In a groundbreaking advancement poised to accelerate the future of electronic materials, researchers from the newly formed Institute of Science Tokyo alongside the Institute of Statistical Mathematics and other prestigious institutions have harnessed the power of machine learning to identify and design liquid crystalline polymers with exceptional thermal conductivity. This pioneering work addresses one of the critical bottlenecks in modern electronic device engineering: the efficient dissipation of heat through polymer materials that can endure extreme conditions without compromising electrical or mechanical integrity.
Liquid crystalline polymers (LCPs), particularly polyimides, have long stood as promising candidates for thermal management applications due to their unique molecular ability to self-organize into highly ordered structures that facilitate heat transfer. Yet, the challenge of predicting which polyimide molecular architectures would reliably exhibit liquid crystalline phases has, until now, been largely unresolved. Traditional design approaches relied heavily on iterative trial and error, hampering the speed of innovation and material optimization. The latest research disrupts this paradigm by introducing a data-driven route that dramatically shortens development cycles through predictive modeling.
At the core of this breakthrough is a sophisticated machine learning model that functions as a binary classifier, precisely forecasting whether a given polymer chemical structure will form a liquid crystalline phase. The model demonstrates an outstanding classification accuracy of 96%, marking a world first in polymer materials science. Developed using an extensive dataset sourced from PoLyInfo—the comprehensive polymer property database maintained by the National Institute for Materials Science—the algorithm assimilates subtle chemical, physical, and structural cues indicative of liquid crystallinity.
The methodological foundation of the study involved compiling a large virtual library of over 115,000 hypothetical polyimide candidates. This library was generated by systematically recombining fundamental building blocks—five core molecular fragments derived from acid dianhydrides and diamines—originally designed by the research team. Each distinct pair in this molecular toolbox represents a potential polyimide chain configuration, encompassing a vast diversity in chemical composition and predicted physical properties.
Subsequent computational screening identified approximately 10,800 candidates with a high likelihood of forming the sought-after smectic liquid crystalline phase, which is characterized by parallel molecular alignment conducive to lateral heat conduction. Experimental synthesis of six representative candidates verified the model’s predictions, with these novel polyimides achieving thermal conductivities up to 1.26 watts per meter-kelvin (W m⁻¹ K⁻¹). This measured performance significantly surpasses conventional polyimides, which typically exhibit lower thermal conductivities, thereby validating the model’s efficacy in guiding material discovery.
The relationship between molecular rigidity, alignment, and thermal transport was elucidated through detailed experimental characterization. More rigid polyimide backbones promote enhanced in-plane molecular order, creating consistent pathways for phonon transport—the primary mechanism of thermal conduction in polymers. These findings provide crucial insight into molecular design principles necessary to engineer next-generation thermally conductive polymers optimized for applications in semiconductor cooling, flexible electronics, and aerospace insulation.
This research constitutes a watershed moment signaling the increasing integration of artificial intelligence tools in fundamental materials science. The ability to predictably tailor polymer properties using machine learning accelerates innovation beyond conventional limitations and showcases the potential for rapid, cost-effective development of polymers with finely tuned thermal, mechanical, and electronic properties. As co-author Professor Teruaki Hayakawa succinctly notes, “Machine learning is transforming polymer design from intuition-driven art into a quantitative science.”
The collaborative nature of the effort, combining synthetic polymer chemistry, computational modeling, and data science, underscores the multidisciplinary approach required to address complex challenges at the intersection of materials and device engineering. Support from key stakeholders like the Japan Science and Technology Agency and the Ministry of Education, Culture, Sports, Science and Technology highlights the strategic importance attributed to these emerging technologies for Japan’s scientific leadership and industrial competitiveness.
Looking forward, the team envisions extending their ML-based framework to other classes of liquid crystalline materials beyond polyimides. The scalability of their approach opens pathways to discover entirely new polymers with programmable functionalities such as enhanced electrical conductivity, optical anisotropy, or biodegradability, thereby impacting a broad spectrum of technological areas. Moreover, this trailblazing work exemplifies how the synergy between chemical intuition and computational power can unlock previously inaccessible material landscapes.
Published in the 11th volume of the esteemed journal npj Computational Materials on July 2, 2025, this study not only sets a new benchmark for polymer thermal material development but also exemplifies how data science-driven discovery reshapes traditional materials research. As demand for more efficient thermal management solutions grows exponentially with advancing electronics miniaturization and performance requirements, such innovations are expected to play a pivotal role.
The research leadership included Principal Investigator Professor Junko Morikawa at Institute of Science Tokyo, with significant contributions from Professors Teruaki Hayakawa and Ryo Yoshida, whose group developed the binary classification model. The hands-on synthesis and thermal characterization efforts were led by Morikawa’s team, including graduate students Hayato Maeda and Shiori Nakagawa, while Associate Professor Stephen Wu spearheaded the collaborative project management from the Institute of Statistical Mathematics.
In essence, this machine learning-enabled discovery exemplifies the future trajectory of functional polymer design, where virtual material libraries and predictive analytics condense years of experimental labor into months or even weeks. This transformative approach promises to unlock advanced polymeric materials custom-tailored across a spectrum of applications, from next-generation electronics to sustainable technologies, fundamentally changing how materials science innovation unfolds.
Subject of Research: Not applicable
Article Title: Discovery of liquid crystalline polymers with high thermal conductivity using machine learning
News Publication Date: 2-Jul-2025
Web References:
https://www.nature.com/articles/s41524-025-01671-w
http://dx.doi.org/10.1038/s41524-025-01671-w
References:
Morikawa, J., Hayakawa, T., Yoshida, R., Wu, S., Maeda, H., Nakagawa, S. (2025). Discovery of liquid crystalline polymers with high thermal conductivity using machine learning. npj Computational Materials, 11.
Image Credits: Institute of Science Tokyo
Keywords:
Machine learning, Artificial intelligence, Polymer chemistry, Materials science, Thermal conductivity, Liquid crystalline polymers, Polyimides, Computational modeling, Data-driven design, Advanced electronics, Adaptive systems, Polymer materials research