Data-Driven Improvement of ZT in SnSe-Based Thermoelectric Systems

Approximately 70% of all energy input is dissipated as waste heat globally.¹ Recovering this considerable and ubiquitous energy loss would help to solve the energy and environmental crises faced by humanity. Semiconductor-based thermoelectric (TE) technology is a possible eco-friendly and sustainable solution to this problem. TE semiconductors can generate electricity if a temperature difference is applied under the Seebeck effect while neither releasing hazardous chemicals nor making mechanical noise. Their conversion performance is typically determined by the thermoelectric figure of merit (<i>ZT</i>), consisting of the electrical conductivity (<i>σ</i>), Seebeck coefficient (<i>S</i>), and thermal conductivity (<i>κ</i>) at a given temperature, given as follows:<br/><i>ZT</i> = <i>σS²T</i>/<i>κ.</i><br/>Although achieving a high <i>ZT </i>value has been the ultimate goal in thermoelectrics, doing so is highly challenging due to the complex interdependence of <i>σ, S, </i>and <i>κ</i>.^2,3 Accordingly, many innovative strategies have been developed to maximize the over the past two decades. Doping and alloying are fundamental strategies to improve the thermoelectric performance of bare materials. However, identifying outstanding elements and compositions for the development of high-performance thermoelectric materials is challenging. In this presentation, we will show a data-driven study using machine learning (ML) to improve the thermoelectric performance of SnSe compounds with various doping and alloying. We built highly accurate predictive models of thermoelectric properties based on the newly generated experimental and computational dataset. A well-designed feature vector consisting of elemental properties and the electronic structures of doped/alloyed materials plays a crucial role in achieving accurate predictions for unknown alloying elements. The predictive models allowed for rapid screening of high-dimensional material spaces and evaluated their thermoelectric properties with various compositions of 59 alloying elements. This screening provided overall strategies to optimize and improve the thermoelectric properties of doped/alloyed SnSe compounds. As a result, we identified several alloying elements leading to a high <i>ZT</i> over 2.0. And we proposed a design principle for improving the <i>ZT</i> by Sn vacancies that differ depending on the alloying elements. We will further discuss various strategies to enhance the predictive power of ML models by introducing new types of descriptors of alloyed materials.<br/>Reference<br/>1. Gingerich, D. B.; Mauter, M. S., Quantity, Quality, and Availability of Waste Heat from the United States Thermal Power Generation. <i>Environ Sci Technol </i><b>2015</b>, <i>49</i> (14), 8297-306.<br/>2. Tan, G.; Zhao, L. D.; Kanatzidis, M. G., Rationally Designing High-Performance Bulk Thermoelectric Materials. <i>Chem Rev </i><b>2016</b>, <i>116</i> (19), 12123-12149.<br/>3. Vaqueiro, P.; Powell, A. V., Recent developments in nanostructured materials for high-performance thermoelectrics. <i>Journal of Materials Chemistry </i><b>2010</b>, <i>20</i> (43).