Tohoku University Debuts AI Materials Map to Speed Thermoelectric Discovery

Overview

• The map visualizes materials as points on a graph where proximity reflects structural similarity and an axis captures predicted zT for thermoelectric performance.
• Researchers combined StarryData2 literature measurements with Materials Project calculations and trained MatDeepLearn using a message‑passing neural network.
• The team says the map helps experimentalists identify analogs of promising materials and reuse established synthesis protocols to reduce trial‑and‑error.
• The study, led by Yusuke Hashimoto with Takaaki Tomai, Xue Jia, and Hao Li at Tohoku University’s AIMR/FRIS, was published in APL Machine Learning on July 28, 2025.
• Planned extensions target magnetic and topological materials with added magnetic, chemical, and topological descriptors toward a broader design platform.