Rapid searching of spectrum image databases for rare events, or finding the needle in the haystack when you don’t even know you’re looking for a needle!

David S. Bright and Dale E. Newbury

Abstract:
We have developed the Maximum Pixel Spectrum, a software tool that enables rapid
searching of SEM x-ray spectrum image databases to detect rare features, even if the
analyst does not know in advance which elements are present. Single pixel features can be
efficiently recognized and highlighted. Applications include contamination identification
and localization and analysis of minority/trace phases.

Purpose: Characterization of the microstructure of materials often requires detection of
rare features, such as naturally occurring minority/trace phases or unintentional particulate
contamination. X-ray mapping in the SEM, a traditional tool for measuring elemental
distributions with micrometer to nanometer spatial resolution, has recently been greatly
enhanced by the development of x-ray spectrum imaging, in which a complete x-ray
spectrum is stored at each pixel location visited by the beam. The silicon drift detector
(SDD), described here in FY03 and whose development was aided by NIST SBIR grants,
combined with digital signal processing enables x-ray count rates above 100 kHz,
permitting recording of useful x-ray spectrum images in 200 s or less. The resulting
stream of 200 Mbyte image databases is creating a demand for software tools that are
quick and efficient at locating features of interest.

Major Accomplishments: We have developed a software tool within the NIST LISPIX
image processing platform (available free at http://www.nist.gov/lispix/) that determines the
MAXIMUM PIXEL SPECTRUM by finding the maximum value within each energy
channel x-y plane and plotting this value versus energy. This new function is compared
with the SUM SPECTRUM, similarly calculated by adding all values within a plane, as
shown in Figure 1(a) for a spectrum image of Raney nickel, a methanation catalyst. Peaks
in the SUM SPECTRUM correspond to common features in the x-ray spectrum image, as
illustrated in Figure 1(b) where the aluminum-rich phases are highlighted.. While these
same peaks are found in the MAXIMUM PIXEL SPECTRUM, additional peaks can be
recognized that correspond to rare events, down to the single pixel level, shown in Figure
1(c) for a chromium contaminant that appears at a single pixel, or 1/51200 for a 256x200
scan. Note that the rare chromium feature has been found despite being completely
unknown to the analyst.

Impact: The MAXIMUM PIXEL SPECTRUM has had an immediate impact in the
microanalysis field. First publicly presented at the SCANNING 04 conference held in
Washington in April, 2004, the method was immediately adopted by a U.S. manufacturer
of microanalysis software systems, appearing in this vendor’s system at the Microscopy
and Microanalysis Conference in August and in their advertisement in Microscopy Today
in September, 2004 (page 27). Other vendors are rapidly incorporating the MAXIMUM
PIXEL SPECTRUM as a feature in their spectrum imaging software. We anticipate that
the combination of SEM SDD x-ray spectrum imaging and derived spectrum image
processing tools will have a broad impact in materials analysis, supporting technology,
physical and biological science, and forensic applications.

Future Plans:
The MAXIMUM PIXEL SPECTRUM and the SUM SPECTRUM are members of a class
of transformations known as “derived spectra” to distinguish them from the true spectra
recorded in the spectrum image. We plan to investigate additional algorithms for derived
spectra to seek software tools that can aid the analyst in other aspects of x-ray spectrum
imaging. Other microanalysis spectroscopies, such as electron energy loss and Auger
electron, may also benefit from derived spectrum tools.

Web Contact micro@nist.gov