Work Function Prediction by Graph Neural Networks for Configurationally Hybridized Boron-Doped Graphene.

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Graphene electrodes are crucial for electronic and optoelectronic devices.
• Work function of graphene significantly impacts device performance.
• Doping is an effective method to tune graphene's work function, but traditional methods are slow.

Purpose of the Study:

• To develop a rapid and accurate method for predicting the work function of doped graphene.
• To establish a structure-property relationship for boron-doped graphene.
• To leverage deep learning for accelerated materials discovery.

Main Methods:

• A dataset of over 30,000 boron-doped graphene compositions and their work functions was generated using density functional theory (DFT) simulations.
• A novel fusion model, GT-Net, combining transformers and graph neural networks (GNNs), was developed.
• Effective GNN-based descriptors were engineered for improved prediction accuracy.

Main Results:

• The GT-Net model achieved high accuracy in predicting graphene work function, with R² = 0.975 and RMSE = 0.027.
• Comparison of three GNN methods demonstrated the superiority of the proposed approach.
• The study validated the performance of GNNs on graph-level tasks.

Conclusions:

• Deep learning, specifically GNNs, offers a powerful tool for predicting material properties like graphene's work function.
• This approach accelerates the design and discovery of novel graphene-based materials for advanced electronic applications.
• The findings enable atomic-level material design for specific desired properties.