This story is featured in the Asia Research News 2026 magazine. Book space in the next issue now.
Get the news in your inbox
In materials science, thin films are microscopic layers of conductive material that are coated onto a surface. These are often used in semiconductors or communication devices, but they are hampered by a major problem. Tree-like structures, called dendrites, can form during or after coating and grow as ions move along the film following paths of least resistance. Dendrites affect the structure and function of thin films, inhibiting the flow of electricity and reducing overall performance — a major thorn in the side of ultra-fast communication technology.
A team of researchers from the Tokyo University of Science, led by Masato Kotsugi, has developed a novel method for understanding how and why these structures form, with a view to one day preventing them. For this interdisciplinary study, the team combined topology (the mathematical study of the properties of a geometric object), physics, and machine learning. Their paper was published in Science and Technology of Advanced Materials: Methods.
Thin films are very thin conductive material coated onto surfaces of semiconductors or communication devices but dendrites resembling a tree or snowflake can form reducing performance, a major problem for ultra-fast communication technology.
The team used a method called persistent homology (PH), a form of topological data analysis that captures the shape of complex surfaces. This enabled them to capture the complex structure of dendrites in ways conventional imaging can’t. Next, they used a machine learning technique known as principal component analysis (PCA) to study how these shapes were related to changes in Gibbs free energy — the energy available in a substance for chemical reactions. They were then able to establish a relationship between structure and process in the growth of dendrites, revealing the energy gradients that drive the branching behaviour.
“What makes this research stand out is the development of a completely new kind of AI model,” explains Kotsugi. “Unlike conventional ‘black-box’ AI, our method is explainable and connects two things that were never directly linked before: the shape of the material’s structure and the energy changes that drive its growth.”
The team hopes this new approach could help improve the development of next-generation electronics, particularly in high-speed communication technologies that could take us beyond 5G. These faster, more efficient communication technologies are needed to support the rapid growth of autonomous machines and Internet of Things sensing devices, such as wearable tech, smart home devices, and autonomous or connected vehicles.
The researchers plan to broaden their framework to take into account more energy forms than just Gibbs free energy so they can create a more universal model for materials design that links process, structure, and function. “This will allow us to analyse a variety of materials in a unified way,” says Kotsugi. “Ultimately, this approach will establish a physics-based, explainable AI for science, paving the way for innovative materials design.”
Read the paper
Science and Technology of Advanced Materials: Methods: https://doi.org/10.1080/27660400.2025.2475735
Further information
Prof Masato Kotsugi
[email protected]
Tokyo University of Science
STAM-M Inquiries
[email protected]
STAM Methods Editorial Office
Download this story
Advertisement
