Search
Close this search box.
Search
Close this search box.

Tokyo University of Science Innovates with Deep Learning to Expedite the Search for Single-Molecule Magnets

Tokyo University of Science Innovates with Deep Learning to Expedite the Search for Single-Molecule Magnets

Science News / Artificial Intelligence & Machine Learning

In an era where the demand for advanced memory storage and quantum computing is rapidly escalating, a team of researchers from Tokyo University of Science (TUS) has made a groundbreaking stride in materials science. By harnessing the capabilities of deep learning, they have significantly sped up the discovery process of single-molecule magnets (SMMs), which are at the forefront of next-generation computing technologies.

The study, led by Professor Takashiro Akitsu alongside Assistant Professor Daisuke Nakane and Mr. Yuji Takiguchi, was published in the International Union of Crystallography Journal (IUCrJ) on February 1, 2024. Their research presents a pioneering deep learning model that can predict SMMs from an extensive database of 20,000 metal complexes, using just the crystal structures as the determining factor.

Prof. Takashiro Akitsu
Prof. Takashiro Akitsu – World Congress on Materials Science & Engineering 2024 (scientificcollegium.net)

The allure of SMMs lies in their unique magnetic relaxation behaviors!

Moreover, observable even at the scale of single molecules. These behaviors are key to the development of ultra-high-density storage media and quantum computing devices. Traditionally, identifying such materials has involved complex and costly experiments, but the TUS team’s approach bypasses these hurdles by employing a machine learning model capable of recognizing the intricate patterns in molecular structures that suggest SMM properties.

This deep learning model, which operates on the basis of a 3D Convolutional Neural Network using the renowned ResNet architecture, was trained using a dataset of crystal structures with salen-type ligands. Remarkably, it achieved an accuracy rate of 70% in categorizing molecules as SMMs or non-SMMs. Furthermore, when applied to the broader dataset, it successfully pinpointed metal complexes previously confirmed as SMMs, thereby validating its efficacy.

The TUS researchers took a meticulous approach, analyzing over 800 research papers published between 2011 and 2021, to compile their dataset.

They enriched their analysis with 3D structural data from the Cambridge Structural Database, ensuring a robust and reliable foundation for their model’s training.

In their findings, the model highlighted several multinuclear dysprosium complexes. Moreover, renowned for their substantial effective energy barriers—essential for stabilizing magnetic moments in SMMs. Despite the success, the researchers emphasize the importance of supplementary experimental work to fully understand the SMM behavior under standardized conditions.

Professor Akitsu’s team is optimistic about the implications of their work. By reducing the dependence on intricate quantum chemical calculations and laborious simulations of magnetism, this approach can streamline the molecular design process. The result is a more efficient pathway to the discovery of functional materials, with considerable savings in time, resources, and expenses.

In conclusion, this impressive leap forward in applying deep learning to material sciences not only sets a precedent for the discovery of SMMs. However, also opens the door for AI to play a more significant role in the innovative design of materials across various disciplines.

Journal: International Union of Crystallography Journal (IUCrJ)

DOI: https://doi.org/10.1107/S2052252524000770

Tokyo University of Science