Large-Scale Raman Spectrum Calculations in Graphene using Deep Learning

Raman spectroscopy is commonly used to assess the quality of post-growth graphene samples or their dopant concentration after a functionalisation treatment. Many ad hoc rules have been developed to analyze the obtained spectrum and give them an underlying physical meaning. We aim to improve our understanding of these spectrum and how they relate to atomic compositions by using deep neural networks based on the Schnet¹ architecture, trained on DFT data, and extending their capacities way beyond a few thousands of atoms, while keeping their near ab initio accuracy. This method allows us to natively calculate the phonon frequencies and modes induced by a large range of defect types and concentrations, which subsequently can be used to predict the Raman response of the simulated samples.

1.Schütt, K. T. Et al. (2018). Schnet–a deep learning architecture for molecules and materials. The Journal of Chemical Physics, 148(24).