A computational framework for post-acquisition stability assessment in scanning tunneling microscopy

Abstract

In high-precision scanning-based measurement systems, the reliability of data obtained at the atomic scale may be limited by hardware-induced geometric instabilities and operator-dependent interpretation processes. In particular, in artificial intelligence-based surface characterization, the lack of quantitative verification of measurement stability increases the risk of error propagation in subsequent analysis and modeling stages. In this study, a modular, traceable, and hybrid computational framework is presented for quantitative and comparative post-acquisition stability assessment in atomic-resolution STM measurements. The proposed approach combines deterministic computer vision-based analyses with modern deep learning components in a hybrid workflow, producing pixel-level drift metrics together with local and surface-level image-similarity indicators based on the structural similarity index measure and normalized cross-correlation. Going beyond conventional drift-correction-oriented approaches, the study puts forward the concept of ‘drift-aware data reliability’ as a systematic framework for post-acquisition measurement stability assessment. The resulting quantitative outputs make the stability profile of the measurement system visible and provide a quantitative and traceable basis for assessing the usability of the data in subsequent analysis, interpretation, and modeling processes.