Probing nanoscale structural perturbation in a WS2 monolayer via explainable artificial intelligence

Abstract

This study investigates the applicability of the machine learning model in correlative spectroscopy to enhance spatial resolution for probing nanoscale structural perturbations. The developed model demonstrates significant enhancement in spatial resolution, achieving up to 50 nm through the integration of Kelvin probe force microscopy and atomic force microscopy data. The predicted nanoscale Raman image reveals abnormal behaviors associated with strain-induced lattice perturbations, such as the presence of compressive and tensile strains within identical nanoscale wrinkles. Afterward, we interpreted the trained model using explainable artificial intelligence techniques, uncovering synergistic contributions to the Raman features across each input dataset within the nanoscale region. Our analysis demonstrates that the model effectively reflects key strain-induced lattice behaviors, highlighting its nanoscale sensitivity to structural perturbations. Finally, we validated these findings using quantum mechanical calculations, which confirmed the strain-induced changes in Raman-active modes. This study offers comprehensive insights into nanoscale structural perturbations, paving the way for innovative approaches to high-resolution spectroscopic analysis in low-dimensional materials.