Predicting thermoelectric properties from chemical formula with explicitly identifying dopant effects

Abstract

Dopants play an important role in synthesizing materials to improve target materials properties or stabilize the materials. In particular, the dopants are essential to improve thermoelectic performances of the materials. However, existing machine learning methods cannot accurately predict the materials properties of doped materials due to severely nonlinear relations with their materials properties. Here, we propose a unified architecture of neural networks, called DopNet, to accurately predict the materials properties of the doped materials. DopNet identifies the effects of the dopants by explicitly and independently embedding the host materials and the dopants. In our evaluations, DopNet outperformed existing machine learning methods in predicting experimentally measured thermoelectric properties, and the error of DopNet in predicting a figure of merit (ZT) was 0.06 in mean absolute error. In particular, DopNet was significantly effective in an extrapolation problem that predicts ZTs of unknown materials, which is a key task to discover novel thermoelectric materials.

Artificial Intelligence‐Based Material Discovery for Clean Energy Future

“…AI-assisted materials design and discovery methods are applied on thermoelectric materials, too. [88][89][90] Since the dopants have a significant role in improving electric performance, Na et al [90] suggested a specific neural network (DopNet) for the prediction of the thermoelectric properties in the doped materials. In this study, they measured the effects of different dopants on the five thermoelectric properties (e.g., Seebeck coefficient, thermal conductivity, electrical conductivity, power factor, and figure of merit).…”

A public database of thermoelectric materials and system-identified material representation for data-driven discovery

“…In a different direction from the conventional approach, machine learning has been studied to efficiently approximate the relationships between the materials and their physical properties 17,18 . Several machine learning methods outperformed the conventional calculation-and simulation-based methods in predicting the physical properties of the materials 19,20 .…”

“…However, its applicability is still limited to the pristine materials because the elemental graph is defined only for the pristine materials. DopNet was proposed to predict the physical properties of the alloy and doped materials from their chemical compositions based on a material space embedding approach 18 . By separately representing the host materials and the dopants, DopNet was able to learn more informative and latent features of the doped materials and consequently achieved stateof-the-art accuracies in predicting thermoelectric properties of the doped materials.…”

A Critical Review of Machine Learning Techniques on Thermoelectric Materials

“…Recently, attempts have been made to predict the TE properties of doped systems by using ML. For example, Na et al proposed a novel neural network architecture called DopNet, which can predict the TE properties of doped systems using only chemical formula following the process demonstrated in Figure . To capture the effects of dopants, DopNet independently describes the host materials and dopants.…”

“…DopNet architecture and its process from chemical formula input to the prediction of target property y . Reproduced with permission from ref . Copyright 2021 the authors.…”