Data-driven discovery of 3D and 2D thermoelectric materials

Area of Science:

• Materials Science
• Computational Materials Science
• Solid State Physics

Background:

• Thermoelectric materials are crucial for energy harvesting and solid-state cooling.
• Discovering new high-performance thermoelectric materials is challenging due to the vast chemical space and complex property relationships.

Purpose of the Study:

• To systematically search for novel high-efficiency three-dimensional (3D) and two-dimensional (2D) thermoelectric materials.
• To develop machine learning models for accelerated screening of thermoelectric materials.

Main Methods:

• Combined semiclassical transport techniques with density functional theory (DFT) calculations for high-throughput screening.
• Utilized the JARVIS-DFT database containing 36,900 3D and 900 2D materials.
• Trained machine learning models (gradient boosting decision trees) using DFT-derived data and force-field inspired descriptors.

Main Results:

• Identified 2932 promising 3D and 148 promising 2D thermoelectric materials based on specific criteria (bandgap, Seebeck coefficient, power factor).
• Investigated trends in thermoelectric performance related to chemical, structural, crystallographic, and dimensionality factors.
• Predicted several classes of efficient 3D and 2D thermoelectric materials, including Ba(MgX)2, X2YZ6, K2PtX2, NbCu3X4, Sr2XYO6, TaCu3X4, and XYN.

Conclusions:

• The study successfully identified a significant number of promising 3D and 2D thermoelectric materials.
• Machine learning models offer a computationally efficient approach to pre-screen materials, guiding future experimental and computational efforts.
• The publicly available dataset and tools facilitate further research in thermoelectric materials discovery.