Apr 23, 2024
4:00pm - 4:15pm
Room 322, Level 3, Summit
Andrea Albino1,Ta-Shun Chou2,Hampus Näsström1,Amir Golparvar1,Theodore Chang1,Alvin Noe Ladines1,Lauri Himanen1,Mohammad Nakhaee1,Andreas Popp2,Jose Marquez Prieto1,Sebastian Brückner2,1,Markus Scheidgen1,Claudia Draxl1,Martin Albrecht2
Humboldt-Universität zu Berlin1,Leibniz-Institut für Kristallzüchtung2
Andrea Albino1,Ta-Shun Chou2,Hampus Näsström1,Amir Golparvar1,Theodore Chang1,Alvin Noe Ladines1,Lauri Himanen1,Mohammad Nakhaee1,Andreas Popp2,Jose Marquez Prieto1,Sebastian Brückner2,1,Markus Scheidgen1,Claudia Draxl1,Martin Albrecht2
Humboldt-Universität zu Berlin1,Leibniz-Institut für Kristallzüchtung2
Data-driven material science has the potential to transform the way we design and develop materials. This emerging field represents a significant departure from traditional trial-and-error methods and empirical approaches that have characterized materials science for decades. Experiments in this area involve a multidimensional parameter space, making analysis challenging and time-consuming. Finding predictive empirical relations that allow for precise control over various aspects of the synthesis process has posed a challenge to human cognitive abilities alone. This becomes even more complex when combining datasets from different labs or from different scientists due to the lack of established standards for data models and methods to capture the large number of experimental details, including elaborate workflows and a large diversity of instruments for characterization. It requires a radical shift in how information is handled and research is performed. Experiment data must be complemented with its rich-metadata context, also covering ontologies and workflows according to the FAIR (findable, accessible, interoperable, reusable) principles [1].<br/>Opening new perspectives towards finding structure, correlations, and novel information with data-analysis and AI tools, therefore, is intimately connected to the challenge of integrating this information into a FAIR infrastructure [2].<br/><br/>Here we present growth optimization of β-Ga<sub>2</sub>O<sub>3</sub> thin films by metalorganic vapor phase epitaxy (MOVPE) on (1 0 0) β-Ga<sub>2</sub>O<sub>3</sub> semi-insulating substrates which is a successful example of applying ML/AI approaches in materials synthesis. Improving the efficiency of this synthesis is highly relevant due to the wide range of electronic device applications based on Ga<sub>2</sub>O<sub>3</sub>. Applying AI modeling on the experimental data related to thin film growth, we effectively improved the growth rate and prediction of doping level in MOVPE-grown Si-doped β-Ga<sub>2</sub>O<sub>3</sub> [3,4].<br/><br/>We will present the implementation of this use case in NOMAD (nomad-lab.eu) [5] and discuss the functionalities developed to digitize the entire data lifecycle in crystal growth and epitaxy, with the ultimate goal of FAIR data for materials growth. We developed and deployed Electronic Laboratory Notebooks (ELN) to document all relevant synthesis procedures. We will show how structured data opens up the potential to create tools that facilitate and automate data management, and the sustainable application of AI-based analytics, dedicated to process optimization in synthesis.<br/><br/>[1] Wilkinson, M., et al. The FAIR Guiding Principles for scientific data management and stewardship. Sci Data. 2016; 3, 160018.<br/>[2] Scheffler, M., et al. FAIR data enabling new horizons for materials research. Nature. 2022; 604, 635-642.<br/>[3] Chou, T.-S. et al. Toward Precise n-Type Doping Control in MOVPE-Grown β-Ga2O3 Thin Films by Deep-Learning Approach. Crystals. 2022; 12(1), 8.<br/>[4] Chou, T.-S. et al. Machine learning supported analysis of MOVPE grown β-Ga2O3 thin films on sapphire. Journal of Crystal Growth. 2022; 126737.<br/>[5] Scheidgen et al. NOMAD: A distributed web-based platform for managing materials science research data. Journal of Open Source Software. 2023; 8(90), 5388.