Machine learning (ML) has been demonstrated by scientists to enable the accurate and efficient computation of fundamental electronic properties of binary and ternary oxide surfaces. Their model, which is based on machine learning, has the potential to be applied to other compounds and properties. This research can assist in the screening of surface properties of materials and the development of functional materials.
The research outcomes can help in evaluating the surface features of materials and creating functional materials. To create new materials with better properties, it is crucial to thoroughly examine their atomic and electronic structures. Parameters like ionization potential (IP) and electron affinity (EA) provide valuable information about electronic band characteristics.The configuration of semiconductors, insulators, and dielectrics’ surfaces is crucial. Determining the exact values of IPs and EAs in these nonmetallic materials can help determine their suitability for use in functional surfaces and interfaces for photosensitive and optoelectronic devices.
Furthermore, the values of IPs and EAs are heavily influenced by the surface structures, adding complexity to their quantification. The traditional method of calculating IPs and EAs involves using precise first-principles calculations to separately quantify the bulk and surface systems. This time-consuming process hinders the accurate quantification of IPs and EAs.Many surfaces require efficient computational approaches for their analysis.
To tackle the various challenges involved in quantifying IPs and EAs of nonmetallic solids, a group of researchers from Tokyo Institute of Technology (Tokyo Tech), led by Professor Fumiyasu Oba, have shifted their focus to machine learning (ML). Their study has been published in the Journal of the American Chemical Society.
Professor Oba explains the reason behind their current research, stating “In recent years, ML has gained significant attention in materials science research. The capability to screen materials virtually has become a key area of interest.”Using machine learning technology is a highly effective method for discovering new materials with exceptional properties. Additionally, the capability to train extensive datasets using precise theoretical calculations enables the accurate prediction of significant surface traits and their practical consequences.”
The scientists utilized an artificial neural network to create a regression model, integrating the smooth overlap of atom positions (SOAPs) as numerical input data. Their model effectively and accurately foresaw the IPs and EAs of binary oxide surfaces by utilizing data on bulk crystal structures and surface termination planes.
Additionally, the ML-based prediction model has the capability of ‘transfer learning,’ which means a model created for a specific purpose can be adjusted to include new datasets and reused for other tasks. The researchers integrated the impacts of multiple cations into their model by creating ‘learnable’ SOAPs and used transfer learning to predict the IPs and EAs of ternary oxides.
Professor Oba’s concluding statement emphasizes that “Our model is not limited to predicting surface properties of oxides, but can also be applied to study other compounds and their properties.
