Highly Deformable Van der Waals Chalcogenides with Superior Thermoelectric Performance from Interpretable Machine Learning

Van der Waals (vdW) chalcogenide-based flexible thermoelectric devices hold great promise for wearable electronics. However, intrinsic vdW chalcogenides that combine high flexibility with superior thermoelectric figures of merit (ZT) remain extremely rare. Consequently, there is an urgent need to develop methods capable of high-throughput screening to identify potential vdW chalcogenides with both robust flexibility and favorable ZT value. In this study, over 1000 vdW chalcogenides are high-throughput screened for their flexibility and ZT values. Flexibility is evaluated using the previously developed deformability factor, while ZT values are predicted using a machine learning model. Several candidates with large deformability and high ZT are successfully identified. Among these, NbSe₂Br₂ emerges as the top-performing material. Further first-principles calculations reveal that it achieves a maximum ZT value of 1.35 at 1000 K, the highest reported so far among flexible inorganic thermoelectric materials. Its power factor value of 8.1 µW cm^-1K^-2 at 300 K also surpasses most organic and inorganic flexible thermoelectric materials. The high ZT_max is mainly contributed by the extremely low thermal conductivity and the high Seebeck coefficient along the out-of-plane direction at high temperatures. The study offers new material options for the development and application of flexible thermoelectric devices based on layered chalcogenides.