AI-enabled construction and prediction of atomic models for thin-film heterostructures via materials genome approach

We conduct an advanced study on the AI-enabled construction and prediction of atomic models for thin-film heterostructures using a materials genome approach. This research employs a novel computational strategy to predict the structural and electronic properties of new thin-film heterostructures, particularly focusing on flexible muscovite mica substrates. A detailed analysis is presented on Gallium Nitride (GaN), a material renowned for its optoelectronic applications, using it as a thin-film crystal. Notably, the Ga-polar GaN/Muscovite model stands out as the most stable configuration encountered in the study, exhibiting the lowest interface energy, calculated at −1.21 eV/Å2. This model demonstrates significant potential for enhanced device performance, primarily due to its ability to form twelve robust GaO bonds at the interface. Our findings not only highlight the utility of the materials genome approach in accurately predicting material properties but also underscore the advantages of AI in accelerating the design and analysis of semiconductor heterostructures. This methodological innovation opens new avenues for the development of advanced materials that could revolutionize optoelectronics and other technology sectors. Research Focus: The research focuses on the evolution of crystallographic and structural databases, starting with the introduction of the Crystallographic Information Framework (CIF) in 1991. The launch of the Inorganic Crystal Structure Database (ICSD) in 1997 provided a foundational repository for inorganic structures. Between 2006 and 2010, tools like PyCIFRW and CIF2Cell enhanced data interoperability and three-dimensional crystal modeling, while the Crystallography Open Database (COD) democratized access to crystallographic data. In 2020, the Heterostructures Open Database (HOD) integrated artificial intelligence to automate heterostructure modeling and predict interfacial energies, revolutionizing material design. These developments highlight the transformative impact of database advancements and AI integration on accelerating semiconductor and materials science research. [Display omitted] •The Heterostructures Open Database (HOD) successfully enabled the identification of the most stable N-polar GaN/K-terminated Muscovite model, with a low interface energy of −0.281 eV/Å2. This aligns with experimental results showing N-polar GaN formation under nitrogen-rich conditions. The study demonstrates the effectiveness of combining Arti