Decoding Chemical Inhomogeneity in Iron Chalcogenides: Insight from Self-Organizing Map Analysis of STM/S Data

Chemical pressure from the isovalent substitution in the epitaxial film iron chalcogenide can effectively tune their properties. However, such substitution during epitaxial growth inherently leads to chemical inhomogeneity, making the determination of alloy composition and substitutional sites challenging. To address this issue, we have employed two-step machine learning solution using K-mean and SVD [1,2] to analyze scanning tunneling microscopy/spectroscopy (STM/S) data. In this study, we introduce a different approach using self-organizing map (SOM), a type of artificial neural network, to discern the Se/S ratio in superconducting single layer FeSe_1-xS_x alloys. Similar to previous methodologies, this unsupervised competitive learning method can determine the Se/S ratio effectively. However, this SOM-based approach offers an improved approach to interpret non-linear STM data, while eliminating the need to pre-specify the number of clusters.

[1] Zou et al., “Deciphering Alloy Composition in Superconducting Single-Layer FeSe_1−xS_x on SrTiO₃(001) Substrates by Machine Learning of STM/S Data”, ACS Appl. Mater. Interfaces 15, 22644−22650 (2023).

[2] Oli et al., “Atomic-scale electronic inhomogeneity in single-layer iron chalcogenide alloys revealed by machine learning of STM/S data”, AIP Adv. 13, 105224 (2023).