Machine Learning-Assisted Synthesis of Two-Dimensional Materials.

Abstract

Two-dimensional (2D) materials have intriguing physical and chemical properties, which exhibit promising applications in the fields of electronics, optoelectronics, as well as energy storage. However, the controllable synthesis of 2D materials is highly desirable but remains challenging. Machine learning (ML) facilitates the development of insights and discoveries from a large amount of data in a short time for the materials synthesis, which can significantly reduce the computational costs and shorten the development cycles. Based on this, taking the 2D material MoS2 as an example, the parameters of successfully synthesized materials by chemical vapor deposition (CVD) were explored through four ML algorithms: XGBoost, Support Vector Machine (SVM), Naïve Bayes (NB), and Multilayer Perceptron (MLP). Recall, specificity, accuracy, and other metrics were used to assess the performance of these four models. By comparison, XGBoost was the best performing model among all the models, with an average prediction accuracy of over 88% and a high area under the receiver operating characteristic (AUROC) reaching 0.91. And these findings showed that the reaction temperature (T) had a crucial influence on the growth of MoS2. Furthermore, the importance of the features in the growth mechanism of MoS2 was optimized, such as the reaction temperature (T), Ar gas flow rate (Rf), reaction time (t), and so on. The results demonstrated that ML assisted materials preparation can significantly minimize the time spent on exploration and trial-and-error, which provided perspectives in the preparation of 2D materials.