RII Track-4: Optimizing the Chemistry of Heterointerfaces in Photovoltaics: A Combination of Electronic Structure Calculations and Machine Learning Approach

材料的性质通常与其内部界面的原子结构和化学性质有关。在内部界面中，异质界面是两种具有不同原子结构和化学性质的材料之间的边界。这些异质界面的原子结构和化学性质是复杂的，很难从形成异质界面的单个材料中预测出来。这种复杂性在多组分异质界面中呈指数增长，因为它涉及界面上大量的化学可能性。因此，设计具有目标材料特性的新型多组分异质界面的化学性质是一项具有挑战性的任务。本研究项目旨在通过结合电子结构计算和机器学习方法，提高我们目前的能力，以确定光伏应用中具有目标电子特性的多组分异质界面的化学性质。机器学习工具可以有效地提取隐藏的化学性质关系，从而大大减少识别满足所需电子性质需求的异质界面的化学性质所需的时间，这是加速发现新型异质界面的关键因素。拟议的项目解决了联邦政府对材料基因组计划的重要授权，其目标是大幅减少发现、制造和部署先进材料的时间和成本。多组分异质界面长期以来一直引起材料科学家和物理学家的兴趣，部分原因是其原子和电子结构和化学的复杂性。这种复杂性使得设计具有目标材料特性的新型多组分异质界面成为一项艰巨的任务。这是因为在界面上的多个元素之间导航巨大的组合化学和构型可能性实在太大了。本研究项目旨在结合电子结构计算和机器学习方法，设计具有目标电子特性的光伏应用异质界面的化学成分。更具体地说，PI计划(a)利用电子结构计算来深入了解决定异质界面电子和原子结构的基础物理；(b)应用机器学习工具来探索界面的隐藏化学-性质关系，以预测期望电子性质的异质界面的化学性质。所提出的方法与传统的耗时且昂贵的爱迪生合成测试实验循环的试错方法不同；因此，它可以大大加快材料的发现。PI预计，在该项目完成后，将产生一个通用的计算模板来研究高度复杂的多组分异质界面的结构-化学-性质关系。最后，机器学习将成为未来材料发现的重要组成部分。从这个项目中获得的知识将有助于培养学生的机器学习工具及其在材料研究中的应用，增加他们在职业生涯早期接触这个不断发展和有影响力的材料科学领域的机会。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Material properties are often linked to the atomic structure and chemistry of its internal interfaces. Among the internal interfaces, heterointerfaces are boundaries separating two materials with different atomic structure and chemistry. The atomic structure and chemistry of these heterointerfaces are complex and cannot be easily predicted from the individual materials that form the heterointerface. This complexity increases exponentially in multi-component heterointerfaces as it involves a vast number of chemical possibilities at the interface. Thus, designing the chemistry of a novel multi-component heterointerface with a targeted material property is a challenging task. This research project aims to enhance our current capability to determine the chemistry of a multi-component heterointerface with a targeted electronic property for photovoltaic applications using a combination of electronic structure calculations and machine learning approach. Machine learning tools can substantially reduce the time needed to identify the chemistry of a heterointerface that meets a desired electronic property need by efficiently extracting hidden chemistry-property relationships, a key factor toward accelerated discovery of novel heterointerface. The proposed project addresses the important federal government mandate of Materials Genome Initiative, the objective of which was to substantially reduce the time and cost to discover, manufacture, and deploy advanced materials.Multicomponent heterointerfaces have long intrigued materials scientists and physicists, in part, because of the sheer complexity of their atomic and electronic structure and chemistry. Such complexity can make designing a novel multi-component heterointerface with a targeted material property a non-trivial task. This is because navigating the vast combinatorial chemical and configurational possibilities between multiple elements at the interface is simply too large. This research project aims to design the chemical composition of a heterointerface for photovoltaic application with a targeted electronic property using a combination of electronic structure calculations and machine learning approach. More specially, the PI plans (a) to develop an in-depth understanding of the underlying physics that determines the electronic and atomic structure of the heterointerface using electronic structure calculations; and (b) to apply machine learning tools to explore hidden chemistry-property relationships of the interface, to predict the chemistry of the heterointerface for a desired electronic property. The proposed approach is a departure from the traditional time-consuming and expensive Edisonian trial-and-error approach of synthesis-testing experimental cycles; thus, it can substantially accelerate materials discovery. The PI anticipates that upon completion of this project a generic computational template to investigate structure-chemistry-property relationship of highly complex multi-component heterointerfaces will be generated. Finally, machine learning will be an integral part of future materials discovery. The knowledge gained from this project will help train student(s) on machine learning tools and their utility in materials research, increasing their exposure early in their careers to this growing and influential materials science field.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.