Where 2D meets ML: Defects and Reaction Kinetics at the atomic monolayer limit

过渡金属二硫属化物（TMDs）是一类有前途的二维材料，具有革命性的纳米电子学和实现二维材料的长期承诺的潜力。贯穿2D材料研究的一个反复出现的主题是缺陷对所观察到的材料性质的贡献的重要性。在TMDs的情况下，已经发现了一系列缺陷，其中置换氧在患病率方面占主导地位。这些缺陷和它们的集合体在这些材料的氧化中也起着至关重要的作用，如WS2所示。氧化可以是一把双刃剑，导致无意的腐蚀，或作为工具，这取决于材料，用于清洁蚀刻（例如WS2）或用于在设备中创建清洁界面（例如HfS2）。缺陷和表面反应的研究，以了解介观，系综测量传统上依赖于高时空分辨率的技术，如扫描隧道显微镜和从头计算。然而，这些方法通常限于研究孤立的、静态的点缺陷，并且常常不能捕获缺陷在实际操作条件下的动态行为。此外，这些技术往往忽略了与操作条件相关的复杂性，导致结果可能缺乏通用性和统计相关性。该项目旨在通过开发一个高通量计算实验框架来弥合这一差距，该框架结合了一种扩展的多尺度建模方法，用于实现对缺陷动力学和氧化的原子理解，由机器学习原子间势提供动力，并采用非侵入性光学椭圆偏振法。最近开发的消息传递神经网络，MACE，训练密度泛函理论计算，将被用来研究缺陷动力学和氧化热力学和动力学障碍。同时，光学椭圆偏振法将提供TMD厚度的非侵入性测量并监测氧化动力学。通过比较实验和计算的动力学速率，我们的目标是提供一个完整的，微观的动态行为的缺陷在TMD图片。

Transition metal dichalcogenides (TMDs) are a class of promising 2D materials with the potential to revolutionize nanoelectronics and fulfill the longstanding promise of 2D materials. A recurring themeacross 2D materials research is the significance of defect contribution to the observed material properties. In the case of TMDs, a constellation of defects have been found, with substitutional oxygen dominating in terms of prevalence. These defects and their ensembles also play a crucial role in the oxidation of these materials, as exemplified by WS2. Oxidation can be a double-edged sword, leading to unintentional corrosion, or serving as a tool, depending on the material, for clean etching (e.g. WS2) or for creating clean interfaces (e.g. HfS2) in devices. The study of defects and surface reactions to understand mesoscopic, ensemble measurements has traditionally relied on high spatiotemporal resolution techniques such as scanning tunneling microscopy and ab initio calculations. However, these methods are typically constrained to studying isolated, static point defects and often fail to capture the dynamic behavior of defects under realistic operating conditions. Furthermore, these techniques often overlook the complexity associated with operating conditions, leading to results that may lack generality and statistical relevance. This project aims to bridge this gap by developing a high-throughput computational-experimental framework that combines an extended, multi-scale modelling method for attaining an atomistic understanding of defect dynamics and oxidation, powered by machine learning interatomic potentials, with the noninvasive optical ellipsometry. The recently developed Message Passing Neural Network, MACE, trained on density functional theory calculations, will be employed to study defect dynamics and oxidation thermodynamic and kinetic barriers. Concurrently, optical ellipsometry will provide non-invasive measurements of TMD thickness and monitor oxidation kinetics. By comparing experimental and computational kinetic rates, we aim to provide an integrated, microscopic picture of the dynamic behavior of defects in TMDs.