Robust combined modeling of crystalline and amorphous silicon grain boundary conductance by machine learning

Abstract

AbstractKnowledge in thermal and electric transport through grain boundary (GB) is crucial for designing nanostructured thermoelectric materials, where the transport greatly depends on GB atomistic structure. In this work, we employ machine learning (ML) techniques to study the relationship between silicon GB structure and its thermal and electric boundary conductance (TBC and EBC) calculated by Green’s function methods. We present a robust ML prediction model of TBC covering crystalline–crystalline and crystalline–amorphous interfaces, using disorder descriptors and atomic density. We also construct high-accuracy ML models for predicting both TBC and EBC and their ratio, using only small data of crystalline GBs. We found that the variations of interatomic angles and distance at GB are the most predictive descriptors for TBC and EBC, respectively. These results demonstrate the robustness of the black-box model and open the way to decouple thermal and electrical conductance, which is a key physical problem with engineering needs.