Overcoming Challenges Associated with Designing of Thermoelectric Materials: DFT and Machine Learning Approaches
Search for clean and renewable energy resources has driven recent interest in designing thermoelectric materials that convert the waste heat to useful electricity. High performance thermoelectric materials require excellent electronic transport and favorable thermal transport, simultaneously. Given the interdependence of various transport parameters, it is daunting to achieve desirable performance. We attempt to address some of these challenges using density functional theory in combination with machine-learning based approaches. We first report the decoupling of Seebeck coefficient and electrical conductivity by tuning the distortion parameter of chalcopyrites leading to complete convergence of bands, thereby resulting in unprecedented enhancement of electronic transport properties. A combination of excellent electronic transport and low thermal conductivity in CdGeAs2 results into a high ZT of 1.67 at 1000K. To find a system with low thermal conductivity, we study the oxychalcogenide system AgBiTeO, demonstrating the unique collective rattling motion hosted by chemical bond hierarchy. The favorable electronic and thermal transport properties result in a maximum ZT of 1.99 at 1200K, which is highest among the existing bulk oxide-based thermoelectric materials. Owing to the complexity and resource extensive calculations involved in determining electron relaxation time (τel), we employed machine learning approach to estimate the τel. The machine learning model uses data available for experimental electrical conductivity and a collection of accessible elemental information. This model with a rmse of 0.22, outperforms the deformation potential model, and performs adequately on the unseen data to predict the relaxation time over a wide range of temperatures. Further, we develop an effective descriptor by using chalcopyrite class of compounds, to guide an accelerated screening of materials with desirable degree of anharmonicity. The high-throughput study corroborates the role of a very simple parameter, “phonon band center”. This can be calculated within the harmonic regime, yet having profound impact on the anharmonicity of the compounds. Since, the performance of thermoelectric devices is limited by the quality of the interface, we explore the role of fundamental parameters, such as surface termination of interface, electronegativity difference and lattice mismatch that influence the interface. Optimization of these parameters will have a significant role in preserving the thermoelectric performance of the materials in devices. The results of our study pave way to overcome some of the critical challenges related to thermoelectrics by effectively addressing electronic and thermal transport problems.