Dataset and Methodology for Material Identification Using AFM Phase Approach Curves

Atomic force microscopy (AFM) phase approach-curves have significant potential for nanoscale material characterization, however, the availability of robust datasets and automated analysis tools has been limited. In this paper, we introduce a novel methodology for material identification using a high-dimensional dataset consisting of AFM phase approach-curves collected from five distinct materials: silicon, silicon dioxide, platinum, silver, and gold. Each measurement comprises 50 phase values obtained at progressively increasing tip-sample distances, resulting in 50x50x50 voxel images that represent phase variations at different depths. Using this dataset, we compare k-nearest neighbors (KNN), random forest (RF), and feedforward neural network (FNN) methods for material segmentation. Our results indicate that the FNN provides the highest accuracy and F1 score, outperforming more traditional approaches. Finally, we demonstrate the practical value of these segmented maps by generating simulated scattering-type scanning near-field optical microscopy (s-SNOM) images, highlighting how AFM phase approach-curves can be leveraged to produce detailed, predictive tools for nanoscale optical analysis.