npj Computational Materials

Interpretable discovery of semiconductors with machine learning

Hitarth Choubisa, Petar Todorovi？, Joao M. Pina, Darshan H. Parmar, Ziliang Li, Oleksandr Voznyy, Isaac Tamblyn & Edward H. Sargent

npj Computational Materials 9: 117(2023)

doi.org/10.1038/s41524-023-01066-9

Published online: 29 June 2023
Abstract| Full Text | PDF OPEN

Abstract: Machine learning models of material properties accelerate materials discovery, reproducing density functional theory calculated results at a fraction of the cost. To bridge the gap between theory and experiments, machine learning predictions need to be distilled in the form of interpretable chemical rules that can be used by experimentalists. Here we develop a framework to address this gap by combining evolutionary algorithm-powered search with machine-learning surrogate models. We then couple the search results with supervised learning and statistical testing. This strategy enables the efficient search of a materials space while providing interpretable design rules. We demonstrate its effectiveness by developing rules for the design of direct bandgap materials, stable UV emitters, and IR perovskite emitters. Finally, we conclusively show how DARWIN-generated rules are statistically more robust and applicable to a wide range of applications including the design of UV halide perovskites.

摘要: 材料性质的机器学习模型加速了材料的发现，以极低成本分再现了密度泛函理论计算的结果。为了弥合理论和实验之间的差距，机器学习预测需要以实验人员可以使用的可解释的化学规则的形式提炼出来。在这里，我们开发了一个框架，通过将进化算法驱动的搜索与机器学习代理模型相结合，来解决这一差距。然后，我们将搜索结果与监督学习和统计测试相结合。这种策略能够有效地搜索材料空间，同时提供可解释的设计规则。我们通过开发用于设计直接带隙材料，稳定紫外发射器和红外钙钛矿发射器的规则来证明其有效性。最后，我们展示了DARWIN生成的规则如何在统计上更加稳健，并适用于广泛的应用，包括紫外卤化物钙钛矿的设计。

Editorial Summary

Interpretable discovery of semiconductors: machine learning

Machine learning has emerged as a powerful tool for accelerating materials discovery, yet it faces challenges in efficient material search and model interpretability. Inverse materials design enables the rapid exploration of vast material spaces to design materials with optimal properties, but it struggles to uncover chemical design rules. On the other hand, interpretable methods can explain the properties of candidate materials but are unable to extract chemical design rules and theories from well-trained machine learning models. To overcome the challenge, this study proposes a machine learning framework that bridges the gap between theory and experiment. A team led by Professor Edward H. Sargent from the Department of Electrical and Computer Engineering at the University of Toronto, has developed a machine learning framework called Deep Adaptive Regressive Weighted Intelligent Network (DARWIN). DARWIN is capable of efficiently searching material spaces and providing interpretable chemical design rules. This framework integrates a machine learning surrogate model, a search algorithm, and knowledge extraction methods. Initially, it generates multiple candidate materials satisfying the desired target properties using surrogate models and search algorithms. Subsequently, statistical techniques and supervised learning are employed to generate and identify chemically interpretable design rules. The effectiveness of DARWIN is demonstrated through the successful design of two classes of direct bandgap materials, including stable UV emitters and infrared perovskite emitters. Furthermore, the study showcases the framework is statistically more robust and applicable to a wide range of applications.

AI探秘：半导体的可解释性发现

机器学习已经成为加速材料发现的强大工具，但还面临着材料高效搜索和机器学习模型可解释性不足的挑战。逆向材料设计可以快速搜索庞大材料空间，以设计具有最佳性质集的材料，但却难以挖掘出化学设计规则。另一方面，可解释性方法可以解释候选材料的性质，但却不能从训练好的机器学习模型中提取化学设计规则和理论。为了克服这一挑战，该研究提出了一种机器学习框架，弥合了理论和实验之间的差距。来自加拿大多伦多大学电气与计算机工程系的Edward H. Sargent教授团队，开发了一种深度自适应回归加权智能网络（DARWIN）机器学习框架，既能够有效地搜索材料空间，也可以提供可解释的化学设计规则。该框架结合了机器学习代理模型、搜索算法和知识提取方法。首先，通过代理模型和搜索算法生成满足所需目标性质的多个候选材料。然后，通过统计技巧和监督学习来生成和试别化学设计规则。他们成功设计两类直接带隙材料，包括稳定紫外发射材料和红外钙钛矿发射材料，证实了DARWIN框架的实用性。此外，他们还展示了该框架生成的规则在统计上更为健壮，并可适用于广泛的应用领域。