Determining alloy composition of single-layer iron chalcogenides by machine learning of STM/STS data

Chemical pressure is an effective tool to tune the properties of quantum materials, in particular at the single-layer limit. In this work, we investigated the effect of chemical pressure in single-layer FeSe films grown on SrTiO₃(001) substrate by molecular beam epitaxy. While both positive and negative pressure can be applied with Se substituted by isovalent S and Te, one of the key parameters that need to be precisely determined is the alloy composition, thus, the magnitude of the pressure. Scanning tunneling microscopy (STM) imaging has been a powerful tool for this task; however, contrasts associated with interfacial inhomogeneity and defects make the determination by visualization challenging, if not impossible. Here, we utilize machine learning to distinguish between S(Te) and Se atoms in STM images. First, defect locations are identified by analyzing spatially dependent dI/dV tunneling spectra using the K-means method. Next, after excluding the defect regions, dI/dV spectra are further analyzed using the singular value decomposition (SVD) method to determine the Se/S(Te) ratio, which was then correlated with the STM images. This approach has been applied to analyze the alloy composition of various Te and S substituted FeSe films, which has played a critical role in the investigation of their superconducting properties.