Download all files (5.7 MB)

Collection

Citation

Description:

In this work, we use a melt process in a vacuum-sealed quartz tube followed by spark plasma sintering to synthesize GeTe compounds. Through the application of active learning with Bayesian optimization, we systematically explore the process parameters to enhance the thermoelectric figure of merit (zT) to achieve a value of zT = 1.14 at 700 K. The results demonstrate an improvement of 25% over the equivalent 2-step process, the same efficiency (within 5%) as the 3-step process while drastically reducing processing time by over 90%, and this within 24 experiments out of more than 18 billion possible combinations. This illustrates the potential of the utilization of GeTe in mid-temperature range applications, and the effectiveness of artificial intelligence assistance to improve materials' properties in the tremendously vast process parameters space. This study is intended to be a textbook case displaying the effectiveness of active learning and Bayesian optimization at performing a full process optimization for enhancing the performance of a material with minimum experimental trials, time and cost.

Rights:

• Creative Commons BY Attribution 4.0 International
  

Keyword: Material process, bayesian optimisation, active learning, chalcogenides, thermoelectrics

Date published: 2025-12-31

Publisher: Informa UK Limited

Journal:

• Science and Technology of Advanced Materials: Methods (ISSN: 27660400) vol. 5 issue. 1 2545174

Funding:

• JSPS PE20752
• CNRS
• NIMS

Manuscript type: Publisher's version (Version of record)

MDR DOI:

First published URL: https://doi.org/10.1080/27660400.2025.2545174

Related item:

Other identifier(s):

Contact agent:

Updated at: 2025-12-10 08:30:23 +0900

Published on MDR: 2025-12-10 08:23:46 +0900