优化。该突触主要是由忆阻器来进行模拟，我们的具体方案是利用二维材料的优异特性来提高器件的性能，从而可以为突触提供更优异的性能。本项目的目的是理解导致阻变现象的主要因素，以及如何控制电流开/关比，循环周期，保留时间，功耗，开关时间和开关电压等。此外，我们同时也将致力于减小不同循环周期和不同器件之间的差异性，这也是忆阻技术面临的最大问题。更重要的是，进一步统计分析高低阻态切换的开关特性，这将帮助阐明该类器件呈现周期性阻态变化的物理和化学机制（即原子重排）。此外，我们将开发器件的更多优异性能，例如可弯曲性和透明性等。该实验将宏观器件（由电子束刻和光刻技术制成）、纳米尺度表征（导电原子力显微镜）以及TEM的原位电学表征相结合， 并利用离子液体来测试器件，从而精确模拟人脑中突触的功能。采集的实验数据，将和计算所得的结果相匹配。

利用二维材料构建忆阻器器件将赋予传统材料基忆阻器所不具备的一些特性，如高透明度、强机械强度、高柔韧性、强散热性能、超低功耗等，并且二维材料基器件可以同时表现双极性阻变和阈值阻变，另外二维材料基忆阻器可以模拟生物脑用作电子突触。本项目围绕“为实现神经态计算而设计、制造、表征和优化电子突触”的总目标开展研究，主要研究内容包括：1）使用2D材料作为开关层和/或电极，制造具有竞争力的电子突触。获得二维六方氮化硼基忆阻阵列器件良品率高达98%，且表现出超低的周期间差异性，低至1.53％和出色的器件间差异性，低至5.74％，实现三种阻态切换、无转移二维材料忆阻器构建等；2）深层阻变物理/化学机制的揭示。利用导电原子力显微镜、截面透射电子显微镜、电子损失能谱等对阻变层中的各种缺陷、导电细丝等进行物理尺寸和化学成份的定性与定量；3）构建出复杂二维材料忆阻器交叉杆神经网络阵列，且结构单元能够满足图像识别精度要求。而且，项目实施过程中成功在二维材料基忆阻器中观察到了随机电报噪声，并将这一现象应用于高级信息加密的一次性密码。本项目研究成果二维材料固态微纳电子中的应用提供基础，为构建面向未来的高密度存储和神经态计算提供了新的思路和方法。