627_1.pdf

Transmission Electron Microscopy: Overview and
Challenges
S. J. Pennvcook1'2. A. R. Lupini1, A. Borisevich,l M. Varela,l Y. Peng,l P. D.
Nellist3, G. Duscher1'4, R. Buczko1'2'5 and S. T. Pantelides1'2,
1
Condensed Matter Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN
'Department of Physics and Astronomy, Vanderbilt University, Nashville, TN
3
Nion Co., 1102 8th Street, Kirkland WA
4
Department of Materials Science and Engineering, North Carolina State University, Raleigh, NC
5
Institute of Physics, Polish Academy of Sciences, Warsaw, Poland
Abstract. We review recent advances in aberration-corrected scanning transmission electron microscopy that allow
sub-Angstrom beams to be used for imaging and spectroscopy, with enormous improvement in sensitivity for single
atom detection and the investigation of interfacial electronic structure. Comparison is made between the electronic and
structural width of gate oxides, with interpretation through first-principles theory. Future developments are discussed.
INTRODUCTION
In recent years commercially available scanning
transmission electron microscopes have become able
to routinely provide atomic-sized electron beams,1'2
allowing simultaneous Z-contrast imaging and electron
energy loss spectroscopy (EELS) as shown in Fig. 1.
This combination is particularly powerful for
electronic materials because it allows direct
comparison of the atomic structure and electronic
properties, as we show for a gate oxide later.
Furthermore, it is now feasible to correct the
inherently large spherical aberration of microscope
objective lenses,3'4' which promises to at least double
the achievable resolution for both the conventional
TEM phase contrast imaging mode, the STEM Zcontrast mode (also referred to as high angle annular
dark field or HAADF imaging) and in addition the
spatial resolution of EELS. The potential benefits for
STEM may turn out to be greater because it is much
less sensitive to chromatic instabilities5, and is
presently the only viable means for obtaining
spectroscopic analysis
at atomic resolution.
Aberration-correction not only increases the available
resolution, it brings single atom sensitivity to both
imaging and EELS.
Here we present recent results from the aberrationcorrected STEMs at Oak Ridge National Laboratory
(ORNL). Although these instruments are no longer
commercially available, aberration-correctors are
currently being fitted to some commercial instruments,
and comparable advances can be anticipated in the
next few years. Our 100 kV VG Microscopes
HB501UX has been fitted'with an aberration corrector
constructed by Nion Co., which improved its
resolution from 2.2 A (full-width-half-maximum probe
intensity) to around 1.3 A, equal to the uncorrected
300 kV HB603U STEM at ORNL. After correction,
the 300 kV STEM is resolving 0.78 A, although the
theoretically achievable probe size in the absence of
instabilities is predicted to be 0.5 A.
The combination of atomic-resolution Z-contrast
microscopy, electron energy loss spectroscopy and
first-principles theory has proved to be a powerful
means for structure property correlations at interfaces
and nanostructures6'7'8'9. The Z-contrast image10'11'12 is
a convenient and intuitive method for revealing atomic
arrangements. Because simulations are not essential
for image interpretation, unexpected configurations
can often be seen.
Examples include isolated
dislocation cores13 and also dislocation arrays that
comprise grain boundaries.14'15 Locating the probe on
CP683, Characterization and Metrology for VLSI Technology: 2003 International Conference,
edited by D. G. Seiler, A. C. Diebold, T. J. Shaffner, R. McDonald, S. Zollner, R. P. Khosla, and E. M. Secula
© 2003 American Institute of Physics 0-7354-0152-7/03/$20.00
627
RECENT RESULTS FROM
ABERRATION-CORRECTED STEM
Figure 2 shows the sensitivity to single atoms
available at 100 kV with the enhanced resolution,
where single Bi atoms on lattice sites within the crystal
are visible. Similar results have been shown also for
Sb atoms in Si using an uncorrected 200 kV
microscope.2 In our case, intensity profiles across the
image reveal which of the two columns of the
dumbbell contain the Bi atom. The density of bright
spots correlates with the known dose of Bi, which was
introduced by ion implantation followed by
recrystallization through solid phase epitaxial growth.
The sample was prepared by standard ion milling
procedures using a final cleaning at 0.5 kV.
FIGURE 1. Schematic showing simultaneous Z-contrast
imaging and EELS in the STEM. The image distinguishes
the sublattice polarity in GaAs.
an individual atomic column selected from the image
allows EELS measurements of local elemental
concentrations and electronic structure.16'17 Very
recently, identification of a single atom within a bulk
crystal has been demonstrated.18
These techniques are ideally complemented by
first-principles total-energy calculations.
The
structures suggested from experiment avoid the need
to search the large number of possible defect
configurations.
Theory can efficiently perform
structural relaxations of configurations suggested from
experiment, to confirm and refine the suggested
structure.
Theory can also provide segregation
energies for impurities and point defects and the local
electronic structure and optical properties.19'20 Finally,
theory can be used to calculate EELS spectra from first
principles, although excitonic effects need to be
included for a good match to experiment.21
628
FIGURE 2. Z-contrast image of Bi-doped Si (110)
revealing the columns containing individual Bi atoms.
Figure 3 compares the performance before and
after fitting a Nion aberration-corrector to the ORNL
300 kV STEM. On the left is the optimum probe
profile for the uncorrected microscope, and an image
and line trace from Si (110). The right hand side
shows the 0.5 A probe predicted after aberrationcorrection, showing the same current squeezed into a
smaller, brighter probe. Such a probe should therefore
give greatly improved contrast and signal to noise
ratio. The resolution obtained in the image, however,
is limited to around 0.9 A due to instabilities.
Nevertheless, the Si (110) image now shows more
contrast, with much deeper dips between the
dumbbells. In addition, the effect of probe tails has
been much reduced. We no longer see the weak
subsidiary maxima in the center of the channels that
are present in the uncorrected image. These features
come from the extended tails on the uncorrected probe.
When this probe is centered on a channel, the tail
overlaps the nearest six columns of Si giving a weak
secondary peak. After correction the tails are reduced
in intensity and brought closer to the central peak so
this spurious feature is no longer present.
13 A
0,5 A
FIGURE 4. Comparison of images of Pt trimers and dimers
on a y-alumina support before (left) and after (right)
aberration correction.
GATE OXIDE CHARACTERIZATION
These new tools will allow more detailed and
precise characterization of gate oxides, not only
comparison of their electronic and structural thickness,
but investigation of impurity atom sites, interfacial
steps, and their effect on local electronic properties.
Structural width
FIGURE 3. Comparison of the performance of the 300 kV
STEM before correction (left) and after installation of a Nion
aberration corrector (right). Panels show, from top to
bottom, the theoretical probe profile, images and intensity
profiles of Si (110).
The use of the Z-contrast STEM for measuring the
structural thickness of a gate oxide has recently been
reviewed.22 The method is based on the fact that to a
good approximation the HAADF image is an
incoherent image, which is simply the true image
blurred by an amount independent of the specimen.
Mathematically, the image intensity is given by
More spectacular is the improved sensitivity to
single atoms. Figure 4 shows Pt trimers and dimers
imaged with the 1.3 A probe of the 300 kV STEM
before correction. This was the first time individual
atoms were visible on a real industrial support
material, but the right hand side shows the greatly
improved contrast and signal to noise ratio after
aberration correction. The reason is that the peak beam
intensity is greatly increased, but at the same time the
area of the probe is reduced. The single atom
therefore shows greater intensity with less background
from the support material. Aberration correction
produces a rapid, non-linear improvement in image
quality.
= O(R) * P2(R)
(1)
Where R represents the two-dimensional coordinates
of the image, O the object function, often represented
simply as a Z2 Rutherford cross section for each atom,
where Z is atomic number, and P2 is the effective
probe intensity profile, including any broadening
within the crystal. The model works surprisingly well
in thin crystals because the electrons tend to channel
along the atom columns, which therefore appear sharp
and bright with a relatively low background between
them. In thicker crystals, beam broadening increases
the background. The contrast from the atomic columns
is reduced, and the effective probe gains a broad
background. The same effect occurs in the amorphous
oxide, but at a different rate. The effective probe can
change from the Si to SiO2, but in a thin specimen the
effect is second order and useful quantitative
measurements can be made assuming incoherent
imaging with a sample independent probe.
629
analysis is certainly possible to better take account of
the tails on the probe, the dechanneling due to strain
and also the effect of specimen thickness.
For an abrupt interface, convolution with the probe
will smooth out the step and the point of inflexion
marks the position of the interface.23 Differentiating
the intensity trace will regenerate the probe profile. If
the interface has structural roughness, this will further
broaden the intensity profile, but the point of inflection
will remain at the mean interface position. This
method is relatively immune to details of the specimen
and microscope parameters. If we further assume the
probe approximates to a Gaussian, for which
convolution means adding the full-width-half-maxima
in quadrature, i.e. peak width2 = effective probe width2
+ interface roughness2), inverting this relationship
gives the interface roughness, which is generally
assumed to be Gaussian in form. An example is
presented in Fig. 5.
Electronic width
The electronic width of the gate oxide may be
significantly narrower than the structural width
apparent from the image, due to sub-oxide bonds at the
interface. Figure 6 shows a Z-contrast image of a
Si/SiO2 interface, a linescan of the image intensity
with EELS spectra recorded simultaneously. The
bright region near the interface is a strain contrast
effect, seen with low inner angles of the annular
detector.10 The EELS spectra show the electronic
effect of the interface extends significantly further than
the geometric extent that is seen in the intensity trace.
Whereas the geometric roughness is again ~ 4 - 5 A,
the EELS edge takes ~ 1 nm for the transition from
bulk Si, which shows a Si L2,3 edge onset at ~ 100 eV,
to full stoichiometric SiO2 with an edge onset at 106
eV. This is not due to delocalization of the inelastic
scattering. Under incoherent conditions (the use of a
large acceptance angle into the spectrometer), quantum
mechanical calculations of the EELS spatial resolution
have shown that the ultimate image resolution is the
geometric size of the initial inner shell orbital24.
It reflects that fact that the full SiO2 band gap is only
seen in fully stoichiometric oxide.
FIGURE 5. HAADF image of a Si-SiO2 interface showing
extraction of interface roughness. The image is obtained
with a VG Microscopes HB603U dedicated STEM with a
probe size of about 0.13 nm, sufficiently small to resolve the
dumbbells. A vertically averaged line trace (grey) is fitted
with a smooth curve (dashed line), differentiated (dotted
line) and the width compared to the probe profile.
The FWHM of the derivative trace is 5.1 A, and
subtracting (in quadrature) the FWHM of the probe
intensity profile of 1.3 A gives a roughness of 4.9 A.
Note, however, that the derivative trace is somewhat
asymmetric. This is due to some dechanneling in the
image of the Si crystal in the last few monolayers,
caused by strain induced by the oxide, which is
discussed further below. The distance from the peak
of the derivative trace to half intensity on the oxide
side of the interface is 2.0 A, which if we double and
subtract the probe FWHM gives an interfacial
roughness of 3.7 A. This illustrates the potential
accuracy and errors in the method. More detailed
1
100
120
ftV)
FIGURE 6. HAADF image (top), intensity trace and EELS
spectra. Shaded region on the interface spectrum reveals
interface states due to suboxide bonding. The electronic
width of the interface is ~ Inm, more than double the
structural width.
630
actually energetically favorable, the reason being that
the Si-Si bond is very stiff, and inserting an additional
O between the two Si atoms provides another degree
of freedom for relieving the large elastic stresses (Fig.
8).
THEORETICAL STUDY OF ATOMIC
AND ELECTRONIC STRUCTURE AT
THE SI/SIO2 INTERFACE
First-principles theory provides excellent insight
into the effect of interface structure on the electronic
properties. In Fig. 7 we compare EELS spectra from
various interfacial bonding configurations calculated
from density functional theory, with core hole effects
included by the Z+l method.21 The relative edge
onsets
were
obtained
from
experimental
photoemission data. The suboxide bonding on the right
interface delays the opening of the band gap and
produces characteristic spectral features. In principle,
the ability to relate specific features in the EELS
spectrum to specific bonding arrangements at the
interface allows experimental spectra to be inverted,
and the nature of the interfacial bonding to be
determined.
FIGURE 8. Schematic showing the increased range of
positions available for a Si atom after oxidation.
FUTURE DIRECTIONS
The increased resolution, contrast and signal to
noise ratio made possible through aberration
correction is clear from the examples given above.
Plans exist to correct even higher order aberrations,25
allowing even finer probes to be formed as shown in
Fig. 9.
s 10*
710s
810*
98 100 102 104 106 108 110
* i M mrad; Cs « 1 mm
* 2S mmg; €§ » 0
» 3S rnrad; Cs * 0
d; Cs*0
610®
(eV)
4 10s
FIGURE 7. Calculated EELS spectra from specific Si
atoms at an abrupt Si/SiO2 interface (left) and at one
containing sub-oxide bonding (right).
The sub-oxide
configurations (dashed) have a significantly reduced band
gap compared to the abrupt interface and are responsible for
the interface states seen in Fig. 6
3 10s
210s
1 10s
0
The limitation to inverting such data at present is
the large number of sites and configurations that are
probed in a real specimen. This is not a limitation
imposed by the probe, but by the extensive intermixing
induced by the oxidation process. If alternative,
gentler, oxidation methodologies could be developed it
might be possible to reduce the intermixing and probe
specific problem sites.
FIGURE 9. Probe intensity profiles for a 300 kV STEM
with spherical aberration under optimum defocus of- 45 nm,
and for various apertures after aberration correction,
normalized to the same total incident intensity through the
aperture.
Although
chromatic
aberration
becomes
increasingly important, transferring intensity from the
central peak into the tails of the probe, the question
arises as to the fundamental resolution limit. For a
single atom, the resolution limit would be just the high
angle components of the atomic potential, thermally
smeared, and the resolution limit would be the mean
square thermal vibration amplitude, a few tenths of an
Angstrom. Based on the Bloch wave analysis, it was
proposed that the fundamental limit in a zone axis
In order to investigate such issues, total energy
calculations were performed for a variety of possible
interfacial structures, with various forms of crystalline
oxide, with and without suboxide bonding.8 The
excess interfacial energy was calculated, to remove the
effects of the stressed bulk oxide in the structure. The
surprising outcome was that abrupt interfaces are
631
With a small aperture the crystal lattice could be
imaged in projection, while 3D tomography could be
achieved with a larger aperture. Furthermore, EELS
data or X-ray fluorescence data could also be collected
and reconstructed in 3D, providing the ultimate
analysis: 3D reconstruction at atomic resolution with
single atom identification.
crystal, in the presence of multiple elastic scattering,
would be the Is state.26 This is borne out by recent
calculations.27
The reason is that in the aberration corrected
STEM the probe becomes so small that it closely
matches the Is states, and therefore, if it enters the
crystal over a column of atoms, a substantial fraction
will become the Is state, and go straight down the
column, scattering to the ADF detector as it
propagates. If the probe enters between the columns,
there is no Is state to couple into, and the probe
decomposes into many less localized states which do
not scatter efficiently into the high angle detector.
Thus it is physically very reasonable that the image
represents a direct image of the Is states under these
conditions. The Is states in GaAs have a width of ~
0.3 A, which therefore represents the quantum
mechanical limit to resolution in the microscope.28
FIGURE 11. Schematic showing depth sectioning in the
STEM. The impurity atoms (lighter) may be identified from
the image or by spectroscopy and the entire specimen
reconstructed at atomic resolution in 3D.
The maximum overlap between the probe and the
As Is state occurs for a probe angle of ~ 20 mrad,
which gives a probe size of- 0.5 A. Figure 10 shows
the calculated image of GaAs with such a probe; the
width of the features is -0.6 A, as expected for a
convolution of the probe with the Is states, and the
intensity shows the expected Z-contrast.
SUMMARY
FIGURE 10. Calculated image of GaAs for a 0.5 A probe.
This raises the question of what the image might
look like at even larger aperture angles. Aberration
correction may open up modes of microscopy that
were never before possible. Confocal imaging has
revolutionized optical microscopy by facilitating 3D
tomography through depth sectioning. ADF STEM is
ideal for tomography since the images show no
contrast reversals with focus, a key requirement. If the
probe-forming aperture is doubled again, to 50 mrad or
above, then most of the probe intensity is contained in
the high angle components of the probe. These are
traveling at sufficiently high angles to the crystal zone
axis that they propagate primarily as plane waves, with
very small Is Bloch state component. Such beams are
scattered weakly by the sample, and therefore come to
a focus at a unique depth in the specimen. Depth
sectioning will become a viable technique in electron
microscopy for the first time, as depicted
schematically in Fig. 11.
632
The correction of aberrations in the electron
microscope removes the major barrier to resolution
that has existed since its invention. The greatly
improved sensitivity and signal to noise ratio is
expected to allow single atom detection not only in
imaging but also in spectroscopy, perhaps even in
three dimensions.
The ultimate sensitivity will
become available for determining dopant profiles, the
influence of individual impurity atoms at a buried
interface, and the nature of interface trap states.
ACKNOWLEDGMENTS
This research was sponsored by the Division of
Materials Sciences, U.S. Department of Energy, under
contract DE-AC05-OOOR22725 managed by UTBattelle, LLC, by NSF under grants no. DMR9803021, and DMR-9803768, by AFOSR MURI
Grant No. F-49620-99-1-0289, DARPA Grant No.
MDA972-98-1-0007, EPRI Grant No. WO806905, the
William A. and Nancy F. McMinn Endowment at
Vanderbilt University, and by an appointment to the
ORNL postdoctoral research program administered
jointly by ORNL and ORISE. Computation time was
partially supported by the National Science
Foundation under DMR990002N and DMR000004N
and utilized the SGI Origin 2000 at the National
Center for Supercomputing Applications, University
of Illinois at Urbana-Champaign and the National
Energy Research Scientific Computing Center,
supported by the Office of Science of the U.S.
Department of Energy under Contract No. DE-AC0376SF00098.
Electron Physics 123 edited by P. W. Hawkes,
NewYork: Academic Press, 2002, pp. 173-206.
13
A. J. McGibbon, S. J. Pennycook, and J. E. Angelo,
Science 269, 519-521 (1995).
14
M. M. McGibbon, N. D. Browning, M. F. Chisholm, A.
J. McGibbon, and S. J. Pennycook, V. Ravikumar, and
V. P. Dravid, Science 266, 102-104 (1994).
15
M. F. Chisholm and S. J. Pennycook. Mater. Res. Soc.
Bull. 22, 53-57 (1997)
16
N. D. Browning, M. F. Chisholm, and S. J. Pennycook,
Nature 366, 143-146(1993).
17
G. Duscher, N. D. Browning, and S. J. Pennycook, Phys.
Stat. Sol. (a) 166, 327-342 (1998).
REFERENCES
1
E. M. James, N. D. Browning, A. W. Nicholls, M.
Kawasaki, Y. Xin, and S. Stemmer, J. Elect. Micr. 47,
561-574(1998).
2
P. M. Voyles, et al., Nature, 416, 826-829 (2002).
18
3
M. Haider, S. Uhlemann, E. Schwan, H. Rose, B. Kabius
and K Urban, Nature 392, 768-769 (1998).
M. Varela et al, Science, submitted (2003).
19
O. L. Krivanek, N. Dellby
Ultramicroscopy,l%, 1-11 (1999).
M. F. Chisholm, A. Maiti, S. J. Pennycook, and S. T.
Pantelides, Phys. Rev. Lett. 81, 132-135 (1998).
20
Y. Yan, M. F. Chisholm, G. Duscher, A. Maiti, S. J.
Pennycook, and S. T. Pantelides, Phys. Rev. Lett. 81,
3675-3678(1998).
21
R. Buczko, G. Duscher, S. J. Pennycook and S. T.
Pantelides, Phys. Rev. Lett. 85, 2168-2171 (2000).
4
and
A.
Lupini,
5
P. D. Nellist and S. J. Pennycook, Phys. Rev. Lett. 81,
4156-4159(1998).
6
S. J. Pennycook, G. Duscher, R. Buczko, M. Kirn, N. D.
Browning and S. T. Pantelides, in Encyclopedia of
Materials: Science and Technology, Elsevier Science
Ltd. 2313(2001).
7
X. Fan, E. C. Dickey, P. Eklund, K. A. Williams, L.
Grigorian, R. S. Buczko, S. T. Pantelides and S. J.
Pennycook, Phys. Rev. Lett. 84,4621-4624 (2000).
8
R. Buczko, S. J. Pennycook, and S. T. Pantelides, Phys.
Rev. Lett. 84, 943-946 (2000).
9
M. Kim, G. Duscher, N.D. Browning, K. Sohlberg, S.T.
Pantelides, and S.J. Pennycook, Phys. Rev. Lett. 86,
4056-4059 (2001).
10
11
12
A. Diebold et al., Microscopy and Microanalysis, in
press (2003).
S. J. Pennycook and P. D. Nellist "Z-Contrast Scanning
Transmission Electron Microscopy" in Impact of
Electron and Scanning Probe Microscopy on Materials
Research edited by D. G. Rickerby, U. Valdre and G.
Valdre, The Netherlands: Kluwer Academic Publisers,
1999, pp. 161-207.
P. D. Nellist and S. J. Pennycook, "The principles and
interpretation of annular dark-field Z-contrast imaging"
in Advances in Imaging and Electron Physics 113 edited
by P. W. Hawkes, New York: Academic Press, 2000, pp.
147-203.
S. J. Pennycook, "Structure determination through Zcontrast microscopy" in Advances in Imaging and
633
23
D.A. Muller, "Gate Dielectric Metrology Using
Advanced TEM Measurements" in Characterization and
Metrology for VLSI Technology 2000, edited by D.G.
Seiler et al., AIP conference Proceedings 550, New
York: American Institute of Physics, 2000, pp 500-505.
24
B. Rafferty, and S. J. Pennycook, Ultramicroscopy, 78,
141-151 (1999).
25
N. Dellby, O. L. Krivanek, P. D. Nellist, P. E. Batson,
and A. R. Lupini, Journal of Electron Microscopy 50,
177-185(2001).
26
S. J. Pennycook, B. Rafferty and P. D. Nellist,
Microscopy and Microanalysis 6, 343-352 (2000).
27
Y. Peng, P. D. Nellist and S. J. Pennycook, J. of Electron
Microscopy, submitted (2003).
28
S. J. Pennycook, A. R. Lupini, M. Varela, A. Borisevich,
Y. Peng, and P. D. Nellist, "Quantum Mechanical
Resolution Limits to Imaging and Spectroscopy in the
Transmission Electron Microscope" in From the Atomic
to the Nano Scale, edited by J. McGuire et al., AIP
Conference Proceedings in press, New York: American
Institute of Physics