A deep learning perspective into the figure-of-merit of thermoelectric materials

In the present work, we employed a deep learning (DL) technique to investigate thermoelectric (TE) materials performance. The figure-of-merit (ZT) determined experimentally for various compositions at a wide range of temperatures with composition features were used to train the DL model. The validation results showed that the built DL model exhibited a reliable accuracy with a cross-validation score R2 value of 0.91. In addition, by this model, the experimental behaviors of TE materials (ZT vs temperature) were successfully reproduced and a general prediction of 300,000 composition have been done.

Automated summary

This study employed deep learning to investigate the performance of thermoelectric materials, achieving reliable and accurate predictions, successfully reproducing experimental behaviors, and making extensive predictions for a large number of compositions.