Journal of Electron Microscopy 50(3): 141–146 (2001)
© Japanese Society of Electron Microscopy
...............................................................................................................................................................................................................................................................................
Full-length paper
High-resolution elemental mapping of titanium oxide /
aluminium oxide multilayer by spectrum-imaging
Hiroki Kurata1,*, Hiroshi Kumagai2 and Kazunari Ozasa2
1Japan
Atomic Energy Research Institute (JAERI), Tokai-mura, Naka-gun, Ibaraki 319-1195 and 2The Institute of
Physical and Chemical Research (RIKEN), 2-1 Hirosawa, Wako, Saitama 351-0198, Japan
*To whom correspondence should be addressed. E-mail: [email protected]
..................................................................................................................................................................................................................
Abstract
A spectrum-imaging technique based on scanning transmission electron
microscopy combined with an electron energy-loss spectroscopy has been
applied for the multilayer of amorphous titanium oxide and aluminium
oxide layers on silicon substrate. We demonstrate the high-resolution elemental mapping and discuss the advantage of this method compared to an
energy-filtering transmission electron microscopy. The main advantage is
the absence of chromatic broadening, which allows the use of a large collection angle to acquire spectrum-image data and a wide energy window to
integrate the core-loss signals. This suggests that the spatial resolution of
elemental maps is mainly determined by the size of the electron probe.
..................................................................................................................................................................................................................
Keywords
spectrum-imaging, electron energy-loss spectroscopy, scanning transmission
electron microscopy, elemental mapping, multilayer
..................................................................................................................................................................................................................
Received
7 November 2000, accepted 26 January 2001
..................................................................................................................................................................................................................
Introduction
One of the most important roles of analytical electron microscopy is to visualize the distribution of a specific element in a
local specimen area. Elemental mapping based on electron
energy-loss spectroscopy (EELS) can be performed with two
kinds of method. One is energy-filtering transmission electron
microscopy (EFTEM) using a fixed electron beam. This
method offers fast two-dimensional imaging by detecting
electron intensity with a specific energy loss. The elemental
maps are obtained by subtracting the background intensities
under core-loss edges by using a two- or three-window technique. Several groups have reported elemental mapping with
better than 1 nm resolution using EFTEM [1–4].
The other method to obtain elemental maps is performed on
a scanning transmission electron microscope (STEM) combined with parallel-EELS, the so-called spectrum-imaging
technique [5,6]. The electron probe is scanned on a specimen
surface and the electrons scattered into high angles are
detected to form an annular dark field (ADF) STEM image. A
sequence of energy-loss spectra is acquired from all probe
positions defined on the ADF image. The resulting threedimensional spectrum-image data are a function of the spatial
coordinates x, y and the energy loss DE. After acquisition, we
can generate elemental maps from spectrum-image data by
subtracting backgrounds from the core-loss spectra above the
specific edges at each pixel in the image. Spectral changes as a
function of the probe position are also available as a spectrumline profile mode. The advantage of this technique is to be able
to correlate the spatial coordinate and the spectral features as
accurately as possible by a posteriori processing. Spatially
resolved EELS analysis has been carried out, in combination
with a high angle ADF imaging, to investigate chemical composition and bonding state at near atomic column resolution
by using a small probe size [7–11]. Owing to the recent technical developments for acquiring spectrum-image data, the
elemental map reaching to near single atom sensitivity has
been achieved [12,13].
In the present work, we demonstrate high-resolution elemental mapping by means of the spectrum-imaging technique
and discuss some advantages of this method compared to
EFTEM. The specimen examined in this work was a multilayer
of 40 pairs of amorphous titanium oxide and aluminium oxide
layers with a thickness period of 2 nm on a silicon substrate
[14]. The optical properties of this multilayer offer high
reflectance of over 30% at a wavelength of 2.734 nm and an
incident angle of 71.8° from normal incidence [15]. This fact
shows that it has promise for use as a soft X-ray reflector at
water-window wavelength region between the oxygen and
carbon K absorption edges at 2.33 and 4.36 nm, respectively.
On the other hand, the interfacial structure of this multilayer
142
J O U R N A L O F E L E C T R O N M I C R O S C O P Y, Vol. 50, No. 3, 2001
is smooth, so this is a suitable material for checking the
performance of high-resolution elemental mapping.
Methods
An alternate multilayer of amorphous titanium oxide (TiO2)
and aluminium oxide (Al2O3) was prepared on a silicon
substrate by an atomic layer deposition method of controlled
growth with sequential surface chemical reactions. Preparation method details and the optical properties of the multilayer have been reported elsewhere [14,15]. Cross-sectional
STEM specimens were prepared by mechanical dimpling and
Ar-ion milling to perforation.
The spectrum-imaging investigations were performed on a
JEOL JEM-2000F microscope equipped with a Schottky type
field emission gun operating at 200 kV. The optimum convergent semi-angle of an incident electron probe, as specified by
Scherzer [16], is aopt = 1.4(l / Cs)¼ = 8 mrad, where l (=
0.0025 nm) is the wavelength of 200 kV electrons and Cs (=
2.3 mm) is the spherical aberration coefficient of the objective
lens. By adjusting the excitation of a second condenser lens
with a free-lens controller, the optimum convergent angle was
set to perform the present experiments.
The electron probe with a diameter of less than 1 nm was
scanned digitally on a specimen surface by a scan generator
(JEOL ASID-20) coupled with a DigiScan (Gatan model 688).
ADF and bright field STEM images were acquired with the
Digital Micrograph software (version 3.3). The EELS spectra
were measured from each probe position by a DigiPEELS
(Gatan model 766) which was controlled by the spectrumimaging package incorporated in the Digital Micrograph. In
this system, the spectrum-images can be acquired from specimen regions selected on an initially recorded ADF image.
Immediately after the selection of region, the acquisition of
the spectrum-image data was started in order to minimize
specimen drift, because the ADF and spectral signals were not
measured simultaneously. This software enables us to correct
for spatial drift during spectrum acquisition by obtaining a
fast ADF image from a preset region of the specimen. Each
ADF image is cross-correlated with the previous one and the
amount of spatial drift is determined. This value is used to
adjust the probe position. In the present experiment, the spatial drift correction was applied at each two or four pixels of
the probe scan. Acquisition time of a spectrum was 0.1 or 0.2 s
per pixel. The spectrum-image data were corrected for readout noise, dark-current and gain variation of the photo-diode
array detector at each pixel.
In order to get a sufficient collection semi-angle for the
spectrometer, the short camera length (3 to 7 cm) was set by
Fig. 1 (a) Cross-sectional ADF image of the multilayer of amorphous
titanium oxide and aluminium oxide on silicon substrate. (b) Sum of
spectra from 900 pixels extracted from the spectrum-image data of the
selected area. (c) ADF image and elemental maps of titanium and
aluminium obtained from the spectrum-image data.
H. Kurata et al. Elemental mapping of multilayer by STEM-EELS
Fig. 2 Intensity profiles of titanium maps observed with three different collection semi-angles. The image contrast (C) of the Ti layer,
which is defined by the ratio between top and bottom intensities in
the oscillating profiles, was shown to estimate the spatial resolution
roughly.
using a free-lens controller attached to the microscope. The
typical collection semi-angle, b, referred to the specimen was
usually 20 to 30 mrad, which could average out diffraction
effects on spectrum intensity. By changing the camera length
or the size of an entrance aperture of the spectrometer, we
observed elemental maps dependent on the collection angle.
After the acquisition of spectrum-image data, the calibration of energy-loss was carried out and then the background
intensity under each core-loss edge was subtracted at each
pixel by using an inverse power law fitting. The extracted
core-loss signals were integrated over an appropriate energy
window to generate elemental maps.
Results and discussion
The cross-sectional ADF image observed from an alternate
multilayer structure of titanium oxide and aluminium oxide
on a silicon substrate is shown in Fig. 1a. The thickness of
each oxide layer is 1 nm and the period of this structure is 2
nm, which agrees with the expected value from preparation
conditions of the specimen. This image was obtained with an
annular detection angle of 40 to 90 mrad, so the intensity in
the image shows the atomic number sensitivity (Z-contrast)
143
[17]. The layers with bright and dark contrast in the multilayer region should correspond to the titanium oxide and
aluminium oxide, respectively. In order to confirm this
assignment, spectrum-image data with a pixel size of 0.25 nm
were acquired from a selected area (4.5 × 12.5 nm) in the
multilayer region. Figure 1b shows the summed spectrum
from 900 pixels of the measured spectrum-image data. The
spectrum-image data consists of the entire spectrum measured with a wide energy range, so core-loss signals from different kinds of element can be obtained by one acquisition, as
shown in Fig. 1b. Therefore, plural elemental maps can be
constructed with low electron dosage compared with the case
of EFTEM. By integration over Ti L2,3-edge and Al K-edge
signals at each pixel after background subtraction, elemental
maps can be constructed as shown in Fig. 1c. The integration
width, dE, for the Ti map was set to 70 eV, but that for the Al
map was extended to 400 eV from the edge position because of
the low excitation probability of Al K-edge compared to the Ti
L2,3-edge. In both maps each oxide layer is resolved individually, which indicates that high-resolution elemental mapping
with the order of 1 nm is attained in both edges. It is quite
clear that the layers with bright contrast in the ADF image
correspond to the titanium oxide layers as inferred above.
It should be noticed that in the Al map the deterioration in
the spatial resolution was not observed, although the wide
energy window was used to integrate the elemental signals. In
the EFTEM observation, the use of a wide energy window led
to blurred maps due to the chromatic aberration of the objective lens [3,18,19]. The maximum broadening caused by the
chromatic aberration characterized by the coefficient Cc can be
given by Ccq0dE / E0, where q0 is the maximum scattering angle
collected by the spectrometer and E0 the incident electron
energy [20]. Due to this effect the typical energy window used
in the high resolution elemental mapping by the EFTEM is
limited to be a few tenth electron volts. In the case of the spectrum-imaging, however, the chromatic broadening was not
detected as shown in the Al map in which the spatial resolution was almost equivalent with that of the Ti map except for
the poor signal-to-noise ratio. This fact is attributed to the
sequential measurement of the spectrum pixel by pixel with a
diffraction mode in the spectrum-imaging, so the spatial resolution is mainly determined by the probe size. Therefore, the
elemental maps obtained by spectrum-imaging can be constructed without loss of resolution by using a wide energy
window, if core-loss peaks coming from other elements are
not included in this energy region. Since using a wide energy
window increases the partial cross-section of the concerned
edge, the detection efficiency of atom should be improved.
This is an advantage of the spectrum-imaging technique.
Another factor restricting the spatial resolution of elemental
maps is the delocalization of inelastic scattering, which also
depends on the collection angle. With small collection angles,
delocalization worsens [18,21–23], which is an opposite
dependence to the chromatic broadening mentioned above.
Therefore, the collection angle must be optimized to attain the
144
J O U R N A L O F E L E C T R O N M I C R O S C O P Y, Vol. 50, No. 3, 2001
Fig. 3 (a) ADF image and elemental maps of silicon, oxygen and titanium of region near interface between multilayer and silicon substrate.
(b) Intensity profiles of ADF image and elemental maps along the direction perpendicular to the interface.
best resolution in EFTEM experiments. However, the absence
of chromatic broadening in elemental maps observed by spectrum-imaging allows us to use a large collection angle. In
order to examine the elemental mapping dependent on the
collection angle, the Ti maps of the multilayer were observed
with three different collection angles (16, 32 and 80 mrad).
Figure 2 shows the intensity profiles of the Ti maps along
the direction perpendicular to the layer structure. These profiles show a simple structure oscillating with a period of 2 nm,
reflecting Ti distributions. Although the intensity at Al2O3 layers does not become zero, which mainly can be attributed to
the probe size comparable to the thickness of TiO2 layers, the
profiles seem to be changed slightly, reflecting the different
spatial resolution. The image contrast, C, of the Ti layer, which
is defined by the ratio between top and bottom intensities in
the oscillating profiles, is shown in this figure. This contrast
value can be used as a rough estimation of the spatial resolution. With an increase of the collection semi-angle from 16 to
H. Kurata et al. Elemental mapping of multilayer by STEM-EELS
145
Fig. 4 Comparison of silicon L2,3-ELNES spectra measured at interface and substrate region, extracted from the spectrum-image data.
32 mrad, the C value becomes large (about 12%), which indicates that the spatial resolution of the elemental map is
improved by reducing delocalization contribution to the overall resolution limit. The contrast values of images obtained
with a collection semi-angle of 32 and 80 mrad are equivalent,
which might mean that the dependence of the delocalization
on the collection angle is quite weak at large angles of Ti L2,3edge. The use of large collection angles can lead to the high
collection efficiency of core-loss signals, which allows the
acquisition of spectrum-image data with a short integration
time, while maintaining good spatial resolution. This is also an
advantage of the spectrum-imaging technique. In order to perform elemental mapping at the best spatial resolution, therefore, the collection angle should be kept as large as possible,
although the energy resolution of the spectrum worsens
because of the third order aberration of the spectrometer.
Next, we measured spectrum-image data from an interface
region between the multilayer and Si substrate. The elemental
maps of Si (L2,3-edge: DE = 120 ± 20 eV), O (K-edge: DE =
552 ± 20 eV) and Ti (L2,3-edge: DE = 474 ± 20 eV), as well as
the corresponding ADF image, are shown in Fig. 3a. Figure 3b
is the intensity profile of each map along the direction perpendicular to the interface. The intensities of the elemental maps
are normalized by the value of the partial cross-section for
each edge, calculated by the Hartree-Slater method [24]
assuming a collection semi-angle of 20 mrad, so that the normalized intensities can be related to the relative number of
atoms. Indeed, the ratio between oxygen and titanium intensities in the titanium oxide layer is almost 2:1, which is con-
sistent with the chemical composition of this oxide layer.
Here, we can observe two distinct features. One is the excess
oxygen distribution found at region 1 indicated in Fig. 3a. In
this region, the Ti intensity is weak compared to that of the
titanium oxide layers far from the interface. The thickness of
this region is about 2 nm. These facts may suggest that at the
initial stage of the fabrication of the multilayer, the chemical
reactions to produce the first two layers of titanium and aluminium oxides have been incomplete.
Another feature is found at the near interface region
marked as layer 2 with the thickness of about 1 nm or less, in
which the intensity of the ADF image is rather weak, compared to that of Si substrate. In this region, some amounts of
oxygen and silicon atoms are detected as shown in the intensity profile of Fig. 3b, suggesting the existence of silicon oxide.
In order to make clear the chemical species of this interface
layer, the silicon L2,3-edge spectra were extracted from the
spectrum-image data to use the near-edge fine structure as a
fingerprint. Figure 4 shows the energy-loss near-edge structure (ELNES) of the silicon L2,3-edge measured at two positions indicated in the Si map. It is clear that the fine structure
of the interface region is completely different from that of the
substrate. This fine structure shows two peaks near the edge
located at higher energy (about 6 eV) than that of the spectrum of Si substrate, which is the typical feature of ELNES of
silicon dioxide [25]. Therefore, the interface layer can be
attributed to silicon dioxide. But the atomic ratio between
oxygen and silicon atoms determined from the intensity
profile of both maps is not 2:1 exactly. The reason for this is
146
J O U R N A L O F E L E C T R O N M I C R O S C O P Y, Vol. 50, No. 3, 2001
probably the size of the incident electron probe. Since it is
comparable to the thickness of the interface layer, the elemental signals at the interface layer are mixed with those from the
adjacent layer 1 and Si substrate. On the side of the Si substrate the atomic concentration of Si is rich, while it is poor on
the other side, but the average atomic ratio between oxygen
and silicon atoms seems to be 2:1.
These results demonstrate the high sensitivity of ELNES for
identifying the chemical species related to specific elements.
Therefore, the use of ELNES, as well as elemental mapping, is
very powerful for the characterization of the local specimen
area. Because the spectrum-image data stored at one acquisition includes many spectral features extended to a wide
energy range, a posteriori processing can offer a wealth of
chemical information, which is also advantage of this technique.
...................................................................................................................................
5
Jeanguillaume C and Colliex C (1989) Spectrum-image: the next step
in EELS digital acquisition and processing. Ultramicroscopy 28: 252–
257.
...................................................................................................................................
6
Hunt J A and Williams D B (1991) Electron energy-loss spectrumimaging. Ultramicroscopy 38: 47–73.
...................................................................................................................................
7
Batson P E (1993) Simultaneous STEM imaging and electron energyloss spectroscopy with atomic-column sensitivity. Nature 366: 727–
729.
...................................................................................................................................
8
Browning N D, Chisholm M F, and Pennycook S J (1993) Atomic-resolution chemical analysis using a scanning transmission electron
microscope. Nature 366: 143–146.
...................................................................................................................................
9
Browning N D, Wallis D J, Nellist P D, and Pennycook S J (1997) EELS
in the STEM: determination of materials properties on the atomic
scale. Micron 28: 333–348.
...................................................................................................................................
10 Duscher G, Browning N D, and Pennycook S J (1998) Atomic column
resolved electron energy-loss spectroscopy. Phys. Stat. Sol. 166: 327–
342.
...................................................................................................................................
11 Muller D A, Sorsch T, Mocciom S, Baumann F H, Evans-Lutterodt K,
and Timp G (1999) The electronic structure at the atomic scale of
ultrathin gate oxides. Nature 399: 758–761.
...................................................................................................................................
Concluding remarks
In this paper we have performed the spectrum-imaging of a
multilayer of amorphous titanium oxide and aluminium
oxide. We have demonstrated high-resolution elemental mapping and discussed the advantage of this technique compared
to EFTEM. The main advantage is the absence of the chromatic broadening effect on elemental maps, so that spectrumimage data can be acquired with a large collection angle,
which can minimize the delocalization contribution to the
total spatial resolution of maps. Moreover, the integration of
core-loss signals to generate the elemental maps can be done
with a wide energy window, which means that the detection
efficiency should be improved while keeping the best spatial
resolution governed mainly by the probe size. Additionally, we
have found the silicon dioxide layer at the interface between
the multilayer and the substrate by analysing the change of Si
L2,3-ELNES. Using the ELNES together with elemental maps
extracted from the spectrum-imaging data is quite useful for
characterization of the local area of specimen.
12 Leapman R D and Rizzo N W (1999) Towards single atom analysis of
biological structures. Ultramicroscopy 78: 251–268.
...................................................................................................................................
13 Suenaga K, Tencé M, Mory C, Colliex C, Kato H, Okazaki T, Shinohara
H, Hirahara K, Bandow S, and Iijima S (2000) Element-selective single
atom imaging. Science 290: 2280–2282.
...................................................................................................................................
14 Kumagai H, Matsumoto M, Kawamura Y, Toyoda K, and Obara M
(1994) Fabrication of multilayers with growth controlled by sequential
surface chemical reactions. Jpn. J. Appl. Phys. 33: 7086–7089.
...................................................................................................................................
15 Kumagai H, Toyoda K, Kobayashi K, Obara M, and Imura Y (1997)
Titanium oxide/aluminum oxide multilayer reflectors for “waterwindow” wavelengths. Appl. Phys. Lett. 70: 2338–2340.
...................................................................................................................................
16 Scherzer O (1949) The theoretical resolution limit of the electron
microscope. J. Appl. Phys. 20: 20–29.
...................................................................................................................................
17 Pennycook S J (1989) Z-contrast STEM for materials science. Ultramicroscopy 30: 58–69.
...................................................................................................................................
18 Krivanek O L, Kundmann M K, and Kimoto K (1995) Spatial resolution in EFTEM elemental maps. J. Microsc. 180: 277–287.
...................................................................................................................................
19 Kurata H (1999) Advantages of elemental mapping by high-voltage
EFTEM. Ultramicroscopy 78: 233–240.
...................................................................................................................................
20 Shuman H, Chang C F, and Somlyo A P (1986) Elemental imaging and
resolution in energy-filtered conventional electron microscopy. Ultramicroscopy 19: 121–134.
...................................................................................................................................
21 Kohl H and Rose H (1985) Theory of image formation by inelastically
scattered electrons in the electron microscope. Adv. Electro. Electron
Phys. 65: 173–227.
...................................................................................................................................
Acknowledgements
This study was partially supported by the Special Coordination Funds for
promoting Science and Technology of the Science and Technology Agency
of the Japanese Government.
22 Muller D A and Silcox J (1995) Delocalization in inelastic scattering.
Ultramicroscopy 59: 195–213.
...................................................................................................................................
23 Rafferty B and Pennycook S J (1999) Towards atomic column-bycolumn spectroscopy. Ultramicroscopy 78: 141–151.
...................................................................................................................................
24 Leapman R D, Rez P, and Mayers D F (1980) K, L, and M shell generalized oscillator strengths and ionization cross sections for fast electron collisions. J. Chem. Phys. 72: 1232–1243.
...................................................................................................................................
References
...................................................................................................................................
1
Berger A, Mayer J, and Kohl H (1994) Detection limits in elemental
distribution images produced by energy filtering TEM: case study of
grain boundaries in Si3N4. Ultramicroscopy 55: 101–112.
...................................................................................................................................
2
Kimoto K, Hirano T, and Usami K (1995) Elemental mapping using a
field emission transmission electron microscope with an imaging filter. J. Electron Microsc. 44: 86–90.
...................................................................................................................................
3
Kurata H, Moriguchi S, Isoda S, and Kobayashi T (1996) Attainable
resolution of energy-selecting image using high-voltage electron
microscope. J. Electron Microsc. 45: 79–84.
...................................................................................................................................
4
Botton G A and Phaneuf M W (1999) Imaging, spectroscopy and spectroscopic imaging with an energy filtered field emission TEM. Micron
30: 109–119.
25 Ahn C C and Krivanek O L (1983) EELS Atras. (Gatan.)