Scanning tunneling microscopy of defect states in the semiconductor Bi2Se3

Abstract

Scanning tunneling spectroscopy images of Bi2Se3 doped with excess Bi reveal electronic defect states with a striking shape resembling clover leaves. With a simple tight-binding model, we show that the geometry of the defect states in Bi2Se3 can be directly related to the position of the originating impurities. Only the Bi defects at the Se sites five atomic layers below the surface are experimentally observed. We show that this effect can be explained by the interplay of defect and surface electronic structure.

Full text

arXiv:cond-mat/0112402v2 [cond-mat.mtrl-sci] 4 Sep 2002
Scanning Tunneling Microscopy of Defect States in the Semiconductor Bi2Se3
S. Urazhdin, D. Bilc, S. H. Tessmer, and S. D. Mahanti
Department of Physics and Astronomy, Michigan State University, East Lansing, MI 48824
Theodora Kyratsi and M. G. Kanatzidis
Department of Chemistry, Michigan State University, East Lansing, MI 48824
Scanning tunneling spectroscopy images of Bi2Se3doped with excess Bi reveal electronic defect
states with a striking shape resembling clover leaves. With a simple tight-binding model we show
that the geometry of the defect states in Bi2Se3can be directly related to the position of the
originating impurities. Only the Bi defects at the Se sites five atomic layers below the surface are
experimentally observed. We show that this effect can be explained by the interplay of defect and
surface electronic structure.
PACS numbers: 68.35.Dv, 68.37.Ef, 73.20.Hb, 71.55.Ht
Understanding the electronic properties of defects and
the ability to control them will be crucial for the perfor-
mance of the future microelectronic devices [1]. Scanning
Tunneling Microscopy (STM) represents a unique tool for
the studies of defects as it combines atomic scale resolu-
tion with local spectroscopic capability. However, STM
observation and analysis of defect states in semiconduc-
tors are complicated by surface effects such as in-gap sur-
face states and reconstruction. These effects are avoided
at the (110) surfaces of a number of III-V semiconduct-
ing systems [2], attracting extensive research [3]–[8]. A
number of point defect types have been observed. How-
ever, positions of these defects with respect to the surface
plane could be inferred only from indirect observations.
The interpretation of such observations is complicated by
the drastic effect the surface proximity may have on the
defect states [9].
Modeling STM measurements of defects in semicon-
ductors is not straightforward: Approximation of the
STM images by maps of the local surface electronic den-
sity of states [10] is justified only if the charge relaxation
rates of defect states significantly exceed the tunneling
rate of electrons between the tip and the sample [11].
Tip-induced effects also need to be taken into account.
These may include both local band bending [3], and
charging of the defect states by the tunneling current, re-
sulting in bias voltage-dependant lattice relaxation in the
vicinity of the defect atoms [8]. Careful analysis is nec-
essary to clearly separate these effects from the intrinsic
defect properties, and the bulk features of the observed
defect states from the surface effects.
In this paper we present cryogenic STM and scanning
tunneling spectroscopy (STS) studies of the layered nar-
row gap semiconductor Bi2Se3, which can be viewed as
a model system for the STM study of near-surface de-
fect states. The bonding scheme of Bi2Se3allows a di-
rect determination of the position of a subsurface defect
atom with respect to the surface: Atomic planes consist-
ing of either Bi or Se hexagonally arranged atoms are
stacked in a close-packing fcc fashion; 5 atomic planes
FIG. 1: a) Structure of Bi2Se3showing atomic ordering in a
layer. Arrow indicates the rhombohedral [111] layer stacking
direction. In the bulk, Se1 and Se3 positions are equivalent,
but we use Se3 notation for the bottom Se atomic plane of
the surface layer (with Se1 position at the surface). (b) A
schematic of the bonding into strongly pp σ interacting chains
of atoms, 5 atoms per layer. Black orbitals represent Bi; white
orbitals represent Se.
with atomic order Se1-Bi-Se2-Bi-Se1 (Fig. 1(a)) form a
layer. The layers are weakly bound to each other by
Se1-Se1 bonds. Both valence and conduction bands are
formed almost exclusively by the 4pand 6porbitals of
Se and Bi respectively [12]. For each atom, the closest
neighbors from the adjacent atomic planes form almost
a regular octahedron, so the bonding can be roughly ap-
proximated by strongly interacting ppσ chains of atomic
p-orbitals (Fig. 1(b)), with a weaker ppπ-type interac-
tion between adjacent chains. A substitutional defect is
therefore likely to produce a perturbation in the elec-
tronic local density of states (DOS) predominantly along
the three ppσ chains passing through the defect atom.
Hence, the defect state should be observed at the sur-
face as three spots of modified DOS around the atoms
terminating these chains at the surface.
We have performed STM and STS measurements using
a custom built low-temperature microscope with direct
immersion in liquid He-4 [13]. Stoichiometric Bi2Se3sin-
gle crystal samples, as well samples doped with 2–5%
excess Bi or Se were grown by a directional solidification
technique. The stoichiometric as-grown Bi2Se3samples
are n-type with carrier concentration of about 1019 cm−3.
Doping samples with excess Bi introduces substitutional
Bi defects at the Se sites (BiS e antisites), which are shal-

2
FIG. 2: (a) A 30 ×30 nm differential CITS map of a Bi-doped
sample. (b),(c) 3.5×3.5 nm topographic maps encompassing
one of the defect features. Sample bias voltages are −0.3 V
(b), and −0.6 V (c). (d) Schematic of atomic positions in (c).
low acceptors. However, because of the low solubility of
Bi in Bi2Se3[14], the Bi-doped samples are n-type due
to the compensating defects. Doping samples with ex-
cess Se introduces shallow donor-type Se substitutional
defects at the Bi sites (SeB i antisites). The samples are
sufficiently inert to obtain atomic resolution in air. How-
ever, to minimize the surface contamination, in the ex-
periments reported here, the samples were cleaved in situ
or in a glove box directly attached to the STM setup in
ultrapure He gas prior to transfer to the STM with sub-
sequent cooling to T=4.2 K.
To map out the defect states, we have performed dif-
ferential current imaging tunneling spectroscopy (CITS)
measurements [15]. The CITS maps were acquired by fix-
ing the tip at each point during topographic imaging and
measuring the differential conductance at various bias
voltages with the feedback loop disabled. Only Bi-doped
samples exhibited an appreciable density of observable
defect states, as shown in Fig. 2(a) for a sample doped
with 5% excess Bi (Bi2Se2.85 ). The map was acquired
with sample bias Vb=−0.2 V in the feedback mode (with
tunneling current set to 50 pA), and Vb=−0.45 V for the
conductivity measurement. The defects appear as reg-
ular clover-shaped bright features, indicating areas with
locally enhanced conductance at Vb=−0.45 V. Irregu-
lar spots in this image (mostly in the upper left corner)
resulted from topographic defects. The data discussed
below were obtained on a more weakly doped Bi2Se2.95
sample, where the defect density was reduced.
FIG. 3: (a) Differential conductance spectra acquired in the
vicinity of a clover-shaped defect at various distances from the
center along one of the lobes. Numbers on the right mark the
distance from the center in nanometers; curves are offset for
clarity. (b) Differential conductance of stoichiometric Bi2Se3
vs. calculation performed in a slab geometry as explained in
the text. (c) Calculated total bulk near-gap density of states
(DOS) vs. total DOS in the slab geometry.
Fig. 2(b),(c) present topographic maps of the sample
area encompassing an isolated clover-shaped defect state.
Topographic image Fig. 2(b), acquired at a sample bias
voltage Vb=−0.3 V, shows a periodic atomic structure,
indicating no significant structural variation associated
with the defect. The height of the atomic corrugations
in Fig. 2(b) is about 30 pm. Fig. 2(c) shows a topo-
graphic image of the same area acquired at Vb=−0.6 V,
where the largest corrugations, locally enhanced by tun-
neling through the defect state, are about 100 pm high.
The highest amplitude of the defect state correlates with
positions of three surface Se atoms (marked with larger
black circles in schematic Fig. 2(d)), forming a regular tri-
angle. These atoms terminate three ppσ-bonded chains
passing through the Se1 site five atomic layers below the
surface, for which we also use notation Se3. Since the ob-
served defects appear only in Bi-doped samples and they
originate from Se sites, we attribute them to the BiSe3
antisites.
Fig. 3(a) presents a series of differential conductance
spectra acquired in the vicinity of an isolated defect.
The spectra were obtained by numerical differentiation
of 60 I-V curves with setpoint parameters Vb=−0.3 V,
I=0.8 nA. At the measurement temperature of 4.2 K,
thermal broadening is negligible on the displayed bias
voltage scale. The defect state appears as a broad reso-
nance (indicated by an arrow) in the energy range where
the differential conductance is suppressed away from the
defect. The established semiconducting gap value is

3
about 0.3 eV [12], [14], therefore the defect levels appear
inside the valence band. Theoretical modeling is nec-
essary to understand this spectroscopic feature, as well
as why only BiSe antisites five atomic planes below the
surface are observed.
First, we performed ab initio calculations in the full
potential relativistic LAPW formalism [16] within LDA
approximation. To model the surface, a supercell geom-
etry was used, with distance between slabs (consisting of
15 atomic planes, or 3 layers, each) increased by 0.5 to
1.5 nm as compared to the bulk crystal structure. The
calculated band structure did not exhibit significant vari-
ation for the slab separation larger than 0.4–−0.5 nm,
therefore we found the slab separation of 0.7 nm, used
for the calculations presented below, sufficient for mod-
eling the surface. The differential conductance spectra
were approximated by the local DOS in the center of
the gap between the slabs [10]. The calculation pre-
sented in Fig. 3(b) was performed for a position above
a Se1 atom, although we found the variation of the cal-
culated spectra with position respective to the surface
atoms to be insignificant. The calculation reproduces
both the finite conductance in the bulk semiconducting
gap, and the suppressed conductance just below the gap.
In Fig. 3(c), the total DOS calculated in the slab ge-
ometry is compared to the calculation of the bulk DOS,
which reproduces the accepted semiconducting gap value
of 0.3 eV [17]. Band structure analysis indicates that
the highest valence band (HVB) states are predominantly
Se1-Se1 antibonding type. As the Se1-Se1 bonds are bro-
ken at the surface, the splitting of these states is reduced,
resulting in the observed suppression of the differential
conductance in the HVB energy range (Fig. 3(b)). The
states that appear in the gap have high dispersion along
the surface. They originate from the rehybridization of
the surface Se1 valence with Bi conduction states, bring-
ing the latter down below the bulk conduction band min-
imum.
ab initio calculations of a single defect state are com-
plicated by the large cluster or supercell size necessary to
model the impurity states without introducing artificial
interaction between defects. Instead we use a LCAO ap-
proximation as a simple model of the system [12]. This
model presents just a qualitative argument and is not
capable of reproducing the detailed electronic structure
or the semiconducting gap value. However, it gives a
surprisingly good qualitative agreement with the experi-
mental observations and first principles calculations. In
a tight binding formalism
Hψ =HX
i
uiφi=X
i
Eiuiφi+X
i6=j
Vij uiφj,(1)
where φiare atomic wave functions and Vij are off-
diagonal matrix elements of H. We approximate Eiby
the atomic term values [18] of Bi and Se, and take into ac-
count only ppσ interaction between the closest neighbors,
as shown in Fig. 1(b). Thus the problem is reduced to
a system of noninteracting one-dimensional chains, with
three matrix elements V1,V2, and V3, corresponding to
Se1-Bi, Se2-Bi, and Se1-Se1 ppσ bonds. Consider first a
5-atom chain Se1-Bi-Se2-Bi-Se3 (we also call it a unit),
representing a single layer in our approximation. Here
Se3 position is equivalent to Se1. The highest energy
filled state is nonbonding
ψ0=1
p2 + (V1/V2)2(φSe1+φSe3−V1/V2φSe2).(2)
We model the bulk by a long chain of Se1-Bi-Se2-Bi-
Se1 units, with the interaction between layers expressed
by V3. Due to the Se1-Se1 interaction, the nonbonding
level is split, and the HVB states become antibonding
in the sense of Se1-Se1 bond character. The splitting is
large, because the state Eq.(2) has a significant weight
on the interlayer Se1 atoms. As the interlayer bonds
are broken at the surface, the HVB states of a chain de-
cay at the surface (the end of the chain), as shown in
Fig. 4(a), where the antibonding character of the valence
band maximum (VBM) state can also be seen. In semi-
classical terms, as illustrated in Fig. 4(b), the surface gap
is larger than the bulk value. This effect is in agreement
with the more accurate first principles calculations and
spectroscopic measurements, Fig. 3(b), (c).
To highlight the importance of these surface effects for
the observation of the defect states, in Fig. 4(c) we plot
the calculated dependence of the near-gap energy levels
on the position of the BiSe antisite. The second layer
(positions 4–6) is only weakly affected by the proxim-
ity of the surface, so the defect level is split from the
VBM and its energy is only weakly dependant on the
position. As the antisite position approaches the surface
(positions 1–3), the defect level energy is reduced as the
surface gap opens up, so that the defect state merges
with the bulk valence states, forming a resonance inside
the bulk valence band. This behavior is supported by
the semiclassical picture shown in Fig. 4(b). Only the
BiSe3state is observed in the experiment, because, as
our model suggests, BiSe1and BiSe2states are so much
lowered in energy by the proximity of the surface that
they form small amplitude broad resonances in the va-
lence band. STM images do not exhibit defect features
associated with BiS e antisite in the second layer. Our
model suggests, that they form bound states in the bulk
gap, which cannot sustain STM current. Surface effects
thus provide a mechanism for the charge relaxation of
near-surface defect states through the bulk valence band.
Similar surface effects should be observable in other
semiconductors, e.g. at (110) surface of GaAs, where the
valence band is, like in Bi2Se3, suppressed at the sur-
face [2]. As a result, in-gap impurity states may become
resonances in the valence band, if the originating impuri-

4
FIG. 4: (a) Amplitudes ui(Eq. (1)) in the VBM state for a
16-unit chain plotted as a function of atomic position along
the chain. Only the first 20 amplitudes are shown. Solid black
circles —Se1 positions, gray circles — Bi positions, open cir-
cles — Se2 positions. (b) Schematic of a qualitative semiclas-
sical picture of the near-surface defect resonance formation.
(c) Calculated near-gap levels of a 16-unit chain plotted as a
function of the position of BiSe antisite. The 1st position is
at the surface.
ties are sufficiently close to the surface [9]. This suggests
an alternative explanation for some of the published re-
sults [5]. It may also be possible to induce the resonant
behavior of near-surface defects by careful control of the
surface band bending with doping and/or external field.
Resonances induced by near-surface defects can be con-
trasted to the bulk-like in-gap states. The origin of the
spectroscopic broadening of the latter [4], and the mech-
anisms of their charge relaxation, allowing their obser-
vation with STM, need further theoretical and experi-
mental studies [11]. Variable temperature studies of the
influence of the local defect distribution on the spectro-
scopic features of defect states may provide insight into
these issues, and Bi2Se3represents a convenient model
system for such studies.
In summary, we have observed clover-shaped defect
states in Bi2Se3doped with excess Bi, which appear
as resonances in the high valence band, and can be at-
tributed to BiSe antisites in fifth atomic layer from the
surface. In the analysis of these defect states, we have
demonstrated the importance of the surface effects for the
electronic structure of near-surface defects. While BiSe
defects in the bulk Bi2Se3form shallow acceptor levels,
the near-surface defects produce resonances in the energy
range of the bulk valence states, which are suppressed at
the surface. We suggest that similar surface effects are
likely to be observable in other semiconducting systems.
We thank J. Nogami, Norman O. Birge, M. I. Dyk-
man, T. Hogan, and S. Lal for helpful discussions. This
work was supported in part by NSF (DMR-0075230) and
ONR/DARPA (N00014-01-1-0728). SHT acknowledges
support of the Alfred P. Sloan Foundation.
[1] See, e.g. B. E. Kane, Nature 393, 133 (1998).
[2] Xuejun Zhu et al., Phys. Rev. Lett. 63, 2112 (1989).
[3] J. F. Zheng et al., Phys. Rev. Lett. 72, 1490 (1994).
[4] R. M. Feenstra, J. M. Woodall, and G. D. Pettit, Phys.
Rev. Lett. 71, 1176 (1993).
[5] B. Grandidier et al, Appl. Phys. Lett. 72, 2454 (1998).
[6] Chr. Wittneven it et al., Phys. Rev. Lett. 81, 5616
(1998).
[7] B. Grandidier et al., Appl. Phys. Lett. 74, 1439 (1999).
[8] Ph. Ebert, K. Urban, and M. G. Lagally, Phys. Rev. Lett.
72, 840 (1994).
[9] R. B. Capaz, K. Cho, and J. D. Joannopoulos, Phys. Rev.
Lett 75, 1811 (1995).
[10] J. Tersoff and D. R. Hamann, Phys. Rev. Lett. 50, 1998
(1983).
[11] X. de la Broise et al., Phys. Rev. B 61, 2138 (2000).
[12] S. K. Mishra, S. Satpathy, and O. Jepsen, J. Phys.: Cond.
Matt. 9, 461 (1997).
[13] S. H. Tessmer, D. J. Van Harlingen, and J. W. Lyding,
Rev. Sci. Instrum. 65, 2855 (1994).
[14] D. R. Lovett, Semimetals and Narrow-Bandgap Semicon-
ductors (Pion Limited, London, UK, 1977).
[15] J. A. Stroscio and R. M. Feenstra, in Scanning Tunneling
Microscopy, Methods in Experimental Physics Vol. 27
(Academic Press, New York, 1993).
[16] David J. Singh, Plane waves, Pseudopotentials and the
LAPW method, (Kluwer Academic Publishers, Boston,
1994).
[17] Although LDA approximation usually underestimates
the semiconducting gap value, it has been rather success-
ful in application to chalcogenide semiconductors. See,
e.g. P. Larson, S. D. Mahanti, and M. G. Kanatzidis,
Phys. Rev. B 61, 8162 (2000).
[18] W. A. Harrison, Electronic Structure and the Properties
of Solids (W. H. Freeman and Co., San Francisco, 1980).