Much of analytical TEM revolves around elemental analysis based on core-shell ionization and its role in electron energy-loss spectroscopy (EELS) and energy-dispersive X-ray spectroscopy (EDS). In these techniques, integrals of the primary or secondary ionization signals (typically over many tens of eV in energy) are used to measure and map the elemental composition of probed sample areas.

In contrast, present-day STEM EELS systems are able to reveal spectral details with resolution in the range 0.1-1.0 eV. This means that EELS provides access to electronic structure and response information that goes beyond the simple elemental composition information of the integrated core-loss signals. Such high-resolution electronic information is expressed in EELS spectra taken from the 0-30 eV low-loss range (valence loss EELS or VEELS) and in fine structure within a similar energy range of core ionization edge thresholds (energy-loss near-edge structure or ELNES).

Our GIF Tridiem line of EELS spectrometers provides direct access to this type of specimen information (when mounted to a TEM/STEM instrument with a beam source of comparable energy resolution). With the addition of Gatan’s STEMPack, you can explore variations in the EELS fine structure in different sample areas in real time. By systematically accumulating collections of such EELS spectra into spectrum image (SI) data sets and applying the analysis tools of our EELS Analysis software, you can begin to form maps of different material phases based on differences in VEELS and ELNES.

Read on for further details about how GIF Tridiem, STEMPack, and the EELS Analysis software tools enable you to isolate and map valence loss and near-edge fine structure signals.

Valence-loss EELS mapping with GIF Tridiem, STEMPack, and EELS Analysis

A key benefit of the GIF Tridiem detector system is its ability to capture EELS spectra with exposure times as short as 1 msec and readout rates as high as 40 spectra per second. This means even low-loss spectra with intense zero-loss peaks can easily be viewed in real time and rapidly collected into complete and detailed spectrum-image data sets. For example, the movie below shows such a data set being explored, post-acquisition, with the Spectrum Picker tool of the Gatan Microscopy Suite (GMS) software:

If you have a problem viewing this movie click here to download a copy

You will notice that this spectrum image data set is unusual for its high degree of spatial information. In fact, it represents a scan of 330 x 190 pixels. Such image detail in STEM SI data becomes a practical possibility for EELS systems with high spectral readout rates, like Gatan’s GIF Tridiem or Enfina spectrometers. No other commercial EELS system even approaches such performance in STEM EELS SI mode. The scan shown above was taken with exposure time of 5 msec per pixel. At the maximum readout rate of 40 spectra per second, the total acquisition time is less than half an hour.

By observing the spectra while scanning the probe around the sample, it becomes clear that each phase has distinct details in the valence-loss range. In some cases, the chief difference lies in the position or width of the plasmon peak. In other cases, it is in the presence or absence of interband transitions below the plasmon peak. In addition to the inherent differences in the VEELS fine structure, the data above unfortunately also reflects variations that are purely related to multiple scattering and variations in effective specimen thickness. These confounding effects hinder the isolation of unique characteristic signals that might be harnessed for chemical phase mapping.

The EELS Analysis tools of the GMS software (included with every GIF or Enfina system) allow the multiple-scattering limitation to be overcome. The Fourier-log deconvolution function can be applied to entire STEM EELS SI data sets, thereby extracting the single-scattering distribution (SSD) at each pixel. Doing so with the data above yields the EELS SSD SI shown in the movie below:

If you have a problem viewing this movie click here to download a copy

MLLS Fitting – generating phase maps from processed STEM EELS data

The shape of the SSD of each material in this particular semiconductor device sample is evidently fairly unique and characteristic of that particular phase. The characteristic SSD profile is stable against thickness variations and thus can form the basis of a signal extraction and mapping approach based on multiple linear least squares (MLLS) fitting to reference spectra.

Taking VEELS spectra from representative sample areas of known composition, performing Fourier-log deconvolution on each one, and normalizing the result yields the following reference SSD spectra:

Using these and the full STEM EELS SI data set above as inputs to the MLLS fitting routine in GMS yields the following maps of the fit coefficient values:

The above maps illustrate both the strengths and weaknesses of the MLLS approach for this type of phase mapping. On the whole, the images above give very good separation of the different phases. This is largely because of the clear differences between the reference spectra, i.e. there is a large degree of orthogonality between the various SSDs. These distinct differences make it possible for the MLLS routine to identify some sample areas where two of the phases (e.g. Si and SiO2) appear to overlap through the thickness of this ion-milled sample. On the other hand, there are clearly areas, especially near interfaces, where artifact contrast appears (e.g. in the epoxy map) because the mixture of two different SSD signals can sometimes closely resemble that of a single non-related material. This demonstrates a key weakness of the MLLS approach, i.e. that it only works well when the reference spectra are truly linearly independent. In complex situations with many phases, there is a good chance that several references will have a good deal of commonality and so cannot be distinguished via the simple MLLS approach. A further weakness is that the MLLS approach is only as good as the reference spectra it is given. If no reference spectrum is provided for an important component of the sample, then that sample area is very likely to show up as artifact contrast in several of the maps generated.

After some further processing of the MLLS fit coefficient maps to reduce artifact contrast and noise, application of the Color Mix tool within GMS yields the following informative map of the major phases in this sample (color coding matches that of the reference SSD spectra above):

Additional mapping techniques using EELS fine structure and MLLS

For further examples of the MLLS approach to signal extraction and mapping, please refer to the article on advanced spectral mapping techniques in issue 14 of our KnowHow publication, accessible by clicking here

More Information

For further details about our GIF Tridiem line of energy filter products, please follow the links below:

GIF Tridiem
GIF Tridiem ER

If you would like to learn more about the research and development effort behind GIF Tridiem, please refer to the following publications:

H. A. Brink, M.M.G. Barfels, R.P. Burgner, B.N. Edwards, Ultramicroscopy 96 (2003) 367.
G. Kothleitner and F. Hofer, Micron 34 (2003) 211.

To learn more about STEM EELS with Gatan’s STEMPack product, please click here.