Diagnostic and Correction of Phase Aberrations in Scanning Transmission Microscopy by Artificial Neural Networks

The ultra-high-resolution of state-of-the-art scanning transmission microscopy (STEM) resides in the ability to correct aberrations of the probe forming lenses. In this way, atomically resolved images with Z-sensitive contrast can be obtained. The idea in this contribution is to test AI methods (artificial neural networks or ANN) to measure the aberrations from a single image of the stationary probe on an amorphous region, or Ronchigram intensity I(q). The aim is to assist the microscopist in his manual tuning or to improve the existing routines available with probe correctors with faster ones. This will contribute in the pursuit of a fully automated alignment of the microscope, permitting the microscopist to focus on the properties of the sample in his research rather than spend time in technical procedures of alignment of the instrument itself. Recent approaches with deep learning techniques were demonstrated to be promising in optimizing the STEM aperture for improving resolution [1,2].