ASAHI
LABORATORY

Research

Materials Informatics

JP

In recent materials research, the amount of data obtained from experiments and calculations has been increasing rapidly. However, the discovery of new materials still requires extensive trial and error. It remains difficult to identify promising materials and conditions from a vast number of candidates.

Materials Informatics (MI) is a research field that integrates materials science and information science. It aims to make materials development more efficient and advanced by using machine learning and optimization techniques. In our group, we combine crystal structure prediction, machine learning, first-principles calculations, and quantum computing techniques to explore new materials and predict their properties.

Figure 1 What is materials informatics?

Figure 1 What is materials informatics?

Data Driven Research on Perovskite Solar Cells

Perovskite solar cells have attracted attention as next generation solar cells that are expected to achieve both high power conversion efficiency and low cost fabrication. However, their performance and durability are strongly affected not only by material composition, but also by fabrication conditions and device structures.

Our group uses experimental data and literature data reported by researchers around the world to predict device performance and evaluate stability using machine learning. [1] In addition, we aim not only to make predictions, but also to understand the factors that affect performance and durability through model interpretation. [2]

Figure 2. Data driven research on perovskite solar cells. Based on [1].

Figure 2. Data driven research on perovskite solar cells. Based on [1].

Crystal Structure Prediction of Perovskite Materials

Material properties are strongly affected not only by chemical composition, but also by the arrangement of atoms and molecules. In particular, in organic inorganic hybrid perovskites, the orientation of organic molecules and differences in crystal structures are known to have a significant influence on material properties.

Our group is developing efficient crystal structure prediction methods for perovskite materials by using crystal structure search algorithms, machine learning potentials, first principles calculations, and annealing machines. [3] We also investigate structure candidates that are difficult to observe by experiments alone through computational approaches, aiming to apply the obtained knowledge to materials design.

Figure 3. Crystal structure prediction of perovskite materials.

Figure 3. Crystal structure prediction of perovskite materials.

Crystal Structure Prediction and Property Design of Organic Molecular Crystals

In organic molecular crystals, even crystals formed from the same molecule can show large differences in optical properties, electronic properties, stability, and other properties, depending on molecular arrangement and crystal structure.

Our group explores stable structures of organic molecular crystals using crystal structure prediction techniques. We also analyze the relationship between the obtained crystal structures and their properties, aiming to establish design guidelines for new functional materials.

Figure 4. Crystal structure prediction and property design of organic molecular crystals. Based on [4].

Figure 4. Crystal structure prediction and property design of organic molecular crystals. Based on [4].

Materials Exploration Using Machine Learning and Optimization Techniques

In materials development, there are many variables, such as composition, crystal structure, and fabrication conditions. Therefore, finding optimal materials becomes a large scale search problem.

Our group develops more efficient methods for materials exploration by using machine learning, Bayesian optimization, annealing, and other optimization techniques. By applying these techniques to crystal structure prediction and materials design, we aim to realize the next generation of data driven materials research.

In addition to perovskite materials and organic molecular crystals, our group studies a wide range of materials, including energy materials, electronic materials, and functional materials. [5-9] Through industry academia collaboration and joint research, we are working to establish new methods for materials development based on computational science and data science.

Asahi Laboratory Publications