Data-driven discovery of van der Waals materials with high optical anisotropy

Large optical anisotropy over a wide spectral range is crucial for effective light control in many photonic devices. This creates a growing need for natural materials with giant anisotropy (Δn > 1) to meet both scientific and industrial demands.
Bulk transition-metal dichalcogenides (TMDCs) are highly promising in this regard due to their intrinsically anisotropic van der Waals (vdW) layered structures, which naturally produce strong intrinsic birefringence.
In our study, we trained an ALIGNN graph neural network to predict birefringence using only crystal structures and elemental compositions (Figure 1). To enable this, we collected a database of known layered vdW materials with crystal structures and optical properties calculated via density functional theory (DFT), supplemented with experimental data for a subset of samples.
We then screened crystalline materials databases (MaterialsProject and GNoME) and identified new candidate materials with high optical anisotropy. Subsequent DFT calculations and experimental measurements validated these predictions, demonstrating the effectiveness of our approach in discovering novel anisotropic materials [L. Bereznikova et al., Materials Horizons, 2025].