Ryunosuke Matsumura is a PhD candidate at the Graduate School of Chemical Sciences and Engineering, Hokkaido University, and a member of the 2D Quantum Materials Group at NIMS-MANA. He is currently working on MOCVD growth of transition metal dichalcogenides and closed-loop optimization using automated experiments and large language models.

Transition metal dichalcogenides (TMDs) are promising for next-generation logic and quantum technologies owing to their atomic-scale thickness, excellent electrostatic controllability, and excitonic/spin–valley properties.1 For such applications, large-area synthesis of high-quality TMDs with controlled layer number, crystal orientation, crystallinity, and uniformity is essential. Metal–organic chemical vapor deposition (MOCVD) is a promising route to reproducible large-area growth, but its many interdependent process parameters make optimization complex and often reliant on empirical trial and error.

We recently reported the application of Bayesian optimization to accelerate MOCVD growth optimization by efficiently identifying promising conditions from a limited number of experimental trials.² However, Bayesian optimization is inherently restricted to predefined parameter spaces and objective functions. In addition, manual growth experiments and film evaluations remain major bottlenecks to experimental throughput. In this work, we developed an automated experimental platform integrated with a Large Language Model (LLM)-assisted framework for knowledge accumulation, interpretation of experimental outcomes, hypothesis generation, and adaptive refinement of exploration strategies.

1. C. Liu et al., Nature 646, 1081(2025) DOI: 10.1038/s41586-025-09621-8
2. F. Zhang, et al., ACS Appl. Mater. Interfaces 16 (43), 59109 (2024) DOI: 10.1021/acsami.4c15275