Accelerating AI-driven materials innovation

AI-driven materials discovery has great potential to accelerate innovation, but it relies on large and diverse datasets. The lack of such data remains a major bottleneck in the field.

Through a collaboration between Murata Manufacturing Co., Ltd., and the National Institute for Materials Science (NIMS), researchers have built a comprehensive new database of dielectric material properties — a specific class of materials necessary for electronics — curated from thousands of scientific papers. Their study was published in Science and Technology of Advanced Materials: Methods.

The researchers used the Starrydata2 web system to collect experimental data on over 20,000 material samples from more than 5,000 publications. The team developed a standardised approach to extract data from graphs, including temperature-dependent properties, which are often omitted in other databases.

"What makes our work unique is the meticulous process of manually tracing graphs and correcting inconsistencies in original research papers to create a clean, high-quality dataset," the researchers say.

The scientists then used machine learning to predict properties of materials and how they would behave electronically.

Going further, the team created visual maps of the data to better understand the models’ predictions, including grouping similar materials to identify patterns and how a material's composition affects its properties.

This work advances our understanding of dielectric materials and moves research beyond traditional trial-and-error approaches. "By curating the largest dataset ever and combining various machine-learning methods, we succeeded in visualising the landscape of the entire compositional space with exceptional scope and range," the researchers explain. 

---

Did you know?

Dielectric materials are essential for modern electronics as energy and data storage, to prevent short circuits, to guide electromagnetic signals and much more.

---

The team plans to make the dataset publicly available this year (2026), allowing scientists worldwide to use it for new discoveries. Future work may involve expanding data collection to include manufacturing methods and processing conditions, enabling more comprehensive predictions that would intricately link production processes to material properties.

"We hope that this foundational work will inspire similar data collection initiatives and new approaches to materials discovery, leading to smarter materials development pathways that benefit society through improved electronic technologies," the researchers conclude.

Read the paper

Science and Technology of Advanced Materials: Methods: https://doi.org/10.1080/27660400.2025.2485018

Further information 

Tomoki Murata
[email protected] 
Murata Manufacturing Co., Ltd.

Yukari Katsura
[email protected] 
National Institute for Materials Science (NIMS)

STAM-M Inquiries 
[email protected] 
STAM Methods Editorial Office

Download this story

More Asia Research News 2026 stories