Accurately and efficiently predicting the optical properties of two-dimensional (2D) materials is crucial for photonics applications, but this task remains challenging due to discrepancies between theoretical and experimental methods. Here, we propose a physics-based machine learning (ML) framework to accelerate the screening of two-dimensional materials. It combines first-principles density functional theory (DFT) calculations with a graph neural network model, along with experimental spectral validation and Cauchy model integration. Within this framework, we have compiled a database of more than 1,000 monolayers of transition metal dichalcogenides (TMDs) and their optical properties.We also proposed a general method for defining a physically meaningful thickness for two-dimensional structures, thereby correcting the optical properties obtained from PBE-based density functional theory. Using the collected database, we developed a machine learning model based on the Cauchy model to calculate the refractive index in the near-infrared region (755–1064 nm). The developed method reflects the correlation between the atomic structure of a single molecular layer and its optical performance, which has been confirmed through extensive testing on an independent two-dimensional material database. Therefore, our machine learning–driven strategy provides a powerful tool for rapidly screening novel monolayer materials with customized optical functionalities, significantly accelerating the discovery and design of next-generation photonic materials. As an application, we further demonstrate how high-refractive-index candidate materials, such as Bi2Te2Se, can achieve enhanced field confinement and long coupling lengths in monolayer waveguides, highlighting their potential for integrated photonics.

The optical properties of two-dimensional materials are crucial for the development of modern photonic devices, particularly because their refractive index can reach as high as 4.0 in the visible and infrared ranges, attracting significant attention. However, accurately and efficiently predicting the optical responses of two-dimensional materials remains a challenge, primarily due to discrepancies between theoretical calculations and experimental measurements. Recently, a study proposed a novel materials informatics framework that combines physics-guided approaches with machine learning, aiming to achieve high-throughput screening and optical property prediction of two-dimensional transition metal chalcogenide materials, providing a powerful tool for the rapid discovery and design of next-generation photonic materials.

Reference News:

DOI: 10.1021/acsnano.5c10644