Neuromorphic devices based on 2D layered materials heterostructures

人工神经网络(ANN)是人工智能应用的核心。然而，人工神经网络通常是由传统的冯·诺伊曼体系结构实现的，其中存储器和逻辑是分开的，需要连续的数据交换，这导致了处理瓶颈和巨大的功率消耗。神经形态计算提供了人工神经网络的另一种方法，它从人脑中汲取灵感，基于人工神经元和突触，记忆和逻辑位于同一位置。本博士项目致力于设计、模拟和测试基于不同2D/层状材料(2DLM)形成的异质结构的新型高效神经形态器件，这些异质结构由不同的2D/分层材料(如石墨烯和过渡金属二卤化物)形成。这种材料具有数十个实验可用的2DLm和超过2000个理论预测的2DLm，并且可以按任意顺序和相互角度堆叠它们，在材料组合、原子精确的界面和缺陷工程方面提供了前所未有的灵活性。该项目涉及不同神经形态设备的设计、(纳米)制造和电气测试。实验活动将通过有限元模拟完成。该项目将探索基于2DLM异质结构的不同垂直器件结构，用于节能的神经形态计算。基于HfS2等2DLM的热氧化和等离子体氧化的最新结果，该项目将首先专注于二维绝缘体的研究，旨在了解氧化过程、如此形成的氧化物和半导体/氧化物界面的性质。然后，重点将转移到器件设计和制造上，并将探索不同的结构和材料组合，并将测试和优化单个器件的性能。最后，几个设备将被组合成阵列，这些阵列将用于执行基本的机器学习任务，如模式识别或分类。在整个项目中，将特别注意材料和设备的可扩展性，以及与现有硅基信通技术的整合。该项目将既是实验性的，也是计算性的。该项目的实验部分将包括通过使用最先进的、专门设计的手套盒系统来制造2DLM异质结构，该系统允许在惰性气氛中精确堆叠不同的2DLM，以及控制界面污染、高温退火、精确引入缺陷和等离子刻蚀。神经形态器件的制造将使用基于洁净室的微制造(光学和电子束光刻、反应离子刻蚀、金属蒸发等)。设备性能将使用探针站和一整套测试设备(源/测量单元、脉冲发生器、阻抗分析仪、矢量网络分析仪、示波器、锁定放大器等)进行测试。设备测试将包括直流扫描、电压脉冲、保持力和耐久性测试。该项目的计算部分将包括有限元模拟，主要通过技术计算机辅助设计(TCAD)工具Synopsys Sentaurus实现。如果需要，COMSOL多物理公司和CST微波工作室将对TCAD模拟进行补充。该项目与EPSRC的不同领域很好地结合在一起，涵盖了多个主题。该项目尤其与“人工智能技术”、“微电子设备技术”和“石墨烯和碳技术”领域保持一致。这项研究将与电阻开关器件领域的顶尖专家Kenyon教授和Mehonic博士(UCL电子电气工程系)密切合作，他们是设计和制造最先进的硅阻随机存取存储器的“固有”衍生公司的创始人和首席技术官/首席技术官。媒体

Artificial neural networks (ANNs) are at the core of artificial intelligence applications. However, ANN are usually implemented by "conventional" von Neumann architectures, where memory and logic are separated, requiring continuous data exchange which causes processing bottlenecks and large power consumption. An alternative approach to ANN is provided by neuromorphic computing, which takes inspiration from the human brain and is based on artificial neurons and synapses, where memory and logic are co-located. This PhD project focusses on design, simulation and testing of novel energy-efficient neuromorphic devices based on heterostructure formed by combining different 2D/layered materials (2DLM) such as graphene and transition metal dichalcogenides. With tens of 2DLM experimentally available and over 2,000 theoretically predicted and the possibility of stacking them in arbitrary order and mutual angle, such materials offer unprecedented flexibility in terms of combination of materials, atomically-precise interfaces and defect engineering. The project involves design, (nano)fabrication and electrical testing of different neuromorphic devices. The experimental activity will be completed by finite-element simulations.The project will explore different vertical device structures based on 2DLM heterostructures to be used for energy-efficient neuromorphic computing. Building on recent results on thermal and plasma oxidation of 2DLM such as HfS2, the project will initially focus on the investigation of two-dimensional insulators, aiming to understand of the oxidation process, the properties of the so-formed oxide and semiconductor/oxide interfaces. The focus will then shift towards device design and fabrication, and different structures and combinations of materials will be explored and the performance of individual devices will be tested and optimized. Finally, several devices will be combined to form arrays, which will be used to perform basic machine learning tasks such as pattern recognition or classification. Throughout the project, particular attention will be paid to scalability of materials and devices, as well as integration with existing silicon-based ICT technology. The project will be both experimental and computational. The experimental part of the project will consist in the fabrication of 2DLM heterostructures by using a state of the art, purposely-designed glovebox system which allows precise stacking of different 2DLM in an inert atmosphere, as well as control of interface contaminations, high-temperature annealing, precise introduction of defects and plasma etching. Fabrication of the neuromorphic devices will be completed using cleanroom-based microfabrication (optical and electron beam lithography, reactive ion etching, metal evaporation, etc). Device performance will be tested using a probe station coupled with an ensemble of testing equipment (source/measure units, pulse generators, impedance analysers, vector network analysers, oscilloscopes, lock-in amplifiers, etc.). Device testing will include DC sweeps, voltage pulses, retention and endurance tests. The computational part of the project will consist of finite-element simulations, achieved mainly via a Technology Computer Aided Design (TCAD) tool, Synopsys Sentaurus. TCAD simulations will be complemented by Comsol Multiphysics and CST Microwave Studio when needed. The project is well aligned to different EPSRC areas, encompassing multiple themes. In particular, the project is aligned with the "Artificial intelligence technologies", "Microelectronic device technology" and "Graphene and carbon technology" areas. The research will be conducted in close collaboration with Prof. Kenyon and Dr Mehonic (UCL Electronic and Electrical Engineering Dpt), leading experts in resistance-switching devices and founders and CSO/CTO of the "IntrinSic" spin-out company, which designs and manufactures state of the art silicon resistive random-access memories. The pr