Emerging machine learning strategies for diminishing measurement uncertainty in SPM nanometrology

Abstract

AbstractScanning probe microscopy (SPM) is an outstanding nanometrology tool for characterizing the structural, electrical, thermal, and mechanical properties of materials at the nanoscale. However, many challenges remain in the use of SPM. Broadly speaking, these challenges are associated with the acquisition of the SPM data and the subsequent analysis of this data, respectively. Both problems are related to the inherent uncertainty of the data obtained in SPM-based measurements due to the nanoscale geometry of the SPM probe tip, the state of the sample imaging region, the data analysis methods themselves, and the experience of the users. Machine learning (ML) approaches have been increasingly applied to address these problems in recent years. In general, ML approaches involve constructing a well-organized and representative SPM dataset from experimental and theoretical trials, and then using the data features of this dataset for ML models to learn and produce appropriate predictions. Herein, this review examines the development of recent ML strategies for reducing measurement uncertainty in SPM-based measurements. The review commences by introducing the ML models and algorithms commonly used in SPM-related applications. Recent approaches for collecting and preprocessing the SPM data to extract significant data features for further ML processing are then introduced. A review of recent proposals for the applications of ML to the improvement of SPM instrumentation and the enhancement of data processing and overall understanding of the material phenomena is then presented. The review concludes by presenting brief perspectives on future opportunities and open challenges in the related research field.