Enhancing MOVPE Synthesis through FAIR Data and ML Integration in NOMAD

Data-driven materials science is revolutionizing materials design and development, moving beyond traditional trial-and-error approaches toward a systematic, predictive model. This shift is particularly powerful in complex synthesis processes such as metalorganic vapor phase epitaxy (MOVPE), where multidimensional parameter spaces and intricate workflows make analysis challenging. Finding reliable empirical relationships for synthesis control often exceeds human cognitive limits, particularly when dealing with data from diverse sources, each with unique methodologies, instruments, and experimental details [1]. The FAIR (Findable, Accessible, Interoperable, Reusable) principles address this by establishing data standards, promoting data-rich research that is consistent and accessible [2].
In this work, we present a case study on optimizing the growth of β-GaO thin films using MOVPE on (100) β-GaO semi-insulating substrates, a valuable material for various electronic applications. By integrating machine learning (ML) into the experimental workflow, we have improved both the growth rate and the prediction accuracy for doping levels in Si-doped β-GaO films. This application demonstrates how ML can effectively advance synthesis control and materials performance [3,4].
Our approach leverages the NOMAD platform (nomad-lab.eu) to implement this MOVPE use case, digitizing the complete data lifecycle from experiment setup to AI-based analysis. Using NOMAD’s Electronic Laboratory Notebooks (ELNs), we document all synthesis procedures in a structured format, ensuring data completeness and accessibility. This structured data enables the development of automated data management tools and sustainable AI-based analytics for process optimization in materials synthesis. Our results show the power of combining FAIR data infrastructure with ML tools to support efficient, reproducible, and optimized materials synthesis.
[1] Scheffler, M., et al., Nature, 604, 635-642 (2022).
[2] Wilkinson, M., et al., Sci Data., 3, 160018 (2016).
[3] Chou, T.-S. et al., Crystal, 12(1), 8 (2022).
[4] Chou, T.-S. et al., 126737 (2022).

This work is funded by the NFDI consortium FAIRmat - Deutsche Forschungsgemeinschaft (DFG) - Project 460197019