Artificial intelligence assisted thermoelectric materials design and discovery

Abstract Materials discovery from infinite earth repository is considered the bottleneck of each revolutionary technological progress. The discovery of the new materials is hindered by the labor-intensive and time-consuming process. Although machine learning techniques shown the excellent capability for speeding up materials discovery, it is still challenging to obtain effective material feature representations and making a precise prediction of the material properties are still challenging. This work focuses on developing an automatic material design and discovery framework enabled by data-driven AI models. We utilize the material's thermoelectric (TE) properties prediction as a baseline to demonstrate the investigation logistic. The framework works in a closed-loop mode. Starting from the well-established TE materials databases, we develop a TE material encoding method and TE material screening model by using machine learning technology. The models show excellent accuracy for TE properties prediction. Furthermore, the developed AI models have identified 6 novel promising p-type TE materials and 8 novel promising n-type TE materials. Evaluated by the Density-functional Theory (DFT) calculation, the prediction results have a good agreement with the realistic material's TE property. The proposed framework well accelerates the design and discovery of the new functional materials.