Driving Thermoelectric Optimization in AgSbTe2 via Design of Experiments and Machine Learning

TL;DR

This paper introduces a combined Design of Experiments and machine learning approach to optimize the thermoelectric properties of AgSbTe2, achieving over 30% improvement in efficiency and providing an open-source tool for broader scientific applications.

Contribution

It presents a novel integrated methodology for thermoelectric optimization using experimental design and machine learning, with a practical open-source implementation.

Findings

Achieved a thermoelectric figure of merit of 1.61 at 600 K

Improved electrical properties and reduced thermal conductivity in optimized material

Over 30% higher performance than previous literature values

Abstract

Systemic optimization of thermoelectric materials is arduous due to their conflicting electrical and thermal properties. A strategy based on Design of Experiments and machine learning is developed to optimize the thermoelectric efficiency of AgSb1+xTe2+y, an established thermoelectric. From eight experiments, high thermoelectric performance in AgSb1.021Te2.04 is revealed with a peak and average thermoelectric figure of merit of 1.61 +/- 0.24 at 600 K and 1.18 +/- 0.18 (300 - 623 K), respectively, which is over 30% higher than the best literature values for AgSb1+xTe2+y. Ag-deficiency and suppression of secondary phases in AgSb1.021Te2.04 improves the electrical properties and reduces the thermal conductivity (~0.4 W m-1 K-1). Our strategy is implemented into an open-source graphical user interface, and it can be used to optimize the methodologies, properties, and processes across different scientific fields.