STEM EDS/EELS for Phase Analysis of Deep-Mantle Rock Assemblages Supported by Machine Learning

Abstract

Scanning transmission electron microscopy (STEM) is now a powerful technique employed in the study of minerals and rocks. In addition to providing high-resolution images, STEM generates a variety of additional signals that are often employed for analyzing geological samples. The two most prominent techniques among them are energy-dispersive X-ray spectroscopy (EDS) and electron energy-loss spectrometry (EELS). EDS is a relatively simple and robust technique to do chemical analysis. EELS is a more challenging technique but is in principle capable of not only measuring elemental composition, but also giving information about chemical bonding and oxidation state, band structure, etc. These two methods are both applicable for the quantitative chemical analysis of mineral phases which do not overlap across the thickness of the TEM sample. In the scenario where phases instead overlap in the projected volume, 2D EDS/EELS is insufficient to obtain accurately the composition of every single phase. Although in principle this problem can be tackled using electron tomography, this technique drastically complicates the data acquisition, and the minerals may experience severe sample degradation under the extensive beam exposure required to record the tomogram.