High Throughput Screening of Ternary Nitrides with Convolutional Neural Networks

Abstract

The development of new materials is a core aspect of advancement in synthesis and application for industry. There is a vast number of possible chemical permutations of the basic elements that can be explored to synthesize materials that possess attractive catalytic, mechanical and electrical properties that may not be easily accessible to traditional experimental methods for various reasons, including cost and time considerations. Nitrides, as examples, require very stringent and precise conditions to successfully synthesize making their experimental exploration very slow. In this paper, we employ the use of machine learning algorithms to predict the bulk properties of Ternary Metal Nitrides (TMN), specifically their bulk modulus which is correlated with the hardness of the material. We were able to develop a consistent model with encouraging accuracy, that was able to predict the bulk moduli of materials that previously did not have computed values. The model was trained on $10^3$ ternary materials with known elastic properties and defined structures, and was able to predict the bulk modulus of $\thickapprox 1,000$ Ternary Metal Nitrides (TMNs) to $\thickapprox 80\%$ accuracy. This approach is orders of magnitude faster than the traditional computational approaches like density functional theory (DFT)\cite{dft-paper} which makes exploratory identification of materials with promising properties fast. We propose that such models be used to select interesting candidates for high throughput computation from first principles.