15

Scanning Tunneling Microscopy in

Surface Science
Peter Sutter

1 Introduction

After its invention in 1982 by Gerd Binnig and Heinrich Rohrer (Binnig
et al., 1982, 1983), who were awarded the 1986 Nobel Price for this
discovery, scanning tunneling microscopy (STM) has rapidly become
a standard technique for high-resolution imaging of conducting
surfaces. As such, it has revolutionized the way surface science is
conducted. Having relied almost entirely on diffraction methods for
determining surface structures, and electron spectroscopy for measur-
ing surface chemistry and electronic structure, surface scientists imme-
diately embraced the powerful imaging and spectroscopy capabilities
of STM. Earlier techniques for high-resolution surface microscopy
include field ion microscopy (Muller and Tsong, 1969), low-energy
electron microscopy (Bauer, 1998), ultrahigh-vacuum scanning elec-
tron microscopy (Venables, 2000), and reflection electron microscopy
(Yagi, 1982). Field ion microscopy, the first surface imaging technique
to routinely provide atomic resolution on metals over small sample
areas, has recently seen a revival in the form of modern atom probes
that provide tomographic images of three-dimensional sample volumes
(Cerezo et al., 2001), nicely complementing the surface imaging capa-
bilities of STM. The other techniques, none of which achieves atomic
resolution, have been developed further for specific applications. As an
example, low-energy electron microscopy has become a powerful tech-
nique for real-time microscopy of fast surface processes, such as epi-
taxial growth or surface reactions (Bauer, 1998). STM itself has recently
been at the center of yet another scientific revolution, enabling system-
atic studies on individual structures composed of a small number of
atoms or molecules, and on the order of a few nanometers in size.
This chapter attempts to provide an introduction of the basic con-
cepts of STM, together with illustrations of applications of the tech-
nique. Over two decades after its invention, the applications of STM
for imaging, spectroscopy, and manipulation at the atomic level have
clearly become too numerous to allow for a comprehensive review. In
view of the wide range of applications and overall maturity of the

969
970 P. Sutter

technique, a historic overview may also be of limited value. Hence, I

have tried to identify representative examples in the current literature,
which are discussed briefly to illustrate the different uses of STM. The
connection to historic developments can in most cases be made easily
via literature references. For a more in-depth discussion of basic con-
cepts and application examples, several dedicated monographs on
STM and other scanning probe techniques (Chen, 1993; Wiesendanger,
1994; Güntherodt and Wiesendanger, 1992; Bonnell, 2001) represent an
invaluable resource.
This chapter is organized as follows. In Section 2 basic principles of
STM imaging are introduced, and the imaging methodology as well
as practical and instrumentation requirements are discussed. The
approach taken in surface imaging by STM is illustrated by the example
of silicon surfaces. Section 3 highlights an application of STM that has
gained ever-increasing importance: atomic scale spectroscopy. It pro-
vides a survey of the spectroscopy capabilities of STM, and introduces
a variety of techniques for local spectroscopy and spectroscopic
imaging. In addition, pathways toward obtaining chemical and element
specificity at the atomic scale—traditionally a weakness of STM—are
discussed. The extension of the operating conditions of STM to high
and low temperatures has opened up new avenues of investigation.
Variable temperature STM of dynamic surface processes as well as
atom and molecule manipulation at cryogenic temperatures are the
topics of Section 4. Section 5 discusses STM imaging and spectros-
copy on subsurface structures, using ballistic electrons to probe
buried interfaces or cross-sectional STM on cleavage faces of III–V
semiconductors to image embedded nanostructures. The chapter con-
cludes with a brief discussion of STM image simulation techniques in
Section 6.

2 Basic Principles of STM Imaging

2.1 Elastic Vacuum Tunneling and

Scanning Tunneling Microscopy
STM is based on the vacuum tunneling of electrons between two
solids, one of which is a sharp tip and the other a sample. To remove
an electron from a solid and bring it into vacuum with zero kinetic
energy requires energy equal to the work function φ = EVac − EF, i.e., the
difference between the vacuum level, Evac, and the Fermi energy of the
material, EF. To move an electron from one solid to another, it has to
cross the same vacuum barrier. If the two solids are separated by a
microscopic distance of the order of the decay length of electronic
states into the vacuum (typically about 1–5 nm), electrons can cross the
barrier by quantum mechanical tunneling (Figure 15–1).
If, say, two metals are brought to within tunneling distance, a rapid
charge transfer takes place between them until an equilibrium state is
established in which their Fermi levels are aligned. Once in equilib-
rium, no net charge is transferred on average. Consider now a slightly
modified situation, in which metal A is biased relative to B by applying
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 971

Figure 15–1. Band diagrams (energy vs. distance along the tunneling direc-
tion) for vacuum tunneling between two metal electrodes, A and B.

a voltage, VA. The bias offsets the Fermi energies on either side of the
vacuum barrier by eVA. Under steady-state conditions, this potential
difference establishes a net tunneling current, I, between A and B.
In a simple planar tunneling model using the Wentzel–Kramers–
Brillouin (WKB) approximation, the tunneling current is given by an
integral over the energy range between the Fermi energies on either
side of the vacuum gap,
eVA
I= ∫ ρA ( E ) ρΒ ( E − eVA )T ( E, eVA ) dE (1)
0

where ρA and ρB denote the local density of states at energy E at the

surface of A and B, respectively, and T(E, eVA) is the tunneling transmis-
sion probability for electrons with energy E and applied bias VA:

 2 z 2m  φ A + φB eVA
T ( E, eVA ) = exp  − + −E (2)

 2 2
Here, φA,B are the work functions of the two solids, z is their separation,
and m is the mass of the electron. Evaluation of T for positive (VA > 0)
and negative bias (VA < 0) shows that the transmission probability is
highest for electrons at the Fermi energy of the material that is nega-
tively biased, and falls off exponentially for lower energies down to a
lower cutoff at the Fermi energy of the positively biased electrode. This
general observation has important consequences in tunneling spec-
troscopy and spectroscopic imaging, as discussed in Section 3.
The tunneling current depends strongly on the separation, z, between
the two solids. To illustrate this fact, we simplify Eq. (1) by assuming
constant densities of states, independent of energy. In the limit of low
bias voltage, VA/φA,B << 1, the tunneling current is then given by

(
I = ρA ρBVA exp −2 m ( φ A + φB ) /
2 z ) (3)

i.e., it depends exponentially on the separation. It is from this strong z

dependence of the tunneling current that STM derives its exquisite
height resolution, typically of the order of 0.1 pm, or below 1/100 mono-
layer for most solids.
972 P. Sutter

Figure 15–2. Schematic tip

trajectory for STM imaging in
constant current and constant
height mode.

In actual STM imaging, an atomically sharp tip is used to obtain

laterally confined tunneling at a well-defined position of a sample. A
piezoelectric positioning element controls the in-plane (x, y) position
and height (z) of the tip with picometer resolution. While a constant
bias voltage is applied between sample and tip, the tip is scanned in a
line-by-line fashion to continuously change the tunneling position and
build up an image of a chosen area of interest on the sample. The strong
dependence of the tunneling current on the tip–sample separation can
be used for two basic modes of STM imaging: constant-current and
constant-height imaging (Figure 15–2).
In constant-current imaging, the tunneling current is measured at
each pixel of the scan and is compared with a chosen current set point.
Deviations between the measured tunneling current and the set point
are corrected by applying an appropriate voltage to the z-piezo, thus
adjusting the tip–sample separation. This feedback mechanism main-
tains a constant tunneling current during the scan, while the trajectory
z(x, y) followed by the tip is used to generate a map of the sample
surface or, more accurately, a map of a particular charge density contour
above the surface.
In constant-height mode, the tip height is not modified during the
scan. The tunneling current is again measured at each image pixel, but
is now used directly as a representation of the sample surface via a
current map, I(x, y). The current maps represent a cut through charge
density contours in the plane in which the tip is scanned above the
sample. Obviously, since the tip–sample separation is uncontrolled
during the scan, constant-height imaging is limited to samples with
low corrugation and/or small scan sizes. A more practical implementa-
tion of constant-height STM uses a slow feedback system to adjust the
tip–sample separation on time scales that are long compared to the
residence time at each pixel, thus compensating for sample surface
topography or sample tilt. As the current comparison and tip z cor-
rection are eliminated and only the tunneling current is measured,
data acquisition can be faster than in constant-current imaging. The
constant-height mode is thus preferred for fast STM image acquisition.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 973

2.2 Inelastic Vacuum Tunneling

The above considerations assumed elastic tunneling between two
materials, e.g., a sharp tip and the sample. In this case, the energy of a
tunneling electron is conserved in the transit through the vacuum
barrier. Only inside the opposite electrode the electron thermalizes
to the Fermi energy by phonon emission. Tunneling electrons can,
however, undergo inelastic scattering during the tunneling process.
Vibrational modes of molecules adsorbed on the sample surface, for
example, can be excited by inelastic scattering of a small fraction of the
tunneling electrons (Figure 15–3).
The onset of inelastic tunneling at characteristic energies can then
be detected and used as a fingerprint to identify individual molecular
bonds. Such STM-based single molecule vibrational spectroscopy will
be discussed in Section 3. In addition, controlled amounts of energy
can be deposited locally into individual adsorbates, important for
the manipulation of adsorbed atoms or molecules, as discussed in
Section 4.

2.3 Practical Requirements: Tips, Samples, and

Operating Environment
Obtaining a highly localized tunneling contact requires very sharp
probe tips, ideally terminated by a single atom or a small cluster.
D-band metals (e.g., W) or alloys (PtIr) are generally thought to be
superior for obtaining high spatial resolution due to the strongly direc-
tional nature of the d-wave function. However, atomic resolution has
also been obtained with Au tips, i.e., tip materials with primarily s-
electrons at the Fermi level.
Electrochemically etched tungsten wires are often used for high-
resolution STM in vacuum, while less reactive PtIr or Au tips are
preferred for imaging under ambient conditions. Sharp W tips are
produced by electrochemical etching in ∼2 M KOH or NaOH solution,
using a stainless steel or Pt foil as a counter electrode. dc or ac etching
at typical bias voltages of 5–10 V can be used. Preferential etching at
the meniscus of the solution leads to a tapering of the immersed wire,
causing one end to eventually break off. To prevent further etching that
would blunt the tip apex, an electronic circuit detects the break-off

Figure 15–3. Band diagrams for elastic

and inelastic electron tunneling.
974 P. Sutter

current and switches the bias to zero within a few microseconds.

Etched tips are rinsed in deionized water to remove traces of the etch
solution, and loaded into ultrahigh vacuum (UHV). Contaminants and
surface oxide are desorbed from the tip by a heat treatment in UHV
prior to mounting the tip into the STM. Electron beam bombardment
of the tip apex provides an efficient local cleaning of the thin part of
the tip, and the applied electric field in this procedure may lead to an
additional tip sharpening (Kuk and Silverman, 1986). Alternative tip
cleaning and modification techniques while in tunneling contact
involve voltage pulses of the order of 4–10 V while scanning the sample,
or tip forming in the field emission regime (i.e., at tip–sample large
separation) with applied bias voltages between 10 and 100 V and cur-
rents of several nA. Ideally, the atomic configuration of the tip would
be characterized by field ion microscopy (FIM) prior to its use in the
STM. Attempts have been made to combine FIM with STM. However,
in practice such efforts are of limited benefit since tip changes occur
frequently during scanning, rendering a complete characterization of
the tip impractical.
To allow the controlled biasing of tip and sample, and to prevent
charging, conventional direct current STM is limited to metal or semi-
conductor samples. Alternating current STM (Kochanski, 1989) has
been demonstrated on insulators and organic layers, but is not widely
used. Clean and ordered crystalline surfaces with low surface rough-
ness are preferred for most studies, in particular for atomic-resolution
imaging. Reactive semiconductor or metal surfaces are best prepared
and imaged in UHV, but metal surfaces are also prepared and imaged
routinely in electrolytes (Gewirth and Niece, 1997). In vacuum, sample
preparation often involves a thermal cleaning step to remove contami-
nants. Flash cleaning to 1200°C for a brief period of time (few seconds)
while keeping the vacuum in the low 10−9 torr range removes the native
oxide and provides atomically clean Si samples with a low density of
SiC contaminants. Metal surfaces (e.g., Cu, Al, Pt, Au), and some semi-
conductors (e.g., Ge) are prepared in a two-step procedure involving
repeated cycles of Ar+ ion sputtering and annealing. In refractory
metals such as Ru, C contamination is removed by many (several
hundred) cycles of oxygen adsorption near room temperature and
flashing to 1500°C. In situ cleavage can be a very effective method for
preparing sample surfaces that cannot be sputtered or heated to high
temperatures. Samples best prepared by cleavage include III–V com-
pound semiconductors (GaAs, InAs), as well as high-TC superconduc-
tors (Yba2Cu3O7−x, BiSr3Cu2O8+x). In situ cleavage is also the preparation
method of choice for cross-sectional STM (see Section 5), commonly
performed on (110) cleavage planes of III–V semiconductors. Bulk insu-
lating metal oxides, most prominent among them TiO2, can be imaged
successfully by STM following Ar+ ion sputtering and annealing, which
generates oxygen vacancies and renders the surface region sufficiently
conductive for stable STM imaging. Au and highly oriented pyrolitic
graphite (HOPG) are among the few materials on which atomic resolu-
tion is obtained under ambient conditions. Well-ordered, (111) oriented
surfaces of Au single crystals and thin films on mica can be prepared
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 975

by flame annealing in air (Robinson et al., 1992). Preparation of layered

materials, notably HOPG, is particularly simple. It involves the removal
of bundles of graphene layers by adhesive tape from the crystal. Since
the exposed graphene sheet has no dangling bonds and is very inert,
such samples can be imaged with atomic resolution in air for extended
time periods.
For many materials, in particular metals and alloys, electrochemical
STM is a powerful alternative to imaging in UHV. In this environment,
almost the entire portion of the tip that is immersed in the solution
must be coated by an insulating layer (e.g., wax or glass) to minimize
the faradaic (i.e., ionic) current, which is typically much larger than the
tunneling current and would otherwise dominate the measured signal.
Only a small fraction near the apex of the tip remains uncoated to
allow for tunneling between the tip and sample.

2.4 STM Instrumentation

In tunneling contact the probe tip is only a distance of the order of
1 nm away from the sample. As the tunneling current is exponentially
sensitive on the tip–sample separation, any mechanical vibrations of
the tip relative to the sample have to be minimized. To this end, a two-
fold strategy is commonly pursued. The microscope head itself is built
as compact and stiff as possible by closely integrating sample and tip
on a rigid platform. This results in a high resonance frequency for
oscillations of the tip relative to the sample. In addition, the entire
microscope is isolated mechanically from outside vibration sources,
e.g., by suspending it on soft extension springs or long bungy cords.
The overall goal is to achieve a maximal mismatch between the mech-
anical modes of a soft suspension system and the high resonance fre-
quency of the stiff microscope head. In addition, efficient vibration
damping is required, which can be achieved by magnetically induced
eddy currents. The springs used in many suspension systems have
their own mechanical modes that can be damped by polymer strips
woven into their coils. Direct coupling of acoustic noise into the micro-
scope, finally, is a problem that is best solved by operating the STM in
vacuum (Figure 15–4).
A few additional components are part of any practical tunneling
microscope. A coarse approach mechanism has to be in place, which
moves the tip from an initial distance of several millimeters into
tunneling range without making physical contact with the sample.
Although this has been achieved by mechanical approach mechanisms
(Smith and Binnig, 1986), piezoelectric driven stick-slip motors are now
preferred since they allow the entire coarse approach to be performed
automatically under computer control. For the positioning and scan-
ning of the tip during STM imaging, a segmented piezoelectric tube
scanner is used (Binnig and Smith, 1986). By applying high voltages to
the individual segments, the scanner is independently actuated in the
(x, y) plane parallel to the sample surface and along the perpendicular
z axis with little crosstalk. The tunneling current in the pA to nA
range is converted into a voltage by a sensitive, low-noise amplifier
976 P. Sutter

Figure 15–4. Schematic illustration of the various components of an STM

system, including piezoelectric tube scanner for tip positioning, and con-
trol electronics (current amplifier, feedback, scan generator, high-voltage
amplifier).

with adjustable transconductance of 107–1010 V/A, which provides the

input signal for the tunneling current feedback electronics. Modern
STM controllers typically use digital feedback and scan modules imple-
mented by specific software programs running on fast digital signal
processor (DSP) chips. The measured analog tunneling current signal
is digitized using an analog-to-digital (A/D) converter, and is com-
pared with the current set point by the feedback program on the
DSP. Necessary corrections in the tip height z are computed, and are
converted into analog voltages by digital-to-analog (D/A) converters.
These voltages may be amplified further by a high-voltage amplifier
stage to levels suitable to drive the tube scanner. The scan signal, i.e.,
independent voltage ramps driving the piezo scanner along the in-
plane x and y directions, is also generated digitally on a DSP, followed
by conversion and amplification steps analogous to those of the z
signal. A program running on a host computer and communicating
with the DSP, finally, acquires, displays, and stores the measured tip
trajectory z(x, y) (in constant-current imaging) or position-dependent
tunneling current I(x, y) (in constant-height mode).

2.5 STM Imaging: Application Examples

High-resolution imaging of surface structure is probably the most
widespread application of STM. As a direct imaging technique, STM
has added significantly to the diffraction techniques used traditionally
to determine the structure of clean and adsorbate covered surfaces.
Examples of typical applications include the determination of the
structure of reconstructed surfaces of clean semiconductors, for
instance vicinal Si surfaces (Erwin et al., 1996) and oxides, such as TiO2
(Diebold, 2003), and of adsorbate-induced reconstructions on metals
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 977

(e.g., O/Ru(0001); Meinel et al., 1997); the mapping of the evolution

of surface morphology during epitaxial growth (e.g., metal epitaxy;
Chambliss et al, 1995), etching (Boland and Weaver, 1998), and ener-
getic particle bombardment (e.g., electrons on Si; Nakayama and
Weaver, 1999); and the imaging and structural identification of non-
periodic structures such as surface defects, steps, as well as surface-
supported solid (e.g., silicide nanowires on Si; Chen et al., 2000)
and molecular nanostructures [e.g., molecular rotors on Cu(100);
Gimzewski et al., 1998]. Although electrically conducting samples are
a prerequisite for stable tunneling, high-resolution imaging is feasible
on ultrathin insulating films (for a recent review, see Schintke and
Schneider, 2004) and self-assembled organic monolayers [e.g., alkane-
thiol monolayers on Au(111); Cygan et al., 1998) supported by metal
substrates.
In view of the large number of materials systems to which STM has
been applied, a comprehensive survey would be beyond the scope of
this chapter. Instead, we discuss in some detail recent contributions of
STM imaging to one specific material: silicon.

2.5.1 STM Imaging—Silicon Surfaces

Due to the important role of Si in electronics, extensive STM studies
on the structure of clean Si surfaces with different orientation have
contributed to making Si one of the best-characterized materials system.
Important early milestones in the characterization of clean Si surfaces
include the determination of surface bonding and reconstructions on
Si(111) (Binnig et al., 1983) and on Si(001) (Tromp et al., 1985) used as
a substrate in microelectronics. STM imaging served to establish the
thermodynamics of terraces and steps (Swartzentruber et al., 1990;
Men et al., 1988) on these surfaces, to study electromigration and step-
bunching (Yang et al., 1996), and to survey the stable facets involved
in the equilibrium shape of Si crystals (Gai et al., 1998).
Figure 15–5 shows an example of constant current STM images on
Si(111), a complex reconstructed surface whose geometric (Binnig and

Figure 15–5. (a) Ball-and-stick model of the Si(111)–(7 × 7) surface. Top: top view; bottom: side view.
a, adatom; r, rest atom; c, corner hole. (b) Filled-state constant-current STM on Si(111)–(7 × 7); V = −1.0 V;
I = 0.4 nA. (c) Empty-state constant-current STM; V = +1.2 V; I = 0.4 nA.
978 P. Sutter

Rohrer, 1983) and electronic (Hamers et al., 1986) structure was solved
in pioneering STM experiments, and which continues to play an impor-
tant role in STM technique development. Figure 15–5a shows a struc-
ture model of the (7 × 7) reconstruction (Takayanagi et al., 1985). Figure
15–5b, obtained at negative sample bias (V = −1.0 V), i.e., with electrons
tunneling from occupied sample states to unoccupied tip states, shows
the corrugation associated with filled states of the sample. Conversely,
Figure 15–5c, obtained at positive sample bias (V = +1.2 V), maps the
empty states of the sample. Both images show atomic resolution, clearly
resolving the twelve adatoms (“a”) per unit cell. In addition character-
istic deep “corner holes” (“c”) bounding the diagonals of the 4.6 nm ×
2.9 nm rhombohedral unit cell are imaged.
While all adatoms are mapped uniformly in the empty state image,
the filled state scan shows one-half of the unit cell somewhat higher
than the other, an effect on the charge density due to the different
stacking sequence of atomic layers in the two halves of the unit cell. A
comparison of the images obtained at opposite bias polarity suggests
that in addition to surface topography, the electronic structure of the
sample surface adds substantially to the contrast observed in STM
imaging. This is implicit also in Eq. (1) via the dependence of the tun-
neling current on the local densities of states of both tip and sample at
the tunneling contact. Note also that the rest atoms (“r”), a second
near-surface species with dangling bonds protruding into the vacuum,
are imaged neither at positive nor at negative sample bias. A detailed
discussion of bias-dependent imaging and other tunneling spectro-
scopy methods used to assess the local electronic structure of a sample
surface is given in Section 3.
Apart from the structure of clean Si surfaces, Si-based surface chem-
istry has been studied widely by STM. Initial studies provided a knowl-
edge base for technological processes, such as reactive ion etching,
doping, and chemical vapor deposition (CVD). STM was used to probe
interactions of Si with halogens (Boland, 1993) and with small mole-
cules involved in surface passivation (H2, H; Laracuente and Whitman,
2001), doping (PH3; Wang et al., 1994a), and CVD growth (Si2H6; Wang
et al., 1994b). More recent research directions include the integration
of molecular electronic elements with Si. Under this perspective, the
covalent bonding of a wide variety of organic molecules on Si has been
studied (for a review, see Wolkow, 1999), including small molecules
such as ethylene (Mayne et al., 1993) and acetylene (Li et al., 1997),
simple alkenes (e.g., propylene; Lopinski et al., 1998) and polyenes (e.g.,
1,3-cyclohexadiene; Hovis and Hamers, 1997), pentacene (Kasaya et al,
1998), and benzene (Borovski et al., 1998). On clean Si(001), adsorption
is from the gas phase in UHV. H-passivated Si can also be modified by
ex situ wet-chemical techniques, and reintroduced into UHV for STM
imaging.
Traditionally, an important aspect of the surface science of semicon-
ductors has been epitaxial growth. STM observations at initial growth
stages, i.e., at coverages of fractions of a monolayer (ML) up to several
ML, can provide direct insight into fundamental processes such as
adatom diffusion, incorporation into steps, and nucleation of mono-
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 979

layer islands (for a review, see Zhang and Lagally, 1998). Observations
of growth and equilibrium structures allow identifying kinetic and
thermodynamic factors affecting the growth process, as well as the role
of defects, of elastic strain, etc.
STM studies of Si epitaxy are commonly performed in either of two
modes. Conceptually preferred is the dynamic observation, via time-
lapse STM “movies,” of the growth surface with the sample held in the
microscope at high temperature and in the presence of a deposition
flux. This difficult technique will be discussed in more detail in Section
4. As an alternative, STM imaging can be performed at room tempera-
ture on samples that have been quenched at key stages in the growth
process. If measures are taken to verify that the quenched-in morphol-
ogy is indeed representative of that at high temperature during growth,
this “quench-and-look” technique offers the advantage of higher spatial
resolution and larger scan sizes over high-temperature STM.
Early STM experiments focused on the initial stages of Si homoepi-
taxy and Ge/Si heteroepitaxy, mostly on the technologically important
(111) or (001) surfaces. Key results include the identification of regimes
of step-flow growth and nucleation, determination of the activation
energy and atomistic pathway of surface diffusion (Mo et al., 1991),
explanation of growth and equilibrium shapes of two-dimensional
(2D) islands (Mo et al., 1989), and exploration of the modification of the
growth process by surfactants (Horn-von Hoegen, 1994). Figure 15–6
illustrates the identification of the initial stage of island formation in

Figure 15–6. STM images of 0.01 monolayers Ge on Si(001). The long diagonal
bands are substrate dimer rows. Filled-state images (A–C; V = −2 V; I = 0.2 nA)
and corresponding empty-state images (D– F; V = +2 V; I = 0.2 nA), showing
rows of symmetric (bean-shaped) and asymmetric (buckled) substrate dimers,
and Ge adatom-induced chain-like structures (marked by arrows). The chain-
like paired adatom structures are seen as metastable precursors to the forma-
tion of monolayer islands in Ge/Si homoepitaxy (Reprinted with permission
from Qin and Lagally, © 1997 AAAS).
980 P. Sutter

Ge/Si(001) epitaxy (Qin and Lagally, 1997). Classic theories of nucle-

ation in thin film growth are based on the notion of a “critical nucleus,”
defined as the structural entity for which the addition of one more atom
will for the first time reduce the free energy. Employing very low Ge
coverages and high-resolution dual bias STM imaging to uncover the
link between monomer adsorption and initial 2D growth islands, this
study shows that island formation may not involve a “critical nucleus,”
but instead may proceed via a family of metastable structures of
varying size, with significant consequences on many growth models
that are based on nucleation and require the size of a critical nucleus
as input.
More recently, interest has concentrated on strained layer heteroepi-
taxial growth, e.g., of Si1−xGex alloys on Si(001), as an elegant way of
producing large arrays of nanostructures. Si and Ge are miscible over
the entire composition range, and the lattice mismatch in the Si1−xGex/
Si(001) system can be tuned between 0 and 4% by alloying. Over a wide
range of compositions, Si1−xGex alloys initially wet the Si substrate, and
at higher coverage develop coherent (i.e., dislocation free) faceted three-
dimensional (3D) islands that lower the free energy by relaxing part
of the lattice mismatch strain. Potentially useful in electronics or opto-
electronics, these faceted nanostructures have shown a strikingly
complex array of growth phenomena, which make them interesting for
fundamental growth studies.
Constant current STM images with atomic resolution typically
encompass fields of view of few tens of nanometers. Combining a drift-
stable microscope with state-of-the-art control electronics, substantially
larger atom-resolved images have become feasible. The ability to obtain
image sizes in excess of 1 µm2 with high resolution is potentially pow-
erful for imaging growth processes, since it provides access to all rele-
vant length scales as well as image statistics far superior to that of
conventional small scans. Figure 15–7 illustrates this capability with
an STM image of 1.5 monolayers Ge on Si(111) with a field of view of
0.75 µm and 0.05 nm pixel size. The large scan provides an excellent
overview of the step and island structures resulting from monolayer
Ge deposition. At progressive zoom into the image, individual terraces,
and, ultimately, single surface defects and reconstruction domains
(here coexisting 7 × 7 and 5 × 5 domains) can be examined.
Figure 15–8 gives an example in which the superior statistics result-
ing from large atom-resolved images was used successfully to pinpoint
the mechanism of periodic surface roughening in Ge/Si(001) growth
(Sutter et al., 2003a). Via a sequence of “quench-and-look” experiments
at different Ge coverages, short-range interactions between surface
steps and vacancy lines, periodic arrays of linear chains of dimer
vacancies, are shown to cause a highly correlated surface roughness in
the form of anisotropic 2D islands with progressively higher aspect
ratio (a–c). Finally, images obtained at a critical Ge coverage of four
atomic layers (d) show the transition from 2D to 3D growth by forma-
tion of the first faceted islands on the rough wetting layer, and identify
the atomic-scale pathway of the 2D–3D transition.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 981

Figure 15–7. Large-area atom-resolved STM. (a) STM image with 0.75-µm
field of view and 0.05-nm pixel size, showing the surface morphology of Si(111)
after deposition of 1.5 ML Ge at 550°C. An array of parallel steps, running
diagonally from the upper left to lower right, coexists with islands nucleated
during Ge deposition. (b) Zoomed-in view (50 nm) of the area marked by a
square in (a), showing one of the islands. (c) Further zoom-in (10 nm), showing
the coexistence of (7 × 7) and (5 × 5) reconstructed domains in the area marked
in (b).

Figure 15–8. Transition from 2D to 3D morphology in the growth of Ge on Si(001). Large-area atom-
resolved STM images (250 nm × 125 nm sections are shown here) illustrate the surface morphology at
(a) 1.5 ML, (b) 2.3 ML, (c) 3.5 ML, and (d) 4.0 ML Ge coverage. The transition to 3D growth is preceded
by the formation of highly correlated, anisotropic surface roughness. With the superior statistics of
large-area high-resolution images, the repulsive interaction between surface steps and defects (dimer
vacancy lines) was identified as the origin of the ordering of this surface roughness. (Reprinted with
permission from Sutter et al., © 2003a by the American Physical Society.)
982 P. Sutter

Figure 15–9. Shapes of coherent (i.e., dislocation-free) 3D islands induced by lattice mismatch strain
during Si1−xGex/Si(001) heteroepitaxy. (a) Unit stereographic triangle for Si, showing the major and
minor stable facets (Gai et al., 1998). (b) STM top view, showing a shallow (105) faceted “hut.” (c) Mul-
tifaceted “dome” shape, terminated by (105), (113), and (15 3 23) facets. (d) “Barn”-shaped 3D island,
observed for low-misfit Si1−xGex alloys on Si(001), transformed from a “dome” by introduction of addi-
tional (111) facets (Reprinted with permission from Sutter et al., © 2004 American Institute of
Physics.).

Initial 3D islands with shallow facets, such as those shown in Figure

15–8d, evolve into more complex shapes involving steeper facets, which
give rise to more efficient strain relaxation (Figure 15–9). For Si1−xGex
epitaxy on Si(001), the facets bounding the islands identified by STM
are to a large extent major stable facets of Si and Ge (Figure 15–9a). At
early stages of 3D growth, shallow (105) faceted “huts” are invariably
observed (Figure 15–9b; Mo et al., 1990). These islands then undergo a
shape transformation to become multifaceted “domes” (Figure 15–9c;
Medeiros-Ribeiro et al., 1998). At later stages in the shape evolution,
elastic strain relaxation increasingly competes with plastic relaxation
via dislocations. For Ge/Si(001) dislocations are introduced before the
“dome” shape can transform further by introducing additional steeper
facets. By reducing the Ge concentration of Si1−xGex alloys, and thus
lowering the lattice mismatch strain, the coherent shape evolution has
been extended beyond the “dome” shape to the end-point in the
stereographic triangle, where steep (111) facets are introduced (Sutter
et al., 2004).

3 Tunneling Spectroscopy
Equation (1) gives an expression for the tunneling current I in the WKB
approximation for planar tunneling. In this approximation the tunnel-
ing current is an integral, over the energy range of width |eV| in which
tunneling can occur, of a product of the densities of states (DOS) of tip
[ρT(E)] and sample [ρS (E)] multiplied by the transmission probability,
T(E, eV). While this simple approximation cannot address the high
spatial resolution of STM—a more elaborate theory capable of address-
ing this aspect will be discussed in Section 6—it demonstrates that the
tunneling current is affected by the electronic structure of both the tip
and the sample. In a large number of applications it would be desirable
to merely map the atomic scale “topography” of a sample. The sensitiv-
ity of the tunneling current to the DOS can then be a disadvantage,
since it often prevents a simple interpretation of a tunneling image as
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 983

a “topographic” map. For many metals with small variations in the

DOS near the Fermi energy such a simple interpretation is often pos-
sible, i.e., height maxima in a constant current STM image are corre-
lated with atomic positions. However, for semiconductors with strongly
varying DOS near EF, the simple picture frequently breaks down.
An example is the (001) surface of Si. Geometrically, this surface
consists of Si dimers, in which each atom is bonded to two atoms in
the second layer and to its dimer partner. Constant-current STM images
obtained on Si(001) at positive sample bias indeed show two maxima
in each dimer (Figure 15–10B), which could be interpreted as the
“atomic positions.” At negative sample bias, however, only a single
broader bean-shaped structure is observed (Figure 15–10A), which
extends across each dimer. In fact, the tunneling current in empty-state
STM (positive sample bias) reflects the charge distribution in a π* anti-
bonding state just above the Fermi energy, whose wave function has a
node at the center of the dimer. Even in this case in which the STM
image shows two maxima per dimer, these maxima do not reflect the
atomic positions in the dimer. Worse yet, in filled-state STM (negative
sample bias) a π bonding state is imaged whose charge density peaks
at the center of the dimer bond, thus giving rise to a single “topo-
graphic” maximum at that position.
The sensitivity to the electronic structure of tip and sample, however,
is not merely a complication in STM imaging. It is also one of the major
strengths of the technique, since it allows measuring and mapping
surface electronic structure with very high spatial resolution well
below 1 nm. While most other techniques, such as ultraviolet or x-ray
photoelectron spectroscopy or Auger electron spectroscopy, average
over macroscopic sample areas, recently developed spectromicroscopy
techniques, such as synchrotron photoelectron emission micro-
scopy, provide spectroscopic imaging with 10 nm spatial resolution (see
Chapter 8, this volume). However, measurements of surface electronic

Figure 15–10. Constant-current STM images of Si(001) at negative [V = −1.6 V

(A)] and positive [V = +1.6 V, (B)]. The (2 × 1) unit cell is outlined, and the loca-
tions of the dimers are shown in the center. (Reprinted with permission from
the Annual Review of Physical Chemistry, Volume 40 © 1989 by Annual
Reviews www.annualreviews.org.)
984 P. Sutter

structure “atom-by-atom,” summarized under the term scanning tun-

neling spectroscopy, are feasible only by STM.

3.1 Bias-Dependent Imaging

The simplest scenario for gathering electronic structure information
involves the successive measurement of constant-current STM images
at different sample bias. As discussed in Section 1, the tunneling trans-
mission probability is generally highest at the Fermi energy of the
negatively biased electrode, and falls off exponentially at lower ener-
gies. For positive sample bias, i.e., empty-state STM, elastic tunneling
occurs mainly from states near the tip Fermi energy into unoccupied
sample states. In this polarity, varying the bias voltage probes different
surface states of the sample, particularly if the electronic structure of
the tip varies slowly in the energy range defined by the bias voltages
used. A series of constant-current STM images at different bias can
thus be used to map the spatial distribution of unoccupied surface
states of the sample.
For negative voltages applied to the sample, i.e., filled-state STM, the
situation is not as fortunate. The tunneling current is now dominated
by states near the Fermi level of the sample. As a result, the occupied
sample state with highest energy is invariably imaged preferentially
(Becker et al., 1989). Even if lower-lying surface states exist, as indicated
schematically in Figure 15–11, constant-current STM imaging is rather
insensitive to those states.
As a possible solution to this problem in bias-dependent filled-state
imaging it has been suggested that metal probe tips with smoothly
varying density of states near the Fermi energy could be replaced by
a semiconductor tip with a strongly modulated DOS. Band structure
effects have long been recognized in the interaction of low-energy
electrons with metal surfaces (Bauer, 1994). As an example, the high
normal incidence reflectivity on W(110) at low electron energies is due
to a (110) projected band gap extending over 5 eV, i.e., a lack of extended
bulk states in the energy range of this gap. For W(100), such a gap does

Figure 15–11. Band diagrams illustrating the bias-dependent tunneling

between tip (T) and sample (S). Sample bias V > 0: empty-state STM. V < 0:
filled-state STM. T(E) shows schematically the transmission probability as a
function of energy for states participating in the tunneling process.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 985

Figure 15–12. Energy-filtered STM. (a) Bulk band structure of monocrystalline InAs probe tips, pro-
jected along the (111) zone axis parallel to the tunneling direction. Ep denotes a projected gap in the
bulk conduction band. (b) Band diagram and STM image for low-bias filled state STM with InAs tips,
demonstrating preferential imaging of the adatom state on Si(111)-(7 × 7). (c) High-bias filled state
InAs-tip STM on Si(111)-(7 × 7): alignment of the adatom state with the projected tip gap reduces its
contribution to the tunneling current, giving rise to preferential imaging of the energetically lower
rest-atom state (Reprinted with permission from Sutter et al., © 2003b by the American Physical
Society).

not exist. Low-energy electrons can propagate into extended states, and
the reflectivity in the same energy range is low. Similarly, it can be
expected that a lack of final states due to projected band gaps in the
STM tip material can significantly affect the tunneling probability in
the energy range spanned by such gaps. This modulation of the trans-
mission probability could be used for efficient energy filtering of the
tunneling current in constant-current STM, specifically in filled-state
imaging.
Calculations by density functional theory (DFT) show a number of
gaps in the [111]-projected band structure of III–V compound semicon-
ductors such as InAs. III–V compound semiconductors cleave easily at
(110) planes, and cleavage corners bounded by {110} and (100) planes
are sufficiently sharp for atomic-resolution STM (Nunes and Amer,
1993). In addition, InAs(110) has the advantage of an unpinned surface
with minimal band bending and slight electron accumulation. The
feasibility of energy-filtered STM has been demonstrated by using a
cleaved InAs tip to selectively image different occupied surface states
on Si(111)-(7 × 7) (Sutter et al., 2003b). In these experiments, summa-
rized in Figure 15–12, a [111]-projected gap in the InAs conduction
band, centered at Γ and about 2.6 eV wide, is used to modulate the
tunneling current in filled-state constant current STM. The “adatom”
state (“a”) on Si(111)-(7 × 7), which lies only 0.35 eV below the Fermi
energy and is imaged by STM using a metal tip, dominates the tunnel-
ing current at low tip–sample bias. At higher bias, this state aligns with
a projected gap in the InAs conduction bands and its tunneling
probability is reduced sharply. Under these conditions the “rest atom”
986 P. Sutter

Figure 15–13. (a) Bias-dependent STM on GaAs(110): selective imaging of Ga and As sublattices at
positive and negative sample bias, respectively (Reprinted with Permission from Feenstra et al., ©1987
by the Armeucion Physics Socuity). (b) Compound STM image of the InP(110) surface, assembled from
separate positive and negative bias scans (Reprinted from Ebert et al., ©1992 with permission from
Elsevier). (See color plate.)

state (“r”), a surface state 0.8 eV below EF that is not accessible in con-
ventional STM imaging, contributes a majority of the tunneling current
and can be imaged selectively. Importantly, these results suggest that
robust bulk band structure effects, which are independent of the atomic
structure of the tip, can be used to modulate the transmission probabil-
ity in vacuum tunneling. Once developed into a routine imaging tech-
nique, energy-filtered STM has the potential to provide rich spectroscopic
contrast within the relatively simple framework of bias-dependent
constant current STM.
A classic example of a scenario in which conventional bias-
dependent imaging provides directly interpretable spectroscopic infor-
mation is imaging of cation and anion sites on (110) cleavage planes of
compound semiconductors such as GaAs (Feenstra et al., 1987) or InP
(Ebert et al., 1992) (Figure 15–13). For (110) surfaces of compound semi-
conductors, such as GaAs, the occupied state density (imaged at nega-
tive sample voltage) is localized on the anions (As, P), while the
unoccupied state density (positive sample voltage) is localized on
the cations (Ga, In). As a result, bias-dependent STM imaging can be used
to selectively image the cation and anion sublattices on these surfaces.

3.2 Local I-V Measurements and Current Imaging

Tunneling Spectroscopy
In many instances, a complete electronic structure map is not required,
but one would like to determine, at one or several positions on a
sample, the surface density of states in an interval of a few eV around
the Fermi energy. Such local measurements can, for example, serve
to determine if a specific surface reconstruction or a small cluster is
semiconducting or metallic, to measure the band gap at the surface
of semiconductor samples, and to determine for a given location the
energies of surface states.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 987

Local current–voltage [I(V)] spectra serve this purpose. The basic

information content of a tunneling I(V) spectrum can be demonstrated
by differentiating Eq. (1) [V > 0]:
eV
dI
= eρs ( eV ) ρT ( 0 ) T ( eV , eV ) + ∫ [ eρ s ( E ) ρT ( E − eV ) ∂T ( E , eV ) /
dV 0
d ( eV ) − eρs ( E ) ρT′ ( E − eV ) T ( E, eV )] dE (4)
For a given energy eV, the tunneling conductance dI/dV reflects the
density of states of the sample at that energy, ρS (eV).
Feenstra et al. (1987) noted difficulties with conductance spectros-
copy, resulting from the fact that the expression for dI/dV diverges
exponentially both in voltage (V) and separation (z). These divergences
can be eliminated by normalizing dI/dV by the conductance of the
tunneling junction, [dI/dV]/[I/V] ≈ d(lnI)/d(lnV). In this form, spe-
ctra obtained at different tip–sample separation z can be compared
directly.
Local spectroscopy is quite simple to perform. The STM tip is posi-
tioned and stabilized over a chosen point on the sample. The feedback
loop is switched off while the sample voltage is varied over the desired
values, and a local measurement of tunneling current as a function of
bias voltage, I(V), is performed. As an alternative, the tunneling con-
ductance, dI(V)/dV, can be measured by ramping the bias voltage with
a small ac voltage added, and measuring the ac component of the tun-
neling current at the modulation frequency of the ac part by lock-in
techniques. After the I(V) or dI(V)/dV spectrum is recorded, the tip
feedback is reactivated, and STM imaging can resume.
Figure 15–14a shows tunneling conductance spectra obtained on p-
and n-type GaAs(110) (Feenstra et al., 1987), which provide a measure

Figure 15–14. (a) Local normalized conductance spectra obtained on n- and p-doped GaAs(110),
showing three peaks (V, D, C) originating from surface states (Feenstra et al., 1987). (b) Normalized
conductance spectra obtained on InP(110), illustrating the valence and conduction band edges delimit-
ing the bulk bandgap (1.4 eV), and the A5 and C3 surface states defining the surface gap (1.9 eV)
(Reprinted from Ebert et al., © 1992 with permission from Elsevier).
988 P. Sutter

of the surface state density. The spectra show three distinct peaks
associated with surface states. Peak C is related to a state in the GaAs
conduction band, localized on the Ga atoms, while peak V stems from
a valence state localized on the As anions. Peak D is associated with
tunneling of dopant-induced carriers within the bulk band gap. The
separation between the leading edges of the C and V peaks is close to
the bulk band gap of GaAs (Eg = 1.43 eV), i.e., in this system tunneling
spectroscopy can be used to determine the band gap. Figure 15–14b)
shows tunneling conductance spectra on the (110) cleavage surface of
another III–V compound, InP. Also for InP, the band gap (∼1.4 eV) is
evident from sharp onsets of significant state density at the valence
and conduction band edges. However, for this system it was argued
that a wider surface band gap (∼1.9 eV) is indirect, delimited by a
surface state C3 at the edge of the surface Brillouin zone and a broad-
ened state A5 at the zone center.
A natural, albeit experimentally much more complex extension of
local I(V) spectroscopy is the measurement of tunneling spectra, as
discussed above, at each image pixel of a constant-current STM scan.
This measuring scheme is called current-imaging tunneling spectro-
scopy (CITS). Since complete I(V) spectra are obtained at each image
point, the corresponding data sets can provide a full range of spectro-
scopic information. As an example, current maps I(x, y) at fixed tip–
sample bias can be produced. Instead, the voltage dependent tunneling
conductance dI(V)/dV can be calculated numerically and mapped as a
function of sample position. If the chosen voltage corresponds to the
energy of a surface state of the sample, conductance maps will provide
a direct image of the spatial distribution of that state. Due to the com-
plexity of the data acquisition, the experimental requirements for CITS
are quite stringent. Since complete I(V) curves are measured at each
image pixel, requiring the stopping of a scan and deactivation of the
feedback loop for a fraction of a second per spectrum, a very high sta-
bility of the tunneling gap and low lateral drift of the tip relative to
the sample are of key importance. These conditions are more easily
fulfilled at cryogenic temperatures, where a z-stability of the order of
1 pm and lateral drift velocities of the order of few Å/hour are possible.
Low temperatures also reduce the thermal broadening of the tunneling
transmission coefficient T(E), and thus narrow the linewidth of fea-
tures in the tunneling spectra.
A classic example of the application of CITS, the measurement of the
electronic structure of Si(111)-(7 × 7), is shown in Figure 15–15 (Hamers
et al., 1986). Shown are a constant current image of the surface at +2 V
sample bias (a), as well as current images acquired during the same
scan at bias voltages of +1.45 V and −1.45 V, showing spatial maps of
unoccupied (+)/occupied (−) sample states at energies up to 1.45 eV
above/below the Fermi energy. CITS difference images, a precursor to
modern dI/dV maps, discussed below, calculated by numerically sub-
tracting current images at energies bracketing those of specific surface
states, allowed the mapping of the adatom state, dangling bond, and
backbond state on this complex reconstructed surface.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 989

Figure 15–15. Current-imaging tunneling spectroscopy on Si(111)-(7 × 7) (Hames et al., 1986). (a)
Constant-current STM image (V = +2 V), with one (7 × 7) unit cell outlined (F, faulted; U, unfaulted
half). (b and c) Current images obtained at +1.45 V and −1.45 V, respectively. Note the strong similarity
between the rest-atom contrast in the current image (c), and the contrast obtained in rest-atom imaging
by energy-filtered constant-current STM (Figure 15–12).

More recently, this complex data acquisition mode has been used,
for example, to map modifications in the electronic structure of indi-
vidual carbon nanotubes arising with the supramolecular assembly of
C60 molecules inside the hollow tube to form nanotube “peapods”
(Hornbaker et al., 2002). dI/dV spectra obtained along the symmetry
axis of C nanotubes show distinct peaks in the density of unoccupied
states with a spatial periodicity corresponding to the average spacing
of embedded C60 molecules observed by transmission electron micros-
copy, suggesting that the insertion of C60 causes a significant modula-
tion of the electronic structure. A control measurement was constructed
by using the STM tip to push the C60 away, and remeasuring dI/dV
spectra along the same nanotube section without embedded C60. The
empty tube obtained in this way shows small and smooth variations
in the density of states along the nanotube axis. Qualitatively, the elec-
tronic structure of the peapods was explained by the single wall C-
nanotube acting as a conduit that enhances the coupling between C60
molecules nested inside it, and the Bragg scattering of nanotube states
due to the periodic arrangement of the C60. Calculations of the peapod
electronic structure confirm the formation of a hybrid electronic band,
which derives its character from both the nanotube states and the C60
molecular orbitals (Figure 15–16).

3.3 Differential Conductance (dI/dV) Mapping

For a given energy eV, the tunneling conductance dI/dV reflects the
density of states of the sample at that energy, ρS (eV). As a result, it can
be expected that high-resolution maps of tunneling conductance can
provide detailed information on spatial variations in sample electronic
structure. For example, by plotting dI/dV at an energy corresponding
to a localized state at the sample surface, the spatial distribution of that
sample state can be mapped.
In a typical experiment a constant current image, at a specific bias
voltage and tunneling current used to stabilize the tunneling gap, is
990 P. Sutter

Figure 15–16. I(V) tunneling spectroscopy on C60/C-nanotube “peapods” (Reprinted with permission
from Hornbaker et al., © 2002 AAAS). (A) Map of an array of full dI/dV spectra along the axis of a C-
nanotube “peapod.” Sample bias voltage is plotted on the horizontal axis and displacement along the
tube on the vertical axis. (B) Representative dI/dV spectra at selected positions along the tube. Large
conductance peaks are found at positions of embedded C60 molecules. (C) Variation of tunneling con-
ductance along the tube axis. (D) Reference spectroscopic map on an empty C-nanotube section
without embedded C60, in which no strong modulation of the tunneling conductance is observed. (See
color plate.)

measured simultaneously with dI/dV maps at one or several voltages/

energies. To obtain the spectroscopic data, the feedback loop is turned
off at each point of the STM scan, and the selected bias voltages are
applied. The tunneling conductance is measured directly using a mod-
ulation technique. At the end of the measurement, the feedback is
reactivated, and the tip is moved to the next image pixel.
If a small modulation voltage dV(t) = a cos ωt with amplitude a and
frequency ω is added to the dc sample bias V0, the resulting tunneling
current
dI
I ( t ) = I 0 + dI ( t ) = I 0 +
V dV ( t ) + 
dV 0
dI
= I0 + s ωt + 
V a cos (5)
dV 0
has a small ac component. Its amplitude at the modulation frequency
ω, which can be measured using a lock-in amplifier, is proportional to
dI/dV|V0. Via measurements at different dc bias V0, a set of tunneling
conductance values at different energies can be determined at each
pixel during an STM scan, and can later be displayed as maps of the
sample density of states at those energies. Although quite simple con-
ceptually, the long dwell times at each image pixel in dI/dV mapping
pose stringent demands on microscope stability, which are best met at
cryogenic temperatures. Under these conditions, thermal broadening
of features in the density of states is also minimized.
dI/dV mapping has been applied in a wide variety of measurements
of electronic states at surfaces and in nanostructures, such as atomic
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 991

chains (Crain and Pierce, 2005) or single molecules (Repp et al., 2005).
A particularly widespread application is the imaging of quasiparticle
interference. In an ideal metal, the Landau-quasiparticle eigenstates are
Bloch wavefunctions with specific wavevectors k and energies ε. Their
dispersion cannot be measured directly by STM. However, in the pres-
ence of disorder, e.g., impurities or crystal defects, elastic scattering
mixes eigenstates with different k that are located on the same quasi-
particle contour of constant energy in k-space. When states k1 and k 2
are mixed by scattering, an interference pattern with wavevector q =
k 2 − k1 appears in the norm of the quasiparticle wavefunction. The
interference leads to modulations in the local density of states with
wavelength λ = 2π/|q|, which can be observed as modulations in the
differential tunneling conductance. Via differential conductance
mapping, interference of electronic eigenstates has been probed in
metals (e.g., Cu(111), Crommie et al., 1993; Au(111), Hasegawa and
Avouris, 1993) and semiconductors (e.g., InAs(110), Wittneven et al.,
1998). When density of states modulations are detected by STM, certain
contours of constant energy in k-space can be reconstructed by analyz-
ing the Fourier transform of the real-space density of states map. Such
Fourier transform spectroscopy simultaneously yields real-space and
momentum-space information on wavefunctions, scattering processes,
and quasiparticle dispersion.
Elastic scattering of Bogoliubov quasiparticles in superconductors
can also give rise to conductance modulations (Hoffman et al., 2002a),
suggesting that STM could be used as a local probe to study electron
correlation in superconductors. Quasiparticle inteference has indeed
been demonstrated for cuprate high-temperature superconductors, such
as Bi2Sr2CaCu2O8+δ (Hoffman et al., 2002a; Vershinin et al., 2004) and
Ca2−xNaxCuO2Cl2 (Hanaguri et al., 2004). On Bi2Sr2CaCu2O8+δ crystals
cleaved at the BiO plane at cryogenic temperature in UHV, a constant-
current image (Figure 15–17A) and differential conductance maps
(Figure 15–17B) are obtained simultaneously in a low-temperature

Figure 15–17. Fourier transform spectroscopy on a cleaved Bi2Sr2CaCu2O8+δ high-temperature super-

conductor. (A) Constant-current STM image (65 nm, 0.13 nm resolution). (B) Simultaneously acquired
differential conductance map (∼ local DOS) at an energy of 12 meV below EF, showing a pronounced
modulation of the electronic structure. dI/dV is measured with 2 mV modulation. (C) Fourier trans-
forms of DOS maps similar to (B), for energies up to 30 meV, representing maps of quasiparticle scat-
tering wave vectors in this energy range. (Reprinted with permission from Hoffman et al., © 2002a
AAAS.)
992 P. Sutter

STM. The conductance map shows a checkerboard-like modulation of

the local density of states. Fourier transforms of dI/dV maps obtained
over a range of energies show dominant q-vectors associated with
quasiparticle scattering. Different spatial patterns and wavelengths are
observed at different energies. Applying a magnetic field perpendicu-
lar to the sample, the quasiparticle scattering can be modified by the
controlled introduction of vortices (Hoffman et al., 2002b). Although
the interpretation of these complex data sets is still somewhat contro-
versial (Vershinin et al., 2004), these experiments clearly demonstrate
the power of an STM-based approach for studying electronic structure
and ordering phenomena in high-temperature superconductors.
A second example in which dI/dV mapping plays a key role is spin-
polarized STM. A magnetic probe tip, e.g., a conventional etched W tip
coated with a thin Fe film and polarized in a magnetic film perpen-
dicular to the tip axis, is used for tunneling on a magnetic sample.
Figure 15–18 illustrates the principle of spin-polarized STM for the
example of a ferromagnetic Gd(0001) sample and a hard magnetic Fe
tip (Bode et al., 1998). In a weak external magnetic field applied parallel
to the sample surface, Gd(0001) has an exchange-split surface state
with majority and minority parts at binding energies of −220 meV and
+500 meV, respectively (Figure 15–18a). In the experiment, the magne-
tization direction of the Fe tip remains fixed independent of the direc-
tion of the applied magnetic field, while the magnetization of the Gd
sample can be switched by reversing the direction of the applied field.
Due to a spin valve effect, the tunneling current of the surface state
spin component parallel to the tip magnetization is enhanced. If the
majority spin is aligned with the tip magnetization (blue curve in
Figure 15–18b) the tunneling conductance of the occupied majority

Figure 15–18. Spin-polarized STM on a Gd(0001) sample with an exchange-split surface state and a
magnetic Fe tip with constant spin polarization close to EF. (a) Due to the spin-valve effect the tun-
neling current of the surface state spin component parallel to the tip magnetization is enhanced. (b)
Illustration of the reversal in the dI/dV signal at the surface state peak position upon switching the
sample magnetically. (c) Experimental observation of this reversal in tunneling into an isolated Gd
island (Reprinted with permission from Bode et al., ©1998 by the American Physical Society). (See
color plate.)
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 993

state will be high, while that of the minority state is low. After reversal
of the applied magnetic field, the minority spin will align with the tip
magnetization. The conductance of the minority state thus increases,
while that of the majority state is reduced. Difference spectra obtained
by subtracting the traces, measured at two polarities of the external
field, provide a direct measure of the sample magnetization, and can
be used to build maps of magnetic order. Spin polarized STM has been
shown to be a very powerful technique for mapping magnetic and
antiferrromagnetic order with atomic resolution, as demonstrated by
studies on 2D antiferromagnetic ordering in Mn/W(110) (Heinze et al.,
2000) and on magnetic hysteresis in Fe/W(110) (Pietzsch et al., 2001).

3.4 Inelastic Tunneling: Vibrational Spectroscopy

A major drawback of conventional STM imaging is its lack of chemical
and element specificity, which stems from the fact that only states
within few eV of the Fermi energy can be probed. As a result, charac-
teristic core levels, which could provide element specificity, are not
accessible. For single molecules, an alternative pathway to chemically
specific imaging employs vibrational signatures, which are character-
istic of specific molecular bonds.
The use of inelastic electron tunneling spectroscopy (IETS) for vibra-
tional fingerprinting on molecules embedded in planar tunnel junc-
tions dates back to the 1960s (Jaklevic and Lambe, 1966). The technique
is based on the detection of a slight increase in tunneling conductance
at an electron energy eV = hω, - corresponding to a vibrational mode
with frequency ω of an embedded molecule. When the applied bias
reaches this threshold, an inelastic channel for tunneling opens up in
addition to the elastic tunneling at lower bias. Inelastically scattered
electrons make up only a small fraction of the total tunneling current
(typically a few percent). Nevertheless, the second derivative of the
tunneling current, (d2 I/dV 2), usually shows clear signatures of inelastic
tunneling in the form of peaks at the characteristic bias voltages cor-
responding to different vibrational modes. As in dI/dV maps, d2 I/dV 2
can be measured by lock-in techniques, by adding a small modulation
voltage to the tunneling bias V0. The signal amplitude measured at
twice the modulation frequency is proportional to d2 I/dV 2|V0.
Conventional IETS in planar oxide tunnel junctions samples large
ensembles of embedded molecules. Soon after the invention of STM, it
was suggested that IETS should be feasible in local tunneling between
a tip and a sample in STM (Binnig et al., 1985). A first convincing
experimental demonstration of inelastic tunneling through few mole-
cules was obtained in a setup of two crossed Au wires with finely
adjustable normal force, allowing the trapping and IET spectroscopy
at cryogenic temperatures on a small number of hydrocarbons
(Gregory, 1990).
IETS on single molecules in a low-temperature STM was finally
demonstrated by Ho and co-workers in 1998 (Stipe et al., 1998a). A
beautiful experiment on single acetylene molecules adsorbed on
Cu(100) not only provides clear evidence for inelastic tunneling at
994 P. Sutter

Figure 15–19. Single molecule vibrational spectroscopy and microscopy. (a) d2 I/dV 2 recorded over
individual C2H2 and C2D2 molecules on Cu(100) at 8 K. (1) Signature of inelastic electron tunneling by
exciting the C—H stretch mode with an energy of 358 meV. (2) Isotope shifted (266 meV) inelastic tun-
neling peak of the C—D stretch mode. (b) Constant current (CC, 4.8 nm) STM of adjacent C2H2 and
C2D2 molecules on Cu(100), along with simultaneously acquired d2 I/dV 2 maps obtained at 358 mV and
266 mV, imaging the C—H and C—D bonds selectively. The symmetric round appearance is attributed
to rotations of the molecules during the experiment. A d2 I/dV 2 map obtained at intermediate energy
(311 mV) shows no vibrational contrast. (Reprinted with permission from Stipe et al., © 1998 AAAS.)

energies close to vibrational frequencies determined in the gas phase,

but it also shows the expected isotope shift between measurements on
normal and deuterated acetylene. Robust increases in differential con-
ductance of the order of 3–12% are observed for this system.
In addition to inelastic tunneling spectra on individual molecular
bonds, as shown in Figure 15–19a, inelastic electron tunneling micros-
copy with molecular resolution was demonstrated on small molecules,
such as acetylene (Figure 15–19b). In constant-current STM imaging,
both C2H2 and C2D2 molecules are imaged as identical depressions, i.e.,
are indistinguishable. On the other hand, C2H2 molecules are selec-
tively imaged in (d2I/dV2) maps obtained at the characteristic energy
of the C—H stretch mode (358 meV), while C2D2 molecules are imaged
at the isotope shifted energy (266 meV). In a control experiment at an
energy between these two characteristic values, none of the two iso-
topes gives rise to contrast.
The development of single-molecule vibrational spectroscopy and
microscopy provides a powerful tool for performing and viewing
chemistry at the ultimate limit of concentration (individual molecules)
and spatial resolution (atomic). Local energy input by inelastic tunnel-
ing allows the precise manipulation of individual molecules (rotation,
translation, see Section 4) and the cleavage and formation of individual
molecular bonds, as well as the characterization of the individual and
final configurations on the surface.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 995

3.5 Element-Specific Imaging

In contrast to chemical specificity at the level of individual small mol-

ecules, which IETS can provide independent of the particular sample
structure, no universal approach has been identified to obtain element
specific contrast in STM. Various techniques have been suggested to
achieve generic element specificity, among them a scanning tunneling
atom probe operated by controlled atom transfer to the STM tip and
desorption into a time-of-flight chamber (Weierstall and Spence, 1998),
and a scanning probe energy loss spectrometer using an STM at high
tip–sample bias in combination with an electron spectrometer for
local probe electron energy loss spectrometry (Festy and Palmer, 2004).
None of the techniques demonstrated to date provided a universal and
practical pathway toward element specific imaging at the atomic scale.
However, a number of sample–tip combinations have been identified,
in which element contrast is indeed achieved on a case-by-case basis.
Prime examples are bimetallic metal surfaces, on which different
atomic species can be discriminated in constant current STM images.
In a few simple cases, a mere difference in atomic size can give rise
to contrast, e.g., for submonolayer coverages of Pb/Cu(111) with a dif-
ference in metallic radii of nearly 50 pm (Nagl et al., 1994). In a broader
class of metal alloys, electronic effects, i.e., differences in local density
of states above different atoms, give rise to element contrast. Examples
of systems with substantial density of states differences near the Fermi
energy include alloys between transition metals with a partially filled
d-shell and noble metals (Au, Ag, Cu). STM images of the (111) surface
of AgPd alloys (Wouda et al., 1998) indeed show the Ag atoms with
lower apparent height than the transition metal atoms, as expected
based on a simple density of states argument. For alloys of transition
metals (e.g., PtRh; Wouda et al., 1996), the prediction of the STM con-
trast is less straightforward, and generally requires ab initio calcula-
tions of the local density of states. For the PtRh(100) surface shown in
Figure 15–20a, a local density of states difference at low tunneling
resistance produces a height difference of about 20 pm between Pt and
Rh, with Rh imaged at larger apparent height. In some systems, where
sufficient differences in local density of states to provide STM contrast
do not exist, such differences can be induced by suitable modification
of the sample surface. In an important semiconductor system, Si1−xGex
alloys, Si and Ge atoms are indistinguishable in conventional constant
current STM, which severely hampers the understanding of epitaxial
growth since crucial local composition information is lacking. Cover-
ing a heterogeneous (111) surface composed of Si- and Ge-rich areas
with 1 atomic layer of Bi induces robust differences in apparent height
of nearly 0.1 nm between the chemically different regions, thus provid-
ing a means for imaging and characterizing lateral heterostructures
of monolayer-thick alternating epitaxial Si and Ge stripes (Kawamura
et al., 2003).
A third mechanism that can give rise to atomic-scale contrast between
different elements is thought to be based on a direct interaction between
tip and sample, in what could be described as a “precursor” chemical
996 P. Sutter

Figure 15–20. Element-specific imaging with atomic resolution by STM. (a)

PtRh(100): for a bulk composition of 50% Pt and Rh, there are approximately
31% Rh (bright) and 69% Pt atoms (darker) at the surface. Black spots are
caused by residual carbon. Pt and Rh atoms tend to cluster in small groups of
the same species (Reprinted with permission from Wouda et al., © 1996 with
permission from Elsevier). (b) Atomically and chemically resolved image of
Ge (bright)/Si (dark) nanowires formed by sequential deposition of submono-
layer amounts of Ge and Si in step flow mode on Bi-terminated Si(111). The
initial Si/Ge boundary (right arrows) is nearly atomically sharp, while inter-
diffusion is observed at the Ge/Si boundary (arrowheads) (Reprinted with
permission from Kawamura et al., © 2003 by the American Physical Society).

bond. With the tip–sample distance too large to allow true chemical
bonding, either atomic relaxation or the local charge density can change
as the tip is moved across different atomic species, which in turn
affects the tunneling current and gives rise to apparent height differ-
ences in constant current STM. Qualitatively, surface atoms with
higher chemical affinity to the tip atom should have higher tunneling
conductance and thus will be imaged at larger apparent height. Several
metal alloy systems, such as PtNi(111) and PtRh(111), are believed
to show element contrast that is based on tip–sample interaction.
Clearly, the nature of the tip apex would be of key importance for this
type of contrast. This may explain why for some systems element
contrast is obtained only after deliberate chemical modification of the
STM tip.

4 STM at High and Low Temperatures

STM combines unique capabilities in both high-resolution imaging and

spectroscopy at surfaces. Some form of controlled environment—such
as UHV, an inert gas or fluid—may be required, mainly to ensure
reproducible sample surface conditions. But many basic microscopy
tasks, including a broad range of imaging and spectroscopy experi-
ments, are performed conveniently at room temperature. Soon after the
invention of STM, however, it was recognized that control over the
sample temperature enables a wide variety of new experiments that
are impossible to perform at room temperature. As an example, the
structure or chemical reactivity of a given surface, e.g., of a catalyst,
may depend on temperature. To characterize its properties in a particu-
lar temperature regime, the sample has to be heated or cooled during
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 997

STM imaging. Further, though imaging of thin film growth in a “quench

and look” mode is possible, as discussed previously, studying growth
processes directly at relevant sample temperatures would often be
preferred. And finally, investigation of surface diffusion, chemical
reactions, or molecular dissociation at surfaces often requires cooling
of the sample to cryogenic temperatures to slow the rates of these
thermally activated processes sufficiently to map them by relatively
slow STM imaging. Probably the best known example of the benefits
of cryogenic STM is atom and molecule immobilization for atomic or
molecular manipulation, which allows artificial structures to be built
from individual atoms and can provide highly idealized structures for
probing chemistry and condensed matter physics at the ultimate spa-
tial limit. Below, the technical challenges involved in temperature-
dependent STM are discussed briefly, and selected examples of the
main classes of STM experiments at high and low temperature—the
imaging of dynamic surface processes via STM movies and the assem-
bly of nanostructures “one atom at a time” by atomic/molecular
manipulation with the STM tip—are presented.

4.1 STM at High and Low Temperatures: Motivation and Issues

Compared to room temperature operation, STM experiments at high,
low, or even variable temperatures involve a number of additional
issues. A primary experimental difficulty arises from thermal drift due
to incomplete thermalization and the resulting variations in thermal
expansion across the microscope, which cause relative motion between
the STM tip and sample. Thermal expansion coefficients are tempera-
ture dependent, and generally approach zero as T → 0. As a result,
thermal drift rates tend to be low and the overall instrument stability
is improved in low-temperature STM. Stable low-temperature micro-
scopes reach lateral drift rates of 0.1 nm/h (Meyer and Rieder, 1997),
and values of vertical stability of the tunneling gap approaching 1 pm
have been reported (Stipe et al., 1998a). Beyond instrument stability,
there are strong additional incentives for operation at cryogenic tem-
peratures. The rates of thermally activated processes, such as surface
diffusion of adsorbed atoms or molecules, can be slowed considerably
at low temperatures. The freeze-out of thermal motion paves the way
for time-lapse movies documenting diffusion processes, and for con-
trolled atom or molecule manipulation using the STM tip. On the other
hand, the thermal broadening of the tunneling distribution and of the
electronic structure can be minimized at cryogenic temperatures,
which results in higher energy resolution in spectroscopy.
STM imaging at high temperatures is generally affected by high drift
rates, which may require active drift compensation between consecu-
tive scans to allow imaging a given region on the sample for extended
periods of time. Intrinsic drift rates of 3 nm/min without and 0.2 nm/
min with drift correction have been reported for dedicated high-
temperature microscopes (Voigtländer, 2001). Additional practical dif-
ficulties involve sample heating, temperature measurement, as well as
the need for high scan speeds to capture dynamic phenomena at high
temperatures.
998 P. Sutter

4.2 Capturing Dynamic Surface Processes: STM Movies and

Atom Tracking

The most popular approach for imaging dynamic processes involves

repeated scanning of the same sample area to generate a time-lapse
STM “movie.” This capability has been used to study a wide variety of
phenomena, including step dynamics (Kuipers et al., 1993), epitaxial
growth (Voigtländer and Zinner, 1993), self-diffusion on semiconduc-
tor (Borovski et al., 1997) and metal surfaces (Horch et al., 1999), as well
as diffusion of adsorbates (Wintterlin et al., 1997) and large molecules
(Schunack et al., 2002). If rates of dynamic processes, such as surface
diffusion, are measured at several temperatures, activation energies of
these processes can be determined. Complex reaction pathways,
including adsorption, diffusion, dissociation, etc., can be elucidated at
the atomic scale, and active sites can be identified.
As a near-field microscopy technique STM is based on raster scan-
ning a probe tip. The scan speed is limited by the resonance frequency
of the scanning element, typically below about 10 kHz, and by the
electronic bandwidths of the tunneling current amplifier and feedback
circuit. Frame rates in STM movies can be expected to be inherently
lower than those in parallel imaging techniques, such as transmission
electron microscopy or low-energy electron microscopy (Bauer, 1998;
Tromp, 2000). To achieve adequate time resolution, the scan sizes in
STM movies tend to be small, of the order of 104 pixels. Additional
measures are typically necessary to deal with slow imaging rates and
capture the detailed time evolution of a given process. Whenever a
phenomenon under consideration permits, cooling of the substrate can
be employed to slow the rates of thermally activated processes. In thin
film growth, slow evaporation rates are employed. In studying surface
reactions involving adsorption from the gas phase, e.g., on catalysts,
low partial pressures of the reactive species in the gas phase are main-
tained. As an alternative to these often rather restrictive provisions, the
development of high-speed microscopes and control electronics has
been a focus of active research recently. As an example, scanners with
mechanical resonances in the range between 50 and 100 kHz have been
developed. Combined with a fast current amplifier (600 kHz band-
width) and feedback loop (1 MHz), atomically resolved images have
been obtained at rates approaching 200 frames/s (Rost et al., 2005).
Such exciting instrument developments may in the future allow the
imaging of new classes of dynamic surface phenomena at relaxed
ambient conditions (higher temperatures, pressure, and growth rates)
by STM.
To illustrate the use of STM movies at cryogenic temperatures to
identify and quantify dynamic surface processes, we discuss the
example of rutile TiO2(110), which has emerged as the prototypical
system for fundamental surface science studies of transition metal
oxides (Figure 15–21). TiO2 has numerous applications in areas as
diverse as heterogeneous catalysis, solar cells, photocatalysis, and
organic waste remediation, and STM plays a key role in elucidating its
fundamental surface processes. The TiO2(110)-(1 × 1) surface consists
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 999

Figure 15–21. STM on Rutile TiO2(110). (a) Structure of the TiO2(110) surface. V, a single oxygen
vacancy. (b) Constant-current STM on TiO2(110). Bright rows are assigned to five-fold coordinated Ti
atoms and dark rows to bridging oxygen atoms. Oxygen vacancies are imaged as protrusions between
the Ti rows. Adsorbed O2 molecules are associated with bright protrusions on the Ti rows. (Reprinted
with permission from Schaub et al., © 2003 AAAS.)

of alternating rows of Ti and O atoms aligned along the [001] direction.

Due to electronic effects, STM images Ti as protruding rows and bridg-
ing oxygen rows, which are geometrically highest on the surface by
about 0.12 nm, as troughs.
The reactivity of TiO2(110) is affected to a great extent by the presence
of oxygen vacancies, which are generated in the process of reducing
the surface by annealing in vacuum. Using STM movies, the reaction
mechanisms at these defect sites were studied for model reactions such
as water dissociation (Brookes et al., 2001; Schaub et al., 2001). The dif-
fusion of oxygen vacancies follows an intriguing mechanism mediated
by O2 molecules (Figure 15–22) (Schaub et al., 2003). Oxygen vacancies

Figure 15–22. O2-mediated diffusion of oxygen vacancies on TiO2(110). (a and b) Consecutive frames
of an STM movie on the motion of oxygen vacancies on TiO2(110) (T = 300 K, 8.5 s/frame). (c) Difference
image of (a) and (b), highlighting changes due to the diffusion of single oxygen vacancies. Vacancies
diffuse perpendicular to the bridging oxygen rows. (d–f) Time-lapse STM of the O2-assisted diffusion
of an individual oxygen vacancy (T = 230 K, 1.1 s/frame). (g) Schematic illustration of the O2-mediated
vacancy diffusion mechanism. (Reprinted with permission from Schaub et al., © 2003 AAAS.)
1000 P. Sutter

are immobile on the adsorbate-free surface. Vacancy diffusion is greatly

enhanced by adsorbed O2. At temperatures sufficiently low that the
diffusion of adsorbed O2 molecules can be captured by STM, the sub-
traction of consecutive frames in time-lapse STM shows that single
oxygen vacancies diffuse along [11̄0] from one bridging O row to the
next, always in the presence of neighboring O2 molecules.
The role of O2 molecules in the vacancy diffusion process is estab-
lished from detailed investigation of single vacancy hops, again based
on time-lapse STM movies at low temperature. As an O2 molecule dif-
fusing along a Ti row approaches an oxygen vacancy, it dissociates and
contributes one oxygen atom toward healing the vacancy, thus creating
a metastable intermediate consisting of a single O atom. The O adatom
is highly reactive, as corroborated in separate experiments involving
dosing of atomic oxygen. It rapidly recombines with a bridging O atom
and emerges as an O2 molecule. If in this process the bridging O atom
is removed from one of the adjacent rows, the net result is a diffusion
jump of an oxygen vacancy by one bridging oxygen row. Given this
O2-mediated mechanism of oxygen vacancy diffusion, the rate of dif-
fusion events is expected to scale linearly with O2 coverage. STM
movies obtained at different O2 exposure show that this is indeed the
case.
While early imaging of dynamic surface processes was performed
almost invariably in UHV, several applications require STM imaging
in what is seen as more “realistic” environments for those applications.
A prominent example is heterogeneous catalysis. It has been recog-
nized that actual reactions under technologically relevant conditions,
often involving elevated temperatures and pressures at or above atmo-
spheric pressure, can involve surface structures and compositions, and
entire reaction mechanisms that differ substantially, even qualitatively,
from those of “simulated” reactions running in UHV, a situation com-
monly termed the “pressure gap” problem of heterogeneous catalysis.
To address the need for imaging with high spatial and temporal resolu-
tion at elevated pressure, a family of dedicated STM instruments was
developed (Rasmussen et al., 1998; Jensen et al., 1999; Lægsgaard et al.,
2001; Rößler et al., 2005). These instruments allow sample preparation
and surface analysis in UHV, followed by exposure to reactants at high
pressure and simultaneous STM imaging. A particularly elegant imple-
mentation of this concept is the “reactor STM,” allowing dynamic STM
imaging of surfaces exposed to reactants in a compact catalytic flow
reactor in combination with the simultaneous analysis of the reaction
products by mass spectrometry.
Figure 15–23 shows an example of a complex dataset obtained during
high-pressure CO oxidation on Pt(110) (Hendriksen and Frenken,
2002). The upper panel traces mass spectrometer signals for O2, CO,
and CO2, showing initial exposure to CO, followed by the introduction
of molecular oxygen into the reactor. The panel below shows represen-
tative STM images obtained at specific stages of the reaction, during
which the sample is kept at a constant temperature of 425 K. Images A,
B, E, F, and H show flat terraces separated by steps, representing the
metallic, CO-covered Pt(110) surface. Image C shows the change in
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1001

Figure 15–23. Simultaneous mass spectrometry and STM imaging in a catalytic flow reactor: CO
oxidation on Pt(110). (Top) Mass spectrometer signals of O2, CO, and CO2 measured at the output of
the flow reactor cell. (Bottom) STM images on the Pt(110) catalyst surface acquired at selected stages
of the process: CO adsorption (A), O2 flow (B–D), and repetition of the sequence (E–H). Note the strong
surface roughening in (D), associated with increased CO2 evolution. (Reprinted with permission from
Hendriksen and Frenken, © 2002 by the American Physical Society.)

surface morphology, a pronounced surface roughening, during a step

in activity giving rise to a sudden increase in CO2 evolution, demon-
strating a direct link between surface roughness and activity for this
surface involving a mechanism that is not observed at low pressure.
Although capable of mapping dynamic surface phenomena, STM
movies have obvious limitations in imaging fast dynamic processes,
such as surface diffusion at room temperature or above. A possible
solution is the development of novel approaches and instruments for
very high-speed STM imaging. As an alternative, frame-by-frame
imaging can be abandoned altogether if only a map of the trajectory
of the diffusing species is desired. The recognition of this fact led to
the development of atom tracking STM (Swartzentruber, 1996). In atom
tracking, the STM tip is locked onto a diffusing surface species using
a 2D lateral feedback mechanism. Once locked, the feedback maintains
1002 P. Sutter

the tip over that species and tracks its coordinates as it diffuses over
the substrate. In the atom-tracking mode, the STM spends all of its time
measuring the diffusion trajectory, which results in substantially
improved time-resolution compared to time-lapse STM movies.
Atom tracking STM has been used to measure the diffusion kinetics
of Si (Swartzentruber, 1996) and SiGe (Qin et al., 2000) dimers on
Si(001) above room temperatures, of water molecules on Pd(111) (Mitsui
et al., 2002), and of Pd atoms in a Pd/Cu(001) surface alloy (Grant
et al., 2001). The latter example is illustrated in Figure 15–24. Pd atoms
in the surface alloy are imaged as protrusions in STM. Locking the
STM tip onto individual Pd atoms, their diffusion pathway can be
tracked. Analyzing the residence time (the time between hops) and
jump length leads to the conclusion that there is no time correlation
between individual Pd diffusion events, and that the diffusion is medi-
ated by surface vacancies rather than Cu adatoms, i.e., involves rapidly
diffusing vacancies visiting Pd atoms in the surface layer. From
the temperature dependence of the Pd hop rate—determined from
temperature-dependent tracking experiments—the activation energy
of the overall process, in this case equal to the sum of the vacancy for-
mation energy and the energy barrier for lateral Pd-vacancy exchange,
can be measured.
In contrast to time-lapse STM, in which the tip is scanned rapidly
across a larger field of view, the tip maintains close contact with the
diffusing entity in atom tracking. Hence, the observed diffusion process
could be affected by tip–sample interactions. While this question has
to be studied on a case-by-case basis, at least one system [SiGe dimer

Figure 15–24. Diffusion kinetics of Pd atoms in the Pd/Cu(001) surface alloy.

(a) Site visitation map of an individual Pd atom, obtained by atom-tracking
STM at a temperature of 62°C. The square array marks the position of the
Cu(001) unit mesh. In this dataset the atom hopped 853 times in a time interval
of 5557 s. (b) Temperature dependence between 31 and 69°C of the average hop
rate of an incorporated Pd atom. The average residence time of Pd atoms
decreases from 145.3 to 5.0 s in this temperature range. The data follow an
Arrhenius form with an activation energy of 0.88 eV and measured prefactor
of 1012.4±0.4 Hz. (Reprinted with Permission from Grant et al., © 2001 by the
American Physical Society.)
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1003

diffusion on Si(001)] has been identified in which the diffusion mecha-

nism depends on the sign of the electric field between tip and sample
(Sanders et al., 2003), indicating that the presence of the probe tip can
indeed affect the measurement.

4.3 Atom and Molecule Manipulation

Atom and molecule manipulation (Hla and Rieder, 2003) experiments
utilize the sharp STM probe tip and the ability to control it laterally
and vertically with picometer resolution to build and modify
nanometer-scale structures, typically in combination with their analy-
sis by STM imaging and tunneling spectroscopy. While conceptually
related to atom tracking, manipulation experiments are performed in
a different regime: the sample is cooled to temperatures low enough
that all thermal diffusion is frozen out, and the tip is typically brought
into close contact to induce strong tip–adatom/molecule interactions.
Since the first demonstration of atomic manipulation of Xe/Ni(110)
by Eigler and Schweizer (1990), manipulation experiments have
branched out considerably, and now encompass scenarios as diverse as
controlled chemical reactions of individual molecules (Lee and Ho,
1999), contacting single molecules with atomic metal wires (Nazin et
al., 2003), construction of mechanical logic gates from adsorbed mole-
cules (Heinrich et al., 2002), and control of excess charge on individual
atoms at insulator surfaces (Repp et al., 2004). While generally based
on some form of tip–sample interaction, atom or molecule manipula-
tion can involve a number of different physical mechanisms: close
proximity and short-range chemical interaction between a tip atom and
adatom to affect the potential landscape seen by the adatom on the
surface, vibrational excitation within adsorbed molecules or between
substrate and adatom, electric field, or direct charging by tunneling
electrons. Different patterns of adsorbate motion can be induced,
including lateral hopping between adsorption sites on the substrate,
vertical transfer between sample and tip, rotation, as well as the con-
trolled making and breaking of individual bonds. Substrate surfaces
with relatively low symmetry, e.g., (110) surface orientations or stepped
surfaces, are often used to establish one-dimensional diffusion path-
ways between substrate adsorption sites that help guide the manipula-
tion process.
Lateral manipulation is based on establishing close proximity
between an adatom and the STM tip to increase the interaction between
them. The approach process is controlled via the tunneling resistance,
which is lowered from typically several hundred MΩ during imaging
to values of the order of 100 kΩ for manipulation. Depending on the
particular system, the lateral force needed to move an adsorbate
between adjacent adsorption sites can be either repulsive or attractive,
giving rise to manipulation by pushing and pulling, respectively. The
manipulation process itself is monitored by measuring the tip height
(or z-piezo voltage) during the lateral motion of the tip, as illustrated
in Figure 15–25 for Pb/Cu(211) (pulling) and CO/Cu(211) (pushing)
(Meyer and Rieder, 1998).
1004 P. Sutter

Figure 15–25. Mechanisms of lateral atom manipulation. (a) “Pulling” manip-

ulation via attractive interaction between a Pb atom on Cu(211) and the W tip.
(b) “Pushing” manipulation via repulsive interaction between a CO adsorbate
and the W tip. In both cases a sawtooth-like tip height profile with the peri-
odicity of the substrate adsorption sites is observed (Reprinted with permis-
sion from Meyer and Rieder, © 1998 The Materials Research Society). (c)
Illustration of the tip and adsorbate motion during manipulation via attractive
interaction. (i) Lateral motion of the tip, accompanied by an approach toward
the substrate. (ii) Hopping of the adsorbate, followed by a sharp retraction of
the tip.

The adsorbate motion in both pushing and pulling modes is strongly

influenced by the preferred adsorption sites defined by the substrate.
A sawtooth-like vertical motion of the tip accompanies each jump of
the adsorbate between neighboring adsorption sites (Figure 15–25a and
b). The origin of this tip motion is illustrated in Figure 15–25c for
the example of an attractive tip–adsorbate interaction. Following the
approach to the adsorbate, the tip is moved laterally. Initially, the adsor-
bate, held in the potential well of a substrate adsorption site, does not
follow. The tunneling gap thus increases, and the tip moves forward
to maintain a constant tunneling current (i). Ultimately, the interaction
with the tip induces a lateral jump of the adsorbate into a neighboring
potential minimum. The tunneling gap closes abruptly and the feed-
back loop causes the tip to retract.
The precision and complexity of structures achievable by lateral
manipulation is illustrated in Figure 15–26 (Heinrich et al., 2002). CO
molecules adsorbed in monomer, dimer, or trimer configurations on
Cu(111) give rise to distinct contrast in constant current STM. While
monomers and dimers are stable at low temperature (5 K in this
example), there are three distinct configurations for trimers: a stable
three-fold symmetric “close-packed” arrangement, and metastable
“straight-line” and “bent-line” (“chevron”) configurations, which relax
with time constants of the order of seconds into the stable state. Complex
structures consisting of up to 545 CO molecules were built with atomic
precision such that any neighboring molecules were initially in stable
dimer configurations, but could be changed to an unstable “chevron”
configuration with a place exchange of a single molecule in the vicinity.
In this way, molecular cascades, similar to toppling rows of standing
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1005

dominoes, could be constructed. Mechanical computation was demon-

strated by molecular cascades set up as logic AND gates, two-input,
and three-input sorters.
Apart from lateral manipulation, vertical manipulation, i.e., transfer
of adsorbates between sample and tip, is possible. Vertical manipula-
tion of CO molecules, for example, has been used for controlled modi-
fication of the STM tip and to induce single molecule reactions (Lee
and Ho, 1999). Vertical transfer is also important for moving adatoms
or molecules across obstactles such as step edges, if they cannot be
surmounted by controlled lateral hops. Manipulation mechanisms
other than direct chemical interaction between the STM tip and adsor-
bates have received increased interest recently. Notably the different
roles of vibrational excitations are actively investigated (Komeda et al.,
2002; Stroscio and Celotta, 2004). Inelastic tunneling can, for instance,
drive controlled rotations of adsorbed molecules (Stipe et al., 1998b).
An intriguing example of atomic manipulation based on inelastic tun-
neling, the controlled manipulation of excess charge on single Au
atoms, is illustrated in Figure 15–27 (Repp et al., 2004). Individual Au

Figure 15–26. Molecule cascades. (Left) (A) Configurations of neighboring CO molecules on Cu(111):
isolated CO molecule, dimer, and trimer in a “chevron” configuration. Large circles mark the positions
of the molecules and small dots indicate the surface layer Cu atoms (1.9 nm scans; T = 5 K). (B) The
same area after one CO molecule in the trimer has hopped to generate a stable close-packed trimer.
(Right) Demonstration of a molecular mechanical AND gate, implemented via cascades of “chevron”-
type CO trimers. (A) Model of the AND gate. (B–D) Sequence of STM images (5.1 nm by 3.4 nm)
showing the operation of the AND gate: (B) Initialization. (C) Result after input X was triggered with
the STM tip. (D) When input Y was triggered, the cascade propagated all the way to the output. (E)
Result of triggering only input Y. (Reprinted with permission from Heinrich et al., © 2002 AAAS.)
1006 P. Sutter

Figure 15–27. Charging of individual Au atoms on ultrathin NaCl/Cu(111). (a) Constant-current STM
on a pair of Au atoms, one of which was charged negatively with the other remaining neutral. (b)
Signature of the charging event. Successful charging, achieved by a voltage pulse while maintaining
the tip above a single Au atom, is indicated by a sudden drop in tunneling current. (c and d) Quasi-
particle scattering: neutral Au adatoms do not scatter NaCl/Cu(111) interface-state electrons (c),
whereas the negatively charged adatom acts as a scatterer (d). (Reprinted with permission from Repp
et al., © 2004 AAAS.)

atoms are deposited on ultrathin, insulating NaCl supported by Cu(111).

Electrons can tunnel between the metal substrate and the STM tip
through the ultrathin insulator, i.e., STM imaging, spectroscopy, and
manipulation are possible for this system. However, the coupling of
the Au adsorbate to the substrate is affected profoundly by the insulat-
ing support. Voltage pulses can be used as a means to reversibly deposit
an excess electron on individual Au atoms. The charging occurs via an
inelastic electron tunneling mechanism enabled by a weak coupling of
the adatom and Cu substrate electronic states. The data suggest that
the coupling is so weak that the lifetime of a negative ion resonance
state of the adatom is in the range of ionic vibrational periods for the
NaCl interlayer, allowing the relaxation of the NaCl lattice, shift of the
negative ion resonance state below the Fermi energy, and capture of
the electron. The excess charge is maintained due to the substantial
relaxation of the underlying NaCl lattice. This stabilization prevents
discharging by tunneling into the metal, and the electron resides on
the Au atoms until removed by a voltage pulse of opposite sign. The
charged Au atom has a distinct signature in constant-current STM
imaging, and the long-range electric field due to the charged Au atom
strongly scatters electrons in interface states at the Cu/NaCl interface.
Charging individual adsorbed atoms is a potentially powerful approach
to tuning their physical properties. For Au/NaCl, significant differ-
ences in surface diffusivity were identified, with the onset temperature
of significant surface diffusion lowered from 60 K (Au) to 40 K (Au−).
Other possible modifications due to controlled charging at the atomic
level include changes in catalytic activity and the controlled manipula-
tion of the net spin magnetic moment of individual atoms.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1007

5 Heterostructures and Buried Interfaces: BEEM,

Quantum Size Effects, and Cross-Sectional STM

The exponential dependence of the tunneling current on the tip–sample

separation makes STM a versatile tool for atomic-resolution micros-
copy and spectroscopy at surfaces. As a result of its unique resolution
and surface sensitivity STM has brought important advances in fields
such as catalyis or epitaxial growth, in which surface processes play a
key role. For many technological applications of semiconductor materi-
als, e.g., in device structures such as transistors, detectors, or lasers, the
active regions encompass complex heterostructures, and device per-
formance is affected much less by the free surface than by buried
interfaces. Therefore, a need arises for a technique capable of probing
subsurface structures, electronic properties, and carrier transport. The
mapping of interfacial and transport properties should occur with
nanometer spatial resolution to provide data that are relevant for semi-
conductor devices with progressively reduced dimensions. Since the
early 1990s, several STM-based techniques have been developed to
provide high-resolution imaging and spectroscopy of buried interfaces
and heterostructures. Ballistic electron emission microscopy (BEEM)
operates in a three-terminal configuration, in which the STM tip, whose
height is controlled by a constant-current feedback loop, injects hot
carriers into a thin film or heterostructure while an additional collector
contact measures the current of carriers that are transmitted through
buried interfaces. Electron interference in a thin metal film, again
with carriers injected from an STM tip and traveling ballistically in
the metal, provides detailed thickness maps of metallic overlayers,
and can—under favorable circumstances—even map the atomic struc-
ture of a buried metal–semiconductor interface. Cross-sectional STM,
finally, is used to image complex heterostructures such as superlat-
tices, embedded self-assembled quantum dots, or substitutional mag-
netic impurities, on nonpolar (110) cleavage planes of III–V compound
semiconductors.

5.1 Probing Buried Interfaces (I)—BEEM

BEEM was invented by Kaiser and Bell (1988) to probe Schottky con-
tacts, i.e., metal–semiconductor interfaces with high spatial resolution
(for a recent review, see Narayanamurti and Kozhevnikov, 2001).
Beyond Schottky barriers, the technique can determine the height of
other subsurface potential barriers and it has also been applied to
measure band offsets between different semiconductors, the energies
of transmission resonances in semiconductor quantum wells and
superlattices, and bound states in buried quantum wires and dots. In
addition, electron scattering at subsurface linear and point defects has
been characterized.
We illustrate the operating principle of BEEM using the example of
a metal–semiconductor junction. The technique operates in a three-
terminal configuration, i.e., adds an additional collector electrode to the
STM tip and sample contact, as shown schematically in Figure 15–28.
1008 P. Sutter

Figure 15–28. Schematic setup

of a typical BEEM experiment
on a metal (M)–semiconductor
(S) heterostructure.

The STM tip is scanned at constant current It, measured between the
tip and base electrode, over the surface of the base layer. The tip serves
as a point source of hot carriers, electrons or holes, injected into the
metal base. If the metal thickness is small compared to the mean free
path for electron–phonon, electron–electron, or impurity scattering,
most carriers reach the metal–semiconductor interface ballistically, i.e.,
with their original energy and momentum distribution. In metals,
the mean free path for electron–phonon scattering of electrons with
energies of a few eV is typically of the order of 10 nm. In significantly
thicker films, electron–phonon scattering would broaden the momen-
tum distribution of the injected carriers, i.e., mostly affect the spatial
resolution of BEEM, while having little effect on the energy distribu-
tion. Hence, the injected ballistic carriers generally have a well-defined
energy distribution and can be used to perform spectroscopic mea-
surements on buried potential barriers.
At the metal–semiconductor interface, ballistic carriers with energies
below a threshold equal to the Schottky barrier height eφB are reflected
back into the metal base, while carriers with energy above the thresh-
old are transmitted into the semiconductor collector (Figure 15–29). If

Figure 15–29. Band diagrams for (a)

ballistic electron and (b) ballistic hole
injection into the metal base on an
n- and p-type semiconductor, respec-
tively. The dark shaded areas indicate
the energy distribution of the injected
carriers, as well as of those transmitted
into the collector. The inset in (a) illus-
trates the onset of the collector current
at a threshold bias corresponding to
the Schottky barrier height, eφB. (After
Bell et al., 1991.)
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1009

other barriers exist in the collector layer, the transport through that
material can involve additional interfacial reflections. Carriers that are
transmitted through the entire collector layer contribute to the collector
current (or BEEM current), which is measured between base and col-
lector contacts.
The energy of the injected carriers is varied by changing the voltage
applied between the STM tip and metal base. For low bias, none of the
injected carriers is transferred across the metal–semiconductor inter-
face. At voltages above the threshold, some carriers have energy above
the Schottky barrier and can cross the interface, causing an increase in
collector current with increasing tip–sample bias above threshold.
From the onset of the BEEM current, the barrier height eφB can be
determined, e.g., by fitting the measured Ic(V) to a theoretical expres-
sion. In a simple one-dimensional theory,
I c (V ) = RIt ∫ dE [ f ( E ) − f ( E − eV )] × θ [ E − ( EF − eV + eφ B )] (6)
where R is a bias independent parameter and f(E) is the Fermi
function.
Figure 15–30 shows experimental BEEM spectra on an Au/n-Si(001)
junction, measured at room temperature in N2 atmosphere, fitted using
Eq. (6) to determine a Schottky barrier eφB = 0.92 eV. Due to its low
reactivity and oxidation resistance, Au was used in many of the early
BEEM experiments as a nonepitaxial metal base on a variety of other
semiconductors. Schottky barrier heights were determined, for example,
for Au/n-GaAs(001) (eφB = 0.70 eV at 77 K) (Bell et al., 1990) and Au/n-
GaP(110) (eφB = 1.41 eV) (Prietsch and Ludeke, 1991). For Au/n-CdTe(001)
(eφB = 0.69–1.07 eV) (Fowell et al., 1990), strong lateral variations in the

Figure 15–30. BEEM spectra obtained on a polycrystalline Au/n-Si(001) junc-

tion. Spectra a–c (symbols) are measured at different tunneling currents. The
calculated spectra (solid lines) correspond to a common Schottky barrier
height value φB = 0.92 eV and R value of 0.045 eV−1. (Reprinted with permission
from Kaiser and Bell, © 1988 by the American Physical Society.)
1010 P. Sutter

measured Schottky barrier height were detected, emphasizing the

need for high-resolution maps of interfacial transport.
Carrier transport in BEEM is affected by all stages of the injection
pathway: (1) tunneling from the tip into the metal base layer, (2) hot
carrier transport through the base, (3) transmission across the metal–
semiconductor interface, and (4) transport in the semiconductor, includ-
ing transmission through additional heterostructure interfaces. In
addition to local Schottky barrier heights, the technique is therefore
sensitive to a number of factors, e.g., surface topography, elastic or
inelastic scattering in the different layers or at interfaces, and the band
structures of the junction partners, which affect not only interface
transmission probabilities but also the spatial resolution obtainable in
BEEM via metal band structure-induced focusing or defocusing.
BEEM current maps can be acquired simultaneously with constant-
current STM images by measuring Ic spatially resolved during a con-
stant-current scan. Figure 15–31 shows STM and BEEM current images
obtained on 2.5 nm epitaxial CoSi2/Si(111) (Sirringhaus et al., 1994). The
lattice mismatch of 1.2% between the metallic silicide and the Si sub-
strate is accommodated by an interfacial dislocation network. In STM
topography, the location of these line defects is detected via their elastic
strain field at the CoSi2 surface. Strikingly, the defects are mapped in
BEEM as sharply localized regions with increased collector current
with a width of only 0.8 nm, as expected for ballistic transport of car-
riers with a strongly forward focused tunneling momentum distribu-
tion, i.e., k|| ∼ 0. While the Schottky barrier was found to be uniform,
as expected for an epitaxial interface, the sharp local increase in trans-
missivity of the CoSi2/Si interface was explained by a significant
increase in in-plane momentum due to scattering by the dislocation
core, facilitating the transmission into the Si conduction band
minima.

Figure 15–31. (a) Constant-current

STM image and (b) corresponding
BEEM image obtained on epitaxial
2.5 nm CoSi2/Si(111). A dislocation
network at the silicide/silicon inter-
face is indicate by dashed lines. In
the BEEM image brighter areas
indicate regions of higher collector
current. Hot electron scattering at
the interfacial dislocations causes a
sharply localized increase in collec-
tor current.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1011

Figure 15–32. BEEM on self-assembled quantum dots. (a) STM and BEEM images of an individual
InAs/GaAs quantum dot under a polycrystalline Au base. (b) Comparison of BEEM spectra obtained
by ballistic electron injection into an InAs dot and into the wetting layer between dots. (Reprinted
with permission from Rubin et al., © 1996 by the American Physical Society.)

In experiments such as those on CoSi2/Si(111), growth and probing

of the entire heterostructure in the same UHV environment is critical
to achieve ordered interfaces and reliable measurements on reactive
metal surfaces. For the microscopy itself, low temperatures have several
practical advantages, such as improved energy resolution due to
reduced thermal broadening of the tunneling distribution, and reduced
thermal drift of the STM. An additional benefit is a lower thermal noise
due to a reduction of thermionic emission across interfacial barriers,
which allows shallower potential barriers to be probed. As a result,
semiconductor band offsets, resonant transport in semiconductor het-
erostructures, and interfacial barriers between an organic layer and a
metal base (Troadec et al., 2005) can be measured successfully by low-
temperature BEEM. Examples are embedded self-assembled quantum
dots (Rubin et al., 1996), lithographically patterned quantum wires
(Eder et al., 1996), double-barrier resonant tunneling structures (Sajoto
et al., 1995), and superlattices (Heer et al., 1998). Figure 15–32 shows
the band profile for carrier injection into an individual InAs quantum
dots in GaAs (Rubin et al., 1996). Also shown are STM and BEEM
images, as well as BEEM spectra obtained with the tip positioned on
top and next to the dot. The semiconductor heterostructure was coated
with a polycrystalline Au base layer, whose morphology largely domi-
nates the contrast in STM. A strong enhancement in BEEM current in
a circular region with 30 nm diameter is associated with carrier trans-
port through a single InAs quantum dot. A comparison of BEEM
spectra obtained on and between dots shows signatures of transport
through two zero-dimensional states of the dot.
Ballistic carrier transport and scattering can, in principle, be used
to probe a wide variety of other systems, beyond Schottky barriers
and semiconductor heterostructures. Examples are metal–insulator–
semiconductor structures (Cuberes et al., 1994) or magnetic multilayers
(Rippard and Buhrmann, 2000). In the latter, an Au/Si(111) Schottky
barrier is used as an analyzer for spin-dependent scattering of carriers
1012 P. Sutter

in a Co/Cu/Co trilayer structure, used in spin-valve devices, inte-

grated on top of the Au base layer. Unpolarized carriers injected from
an STM tip into this magnetic multilayer base will undergo spin-
dependent scattering in the ferromagnetic Co layers. In areas in which
the Co layers couple ferromagnetically, only one spin component will
be scattered heavily and a large fraction of unscattered carriers is
transmitted across the Au/Si interface. If the magnetization direction
of the two Co layers is misaligned, both spin components are strongly
attenuated by scattering, causing a sharp drop in collector current.

5.2 Probing Buried Interfaces (II)—Quantum Size Effects

In BEEM hot carriers, locally injected into a metal/semiconductor het-
erostructure, are used to measure the transmitted current across a
buried interface. Instead, carriers that impinge with energies below
threshold and are reflected back into the metal film can also be con-
sidered. Due to the long mean free path, these carriers can set up
standing waves in the metal film. An STM tip, probing the evanescent
tails into the vacuum, can then be used to image fringes due to inter-
ference of these electrons. The concept of electron interference in metal
films dates back to Jaklevic et al. (1971), and interference effects have
been observed by angle-resolved photoemission, low-energy electron
diffraction and low-energy electron microscopy.
An implementation of an STM experiment to probe interference
effects in the system Pb/Si(111) is shown in Figure 15–33. Pb evapo-
rated onto a stepped Si(111) substrate forms (111)-oriented islands
whose surface is atomically flat. Due to the substrate vicinality, the Pb

Figure 15–33. Electron interference in a Pb “quantum wedge” on Si(111). (Left) Geometry of the
Pb wedge on a stepped Si(111) substrate. The Pb thickness varies in increments of roughly one Si(111)
step height, giving rise to electron interference fringes in the direction of the substrate steps. (Right)
Constant-current STM images (730 nm × 1100 nm) at −5 V (a) and +5 V (b) sample bias. The image
obtained at positive bias shows apparent height changes at the surface of the wedge due to electron
interference in the Pb film. (Reprinted with permission from Altfeder et al., © 1997 by the American
Physical Society.)
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1013

island is wedge shaped, with a thickness that changes in integer mul-

tiples of the Si(111) step height from one substrate terrace to the next.
While STM at positive sample bias, i.e., injection of electrons from the
tip, images the atomically smooth Pb surface, STM images at opposite
bias show bands of apparent terraces and steps, aligned with steps of
the Si substrate, at the surface of the Pb wedge. These images, and
associated tunneling spectra, are interpreted as signatures of quantum
well states in the Pb wedge (Altfeder et al., 1997), and can be used to
determine the position of subsurface steps as well as the position-
dependent absolute thickness of the Pb film. Bands with constructive
and destructive interference alternate with a thickness change d0 of one
Pb(111) monolayer if d0 ≈ λF/4, where λF denotes the Fermi wavelength.
Similar interference effects were also observed by STM for epitaxial
silicide layers, such as CoSi2/Si(111) (Lee et al., 1994) and NiSi2/Si(111)
(Kubby and Greene, 1992).
Strikingly, electron interference can even be used to image interfacial
atomic structures buried under as much as 10 nm of metal. In the Pb/
Si(111) system, the (7 × 7) reconstruction of the Si(111) surface remains
essentially intact upon low temperature evaporation of Pb, except for
some intermixing by replacing Si adatoms by Pb (Altfeder et al., 1998).
Due to the topology of the Fermi surface of Pb, in particular a large
mismatch between the electron effective mass in in-plane and normal
directions, the quantized electron states in the Pb film can be used to
map interfacial structure with a resolution of 0.6 nm for overlayer thick-
nesses exceeding 10 nm, or roughly 10 times the Fermi wavelength in
the metal.

5.3 Imaging Buried Heterostructures—Cross-Sectional STM

The strong interest in low-dimensional semiconductor structures—
quantum wells, wires, and dots—has stimulated widespread activity
in nanoscale imaging of electronic materials with reduced dimension-
ality. Recent efforts have focused on self-assembled quantum dots,
generated by lattice mismatched heteroepitaxial growth. Semiconduc-
tor quantum dots with lateral size in the 10–100 nm range are readily
imaged by STM if they are exposed as islands on a free surface. Con-
sequently, a large number of studies have been devoted to studying
epitaxial growth and quantum dot self-assembly by conventional STM
imaging. However, almost any technological applications of self-
assembled quantum dots require embedding in a matrix, often consist-
ing of the substrate material. The embedding process causes significant
modifications to the dots that include segregation and intermixing,
shape changes, dopant redistribution, and adjustments to the local
strain field in the dot and in the surrounding material. All these
factors make it desirable to image embedded rather than exposed
nanostructures.
Cross-sectional STM (X-STM), originally demonstrated by Feenstra
et al. (1987) for imaging and spectroscopy on (110) surfaces of III–V
compound semiconductors, offers an elegant solution to this chal-
lenge. Semiconductors that cleave easily, as most III–V compounds do,
1014 P. Sutter

are used as a substrate for the growth of embedded self-assembled

quantum dots. A sample is then cleaved in UHV, and STM imaging is
performed on the cleavage face, typically a nonpolar (110) plane, which
provides large step-free areas for atomic resolution imaging. In this
geometry, X-STM gives access to subsurface structures over the entire
thickness of the epitaxial layer and the underlying substrate.
X-STM has been used to image a variety of buried semiconductor
heterostructures, such as superlattice structures in GaAs/AlGaAs
(Salemink and Albrektsen, 1991), InAs/GaSb (Feenstra et al., 1994), and
GaN/GaAs (Goldman et al., 1996), showing atomic-scale composition
fluctuations, interface roughness, lateral stain variations, and phase
separation in these systems. The technique has seen a strongly revived
interest with the advent of self-assembled quantum dots and quantum
dot superlattices (Legrand et al., 1998). Specifically for the imaging of
quantum dots, X-STM relies on the fact that the dots are small and form
rather dense populations. Hence, a random cleavage will cut through
a large number of these nanostructures and provide a cross-sectional
view of their atomic structure on the cleavage plane (Figure 15–34).
Much of the power of X-STM imaging derives from the fact that bias-
dependent imaging provides chemical contrast on (110) cleavage faces
of III–V compounds (Feenstra et al., 1987). Empty-state imaging gives
atomically resolved maps of the cation sublattice and allows a direct
identification of atomic species, e.g., indium atoms in a GaAs matrix
(Pfister et al., 1995). This electronic structure effect not only provides
strong contrast to determine the shape of individual buried InAs
quantum dots and their stacking in multilayer structures, but can also
be used to quantify interface roughness, intermixing, and surface seg-
regation, near the dots and the wetting layer, with atomic precision.
To access this information in high-resolution images, a background
subtraction has to be performed to remove topographic contrast of the

Figure 15–34. Cross-sectional STM on self-assembled quantum dots. (Left) Illustration of the possible
apparent geometries observed due to cleavage at random positions through quantum dots with pyra-
midal and truncated pyramidal shape. (Right) Filled-state constant-current STM of an InAs quantum
dot embedded in GaAs. Part of the image in (a) is treated by a local mean equalization filter to accen-
tuate the atomic corrugations in the dot and the surrounding GaAs matrix, as shown in (b). (Reprinted
with permission from Bruls et al., © 2002 American Institute of Physics; reprinted with permission
from Gong et al., © 2004 Amercian Institute of Physics.)
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1015

Figure 15–35. Imaging of individual dopants by X-STM. (a) Filled-state constant-current STM image
(V = −1.6 V) of cleaved GaAs(110). Individual SiGa substitutional donors appear as protrusions of dif-
ferent apparent height. (b) Band diagram illustrating the electron tunneling process between a metal
tip and the GaAs(110) surface in the presence of tip-induced band bending. The Coulomb potential of
a donor ion locally alters the band bending, causing an increase in tunneling current above the donor.
(Reprinted with permission from Zheng et al., © 1994 by the American Physical Society.)

pronounced local deformation of the cleavage plane due to the relax-

ation of the strained dots and matrix (Davies et al., 2002). Conversely,
when combined with modeling, the elastic relaxation at the free cleav-
age surface itself can be used to quantitatively determine the strain
field in and around individual dots.
Apart from the imaging of buried semiconductor quantum struc-
tures, an application of X-STM that has been receiving increasing atten-
tion is the mapping of electronic dopants for electronics (Figure 15–35)
and, more recently, of magnetic dopants for spintronics applications.
Buried substitutional Zn and Be acceptors in p-GaAs have been imaged
as protrusions in filled-state constant-current STM via a contrast mech-
anism assigned to an increased local state density directly above a
dopant (Johnson et al., 1993). Si donors at Ga sites (SiGa) in GaAs have
been imaged on GaAs(110) as protrusions, a few nanometers in size
and with discrete values of apparent height, superimposed on the
background lattice. The observed contrast in filled-state constant
current STM was attributed to a local perturbation of the near-surface
band bending by the Coulomb potential of the SiGa. The discrete appar-
ent heights were interpreted as a consequence of a distribution of the
donor atoms over five subsurface layers beneath the cleavage surface
(Zheng et al., 1994).
Deep acceptor states due to magnetic impurities in III–V semicon-
ductors, such as substitutional MnGa in GaAs, are expected to play an
important role in the hole-mediated coupling between magnetic impu-
1016 P. Sutter

rities, and thus determine the magnetic properties of hole-mediated

ferromagnetic semiconductors such as Ga1−xMn xAs, important for
emerging spintronics applications. X-STM at room temperature was
used to map the wave function of the hole bound to an individual Mn
acceptor in GaAs (Yakunin et al., 2004). Via bias-induced changes to
the local band bending, the acceptor could be imaged in both the
neutral (US = +0.6 V) and ionized state (US = −0.7 V). The acceptor
ground state has a highly anisotropic structure due to a significant
contribution of d-wave envelope functions, which is well-reproduced
by simulated images based on a tight-binding model of the Mn accep-
tor structure (Tang and Flatté, 2004).

6 STM Image Simulation

Although STM is a very powerful experimental technique in its own

right, geometric and spectroscopic contrast tend to be difficult to sepa-
rate, as discussed previously. In many cases, the unequivocal identifi-
cation of surface structures or of adsorption geometries requires a
comparison of experimental STM images with contrast simulations.
The conventional approach to STM image simulations follows a two-
step process. Relaxed atomic positions of candidate surface structures
are calculated by ab initio theoretical methods, such as density func-
tional theory. The actual STM contrast is then computed using the
Tersoff–Hamann theory of STM (Tersoff and Hamann, 1983, 1985).
By combining bias-dependent atomic-resolution STM images with
simulated images for various candidate structures, even completely
unknown surface structures can in principle be “solved” on the basis
of STM imaging alone. The difficulty lies less with the microscopy than
with the identification of plausible candidate surface structures. In the
past, this key step has typically been approached intuitively, i.e., by
guessing structures based on minimizing the density of dangling
bonds, starting from a truncated bulk structure. Recently developed
systematic techniques, based on stochastic optimization, for generating
comprehensive sets of candidate structures promise to become a pow-
erful tool for solving a wide range of surface structures by STM imaging
and image simulation (Ciobanu and Predescu, 2004).
The Tersoff–Hamann theory, the basis of most STM image simula-
tions to date, provides a transfer matrix formalism to calculate the
tunneling matrix element and tunneling current for realistic configura-
tions of tip and sample. For values of tip–sample separation typical in
STM imaging (∼1 nm), the coupling between tip and sample wave func-
tions is weak, and the tunneling process can be treated by first-order
perturbation theory. Within this framework the tunneling current is
given by
2 πe
I= ∑ [ f (Eµ ) − f (Eν )] Mµν δ (Eν + V − Eµ )

µ,ν
2π 2
e V ∑ Mµν δ ( Eµ − EF ) δ ( Eν − EF )
2
≈ (7)

µ,ν
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1017

where f(E) is the Fermi function, V denotes the tunneling bias, Mµv the
tunneling matrix element between a tip state ψµ and a sample state ψv,
and Eµ(v) the energy of state ψµ(v). In the second line, the Fermi function
was replaced by a step function (zero-T approximation), and the limit
of low tunneling bias has been assumed. If, after Bardeen (1961), the
tunneling matrix element is written as a surface integral, and the
sample and tip wavefunctions are expanded in plane waves, one
obtains
4 π 2
2
d q a qb q* e − κ q z t e iq ⋅x t
m ∫
M µv = − (8)

Here q is the Fourier wavevector and aq, bq are expansion coefficients

in the plane wave expansion of the sample and tip wavefunctions, xt
and zt denote the lateral and vertical tip position, and κq is the decay
constant. Given the wavefunctions of sample and tip, this expression
provides the tunneling matrix element, from which the tunneling
current can be calculated via Eq. (7).
In practice, a difficulty arises from the fact that the atomic-scale
geometry and chemical composition of the tip are unknown. Thus,
assumptions have to be made as to the tip wavefunctions (i.e., bq). For
an s-wavefunction of the tip (as for an ideal “point”–tip), the tunneling
current is
I ∝ ∑ ψ ν ( rt ) δ (E v − E F ) ≡ ρ ( rt , E F )
2
(9)
i.e., is proportional to the local density of states of the sample at the
Fermi energy, a property of the sample surface alone! For finite tun-
neling bias, V, the tunneling current can be written as an integral over
the energy range between the Fermi energy EF and EF + V:
eV
I= ∫ ρt (E ) ρs (E − e V )T (E , e V )d Et (10)
0

and the tunneling conductance at bias V is proportional to the posi-

tion-dependent density of states of the sample at energy eV from the
Fermi energy
dI
( )
dV V
≈ ρS (e V ) ρT (0)T (e V ,V ) (11)

Contrast calculations thus involve the computation of the position- and

energy-dependent sample DOS to obtain simulated maps of tunneling
conductance dI/dV(x,y) or tunneling current I(x,y).
Although originally derived for small voltages, as used for STM
imaging on metals, with this extension the Tersoff–Hamann formalism
can be applied to simulate images at higher bias as well, and has
proven a powerful simulation tools for a wide range of imaging sce-
narios. Tromp et al. (1986) were the first to show that it could be applied
to semiconductor surfaces, as demonstrated by their successful contrast
simulation for the Si(111)-(7 × 7) reconstruction. The systems simulated
since then include surface structures of semiconductors (Fujikawa
et al., 2002; Klijn et al., 2003), ultrathin insulators (Olsson et al., 2005),
1018 P. Sutter

as well as adsorbed atoms (Repp et al., 2004) and molecules (Olsson et

al., 2003; Kühnle et al., 2002).
Figures 15–36 and 15–37 illustrate two of the numerous examples of
this approach in the literature.
Shallow Ge quantum dots self-assembled during heteroepitaxy on
Si(001) invariably have surface termination by (105) facets, with surface
normal just 11° away from [001]. These (105) facets appear extremely
stable, likely due to a low density of dangling bonds and additional
strain stabilization by surface strain compensating for some of the 4%
lattice mismatch strain between the Ge overlayer and the Si substrate.
To calculate the surface energy, and thus explain the stability of the
facet, a detailed knowledge of the surface structure is necessary. Dual
bias constant-current STM combined with STM contrast calculations
was used to identify a best match with one of two proposed structures
of the (105) surface: “paired dimer” (Structure A) and “rebonded step”
(Structure B) (Fujikawa et al., 2002). The STM contrast is clearly identi-
fied as that of a “rebonded step” structure, the structure that not only
has the lowest dangling bond density but also causes tensile surface
strain, key to the strain stabilization of the (105) facet.

Figure 15–36. Identification of the surface structure of Ge(105). Empty- and

filled-state constant-current STM images are compared with two candidate
structures, the paired dimer Structure A and rebonded step Structure B (c and
d). A Tersoff–Hamann calculation of the STM contrast for these two structures
clearly shows that the experimental STM images arise from the rebonded step
structure. (Reprinted with permission from Fujikawa et al., © 2002 by the
American Physical Society.)
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1019

Figure 15–37. Identification of the molecular conformation of Cu-DTBPP on

Cu(211). (a) Molecular model of Cu-DTBPP, with the four-lobed pattern
observed by STM marked in yellow. (b–e) STM contrast calculation for differ-
ent angles between the four legs and the substrate: (b) 60°, (c) 45°, (d) 30°, and
(e) 10°. (f) Experimental STM image of the molecule. (Reprinted from Moresco
et al., ©2002 with permission from Elsevier.) (See color plate.)

Simulations of STM contrast of adsorbed molecules pose a somewhat

different problem than simulations of surface structures. For solid
surfaces, the combination of STM imaging with image simulation can
discriminate between several candidate structures, as discussed in the
previous example. For molecules, the structure is typically well known.
However, the adsorption geometry (site, orientation) as well as the
conformation for larger, more flexible molecules enter as unknowns
into the simulation. A relatively simple example is the stiff molecular
cage structure of fullerene C60 (Hou et al., 1999; Pascual et al., 2000; Lu
et al., 2003). In this case, the structure and conformation are well
known, and the only parameter of the contrast calculation is the ori-
entation. A more complex system, Cu-tetra(3,5-di-tert butyl phenyl)
porphyrin (Cu-DTBPP) on Cu(211), is illustrated in Figure 15–37.
Cu-DTBPP consists of a central porphyrin ring with four symmet-
rically attached di-tert-butyl phenyl (DTBP) groups (Figure 15–37a).
These four bulky groups determine the shape of the molecule, which
is imaged in STM as a four-leaf clover structure, and define the interac-
tion with the metal substrate (Jung et al., 1996). The orientation of the
side groups is the main parameter that needs to be optimized to deter-
mine the conformation of the adsorbed molecule, and to simulate the
experimental STM images. Figure 15–37 shows several candidate con-
formations, with leg angles of 60° (b), 45° (c), 30° (d), and the optimized
angle of 10° relative to the substrate (e), which was found to best repro-
duce the experimental STM image of the molecule.
1020 P. Sutter

Acknowledgments. The compilation of this chapter was supported

by the U.S. Department of Energy under Contract No. DE-AC02-
98CH10886.

References
Altfeder, I.B., Matveev, K.A. and Chen, D.M. (1997). Phys. Rev. Lett. 78, 2815.
Altfeder, I.B., Chen, D.M. and Matveev, K.A. (1998). Phys. Rev. Lett. 80, 4895.
Bardeen, J. (1961). Phys. Rev. Lett. 6, 57.
Bauer, E. (1994). Rep. Prog. Phys. 57, 895.
Bauer, E. (1998). Surf. Rev. Lett. 5, 1275.
Becker, R.S., Swartzentruber, B.S., Vickers, J.S. and Klitsner, T. (1989). Phys. Rev.
B 39, 1633.
Bell, L.D., Hecht, M.H., Kaiser, W.J. and Davis, L.C. (1990). Phys. Rev. Lett. 64,
2679.
Bell, L.D., Kaiser, W.J., Hecht, M.H. and Davis, L.C. (1991). J. Vac. Sci. Technol.
B 9, 594.
Binnig, G. and Rohrer, H. (1983). Surf. Sci. 126, 236.
Binnig, G., Rohrer, H., Gerber, Ch. and Weibel, E. (1982). Appl. Phys. Lett. 40,
178.
Binnig, G., Rohrer, H., Gerber, Ch. and Weibel, E. (1983). Phys. Rev. Lett. 50,
120.
Binnig, G., Garcia, N. and Rohrer, H. (1985). Phys. Rev. B 32, 1336.
Binnig, G. and Smith, D.P.E. (1986). Rev. Sci. Instrum. 57, 1688.
Bode, M., Getzlaff, M. and Wiesendanger, R. (1998). Phys. Rev. Lett. 81, 4256.
Boland, J.J. (1993). Science 262, 1703.
Boland, J.J. and Weaver, J.H. (1998). Phys. Today 51, 34.
Bonnell, D.A., Ed. (2001). Scanning Probe Microscopy and Spectroscopy—Theory,
Techniques, and Applications, 2nd ed., Wiley-VCH, New York.
Borovski, B., Krueger, M. and Ganz, E. (1997). Phys. Rev. Lett. 78, 4229.
Borovski, B., Krueger, M. and Ganz, E. (1998). Phys. Rev. B 57, 4269.
Brookes, I.M., Muryn, C.A. and Thornton, G. (2001). Phys. Rev. Lett. 87,
266103.
Bruls, D.M., Vugs, J.W.A.M., Roenraad, P.M., Salemink, H.W.M., Wolter, J.H.,
Hopkinson, M., Skolnick, M.S., Long, F. and Gill, S.P.A. (2002). Appl. Phys.
81, 1708.
Cerezo, A., Larson, D.J. and Smith, G.D.W. (2001). MRS Bull. 26, 102.
Chambliss, D.D., Wilson, R.J. and Chiang, S. (1995). IBM J. Res. Dev. 39, 639.
Chen, C.J. (1993). Introduction to Scanning Tunneling Microscopy (Oxford Uni-
versity Press, Oxford).
Chen, Y., Ohlberg, D.A.A., Medeiros-Ribeiro, G., Chang, Y.A. and Williams,
R.S. (2000). Appl. Phys. Lett. 76, 4004.
Ciobanu, C.V. and Predescu, C. (2004). Phys. Rev. B 70, 085321.
Crain, J.N. and Pierce, D.T. (2005). Science 307, 703.
Crommie, M.F., Lutz, C.P. and Eigler, D.M. (1993). Nature 363, 524.
Cuberes, M.T., Bauer, A., Wen, H.J., Prietsch, M. and Kaindl, G. (1994). J. Vac.
Sci. Technol. B 12, 2646.
Cygan, M.T., Dunbar, T.D., Arnold, J.J., Bumm, L.A., Shedlock, N.F., Burgin,
T.P., Jones, L., Allara, D.L., Tour, J.M. and Weiss, P.S. (1998). J. Am. Chem.
Soc. 120, 2721.
Davies, J.H., Bruls, D.M., Vugs, J.W.A.M. and Koenraad, P.M. (2002). J. Appl.
Phys. 91, 4171.
Diebold, U. (2003). Surf. Sci. Rep. 48, 53.
Ebert, Ph., Cox, G., Poppe, U. and Urban, K. (1992). Surf. Sci. 271, 587.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1021

Eder, C., Smoliner, J. and Strasser, G. (1996). Appl. Phys. Lett. 68, 2876.
Eigler, D.M. and Schweizer, E.K. (1990). Nature 344, 524.
Erwin, S.C., Baski, A.A. and Whitman, L.J. (1996). Phys. Rev. Lett. 77, 687.
Feenstra, R.M., Stroscio, J.A., Tersoff, J. and Fein, A.P. (1987). Phys. Rev. Lett.
58, 1192.
Feenstra, R.M., Collins, D.A., Ting, D.Z.-Y., Wang, M.W. and McGill, T.C.
(1994). Phys. Rev. Lett. 72, 2749.
Festy, F. and Palmer, R.E. (2004). Appl. Phys. Lett. 85, 5034.
Fowell, A.E., Williams, R.H., Richardson, B.E. and Shen, T.-H. (1990). Semicond.
Sci. Technol. 5, 348.
Fujikawa, Y., Akiyama, K., Nagao, T., Sakurai, T., Lagally, M.G., Hashimoto,
T., Morikawa, Y. and Terakura, K. (2002). Phys. Rev. Lett. 88, 176101.
Gai, Z., Li, X., Zhao, R.G. and Yang, W.S. (1998). Phys. Rev. B 57, 15060.
Gewirth, A.A. and Niece, B.K. (1997). Chem. Rev. 97, 1129.
Gimzewski, J.K., Joachim, C., Schlittler, R.R., Langlais, V., Tang, H. and
Johannsen, I. (1998). Science 281, 531.
Goldman, R.S., Feenstra, R.M., Briner, B.G., O’Steen, M.L. and Hauenstein, R.
J. (1996). Appl. Phys. Lett. 69, 3698.
Gong, Q., Offermans, P., Nötzel, R., Koenraod, P.M. and Loolter, J.H. (2004).
Appl. Phys. 85, 5697.
Grant, M.L., Swartzentruber, B.S., Bartelt, N.C. and Hannon, J.B. (2001). Phys.
Rev. Lett. 86, 4588.
Gregory, S. (1990). Phys. Rev. Lett. 64, 689.
Güntherodt, H.-J. and Wiesendanger, R., Eds. (1992). Scanning Tunneling
Microscopy Vol. I & II (Springer, Berlin).
Hamers, R.J. (1989). Annu. Rev. Phys. Chem. 531.
Hamers, R.J., Tromp, R.M. and Demuth, J.E. (1986). Phys. Rev. Lett. 56, 1972.
Hanaguri, T., Lupien, C., Kohsaka, Y., Lee, D.-H., Azuma, M., Takano, M.,
Takagi, H. and Davis, J.C. (2004). Nature 430, 1001.
Hasegawa, Y. and Avouris, Ph. (1993). Phys. Rev. Lett. 71, 1071.
Heer, R., Smoliner, J., Strasser, G. and Gornik, E. (1998). Appl. Phys. Lett. 73,
3138.
Heinrich, A.J., Lutz, C.P., Gupta, J.A. and Eigler, D.M. (2002). Science 298, 1381.
Heinze, S., Bode, M., Kubetzka, A., Pietzsch, O., Nie, X., Blügel, S. and Wiesen-
danger, R. (2000). Science 288, 1805.
Hendriksen, B.L.M. and Frenken, J.W.M. (2002). Phys. Rev. Lett. 89, 046101.
Hla, S.-W. and Rieder, K.-H. (2003). Annu. Rev. Phys. Chem. 54, 307.
Hoffman, J.E., McElroy, K., Lee, D.-H., Lang, K.M., Eisaki, H., Uchida, S. and
Davis, J.C. (2002a). Science 297, 1148.
Hoffman, J.E., Hudson, E.W., Lang, K.M., Madhavan, V., Eisaki, H., Uchida, S.
and Davis, J.C. (2002b). Science 295, 466.
Horch, S., Lorensen, H.T., Helveg, S., Lægsgaard, E., Stensgaard, I., Jacobsen,
K.W., Nørskov, J.K. and Besenbacher, F. (1999). Nature 398, 134.
Hornbaker, D.J., Kahng, S.-J., Misra, S., Smith, B.W., Johnson, A.T., Mele, E.J.,
Luzzi, D.E. and Yazdani, A. (2002). Science 295, 828.
Horn-von Hoegen, M. (1994). Appl. Phys. A 59, 503.
Hou, J.G., Jinlong, Y., Haiqian, W., Qunxiang, L., Changgan, Z., Hai, L., Wang,
B., Chen, D.M. and Qinshi, Z. (1999). Phys. Rev. Lett. 83, 3001.
Hovis, J.S. and Hamers, R.J. (1997). Surf. Sci. 402, 1.
Jaklevic, R.C. and Lambe, J. (1966). Phys. Rev. Lett. 17, 1139.
Jaklevic, R.C., Lambe, J., Mikkor, M. and Vassell, W.C. (1971). Phys. Rev. Lett.
26, 88.
Jensen, J.A., Rider, K.B., Chen, Y., Salmeron, M. and Somorjai, G.A. (1999). J.
Vac. Sci. Technol. B 17, 1080.
1022 P. Sutter

Johnson, M.B., Albrektsen, O., Feenstra, R.M. and Salemink, H.W.M. (1993).
Appl. Phys. Lett. 63, 2923.
Jung, T.A., Schlittler, R.R., Gimzewski, J.K., Tang, H. and Joachim, C. (1996).
Science 271, 181.
Kaiser, W.J. and Bell, L.D. (1988). Phys. Rev. Lett. 60, 1406.
Kasaya, M., Tabata, H. and Kawai, T. (1998). Surf. Sci. 400, 367.
Kawamura, M., Paul, N., Cherepanov, V. and Voigtländer, B. (2003). Phys. Rev.
Lett. 91, 096102.
Klijn, J., Sacharov, L., Meyer, C., Blügel, S., Morgenstern, M. and Wiesendan-
ger, R. (2003). Phys. Rev. B 68, 205327.
Kochanski, G.P. (1989). Phys. Rev. Lett. 62, 2285.
Komeda, T., Kim, Y., Kawai, M., Persson, B.N.J. and Ueba, H. (2002). Science
295, 2055.
Kubby, J.A. and Greene, W.J. (1992). Phys. Rev. Lett. 68, 329.
Kühnle, A., Linderoth, T.R., Hammer, B. and Besenbacher, F. (2002). Nature
415, 891.
Kuipers, L., Hoogeman, M.S. and Frenken, J.W.M. (1993). Phys. Rev. Lett. 71,
3517.
Kuk, Y. and Silverman, P.J. (1986). Appl. Phys. Lett. 48, 1597.
Lægsgaard, E., Österlund, L., Rasmussen, P.B., Stensgaard, I. and Besenbacher,
F. (2001). Rev. Sci. Instrum. 72, 3537.
Laracuente, A. and Whitman, L.J. (2001). Surf. Sci. 476, L247.
Lee, E.Y., Sirringhaus, H. and von Känel, H. (1994). Phys. Rev. B 50, 5807.
Lee, H.J. and Ho, W. (1999). Science 286, 1719.
Legrand, B., Grandidier, B., Nys, J.P., Stiévenard, D., Gérard, J.M. and Thierry-
Mieg, V. (1998). Appl. Phys. Lett. 73, 96.
Li, L., Tindall, C., Takaoka, O., Hasegawa, Y. and Sakurai, T. (1997). Phys. Rev.
B 56, 4648.
Lopinski, G.P., Moffatt, D.J., Wayner, D.D.M. and Wolkow, R.A. (1998). Nature
392, 909.
Lu, X., Grobis, M., Khoo, K.H., Louie, S.G. and Crommie, M.F. (2003). Phys.
Rev. Lett. 90, 096802.
Mayne, A.J., Avery, A.R., Knall, J., Jones, T.S., Briggs, G.A.D. and Weinberg,
W.H. (1993). Surf. Sci. 284, 247.
Medeiros-Ribeiro, G., Bratkovski, A.M., Kamins, T.I., Ohlberg, D.A.A. and
Williams, R.S. (1998). Science 279, 353.
Meinel, K., Wolter, H., Ammer, Ch., Beckmann, A. and Neddermeyer, H.
(1997). J. Phys. Condens. Matter 9, 4611.
Men, F.K., Packard, W.E. and Webb, M.B. (1988). Phys. Rev. Lett. 61, 2469.
Meyer, G. and Rieder, K.-H. (1997). Surf. Sci. 377–379, 1087.
Meyer, G. and Rieder, K.-H. (1998). MRS Bull. 23, 28.
Mitsui, T., Rose, M.K., Fomin, E., Ogletree, D.F. and Salmeron, M. (2002).
Science 297, 1850.
Mo, Y.-W., Swartzentruber, B.S., Kariotis, R., Webb, M.B. and Lagally, M.G.
(1989). Phys. Rev. Lett. 63, 2393.
Mo, Y.-W., Savage, D.E., Swartzentruber, B.S. and Lagally, M.G. (1990). Phys.
Rev. Lett. 65, 1020.
Mo, Y.-W., Kleiner, J., Webb, M.B. and Lagally, M.G. (1991). Phys. Rev. Lett. 66,
1998.
Moresco, F., Meyer, G., Rieder, K.-H., Ping, J., Tang, H. and Joachim, C. (2002).
Surf. Sci. 499, 94.
Muller, E.W. and Tsong, T.T. (1969). Field Ion Microscopy (Elsevier, New York).
Nagl, C., Haller, O., Platzgummer, E., Schmid, M. and Varga, P. (1994). Surf.
Sci. 321, 237.
 Chapter 15 Scanning Tunneling Microscopy in Surface Science 1023

Nakayama, K. and Weaver, J.H. (1999). Phys. Rev. Lett. 82, 980.
Narayanamurti, V. and Kozhevnikov, M. (2001). Phys. Rep. 349, 447.
Nazin, G.V., Qiu, X.H. and Ho, W. (2003). Science 302, 77.
Nunes, G., Jr. and Amer, N.M. (1993). Appl. Phys. Lett. 63, 1851.
Olsson, F.E., Lorente, N. and Persson, M. (2003). Surf. Sci. 522, L27.
Olsson, F.E., Persson, M., Repp, J. and Meyer, G. (2005). Phys. Rev. B 71,
075419.
Pascual, J.I., Gomez-Herrero, J., Rogero, C., Baro, A.M., Sanchez-Portal, D.,
Artacho, E., Ordejon, P. and Soler, J.M. (2000). Chem. Phys. Lett. 321, 78.
Pfister, M., Johnson, M.B., Alvarado, S.F., Salemink, H.W.M., Marti, U., Martin,
D., Morier-Genoud, F. and Reinhard, F.K. (1995). Appl. Phys. Lett. 67, 1459.
Pietzsch, O., Kubetzka, A., Bode, M. and Wiesendanger, R. (2001). Science 292,
2053.
Prietsch, M. and Ludeke, R. (1991). Phys. Rev. Lett. 66, 2511.
Qin, X.R. and Lagally, M.G. (1997). Science 278, 1444.
Qin, X.R., Swartzentruber, B.S. and Lagally, M.G. (2000). Phys. Rev. Lett. 85,
3660.
Rasmussen, P.B., Hendriksen, B.L.M., Zeijlemaker, H., Ficke, H.G. and Frenken,
J.W.M. (1998). Rev. Sci. Instrum. 69, 3879.
Repp, J., Meyer, G., Olsson, F.E. and Persson, M. (2004). Science 305, 493.
Repp, J., Meyer, G., Stojkovic, S.M., Gourdon, A. and Joachim, C. (2005). Phys.
Rev. Lett. 94, 026803.
Rippard, W.H. and Buhrmann, R.A. (2000). Phys. Rev. Lett. 84, 971.
Robinson, K.M., Robinson, I.K. and O’Grady, W.E. (1992). Surf. Sci. 262, 387.
Rößler, M., Geng, P. and Wintterlin, J. (2005). Rev. Sci. Instrum. 76, 023705.
Rost, M.J., Crama, L., Schakel, P., van Tol, E., van Velzen-Williams, G.B.E.M.,
Overgauw, C.F., ter Horst, H., Dekker, H., Okhuijsen, B., Seynen, M., Vijft-
igschild, A., Han, P., Katan, A.J., Schoots, K., Schumm, R., van Loo, W.,
Oosterkamp, T.H. and Frenken, J.W.M. (2005). Rev. Sci. Instrum. 76,
053710.
Rubin, M.E., Medeiros-Ribeiro, G., O’Shea, J.J., Chin, M.A., Lee, E.Y., Petroff,
P.M. and Narayanamurti, V. (1996). Phys. Rev. Lett. 77, 5268.
Sajoto, T., O’Shea, J.J., Bhargava, S., Leonard, D., Chin, M.A. and Narayana-
murti, V. (1995). Phys. Rev. Lett. 74, 3427.
Salemink, H. and Albrektsen, O. (1991). J. Vac. Sci. Technol. B 9, 779.
Sanders, L.M., Stumpf, R., Mattson, T.R. and Swartzentruber, B.S. (2003). Phys.
Rev. Lett. 91, 206104.
Schaub, R., Thostrup, P., Lopez, N., Stensgaard, I., Nørskov, J.K. and
Besenbacher, F. (2001). Phys. Rev. Lett. 87, 266104.
Schaub, R., Wahlström, E., Rønnau, A., Lægsgaard, E., Stensgaard, I. and
Besenbacher, F. (2003). Science 299, 377.
Schintke, S. and Schneider, W.-D. (2004). J. Phys: Cond. Matter 16, R49.
Schunack, M., Linderoth, T.R., Rosei, F., Lægsgaard, E., Stensgaard, I. and
Besenbacher, F. (2002). Phys. Rev. Lett. 88, 156102.
Sirringhaus, H., Lee, E.Y. and von Känel, H. (1994). Phys. Rev. Lett. 73, 577.
Smith, D.P.E. and Binnig, G. (1986). Rev. Sci. Instrum. 57, 2630.
Stipe, B.C., Rezaei, M.A. and Ho, W. (1998a). Science 280, 1732.
Stipe, B.C., Rezaei, M.A. and Ho, W. (1998b). Science 279, 1907.
Stroscio, J.A. and Celotta, R.J. (2004). Science 306, 242.
Sutter, P., Schick, I., Ernst, W. and Sutter, E. (2003a). Phys. Rev. Lett. 91,
176102.
Sutter, P., Zahl, P., Sutter, E. and Bernard, J.E. (2003b). Phys. Rev. Lett. 90,
166101.
Sutter, E., Sutter, P. and Bernard, J.E. (2004). Appl. Phys. Lett. 84, 2262.
1024 P. Sutter

Swartzentruber, B.S. (1996). Phys. Rev. Lett. 76, 459.

Swartzentruber, B.S., Mo, Y.-W., Kariotis, R., Lagally, M.G. and Webb, M.B.
(1990). Phys. Rev. Lett. 65, 1913.
Takayanagi, K., Tanishiro, Y., Takahashi, S. and Takahashi, M. (1985). Surf. Sci.
164, 367.
Tang, J.-M. and Flatté, M.E. (2004). Phys. Rev. Lett. 92, 047201.
Tersoff, J. and Hamann, D.R. (1983). Phys. Rev. Lett. 50, 1998.
Tersoff, J. and Hamann, D.R. (1985). Phys. Rev. B 31, 805.
Troadec, C., Kunardi, L. and Chandrasekhar, N. (2005). Appl. Phys. Lett. 86,
72101.
Tromp, R.M. (2000). IBM J. Res. Dev. 44, 503.
Tromp, R.M., Hamers, R.J. and Demuth, J.E. (1985). Phys. Rev. Lett. 55, 1303.
Tromp, R.M., Hamers, R.J. and Demuth, J.E. (1986). Phys. Rev. B 34, 1388.
Venables, J.A. (2000). Introduction to Surface and Thin Film Processes (Cambridge
University Press, Cambridge).
Vershinin, M., Misra, S., Ono, S., Abe, Y., Ando, Y. and Yazdani, A. (2004).
Science 303, 1995.
Voigtländer, B. (2001). Surf. Sci. Rep. 43, 127.
Voigtländer, B. and Zinner, A. (1993). Appl. Phys. Lett. 63, 3055.
Wang, Y., Chen, X. and Hamers, R.J. (1994a). Phys. Rev. B 50, 4534.
Wang, Y., Bronikowski, M.J. and Hamers, R.J. (1994b). Surf. Sci. 311, 64.
Weierstall, U. and Spence, J. (1998). Surf. Sci. 398, 267.
Wiesendanger, R. (1994). Scanning Probe Microscopy and Spectroscopy—Methods
and Applications (Cambridge University Press, Cambridge).
Wintterlin, J., Trost, J., Renisch, S., Schuster, R., Zambelli, T. and Ertl, G. (1997).
Surf. Sci. 394, 159.
Wittneven, Chr., Dombrowski, R., Morgenstern, M. and Wiesendanger, R.
(1998). Phys. Rev. Lett. 81, 5616.
Wolkow, R.A. (1999). Annu. Rev. Phys. Chem. 50, 413.
Wouda, P.T., Niewenhuys, B.E., Schmid, M. and Varga, P. (1996). Surf. Sci. 359,
17.
Wouda, P.T., Schmid, M., Nieuwenhuys, B.E. and Varga, P. (1998). Surf. Sci.
417, 292.
Yagi, K. (1982). Scan. Electron Microsc. 4, 1421.
Yakunin, A.M., Silov, A.Yu., Koenraad, P.M., Wolter, J.H., Van Roy, W., De
Boeck, J., Tang, J.-M. and Flatté, M.E. (2004). Phys. Rev. Lett. 92, 216806.
Yang, Y.-N., Fu, E.S. and Williams, E.D. (1996). Surf. Sci. 356, 101.
Zhang, Z.Y. and Lagally, M.G. (1998). Science 276, 377.
Zheng, J.F., Liu, X., Newman, N., Weber, E.R., Ogletree, D.F. and Salmeron, M.
(1994). Phys. Rev. Lett. 72, 1490.