Machine learning (ML) enables accurate and efficient calculation of the fundamental electronic properties of binary and ternary oxide surfaces, as demonstrated by scientists. Their ML-based model could be extended to other compounds and properties. The results of the present research can help in the evaluation of the surface properties of materials, as well as in the development of functional materials.
The design and development of new materials with superior properties requires an exhaustive analysis of their atomic and electronic structures. Electronic energy parameters, such as ionization potential (IP), the energy required to remove an electron from the maximum of the valence band, and electron affinity (EA), the amount of energy released when an electron binds to the minimum of the conduction band, reveal important information about the electronic band structure of semiconductor, insulator and dielectric surfaces. Accurate estimation of IP and EA in such non-metallic materials can indicate their applicability for use as functional surfaces and interfaces in photosensitive equipment and optoelectronic devices.
Furthermore, PIs and EAs depend significantly on surface structures, which adds another dimension to the complex procedure of their quantification. Traditional calculation of IP and EA involves the use of precise first-principles calculations, where bulk and surface systems are quantified separately. This time-consuming process prevents quantification of PIs and EAs for many surfaces, requiring the use of computationally efficient approaches.
To address the various issues affecting the quantification of IP and EA of non-metallic solids, a team of scientists at the Tokyo Institute of Technology (Tokyo Tech), led by Professor Fumiyasu Oba, has turned its attention to machine learning (ML). ). The results of their research have been published in theJournal of the American Chemical Society.
Professor Oba shares the motivation behind the present research: “In recent years, ML has gained a lot of attention in materials science research. The ability to virtually examine materials based on ML technology is one way very efficient of exploring novel materials with superior properties. Furthermore, the ability to train large data sets using precise theoretical calculations allows for the successful prediction of important surface characteristics and their functional implications.”
The researchers used an artificial neural network to develop a regression model, incorporating soft superposition of atomic positions (SOAP) as numerical input data. Their model accurately and efficiently predicted the PIs and EAs of binary oxide surfaces using information on bulk crystal structures and surface termination planes.
Additionally, the ML-based prediction model could “transfer learning,” a scenario in which a model developed for a particular purpose can incorporate newer data sets and be reapplied for additional tasks. The scientists included the effects of multiple cations in their model by developing “learnable” SOAPs and predicted the PIs and EAs of ternary oxides using transfer learning.
Professor Oba concludes by saying: “Our model is not limited to predicting the surface properties of oxides, but can be extended to study other compounds and their properties.”