Coupling and decoupling of phonons upon thermal conductions for thermoelectrics

Abstract

Thermoelectric energy conversion which allows us to convert waste heat into electricity requires materials to simultaneously acquire multiple properties for higher conversion efficiency, which is often quantified by the figure of merit. To increase the figure of merit, a product of Seebeck coefficient squared and electronic conductivity needs to be maximized even though they are in a trade-off relationship in terms of carrier concentration. Thermal conduction needs to be impeded as much as possible to allow converted electronic conduction for electricity, even though both electrons and phonons are quantum waves and thus scattering one inevitably tends to scatter another according to the basic theories. To overcome the inherent difficulties, we have attempted to uncover the underlying mechanisms behind thermal conduction beyond existing theories.

To do that, we used two separate approaches. One is based on lattice dynamics (LD) in conjunction with DFT calculations and another is based on molecular dynamics (MD) where interatomic potentials can be obtained through machine learning with DFT results as training data. The LD is more advantageous than MD in that the results can be readily interpreted in terms of phonons as it is based on the theories of phonons. At the same time, it has limitations because of the same reason, being unable to capture thermal conduction beyond the phonon theories, or amorphous theories. On the other hand, the MD can capture any contribution to thermal conduction as it is based on fundamental dynamics alone while extra analyses are needed for the interpretations of computed results. To bridge the two approaches, we use the perturbed molecular dynamics (PMD) instead of conventional approaches through MD including Green-Kubo equations. This enabled us to further analyze governing factors of thermal conduction.

The combined approaches mentioned above enabled us to analyze the mechanisms behind thermal conduction in many kinds of solids. The findings through these combined approaches provide missing key information that are essential for high-throughput computations for the next step of materials explorations.