Title: Leveraging Multimodal Learning for Property Prediction in Stacked Bilayer Systems

Abstract: The integration of artificial intelligence into materials science is a pivotal development within the broader "AI for Science" paradigm, driven by the goal of expediting materials discovery and ensuring high-fidelity property predictions. The strategic stacking of bilayer two-dimensional (2D) materials is a fundamental technique for uncovering novel functionalities and intrinsic physical phenomena, thereby facilitating the design of 2D bilayers suited for a wide array of practical applications. While both experimental and computational research into bilayer van der Waals (vdW) materials has advanced significantly—marked by the successful synthesis of various bilayer structures and the compilation of extensive computational databases through high-throughput computing—the application of AI to model bilayer stacking and forecast new properties remains largely unexplored, highlighting a clear need for further investigation. This study introduces a novel multimodal learning framework designed to analyze interfaces between dissimilar materials, which can collectively yield new or multiple functions. Additionally, the model predicts properties emerging from the vertical integration of distinct functional material layers under specific configurations. Extensive experiments validate the superior effectiveness and efficiency of our proposed method when compared to existing baseline techniques. The source code for this work is publicly accessible at https://github.com/AnVuong123/bimatml.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC