104 2 Energy-Loss Instrumentation
(due to the width of the energy-selecting slit) but these can be addressed by FFT
interpolation and deconvolution methods (Lo et al., 2001).
The main attraction of the spectrum image concept is that more spectroscopic
data are recorded and can be subsequently processed to extract information that
might otherwise have been lost. Examples of such processing include the calculation
of local thickness, pre-edge background subtraction, deconvolution, multivariate
analysis, and Kramers–Kronig analysis. The resulting information can be displayed
as line scans (Tencé et al., 1995) or two-dimensional images of specimen t hickness,
elemental concentration, complex permittivity, and bonding information (Hunt and
Williams, 1991; Botton and L’Esperance, 1994; Arenal et al., 2008). In addition,
instrumental artifacts such as gain nonuniformities and drift of the microscope high
voltage or beam current can be corrected by post-acquisition processing.
The acquisition time of a spectrum image is often quite long. In the past, the min-
imum pixel time has been limited by the array readout time, but that has recently
been reduced from 25 to 1 ms (Gubbens et al., 2010). A line spectrum, achieved by
scanning an electron probe in a line and recording a spectrum from each pixel, can
be acquired more rapidly and is often sufficient for determining elemental profiles.
Similar data can be obtained in fixed-beam TEM mode, with broad beam illumi-
nating a slit introduced at the entrance of a double-focusing spectrometer. The long
direction (y) of the slit corresponds to the nondispersive direction in the spectrome-
ter image plane, allowing the energy-loss intensity J(y,E) to be recorded by a CCD
camera. One advantage here is that all spectra are acquired simultaneously, so spec-
imen drift does not distort the information obtained, although it may result in loss
of spatial resolution.
A common form of energy-filtered image involves selecting a range of energy
loss (typically 10 eV or more in width) corresponding to an inner-shell ioniza-
tion edge. Since each edge is characteristic of a particular element, the core-loss
image contains information about the spatial distribution of elements present in
the specimen. However, each ionization edge is superimposed on a spectral back-
ground arising from other energy-loss processes. To obtain an image that represents
the characteristic loss intensity alone, the background contribution I
b
within the
core-loss region of the spectrum must be subtracted, as in the case of spectroscopy
(Section 4.4). The background intensity often decreases smoothly with energy loss
E, approximating to a power law form J(E) = AE
−r
, where A and r are parameters
that can be determined by examining J(E) at energy losses just below the ionization
threshold (Section 4.4.2). Unfortunately, both A and r can vary across the specimen,
as a result of changes in thickness and composition (Leapman and Swyt, 1983;
Leapman et al., 1984c), in which case a separate estimation of I
b
is required at each
picture element (pixel).
In the case of STEM imaging, where each pixel is measured sequentially, local
values of A and r can be obtained through a least-squares or two-area fitting to the
pre-edge intensity recorded over several channels preceding the edge. With electro-
static deflection of the spectrometer exit beam and fast electronics, the necessary
data processing may be done within each pixel dwell period (“on the fly”) and the
system can provide a live display of the appropriate part of the spectrum (Gorlen