Improving magnetic differential phase contrast imaging by utilizing electron beam precession and novel big data processing algorithms

Abstract

With the continuous miniaturization of novel materials and technologies, magnetism plays an ever-growing role. As the dimensions shrink, magnetic properties behave uniquely, the characterization of which becomes increasingly important. Scanning transmission electron microscopy (STEM) enables high-resolution characterization of materials, with differential phase contrast (DPC) being one of the most powerful tools for direct magnetic imaging. This work encompasses the adaptation of a high-resolution TEM for magnetic near field-free STEM-DPC imaging.

Having established a framework for magnetic characterization using STEM, electron beam precession is applied to enhance the magnetic contrast being imaged. By continuously rotating the electron beam around the optical axis, contrast from electron diffraction is averaged out, highlighting the magnetic contrast. Albeit successful, the technique suffers a loss in resolution while precessing due to the aberrations of the electromagnetic lenses. Without direct access to aberration correction, a simplistic methodology was developed to segment the precessing beam into parts, and following some data processing steps, the impact of lens aberrations is reduced leading to a restoration of the lost resolution. The combined effects of the methodologies are studied and presented, and the overall effect on the quality of magnetic imaging is reported.

With novel developments in STEM detector technology, the recorded datasets are four-dimensional, and often several tens of gigabytes in size. Part of this work explores the use of machine learning algorithms, specifically supervised neural networks, for efficient and robust processing of the recorded big data.