Quantitative EELS: Instrumentation, Techniques and Applications

Session

EMAG - EM Spectroscopy

Authors

Professor Alan Craven (1)

Affiliations

1. School of Physics and Astronomy, University of Glasgow

Keywords

EELS Instrumentation

EELS Analysis

Correction of Artifacts in EELS spectra

Precipitates in Steel

Spectrum Imaging

DualEELS

EELS experimental cross-sections

Abstract text

As would be expected over the course of half a century working on STEM-EELS, lot has changed.   Lateral atomic lateral resolution is now available.   Energy resolution has gone from a few eV to a few meV.   Developments in these areas have led to much improved power supply stability, thermal stability and specimen stage stability, all of which are crucial for real experiments.   The fraction of the EELS signal collected has gone from ~0.001% to a significant fraction of 100%.

Two key developments made the last improvement possible.   The first was the introduction of post-specimen lenses into the STEM column to compress the angular distribution of the electrons leaving the specimen.  The minimum requirement is that angles up to the probe angle should enter the much smaller acceptance angle of the spectrometer.     This issue became more critical as the spatial resolution improved because that required a bigger probe angle but it was helped by aberration correction of the spectrometer allowing a bigger acceptance angle.

The second was the switch from serial scanning of the spectrum across a narrow slit to the use of electronic detectors, first photodiode arrays, then CCD cameras and now single electron detection cameras.   Extending the effective dynamic range of the CCD camera in the form of DualEELS was a further big step forward but something that should not be necessary with single electron detector cameras.

With all these wonderful developments, there is a temptation to assume that an instrument performs ideally.    However, there are always departures from ideality and these can cause problems.   For instance, TEM-STEM columns are optimised for elastically scattered electrons and not for electrons with the significant energy losses studied by EELS.   This can result in variation of collection efficiency with energy loss.   The multipole post-spectrometer optics are set up to give the best compromise between a range of parameters and this can leave some spectral distortion present.   Electrons hitting a solid surface don’t magically disappear but give rise to backscattered and secondary electrons which can result in spectral artefacts.   Finally, detectors themselves have artifacts which e.g. readout noise in CCDs or beam damage causing a change of detection efficiency.

Thus, it is always worth looking at the instrumental performance in the simplest possible way and asking whether the observations match the idealised performance.   If the artefacts can be characterised, then the as-recorded spectral data can be corrected in many cases.

The applications in the talk cover the quantitative analysis of multiphase systems.   To discuss this, it is normal to divide the spectrum into two regions: the high loss region with losses >100eV covering the ionisation edges resulting from excitations from core-level atomic states and the low loss region (0 – 100eV) covering the zero-loss peak, the collective excitations (or plasmons) and the semi-core states.   Multiple scattering couples the two regions together and deconvolution can be used to remove it.   DualEELS allows the whole spectrum to be obtained to allow this deconvolution to be made.   Combined with the spectrum imaging technique, it becomes possible to extract and analyse the various phases present.

The high loss region is in many ways the simpler of the two regions to process.   The background under an ionisation edge is frequently “smooth” and follows an AE^-r form where E is the energy loss and A and r are constants.   This shape can be fitted to the pre-edge background and the underlying background subtracted from the edge.   Hartree-Slater cross-sections are available which allow the edge intensity to be converted to the number of atoms per unit area.

The Hartree-Slater cross-sections are for isolated atoms and the surrounding atoms in the specimen produce fine structure which is normally considered as electron loss near edge structure (ELNES) and extended energy loss fine structure (EXELFS).   

ELNES changes the ionisation cross-section within ~50eV of the edge, which affects the quantification.    When two ionisation edges are close together, the ELNES from the lower one severely perturbs the background in front of the higher one, preventing the use of the AE^-r form for background subtraction.    However, the ELNES shapes are characteristic of the atomic surroundings and so very useful when the same atom is present in different environments, allowing the contributions to be separated.

EXELFS on the other hand causes relatively weak, long period oscillations in the background over a very large energy loss range.    The oscillations prevent good background subtraction under weak ionisation edges and are a particular problem for ionisation edges in the 100 -1000eV range when there are strong edges from semi-core states e.g. when analysing small (Ti,V,Nb)C precipitates in a steel matrix.

One way to tackle these problems is to use multiple linear least squares (MLLS) fitting of experimental cross-sections for the phases of interest together with experimental spectral shapes for the matrix, obtained from the spectrum imaging dataset itself.    In this way the fine structure effects are incorporated and the phases of interest can be extracted and quantified.

The signal level in the high-loss region is relatively low, requiring a long time to record the data.     The signal is much higher in the low loss region but the data are much less easy to interpret.   The edges are close together meaning they are difficult to separate and there is coupling between them and the valence electrons, causing changes of both intensity and position, depending on composition.   However, for the case of precipitates in steel where the various classes of precipitates have been analysed using the high loss region, the corresponding low loss shapes and cross-sections can also be extracted from the data.   MLLS fitting can then be used to analyse large area datasets recorded relatively quickly using only the low-loss region of the spectrum.   In this way, quantities such as the precipitate volume fraction, the number of precipitates per unit size per unit specimen volume and shape distributions can be found.