You are on page 1of 9

Ultramicroscopy 123 (2012) 90–98

Contents lists available at SciVerse ScienceDirect

Ultramicroscopy
journal homepage: www.elsevier.com/locate/ultramic

Scanning transmission electron microscopy:


Albert Crewe’s vision and beyond
Ondrej L. Krivanek a,n, Matthew F. Chisholm b, Matthew F. Murfitt a, Niklas Dellby a
a
Nion Company, 1102 8th Street, Kirkland, WA 98033, USA
b
Oak Ridge National Laboratory, Materials Science and Technology Division, Oak Ridge, TN 37831-6069, USA

a r t i c l e i n f o a b s t r a c t

Available online 17 May 2012 Some four decades were needed to catch up with the vision that Albert Crewe and his group had for the
Keywords: scanning transmission electron microscope (STEM) in the nineteen sixties and seventies: attaining 0.5 Å
STEM resolution, and identifying single atoms spectroscopically. With these goals now attained, STEM
ADF imaging developments are turning toward new directions, such as rapid atomic resolution imaging and
EELS exploring atomic bonding and electronic properties of samples at atomic resolution. The accomplish-
Aberration correction ments and the future challenges are reviewed and illustrated with practical examples.
Single atom imaging & 2012 Elsevier B.V. All rights reserved.
Single atom spectroscopy

1. Introduction results that they have brought, it is not always remembered that
the first time matter was investigated with an atom-sized probe
The invention and successful implementation of the modern that could be manipulated at will was in the scanning transmis-
scanning transmission electron microscope (STEM) by Albert sion electron microscope built by Albert Crewe’s group in
Crewe and his coworkers and students in Chicago in the 1960s Chicago.
and 1970s were nothing short of revolutionary. For the first time In the field of electron microscopy, the achievements and
in human history, it became possible to analyze solid matter with dreams of the Chicago group are usually given due credit, and
a real-space probe of atomic dimensions [1–5]. The STEM pro- they have been followed up energetically. There were many major
duced the first images of single atoms not subject to intense milestones on the journey, such as improving the attainable
electric fields (which are inevitable in a field ion microscope), in resolution by about 2x by going to higher operating voltages,
their native environment [6]. It also pointed the way to quanti- improving the performance of the cold field emission electron
tative microanalysis with single-atom sensitivity, in which the gun, improving the resolution by a further 2–3x through the
inelastic interaction between the electrons in the atom-sized introduction of working aberration correctors, improving the
probe and the sample is used to determine the chemical type of energy resolution and the collection efficiency of electron energy
the atoms in the sample [7]. Also interesting to note, the loss spectroscopy (EELS), improving the efficiency of X-ray
investigations with the Chicago STEM were all done at primary energy-dispersive spectroscopy, monochromating the primary
energies r45 keV, thus minimizing the knock-on damage in the beam for improved EELS energy resolution, and returning to
specimen, and allowing phenomena like thermally-activated lower operating energies, in order to eliminate knock-on radiation
atomic hopping to be investigated [8,9]. damage. Most of these developments have been thoroughly
Four decades later, we are thoroughly accustomed to the idea reviewed, e.g. in Pennycook’s ‘‘Scan through the History of STEM’’
of probing matter and manipulating it on an atomic scale, chapter that gives over 500 references [13].
typically via a single atom that protrudes from the tip of a In this article we review progress in two of the research areas
scanning tunneling microscope (STM), or of an atomic-force of aberration-corrected STEM we have been involved with espe-
microscope (AFM) [10,11]. We have also learned how to form cially closely: rapid imaging made possible by increasing the
atom-size probes from ions, which have some important advan- electron current contained in an ultra-small probe, and low keV
tages over electron probes [12]. Unfortunately, perhaps due to the imaging and analysis.
large number of such techniques now available and the wealth of

2. Rapid imaging of single atoms


n
Corresponding author.
E-mail addresses: krivanek@nion.com, The framework for the imaging of single atoms in the STEM
krivanek.ondrej@gmail.com (O.L. Krivanek). was set out by Crewe, Wall and Langmore in their landmark

0304-3991/$ - see front matter & 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ultramic.2012.04.004
O.L. Krivanek et al. / Ultramicroscopy 123 (2012) 90–98 91

Science paper [6]. The main differences between their work and order qo  b1/l o1 Å  1 the speckle is partially coherent and g can
the present are: reach 2 and more, but that for homogeneous films and
qo 42.5 Å  1, the substrate image is largely incoherent and g
(a) We now have electron probes of o1 Å in diameter, whereas decreases towards 1.
only a 5 Å probe was available to Crewe and coworkers when The visibility criterion for the single heavy atoms is therefore:
they first imaged single atoms. (A 2.5 Å probe became avail- SX uN x 4SC þ vN c ð4aÞ
able in their microscope a few years later.)
(b) Significant progress has been made in the available probe
ntsX =Au½ntðsX =A þ rC t sÞ0:5 4vðgntrC t sC Þ0:5 ð4bÞ
current: whereas 30 pA was a typical probe current available
in the Chicago microscope, we can now operate with probe The criterion will be attained if the electron dose Dx that is
currents of the order of 1 nA in probes of 1.4–2 Å [14]. delivered to each atom X is greater than:
(c) We now pay more attention to suppressing the coherent
DX ¼ nt 4ðA=sX Þ2 ½uðsX =A þ rC t sÞ0:5 þ vðgrc t sC Þ0:5 2 ð5Þ
component of annular dark field (ADF) images of materials
that arises when low scattering angles contribute to the which becomes possible when the time the probe spends over the
image [15–17], and which can produce strong maxima atom:
(‘‘coherent speckle’’) in images of amorphous materials used
t 4ðA=sX Þ2 ½uðsX =A þ rC tsÞ0:5 þ vðgrc tsC Þ0:5 2 =n: ð6Þ
as substrates for heavy atoms. Most ADF images of heavy
atoms are therefore formed using ADF detector inner angles The choice of u determines the probability of false negatives
450 mr, for which the coherent component is considerably (missing actual heavy atoms); the choice of v determines the
reduced. Interestingly, the use of such detectors was first probability of false positives (identification of heavy atoms where
suggested by Crewe and coworkers [18], though for different none exist). Selecting u ¼3 gives a probable rate of false negatives
reasons. of 1.3  10  3, i.e. one out of about 700 heavy atoms being missed.
(d) More accurate and more detailed cross-section data for the Selecting v ¼5 gives a probable rate of false positives of
atomic scattering are now available, e.g. [19]. o3  10  7, allowing large areas of the sample to be examined
without mistaking random maxima for heavy atoms.
To detect single heavy atoms reliably, we must make sure that The above expressions ignore inelastic scattering, whose con-
the weakest signal coming from each heavy atom is larger than tribution to the ADF signal is considerably weaker than the elastic
the strongest signal from the substrate away from the atoms. For one even for the carbon substrate, and much more so for the
an isolated heavy atom X lying on an amorphous carbon substrate heavy atom. They assume that there are no major probe tails, and
and illuminated by a beam of cross-sectional area A, the signal SX, that the atoms stay still while under the beam. They also assume
i.e. the number of electrons scattered while the beam is over the that the ADF detector is 100% efficient, and they do not specifi-
atom and detected by the ADF detector will be: cally account for the noise term that will arise if the carbon film
thickness varies randomly (although this can be included by
SX ¼ ntðsX =A þ rC t sC Þ: ð1Þ
increasing the speckle factor g).
The first term in the bracket is the contribution from the heavy Using Eq. (6) with u ¼3 and v ¼5, and the cross-section data
atom, the second term the contribution from the substrate. n is tabulated for various detector geometries by Treacy [19] for two
the probe current expressed as the number of electrons specific cases:
per second, t is the time the probe spends over the atom, sX is
the ADF cross-section for atom X, rC is the density of the carbon (A) single Au atoms being detected in an optimized microscope
film (in atoms/Å3), t is the film thickness, and sC is the cross- with a bright CFEG at 200 kV, on a non-optimized
section for the carbon atom. The cross sections depend on the substrate, and
atomic type, the ADF lower and upper cut-off angles b1 and b2, (B) single Au atoms being detected in an optimized microscope
the illumination semi-angle a and the primary beam energy Eo. with a bright CFEG at 60 kV, on an optimized substrate,
They can be worked out in many different ways, at least two of
which give nearly identical results [19]. we arrive at the values listed in Table 1, in which the important
The weakest signal collected from a heavy atom will be: results are shown in bold. The table predicts that in both cases,
o40 scattered electrons detected on the ADF detector are
SX 2uN x ¼ nt½sX =A þ rC t su½ntðsX =A þ rC t sÞ0:5 ð2Þ
sufficient to detect each Au atom, and that such a signal can be
where Nx is the mean statistical noise in the image intensity of the collected in about 0.5 ms in the non-optimized case A, and about
heavy atoms (plus the underlying substrate), and we model the 0.1 ms in the optimized case B. The short times needed to detect
peak deviation as being u times larger than the mean noise. single heavy atoms have important implications for rapid sequen-
The strongest signal from an area of the substrate equivalent cing of DNA by STEM imaging, as discussed below.
to the image area of individual heavy atoms will be: For both cases, we kept the same illumination aperture and
ADF detector angles. The illumination angle is typically not varied
SC þvN C ¼ nt rC t sC þvðgnt rC t sC Þ0:5 ð3Þ
greatly as the primary beam energy is changed in Cc-limited
where SC is the average signal from the substrate, NC is the mean systems such as the Nion UltraSTEM [14,20], in which the
statistical noise in the image intensity of the substrate, the peak increase in sensitivity to Cc at lower primary energy Eo mostly
deviation in the substrate image is modeled as being v times offsets the fact that larger absolute deviations of the aberration
larger than the mean noise, and g accounts for the fact that the function (expressed in units of length) can be tolerated at lower
coherent contribution to the random speckle in a dark field image energies than at high ones. The detection angles b1 and b2 should
of an amorphous carbon film will cause the speckle to be stronger ideally be adjusted to keep the scattering wavevectors approxi-
than the incoherent shot-noise variation in the number of mately constant at any primary energy. But in the present
detected electrons. The magnitude of g depends on how coherent example, in which the C substrate is thinner for the 60 kV case,
the dark field image is, which in turn depends on the ADF thus making the coherent speckle contribution less bothersome, it
collection geometry. Our estimate is that for inhomogeneous is acceptable to keep the scattering angles roughly the same for
substrate films and small inner ADF cut-off wavevectors of the the two cases. Because of the shorter electron wavelength l in the
92 O.L. Krivanek et al. / Ultramicroscopy 123 (2012) 90–98

Table 1
Two scenarios for detecting a single Au atom on a carbon substrate. Estimated parameters are shown in italic script, the important results in bold script.

Case A: 200 keV, 200 Å thick C substrate Case B: 60 keV, 20 Å thick C substrate

Primary energy Eo 200 keV Primary energy Eo 60 keV


Probe semi-angle a 30 mr Probe semi-angle a 30 mr
Zero current probe size do 0.6 Å Zero current probe size do 1.0 Å
Coherent probe current Ic 0.3 nA Coherent probe current Ic 0.3 nA
Actual probe current Ip 1.2 nA Actual probe current Ip 0.6 nA
Actual probe size dp ¼do(1þIp/Ic)0.5 1.4 Å Actual probe size dp ¼do(1þIp/Ic)0.5 1.7 Å
ADF lower cut-off (b1) 60 mr ADF lower cut-off (b1) 60 mr
ADF upper cut-off (b2) 200 mr ADF upper cut-off (b2) 200 mr
Min. scat. wavevector qo 2.39 Å  1 Min. scat. wavevector qo 1.23 Å  1
sAu 0.01284 Å2 sAu 0.11747 Å2
sC 0.00010 Å2 sC 0.00120 Å2
C film density r 0.12 atm./Å3 C film density r 0.12 atm./Å3
C support film thickness t 200 Å C support film thickness t 20 Å
Speckle factor g 1.2 Speckle factor g 2
s 5.2  10  7 s s 1.1  10  7 s
Electrons incident on Au atom during time s 3926 electrons Electrons incident on Au atom during time s 347 electrons
Average number of ADF electrons detected per Au atom 34 electrons Average number of ADF electrons detected per Au atom 21 electrons

200 keV case A, the inner cut-off (minimum) scattering wave- be clearly visible. Along with the statistical variation in the
vector qo is then higher than in case B: 2.39 Å  1 vs. 1.23 Å  1. This intensity of the atomic images, the atomic motion is also
plus the strong dependence of the cross-section on the primary responsible for the varying intensity of the same atoms in
energy (cross-sections scale as l  2 for equivalent scattering different images: atoms that move about give less visible peaks
geometries) has made both the Au and the C cross-sections about than atoms that stay still.
10x higher in the 60 keV case B than in the 200 keV case A. When individual frames of the original sequence are exam-
For case A, we have assumed a non-optimized carbon sub- ined, individual atoms are visible too, but they disappear much
strate 200 Å thick, i.e. a substrate that is readily purchased more frequently. When the successive frames are combined into a
commercially, for case B, we have assumed an optimized carbon movie and played back suitably slowed down [21], the impression
substrate 20 Å thick, i.e. a substrate of the type that Crewe and one gets is that the atoms are ‘‘swimming about’’ rather like fish
coworkers developed. The smaller inner ADF collection angle in an aquarium—randomly, slowly and steadily, with only rare
(expressed in Å  1) of case B means that the substrate speckle large sudden movements. This may seem surprising at first, as
will be more important for the assumed 60 keV set-up than for there is no obvious direct mechanism that would displace the
the 200 keV one, and we have taken account of this by setting the heavy atoms continuously by small amounts. The explanation
speckle factor g to 1.2 and 2 for A and B, respectively. probably lies in the substrate: 200 keV electrons can transfer up
Had identical support films been assumed for the two cases, to 43 eV energy to a carbon atom, and the carbon bonding energy
the minimum times t per heavy atom would have been within a is only about 15 eV. This means that C atoms are continuously
factor of 2 for the two cases considered. This shows that once the displaced and even knocked away from the sample under 200 keV
sample and the microscope have been optimized, the variations irradiation. An amorphous carbon film that appears to ‘‘swim’’, i.e.
available when adjusting the primary energy, illumination and to change continuously under an electron beam, is a familiar sight
signal collection geometries are not very strong. to most high resolution electron microscopists. The ‘‘swimming’’
Fig. 1 illustrates that performance similar to that documented is quite energetic at 200 keV. The Au atoms, riding mostly on top
in Table 1 is now indeed possible. It shows ADF imaging at of the C film, therefore simply join in the collective ‘‘swim’’, i.e.
200 keV of single Au atoms on a carbon substrate about 200 Å they move about in response to the changes in the supporting
thick, with probe formation and ADF signal collection conditions film. At 60 keV, the ‘‘swimming’’ motion of heavy atoms is largely
essentially as given by case A. The probe current was 1.2 nA, and absent, but other, less frequent types of atomic motion can be
the probe diameter was about 1.5 Å. Several 512  512 images observed [22].
were acquired unusually rapidly for STEM: the dwell time was Because of the induced atomic motion, 200 keV is not a good
just 0.167 ms per pixel, and the pixels were 0.8 Å  0.8 Å. Just primary energy for localizing Au atoms with sub-nm precision.
4 pixels therefore spanned each Au atom, and the dwell time per However, the substrate atomic motion mostly stops at 60 keV,
atom was only about 0.7 ms. The frame repetition rate was about and appears to stop completely at 40 keV.
23 frames/s, i.e. the imaging was done at close to TV rate. The above discussion indicates that rapid imaging of Au and
Fig. 1(a)–(g) shows the same 45  170 pixel sub-area in other heavy atoms on 20 Å thick C substrates is now possible very
7 consecutive time-averaged frames, each one of which is a sum much as shown in Table 1, i.e. at o1 ms/atom. Being able to image
of 4 consecutive frames. It shows the evolution of the sample over atomic motion at a fast rate is a welcome new capability for
about 1 s. Single Au atoms are clearly visible in each of the frames, atomic resolution STEM, which has typically used slower frame
but they disappear in some frames and come back in other refresh rates than shown here. The new capability may be useful
frames. For instance, the three Au atoms indicated by vertical for studying rearrangements of small crystals and sample edges in
arrows in (a) stay in the same place throughout the sequence, materials such as small catalytic particles and graphene.
whereas the two atoms indicated by horizontal arrows in The single most exciting potential application of the rapid
(g) move around. The bottom one of the two atoms is present imaging of single heavy atoms is however to be found in biology
in frames (c) and (d), disappears in frame (e), and comes back in rather than materials science: sequencing DNA. In this approach,
frame (f), in a slightly different place. The disappearance is labels containing single (or multiple) heavy atoms would be
probably due to the atom moving too much in the frame (e) to attached to a specific base type, preferably using a single uncoiled
O.L. Krivanek et al. / Ultramicroscopy 123 (2012) 90–98 93

lengths [28] than possible when using the so-called shot-gun


sequencing approach, in which the DNA is cut up into segments at
most a few tens of bases long. But even with this advantage, the
EM-based approach will need to be accurate and fast, if it is to be
competitive with the many different DNA sequencing methods
that are now available or being developed [28].
The theoretical and experimental results presented above
show that single heavy atoms should be detectable if the electron
probe spends just 0.1–1 ms over them. Assuming that ways can be
found that allow the probe to progress along each DNA strand and
not wonder pointlessly in the ‘‘wilderness’’ between the strands,
e.g. using a strategy whereby the positions of the strands are
mapped at lower magnification and the probe is then made to
advance only over the strands at the higher magnification, it
therefore appears that around 106 bases/s could potentially be
read using the ADF STEM approach. The human genome has about
3  109 base pairs, which means that the above approach may be
able to cover the whole genome in about 1 h per one type of label.
The ‘‘read’’ would then need to be repeated using heavy atom
labels attached to one of the non-complementary bases, i.e. either
adenine or thymine would need to be read out plus either guanine
or cytosine. Further, ‘‘overcoverage’’ would be needed to build up
enough statistics to allow correcting various mistakes that are
likely to occur in the chemical labeling, the laying out of the DNA,
and the imaging. Picking a 20x overcoverage as a likely possibility
indicates that the above approach may be able to sequence the
human genome in about 40 h, and perhaps as little as 10 h with
further optimization.

3. Single atom imaging and spectroscopy below the knock-on


damage threshold

For primary energies less than about 100 keV, the maximum
energy Emax that can be transferred by a primary beam electron to
an individual atom in the sample is approximately proportional to
Eo/A, where Eo is the primary energy and A is the atomic mass
number [29,30]. More specifically, a maximum energy transfer of
15 eV is possible for hydrogen atoms (1H) at a primary energy
Eo ¼6.9 keV, boron (10B) at Eo ¼65 keV, carbon (12C) at Eo ¼78 keV,
silicon (28Si) at Eo ¼167 keV, and gold (197Au) at Eo ¼774 keV.
Binding energies of the order of 15 eV are common in many
different types of crystals. The short list just given therefore
shows that hydrogen displacement cannot be avoided in a typical
transmission electron microscope with a lowest primary energy
setting of 20 keV, even in the absence of ionization damage, that
imaging carbon without knock-on damage requires Eo smaller
Fig. 1. Rapid STEM ADF imaging of single Au atoms. The images in the sequence
than 100 keV, that imaging silicon without knock-on requires Eo
were collected 0.17 s apart. Atoms marked by vertical arrows stayed approxi-
mately stationary, atoms marked by horizontal arrows moved about. Ultra- smaller than 200 keV, and that strongly bonded heavy atoms can
STEMTM200, Eo ¼200 keV, probe current¼ 1.2 nA. be imaged without knock-on damage at 300 keV and even higher.
In a previous article [22], we have called STEM imaging free of
knock-on damage ‘‘gentle STEM’’. For light atoms such as B, C, N
strand of the DNA, the DNA with the labels would be laid down on and O, gentle STEM can typically be done at 60 keV primary
a thin substrate such as a very thin amorphous carbon film, and energy [31], except at the sample’s edge or in disordered materi-
the heavy atoms constituting the labels would be imaged. This als, where the atoms are bonded less strongly.
should reveal where the bases of the labeled type are located. Atomic resolution is more difficult to attain with gentle STEM
Repeating the experiment with at least one other type of bases than when operating at higher energies. Prior to aberration
labeled (and using the fact that in the double strands, adenine is correction, STEM spatial resolution was typically not good enough
always paired with thymine, and guanine with cytosine) should to resolve atomic columns in regular crystals when the operating
then allow the complete base sequence to be ‘‘read out’’. energy was less than 100 keV. Interestingly, the microscope that
Perhaps not surprisingly, Crewe’s group originated this tech- came closest to giving atomic resolution at low primary energies
nique [23,24]. It is now being pursued again [25–27], with better was the Chicago high resolution STEM [32], operated at up to
chemical labeling methods, better specimen preparation techni- 45 keV. It had an objective lens with low spherical and chromatic
ques, and more sophisticated electron microscopes. A unique aberration coefficients (Cs ¼0.46 mm, Cc  0.8 mm,), and it was
advantage of the EM-based approach should be longer read eventually able to produce probe sizes of about 2.5 Å at 40 keV.
94 O.L. Krivanek et al. / Ultramicroscopy 123 (2012) 90–98

Present-day aberration-corrected STEMs are able to improve clearly resolved STEM images of nearest neighbor pairs (as well as
on such performance by 2–2.5x [14,33]. Their probe size at various topological defects such as 5-fold rings and a single
energies lower than about 200 keV is typically limited by chro- carbon atom and a single impurity atom hanging off the edge of
matic aberration rather than geometric aberrations. For a negli- graphene) were published in early 2010 [22].
gibly small source size (i.e., very small beam current), the smallest As an interesting aside, a subsequent paper by Jinschek et al.
attainable probe size is given by [34]: [37], which used through-focus reconstruction in a conventional
TEM (CTEM) to resolve C atoms in a single and double graphene
dchrom ¼ 0:5ðlC c dE=Eo Þ0:5 ð7Þ
sheet (but not at the graphene’s edge), made no mention of the
where l is the electron wavelength, Cc is the coefficient of STEM work. A News and Views article in Nature Materials then
chromatic aberration, dE is the energy spread and Eo is the wrongly attributed the first atomic-resolution electron micro-
primary energy. scope images of graphene to the CTEM work [38]. The episode
Since the electron wavelength l is proportional to Eo 0.5 at low shows that some parts of the electron microscopy community
primary energies, Eq. (7) shows that dchrom is proportional to unfortunately have yet to acknowledge the contribution of the
Eo 0.75. This means that the resolution grows worse quite quickly STEM to our knowledge of materials, even 40 years after Albert
as Eo is lowered. For 60 keV electrons, Cc ¼ 1.2 mm and dE¼ 0.4 eV Crewe’s pioneering results.
(as is readily attainable with a cold field emission gun (CFEG)) and Fig. 2 shows medium-angle annular dark field (MAADF)
negligibly small source size, the equation predicts dchrom ¼1.0 Å. images of graphene acquired at 60 keV, with a beam current of
With a 60 pA probe current coming from a source giving a 50 pA and a dwell time of 64 ms per each pixel, 0.06  0.06 Å large.
coherent current Ic of 150 pA, the finite size of the electron source The probe semi-angle was 30 mr, and the MAADF detector
increases the probe size to 1.2 Å [34]. For 30 keV electrons, extended from 55 mrad semi-angle up to about 200 mrad. The
Cc ¼0.8 mm and dE ¼0.4 eV, dchrom ¼1.4 Å with a negligible beam images were filtered by a rotationally symmetric two-Gaussian
current, and 1.6 Å for a 60 pA probe. (Note that probe size is a filter that nulled the probe tail at 1.42 Å from the probe center
more stringent performance criterion than the transfer of a given (i.e., at the nearest-neighbor distance), and also removed image
spatial frequency into the image: for instance, a probe of 1.2 Å shot noise with spacing smaller than about 1 Å (i.e., noise at
full-width at half-maximum (FWHM) can easily transfer spatial higher spatial frequencies than those meaningfully transferred to
frequencies higher than (1 Å)  1.) the image), as explained in Ref. [22]. Here we simply note that
In practice, there are several other influences, besides the whenever the probe tail contribution at the near neighbor
chromatic aberration and the finite source size, which can poten- distance (and beyond) is not zero, images of individual atoms
tially become resolution-limiting. These include poorly tuned contain significant contributions from the atom’s neighbors even
geometric aberrations, finite size of the atoms, statistical noise in in monolayer samples. Accurate analysis of the image intensity of
the images, and instrumental instabilities. They are discussed in different types of atoms that may be present as substitutional
[34–36]. Even with these limitations, present-day aberration- impurities is then not possible.
corrected STEMs are readily able to resolve all the individual atoms There was another image defect present: the graphene lattice
in graphene and single sheet BN at 60 keV, including nearest was imaged slightly distorted, probably because the graphene
neighbors that are separated by o1.5 Å in these two materials. sheet was not precisely perpendicular to the incoming electron
It is useful to remember that due to the hexagonal nature of beam. This is especially visible in Fig. 2(b). It does not affect the
graphene and monolayer BN sheets, the reflections that must be interpretation of the images greatly, and it was not corrected.
transferred strongly in order to clearly resolve the near-neighbor The imaging conditions were nearly identical to those we have
atomic pairs in these materials, the (11–20)-type ones, have a used in the past for imaging BN monolayers with substitutional
smaller spacing than the nearest-neighbor distance (d(11–20) ¼ impurities [31]. The same power law relating the intensity Iz of
1.23 Å in graphene vs. 1.42 Å nearest-neighbor (nn) distance, the atom’s image to its atomic type Z in tail-subtracted images
and 1.26 Å in hexagonal BN vs. 1.45 Å nn distance). In other was found once more:
words, in order to achieve atomic resolution in graphene and
Iz ¼ Icarbon ðZ=6Þ1:64 ð8Þ
BN, a probe size that is 15% smaller than the nearest neighbor
separation is very useful. A strong transfer of the (11–20) reflec- For Si, Isilicon ¼4.0  Icarbon. The medium-intensity impurity
tions was first achieved with a STEM for graphene in 2009, and atoms seen in several places in Fig. 2 were found to have

Fig. 2. MAADF images of monolayer graphene containing Si substitutional atoms and other defects. (a) Topologically perfect, but distorted graphene lattice containing a
single Si atom, (b) Si atom bonded in a 9-fold ring, and (c) Si atoms at graphene edge. UltraSTEMTM100, Eo ¼60 keV.
O.L. Krivanek et al. / Ultramicroscopy 123 (2012) 90–98 95

3.8–4.2x carbon atom intensity, demonstrating that the impuri- sample. This applies to tail-subtracted images of flat monolayers,
ties were Si. The presence of Si in the sample was further of atoms dangling off the sample’s edge, and of linear chains of
confirmed by EELS analysis of large aggregates of the impurity atoms. The main attraction of the technique is that it can provide
atoms, which produced the Si L2,3 edge signature. an unambiguous identification of single atoms at smaller electron
The preparation of monolayer graphene often involves close doses than what is required for spectroscopic identification.
contact with materials such as SiO2 or SiC, and silicon is a Electron doses of about 5  106 electrons/Å2 are sufficient for
common contaminant of graphene. In the present sample, Si the MAADF images to be sufficiently noise-free so that atoms
atoms were detected in several types of lattice sites, and Fig. 2 differing by DZ¼ 7 1 can be reliably distinguished from each
shows a selection of them. Some Si atoms were incorporated into other [31], whereas single atom identification by EELS requires
the perfect graphene lattice, as demonstrated by the Si atom in doses of the order of 108 electrons/Å2 and higher [22]. For more
Fig. 2(a), and the right atom in Fig. 2(b). The preferred length of complicated samples, e.g. double and thicker layers, monolayers
the single C–Si bond is 1.85 Å [39], i.e. considerably more than the covered by adatoms or contaminants, nanotubes and nanopods,
1.42 Å C–C distance in graphene. The result is that the silicon’s etc., some degree of overlap of the atomic images cannot be
near neighbor carbon atoms are pushed away, and this can be avoided. The simple relationship between the observed intensity
seen directly in the image. Density functional theory (DFT; [40]) and the atomic type is then no longer available. 3D reconstruction
calculations also show that the graphene surface around the Si from multiple projections may be possible, but it will not be easy
atom becomes buckled and the Si atom stick outs a small distance in the presence of possible radiation damage and other changes in
above (or below) the graphene plane [41], i.e. that it forms the the sample from exposure to exposure. The reconstruction will
apex of a very small triangular pyramid. also need to account for the fact that because of the partially
The buckling is hinted at in the image of the right Si atom in coherent nature of MAADF imaging (in the beam direction), the
Fig. 2(b). The foreshortening of the average graphene hexagon in intensity of an MAADF image maximum due to atoms that
the image indicates that the graphene sheet was tilted along an precisely overlap each other in projection is typically slightly
axis running approximately from bottom left to top right, which more than the simple sum of the individual images of the two
means that the projected faces of a pyramid sticking out of the overlapping atoms [36]. In all such cases, element-specific spec-
sheet should have different sizes. The three hexagons that include troscopic data will be invaluable.
the Si atom are indeed unequal in size: the top left hexagon is
smaller than the other two, and this indicates that the Si atom
was either above or below the plane of the graphene sheet. 4. Electron energy loss spectroscopy (EELS) of single atoms
Some Si atoms preferred to segregate to topological defects, at
which the longer Si–C bond is more easily accommodated, as Single atoms of uranium (Z¼92) were tentatively identified by
shown by the left Si atom in Fig. 2(b). Finally, the edge of the EELS in a non-corrected STEM already in 1991 [45], and single
graphene sheet was often decorated by dense one-dimensional atoms of Gd (Z ¼64) were identified much more unambiguously
arrays of Si atoms. Some of these atoms were incorporated into in 2000 [46]. Subsequent demonstrations of single atom spectro-
5-fold rings located at the graphene edge, where they substituted scopy, obtained in aberration-corrected electron microscopes,
for two carbon atoms that complete similar rings in the usual using nanopod-embedded single atoms of Er (Z ¼68) at 60 keV
‘‘armchair’’ graphene edge termination. This was the case for the [22], and of La (Z ¼57) and Ce (Z ¼58) at 30 keV [47] were even
2nd and 3rd Si atoms (counting from the top) in Fig. 2(c). Other Si clearer. N4,5 and O4,5 edges, with relatively large cross sections,
atoms simply dangled off the graphene edge, attached to just one were used in the above experiments. A single La atom present in a
C atom in the graphene (atoms 4 and 5 in Fig. 2(c)). The average CaTiO3 crystal was also identified in an aberration-corrected
distance of the Si atoms from their nearest neighbor carbon atoms STEM, using the La M4,5 edge that was highly visible due to its
was about 1.9 Å for both the doubly bonded and the singly prominent threshold peaks [48].
bonded Si atoms, confirming that the preferred C–Si bond length Improvements in the beam current available in an aberration-
is indeed much larger than the C–C one. corrected STEM [14], in probe stability relative to the sample [34],
There was also a single nitrogen atom, whose intensity was and in the efficiency of the coupling of the EELS signal into an
1.3x the average carbon atom intensity. It is shown in Fig. 2(b), energy loss spectrometer [20] have now made it possible to
located at a 7-fold ring. N along with B were intentionally extend single-atom spectroscopy studies to single light atoms,
incorporated in the graphene sample, which was prepared by using K-edges with much smaller cross-sections.
reducing doped graphene oxide [42], but they were observed less Fig. 3 shows the results of a spectrum-imaging (SI) experiment
frequently than the Si atoms. performed on B and N substitutional impurities in a graphene
Fig. 2 makes it clear that a large amount of information can be sheet, using 60 keV electrons, in which the EELS signal from
obtained by STEM ADF imaging on monolayer materials consist- individual light atoms is unmistakable. Fig. 3(a) shows an MAADF
ing of light atoms. The STEM is able to image all the atoms, clearly STEM image of a different area of the same type of sample as was
resolved, in a single exposure, and it identifies the atomic number used for Fig. 2. The area contained single N and B impurities: the
Z of the atoms simply by the image intensity of the individual image shows one more intense and one less intense atom.
atoms. The resultant information is similar to the information Fig. 3(b)–(d) show boron, nitrogen and carbon elemental maps
about a BN monolayer sheet that incorporated C and O impurities, computed from a spectrum-image (SI) of this area acquired with a
which we were able to acquire by ADF STEM earlier [31], but the Gatan UHV Enfina spectrometer. The SI acquisition parameters
detailed types of the observed structures are very different. For a were: map size 7  11 pixels, pixel size 0.66 Å, time per pixel 1 s,
thorough understanding, modeling the observed structures, e.g. 1340 energy channels, each energy channel 0.3 eV wide. The
with density-functional theory, is of course essential. STEM work spectrometer acceptance half-angle was about 50 mrad, i.e. it was
focused on understanding the materials science behind the significantly larger than the radius of the bright field disk. The
addition and incorporation of foreign atoms on and in graphene elemental maps were computed using power law background
sheets can also be found in Refs. [43,44]. extrapolation and subtraction.
Determining the atomic type from the atom’s MAADF image The boron map shows a strong peak where the less intense
intensity is straightforward if there is no major contribution to atomic image is in the MAADF image; the nitrogen map shows a
the image of each atom from the other atoms present in the strong peak where the more intense atomic image is. This
96 O.L. Krivanek et al. / Ultramicroscopy 123 (2012) 90–98

Fig. 3. MAADF image of graphene containing impurities and elemental maps. (a) MAADF image (acquired just before the spectrum-image used for the elemental maps),
(b) boron K-edge elemental map, (c) nitrogen K-edge elemental map, and (d) carbon K-edge elemental map. UltraSTEMTM100, Eo ¼ 60 keV.

confirms that the atoms were indeed boron and nitrogen, respec-
tively. The carbon map shows a decrease in the carbon signal
where the B and N atoms were located, which confirms that the
atoms were substitutional.
The carbon lattice is not resolved in the carbon-K EELS
map—not even the ‘‘holes’’ in the graphene sheet are visible. This
means that the spatial frequency of the (10–10) graphene reflec-
tion at (2.13 Å)  1 was not transferred into the carbon map. Since
the probe was about 1.2 Å in diameter, it means that the
delocalization of the carbon K-edge image was 41.7 Å (worked
out by adding the delocalization and the probe contributions to
the resolution in quadrature).
Fig. 4(a) shows spectra obtained by summing SI data from
9 pixels over the nitrogen atom (upper spectrum) and 9 pixels
over the boron atom (lower spectrum). The EELS edges due to the
single atoms are clearly identifiable. Fig. 4(b) shows an energy
loss spectrum from a single-layer BN with minor hydrocarbon
contamination obtained in a separate experiment but under
similar acquisition conditions. There is a clear difference between
the boron edge from the single B atom and from BN: the sharp pn
and sn threshold peaks that are strongly present in the BN
spectrum are largely absent for the single boron atom. The
difference between the two nitrogen K-edges is more subtle, with
the pre-edge peak appearing weaker for the single-atom spec-
trum than for BN. The carbon K edge in the BN spectrum is due to
amorphous carbon/hydrocarbon contamination. It has less pro-
nounced white lines than the graphene spectrum.
It is fascinating to note that EELS data recorded from single
Fig. 4. Comparison of EELS data from the B and N-containing graphene with a
atoms shown here and in [47] demonstrates that the Chicago spectrum from monolayer BN. (a) (top) EEL spectrum extracted from 9 pixels over
group’s 1975 prediction that a single atom of fluorine should be the boron atom in the EELS spectrum-image used for the maps shown in Fig. 3,
detectable by EELS in a well-designed STEM–EELS instrument [7] and (bottom) spectrum extracted from 9 pixels over the nitrogen atom, and
(b) spectrum from BN monolayer acquired in a separate experiment. Ultra-
has now been met, and extended to the point where the fine
STEMTM100, Eo ¼ 60 keV.
structure of EELS edges coming from single atoms can be studied
for many types of atoms, heavy and light. The fine structure
studies promise to provide a sensitive probe of the local atomic Experimental results indicating that delocalization may indeed
environment of individual atoms, and hence a versatile tool for improve at 30 keV have been published [47], but a systematic
achieving a detailed understanding of materials’ properties at the study of delocalization as a function of primary energy remains
atomic scale. to be done. Concentrating on K-edge imaging of the graphene
Powerful as the EELS approach is, it does have its drawbacks. (10–10) reflection, which should be just possible with a small
The main ones are the increased dose necessitated by the small probe and a primary energy of 30 keV, may prove fruitful for such
EELS cross-section, and the fact that the EELS interaction is experiments.
slightly delocalized, even with an optimized detection geometry
that accepts high scattering angles. The delocalization is expected
to decrease at lower primary energies as [30,34] 5. Discussion
3=4
ddeloc: ¼ 0:7  l=ðDE=Eo Þ ð9Þ
When optimizing the imaging of heavy atoms attached to DNA
E1/4
i.e., aso .This is not a strong dependence—lowering Eo from 100 strands, it may be useful to explore signals other than electrons
to 20 keV should improve the delocalization by 31%, and lowering scattered elastically with high momentum transfers, which was
it from 60 keV to 30 keV should give just a 15% improvement. the only signal used here.
O.L. Krivanek et al. / Ultramicroscopy 123 (2012) 90–98 97

Crewe and collaborators were once more the pioneers in A quadrupole gun lens is in fact built into every Nion Ultra-
combining the different signals available to maximum advantage. STEM200 [14], and it may be interesting to experiment with using
They typically enhanced the visibility of single atoms by dividing the lens for this kind of imaging.
the ADF signal by the low-angle inelastic scattering signal Iin. The The jury is still out on whether DNA sequencing by imaging
Z dependence of the ADF image signal was modeled by them as labeled strands in the STEM will prove practical or not. However,
IADF,ZpZ1.33, the dependence of the inelastic signal as Iin,ZpZ0.33, there is no doubt that the single atom imaging and spectroscopy
giving the intensity of the ratio image Ir as at reduced primary energies shown in Figs. 2–4 will be very useful
in untangling the various morphologies formed by dopants and
Ir ¼ IADF,Z =Iin,Z pZ ð10Þ adatoms in and on graphene. Using this approach plus also
simultaneous bright field (BF) STEM imaging, Zan et al. have been
It is worth noting that even though the Z exponent used by able to show several highly relevant results, such as that Cr
Crewe et al. was too low to correctly model incoherent ADF attaches to graphene more strongly than Au [43,44], a fact that is
imaging that only uses electrons scattered to high angles, which used when fabricating electrical contacts to graphene. In future,
gives an exponent 41.5, the exponent of their model of inelastic with further improvements in low keV imaging and analysis,
scattering was probably too low as well, and the relationship single atom spectroscopy should become nearly as routine as
IADF,Z/Iin,ZpZ may thus actually be approximately valid [19]. column-by-column spectrum-imaging has become for crystalline
The ratio method was able to suppress the variation in the materials in the last few years [49–51].
signal arising from variations in the thickness of the supporting It is also increasingly clear that STEM imaging and analysis
film. With a 5 Å probe, such variations can look similar to the offer many advantages over phase-contrast imaging in a conven-
signal from the single atoms. They modify the ADF and the tional TEM (CTEM). Instead of needing to combine many images
inelastic images in much the same way, and dividing the two of a through-focus series in order to compute an exit wave
signals therefore does a good job of suppressing the variations. reconstruction, as was done for instance in [37], the STEM is able
With a 1–2 Å probe, however, the ADF signal due to a single atom to show the exact atomic structure of the sample in a single
is typically much sharper than the variations in the inelastic image, recorded at a specific point in time. It is also able to
image of the support film, which are broadened by the delocaliza- identify the type of atoms present in the sample, both by the
tion of the inelastic signal, and the danger of confusing the two is recorded image intensities, and by the spectroscopic signature of
much less. Another consideration then becomes dominant, the individual atoms. It is even able to explore issues such as
namely that the inelastic signal is an important source of bonding types and charge transfer, by revealing the fine structure
statistical noise, and that using the division method typically and the exact energy of the single-atom inner shell loss spectra.
leads to an increase in the overall noise with no other clear This is usefully contrasted with a recent exploration of graphene
benefit. impurities [52] (using the most advanced CTEM instrument
The ratio method may nevertheless be useful for suppressing presently available), which was unable to identify the impurities
the background variation in single atom images taken from unambiguously.
samples of widely varying thickness. But the method is not often
used nowadays, almost certainly because of a practical problem:
most present-day electron energy loss spectrometers use CCD 6. Conclusion
detectors to record the inelastic scattering data, and do not have
the fast silicon surface barrier detector of the inelastic signal that STEM imaging and analysis have come a long way since they
Crewe et al.’s STEM had [32] (or a scintillatorþPMT detector [20] were introduced as practical electron microscopy techniques by
which is typically even faster). This means that they cannot Crewe and co-workers. Thanks to aberration correctors, bright
record the data needed for the ratio method with a sufficient cold field emission electron sources, and ultra-stable microscopes,
speed. It may therefore be useful in the future to provide fast single atoms as light as boron can now be readily imaged, and
single-channel detectors of inelastic electrons once more. almost all single atoms can give highly informative electron
Detailed considerations of exactly how to optimize the ima- energy-loss spectra, provided of course that they sit still while
ging parameters to maximize the speed of single atom imaging is being illuminated. Single atoms have recently also been identified
outside the scope of this paper. However, it may be useful to using energy-dispersive X-ray spectroscopy [53,54]. It therefore
speculate briefly about what approaches might lead to even faster seems safe to state that most of Crewe’s goals for STEM instru-
DNA sequencing than the performance discussed above. A speed- mental performance have now been attained.
ing up might for instance be possible if the probe was made To fulfill Crewe’s vision completely, however, more may need
intentionally elongated, in such a manner that it would be narrow to be done in the area of biological application. Albert had high
along the direction of the DNA strands, and more extended ambitions for the STEM in biology, and it seems that the last few
perpendicular to them. Such a probe shape can in principle be years have brought many more advances in the applications of
obtained by using quadrupoles in the condenser part of the STEM STEM in inorganic materials, and that biology may have some
to give different source demagnifications in the two perpendicu- catching up to do. If the sequencing of DNA by atomic resolution
lar directions. The convergence angle and the ADF detection imaging in the STEM actually works out and goes on to find
angles would be kept cylindrically symmetric—only the source important applications because of its long read lengths, it will be
projected into the sample plane would be made ‘‘elliptical’’. especially satisfying to those of us interested in Albert’s legacy.
Additionally, the probe itself could then be made deliberately In the area of instrumentation, incremental advances in areas
astigmatic, so that it would be precisely focused in the direction such as corrector and electron source design will no doubt
along the DNA strand (the ‘‘narrow’’ direction of the probe), but continue to improve the resolution available, particularly at low
slightly defocused in the perpendicular, ‘‘long’’ direction. The primary energies, where better resolution, say of the order of 1 Å
Ronchigram pattern, possibly energy-filtered to show only inelas- at 10–20 keV, would be very useful. The largest advances in this
tically scattered electrons, would then contain real space infor- direction are likely to come from chromatic aberration correctors,
mation about how the probe was centered on the DNA strand, and an area that Albert probably thought too far in the distance to
this information could most likely be used to steer the probe so worry about personally. But they are not likely to produce as
that it remained centered on the strand as it was scanned along it. fundamental improvement of microscope performance as Cs and
98 O.L. Krivanek et al. / Ultramicroscopy 123 (2012) 90–98

C5 correctors did, and hence will not propel us far into the ‘‘post- [21] /www.nion.comS.
Crewe’’ electron microscopy era. [22] O.L. Krivanek, N. Dellby, M.F. Murfitt, M.F. Chisholm, T.J. Pennycook,
K. Suenaga, V. Nicolosi, Ultramicroscopy 110 (2010) 935.
One inviting path that was not explored by Albert and his [23] J. Wall, J.P. Langmore, in: Proceedings of the 28th EMSA Meeting, Houston,
coworkers is to improve the energy resolution of electron energy Texas, 1970, p. 252.
loss spectroscopy. This path is especially promising because it is [24] J.P. Langmore, A.V. Crewe, in: Proceedings of the 32nd EMSA Meeting, St.
very open-ended. Present day monochromators and spectro- Louis, Miss., 1974, p. 376.
[25] D.C. Bell, W. Thomas, K. Murtagh, W.R. Glover, Microscopy and Microanalysis
meters allow energy resolution of about 0.1 eV (on sample spectra 16 (Suppl. S2) (2010) 1768.
rather than just on the zero loss peak), typically at the cost of a [26] A.L. Bleloch, C.S. Own, M. Hamalainen, J. Hershleb, K. Kemmish, R. Koene,
broadened electron probe. Much new information about the H. Stark, J. Stark, M. Andregg, W. Andregg, Microscopy and Microanalysis 17
(Suppl. S2) (2011) 1274.
sample would become available if the EELS data could be acquired
[27] D.C. Bell, W.K. Thomas, K. Murtagh, W. Glover, Microscopy and Microanalysis
at 30, 10, 3 and ultimately even 1 meV energy resolution, 17 (Suppl. S2) (2011) 1276.
preferably with an atom-sized electron probe. We are now [28] E.E. Schadt, S. Turner, A. Kasarskis, Human Molecular Genetics 19 (Review
pursuing this direction [55], and we look forward to being able Issue 2) (2010) R227.
[29] A. Zobelli, A. Gloter, C.P. Ewels, G. Seifert, C. Colliex, Physical Review B 75
to report on what the post-Crewe landscape looks like along (2007) 245402.
this path. [30] R.F. Egerton, EELS in the Electron Microscope, 3rd edition, Springer, New York
and Heidelberg, 2011.
[31] O.L. Krivanek, M.F. Chisholm, V. Nicolosi, T.J. Pennycook, G.J. Corbin,
Acknowledgments N. Dellby, M.F. Murfitt, C.S. Own, Z.S. Szilagyi, M.P. Oxley, S.T. Pantelides,
S.J. Pennycook, Nature 464 (2010) 571.
[32] J. Wall, J. Langmore, M. Isaacson, A.V. Crewe, Proceedings of the National
We are grateful to our colleagues at Nion for their part in the Academy of Sciences of the United States of America 71 (1974) 1–5.
design and construction of the microscopes used here, to Profes- [33] T. Sasaki, H. Sawada, F. Hosokawa, Y. Kohno, T. Tomita, T. Kaneyama,
Y. Kondo, K. Kimoto, Y. Sato, K. Suenaga, Journal of Electron Microscopy 59
sor H. Dai of Stanford U. for kindly supplying the graphene
(Supplement) (2010) S7.
sample, and to Prof. S.J. Pennycook for provision of facilities. [34] O.L. Krivanek, M.F. Chisholm, N. Dellby, M.F. Murfitt, in: S.J. Pennycook,
Research at Oak Ridge National Laboratory (MFC) was sponsored P.D. Nellist (Eds.), Scanning Transmission Electron Microscopy, Springer,
by the Division of Materials Sciences and Engineering, Basic 2011, p. 615.
[35] Earl J. Kirkland, Ultramicroscopy 111 (2011) 1523.
Energy Sciences, of the US Department of Energy. [36] O.L. Krivanek, W. Zhou, M.F. Chisholm, N. Dellby, T.C. Lovejoy, Q.M. Ramasse,
J.-C. Idrobo, in: D.C. Bell, N. Erdman (Eds.), Low Voltage Transmission Electron
References Microscopy, RMS Wiley, in print.
[37] J.R. Jinschek, E. Yucelen, H.A. Calderon, B. Freitag, Carbon 49 (2011) 556.
[38] K. Urban, Nature Materials 10 (2011) 165–166.
[1] A.V. Crewe, J. Wall, L.M. Welter, Journal of Applied Physics 39 (1968) 5861. [39] /http://www.wiredchemist.com/chemistry/data/bond_energies_lengths.
[2] A.V. Crewe, M. Isaacson, D.E. Johnson, Review of Scientific Instruments 40 htmlS.
(1969) 241. [40] G. Kresse, J. Furthmüller, Physical Review B54 (1996) 11169.
[3] A.V. Crewe, J. Wall, Journal of Molecular Biology 48 (1970) 375. [41] J.-C. Idrobo, Private Communication.
[4] A.V. Crewe, Microscopy and Microanalysis 10 (2004) 414. [42] M.F. Chisholm, et al., in preparation.
[5] A.V. Crewe, Cold field emission and the scanning transmission electron [43] R. Zan, U. Bangert, Q. Ramasse, K.S. Novoselov, Nano Letters 11 (2011) 1087.
microscope, in: P.W. Hawkes (Ed.), Advances in Imaging & Electron Physics, [44] R. Zan, U. Bangert, Q. Ramasse, K.S. Novoselov, Small 7 (2011) 2868.
vol. 159, Elsevier, Amsterdam, 2009, p. 1. [45] O.L. Krivanek, C. Mory, M. Tence, C. Colliex, Microscopy Microanalysis
[6] A.V. Crewe, J. Wall, J. Langmore, Science 168 (1970) 1338. Microstructures 2 (1991) 257.
[7] M. Isaacson, D.E. Johnson, Ultramicroscopy 1 (1975) 33. [46] K. Suenaga, M. Tence, C. Mory, C. Colliex, H. Kato, T. Okazaki, H. Shinohara,
[8] A.V. Crewe, in: Proceedings of the 47th Nobel Symposium, Chemica Scripta K. Hirahara, S. Bandow, S. Iijima, Science 290 (2000) 2280.
14 (1979) 17. [47] K. Suenaga, Y. Iizumi, T. Okazaki, European Physical Journal Applied Physics
[9] M. Isaacson, D. Kopf, M. Utlaut, N.W. Parker, A.V. Crewe, Proceedings of the
54 (2011) 33508.
National Academy of Sciences of the United States of America 74 (1977)
[48] M. Varela, S.D. Findlay, A.R. Lupini, H.M. Christen, A.Y. Borisevich, N. Dellby,
1802.
O.L. Krivanek, P.D. Nellist, M.P. Oxley, L.J. Allen, S.J. Pennycook, Physical
[10] G. Binnig, H. Rohrer, Reviews of Modern Physics 59 (1987) 615.
Review Letters 92 (2004) 095502.
[11] D.M. Eigler, E.K. Schweizer, Nature 344 (1990) 524.
[49] D.A. Muller, L. Fitting Kourkoutis, M.F. Murfitt, J.H. Song, H.Y. Hwang, J. Silcox,
[12] B. Ward, J.A. Notte, N.P. Economou, Photonics Spectra (2007). Available
N. Dellby, O.L. Krivanek, Science 319 (2008) 1073.
from:/http://www.photonics.com/Article.aspx?AID=30461S.
[50] G.A. Botton, S. Lazar, C. Dwyer, Ultramicroscopy 110 (2010) 926.
[13] S.J. Pennycook, in: S.J. Pennycook, P.D. Nellist (Eds.), Scanning Transmission
[51] M. Varela, J. Gasquez, T.J. Pennycook, C. Magen, M.P. Oxley, S.J. Pennycook, in:
Electron Microscopy, Springer, 2011, p. 1.
[14] N. Dellby, N.J. Bacon, P. Hrncirik, M.F. Murfitt, G.S. Skone, Z.S. Szilagyi, S.J. Pennycook, P.D. Nellist (Eds.), Scanning Transmission Electron Micro-
O.L. Krivanek, European Physical Journal: Applied Physics 54 (2011) 33505. scopy, Springer, 2011, p. 429.
[15] M.M.J. Treacy, A. Howie, C.J. Wilson, Philosophical Magazine A 38 (1978) 569. [52] R. Erni, M.D. Rosell, M.-T. Nguyen, S. Blankenburg, D. Passerone, P. Hartel,
[16] A. Howie, Journal of Microscopy 117 (1979) 11. N. Alem, K. Erickson, W. Gannett, A. Zettl, Physical Review B82 (2010)
[17] P. Hartel, H. Rose, C. Dignes, Ultramicroscopy 63 (1996) 93. 165433.
[18] A.V. Crewe, J.P. Langmore, M.S. Isaacson, in: B.M. Siegel, D.R. Beaman (Eds.), [53] T.C. Lovejoy, Q. Ramasse, M. Falke, A. Kaeppel, R. Terborg, R. Zan, N. Dellby,
Physical Aspects of Electron Microscopy and Microbeam Analysis, Wiley, O.L. Krivanek, Applied Physics Letters 100 (2012) 154101.
New York, 1975, p. 47. [54] K. Suenaga, K. Okazaki, E. Okunishi, S. Matsumura, Nature Photonics, in print.
[19] M.M.J. Treacy, Microscopy and Microanalysis 17 (2011) 847. [55] O.L. Krivanek, J.P. Ursin, N.J. Bacon, G.J. Corbin, N. Dellby, P. Hrncirik,
[20] O.L. Krivanek, G.J. Corbin, N. Dellby, B.F. Elston, R.J. Keyse, M.F. Murfitt, M.F. Murfitt, C.S. Own, Z.S. Szilagyi, Philosophical Transactions of the Royal
C.S. Own, Z.S. Szilagyi, J.W. Woodruff, Ultramicroscopy 108 (2008) 179. Society of London A 367 (2009) 3683.

You might also like