AFM Super-Resolution Reconstruction Neural Network for Imaging Nanomaterials

Nanomaterials hold great significance in fields such as physical, chemistry, and semiconductors. Atomic Force Microscopy (AFM), a widely employed tool for characterizing the surface morphology of nanoscale materials, suffers from a time-consuming imaging process due to its raster scanning method. To accelerate AFM imaging, we proposed an AFM super-resolution imaging method that reconstructs low-resolution AFM images into high-resolution ones, enhancing the AFM imaging speed by 3.5–7.5 times while ensuring imaging quality. We introduced a More Rational Transformer (MRT) as the super-resolution reconstruction neural network. This network enhances the attention mechanism of the Transformer and dynamically integrates the attention mechanism with a depth-wise convolution (DW-Conv), thus better adapting to the processing of AFM images of nanoscale materials. After training and testing on a data set containing common materials and devices for integrated circuits, our method demonstrates superior imaging quality compared to other super-resolution imaging methods. In general, our method is an effective way to accelerate the characterization of nanomaterials.