Machine learning-assisted shape prediction for two-dimensional transition metal dichalcogenides growth

The controlled growth of two-dimensional (2D) transition metal dichalcogenides (TMDs) with precise morphological features remains a significant difficulty in chemical vapor deposition (CVD). However, the conventional trial-and-error approach of empirically varying CVD parameters to control the growth shape is time-consuming and lacks reproducibility. This paper proposes a machine learning-based shape recognition method to guide the growth of 2D TMDs, which can reduce optimization time by several orders of magnitude. Four machine learning (ML) algorithms, random forest (RF), adaptive boosting (AdaBoost), support vector machine (SVM), and k-nearest neighbor (KNN), were used to build a model for predicting the growth shape of 2D TMDs prepared by chemical vapor deposition. By systematically evaluating the performance metrics of each model, including recall, F1_score, accuracy, receiver operating characteristic curve, and area under curve value. The KNN model performed best, with a prediction accuracy of 84%, a recall of 0.84, an F1_score of 0.84, and an AUC value of 0.83. ML can directly predict the shape of TMDs from process parameters, thereby overcoming the limitations of traditional trial-and-error methods. The results of the feature significance analysis indicated that the S/Se evaporation temperatures (T_s/T_se) and the reaction temperatures (T_g) were the key process parameters affecting the growth shapes of TMDs. T_s/T_se and T_g have opposite effects on the evolution of growth shape. These findings provide an important basis for optimizing the preparation process of 2D materials and help to realize the controlled synthesis of the growth shape.

Graphical abstract