| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Thin Films and Coatings: Properties, Processing and Applications
|
| Presentation Title |
Active Learning and Transfer Learning for Rapid Targeted Synthesis of Compositionally Complex Thin Film Alloys |
| Author(s) |
Nathan S. Johnson, Apurva Mehta, Aashwin Ananda Mishra |
| On-Site Speaker (Planned) |
Nathan S. Johnson |
| Abstract Scope |
The field of materials science is increasingly turning to machine learning to assist in synthesizing advanced materials. This study leverages active learning (AL) to expedite the synthesis of thin-film compositionally complex alloys (CCAs) for thermoelectric applications. Traditional thin film synthesis methods are labor-intensive and time-consuming, necessitating an approach that enhances efficiency without sacrificing accuracy. This study employed a high-throughput combinatorial sputtering method to prepare a wide range of alloy compositions. Active machine learning was used to guide a human researcher in finding correct synthesis conditions for a family of 5-element alloys. Our results demonstrated that the AL approach outperforms traditional human-driven sampling in optimizing synthesis parameters. Furthermore, this study used transfer learning to pre-train models and enhance efficiency further. Models pre-trained on lower-dimensional alloy systems were applied to higher-dimensional systems and showed immediate prediction improvement. This approach is adaptable to a wide range of thin film materials and synthesis techniques. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Thin Films and Interfaces, Machine Learning, High-Entropy Alloys |