- Address
- 305-0047 1-2-1 Sengen Tsukuba Ibaraki JAPAN [Access]
Accepting Students
Research
- Keywords
Materials Informatics, Machine Learning, AI
PublicationsNIMS affiliated publications since 2004.
Published patent applications
Society memberships
The Physical Society of Japan, Japanese Society of Applied Physics
Center for Basic Research on Materials
Development of Materials Exploration Algorithms and Self-Driving Laboratory Systems
Materials Informatics, AI for Materials Discovery, Black-Box Optimization, Self-Driving Laboratories
Overview
In materials science, vast amounts of data are generated every day. Analyzing this data with data-driven methods makes it possible to propose materials with desired properties and to build guiding principles for materials design. However, simply applying existing methods is not sufficient; devising and applying algorithms tailored to specific research needs is essential for accelerating the exploration and development of new materials. Based on this idea, we have developed numerous data-driven algorithms specialized for materials research, including black-box optimization methods that can be used as artificial intelligence (AI) for materials exploration, and have released them as open-source software. We have also developed NIMO (https://www.nims.go.jp/nimo/), a platform for autonomous materials exploration that provides access to a variety of AI algorithms. Furthermore, by connecting NIMO to robotic experimental equipment, we are also focusing on building Self-Driving Laboratories that require no human intervention.
Novelty and originality
● Development of NIMO, a software platform for autonomous materials exploration
● Development of PHYSBO, a Python package for Bayesian optimization
● Development of PDC, a method for efficient phase diagram construction, and release of AIPHAD, a web application implementing it
● Development of materials exploration methods using quantum annealing and Ising machines
● Development of LLM-based methods for materials exploration and experimental planning (LLMEP, aLLoyM, etc.)
Details
In autonomous materials exploration, we start from a small amount of data and gradually increase it by alternately repeating property prediction based on machine learning and AI techniques with experiments based on that prediction, thereby optimizing material properties. This research approach is called black-box optimization. This can be used as artificial intelligence (AI) for materials exploration, proposing which material to test next. Among black-box optimization methods, we have developed PHYSBO, a Python package that makes it easy to run Bayesian optimization, the method most widely used in the materials research field, and have released it as open-source software. However, when the number of candidate materials is enormous, Bayesian optimization can require long computation times. To solve this problem, we have also developed FMQA, a black-box optimization method that uses quantum annealing and Ising machines. Furthermore, materials research often calls not only for improving material properties but also for constructing detailed phase diagrams. To meet this need, we have developed PDC (Phase Diagram Construction), a black-box optimization method for phase diagram construction, and have also released AIPHAD, a web application that makes it easy to run PDC. By connecting these AI systems for materials exploration with robotic experiments, Self-Driving Laboratories that require no human intervention can be realized. To this end, we have developed and released NIMO (https://www.nims.go.jp/nimo/) as middleware for realizing such systems. In NIMO, robotic experiments and AI for materials exploration are each treated as modules, allowing Self-Driving Laboratories to be implemented with various combinations of components. NIMO is now used not only at NIMS but also at research institutions in Japan and abroad, and its deployment as a foundational technology connecting Self-Driving Laboratories around the world is progressing.
Summary
Using the data-driven methods we have developed, we have conducted application research on a wide range of materials, including alloys, magnetic materials, organic molecular materials, polymer materials, metamaterials, Li-ion battery materials, solar cell materials, and electrolyte materials. We have also worked on applications that go beyond materials exploration itself, such as the development of olfactory sensors and the automation of Rietveld analysis. In addition, we are focusing on building Self-Driving Laboratories centered on NIMO, and, in collaboration with research institutions in Japan and abroad, are working toward realizing materials development platforms that require no human intervention. All of these efforts are grounded in the concept of AI for Materials Science, that is, embedding AI directly into the research process of materials science itself, and through the development and implementation of data-driven methods, we hope to continue contributing to innovative materials development.
