Gas source homoepitaxy on Si(113) - The interrelation of H-induced reconstructions and growth morphology

Article abstract of DOI:10.1016/S0039-6028(98)00071-5

High-temperature, time-lapsed STM-studies are used to investigate the surface morphology and the growth mode of gas source MBE of Si on Si(113) with disilane (Si₂H₆) as a precursor. A varying hydrogen coverage on the growing surface caused by the dynamic equilibrium between hydrogen supply from the precursor and thermal H desorption at 480°C results in several different reconstructions. Phases with (2 x 7) and (2 x 5) at a low flux of disilane and a (2 x 2) at high flux were distinguished that consist of only two basic elements: a low-H-coverage double-row structure and the gradual insertion of high-H-coverage (2 x 2)-like heavy domain walls. Above H coverages resulting in the (2 x 5) larger domains of (2 x 2) separate, for which a model is suggested. Besides an anisotropic shape of the Si islands in all stages of growth, a strong influence of the H-induced reconstructions on the silicon growth mode is found. Whereas step flow dominates for the lower H-coverage structure (2 x 7) and (2 x 5), the formation of the (2 x 2) leads to a massive nucleation of Si islands arranged in a quasi-periodic way. This difference in nucleation behaviour results in a mechanism for a layer-by-layer-like growth mode, even on a surface with a very high hydrogen coverage in contrast to Si(111).

Full text of DOI:10.1016/S0039-6028(98)00071-5

Surface Science 401 (1998) L375–L382
Surface Science Letters
Gas source homoepitaxy on Si(113) – the interrelation of
H-induced reconstructions and growth morphology
V. Dorna *, Z. Wang 1, U. K o¨ hler
Institut f u¨ r Experimentalphysik IV, AG Oberfl a¨ chen, Ruhr-Universit a¨ t Bochum, 44780 Bochum, Germany
Received 10 November 1997; accepted for publication 12 January 1998
Abstract
High-temperature, time-lapsed STM-studies are used to investigate the surface morphology and the growth mode of gas source
MBE of Si on Si(113) with disilane (Si H ) as a precursor. A varying hydrogen coverage on the growing surface caused by the
2
6
dynamic equilibrium between hydrogen supply from the precursor and thermal H desorption at 480°C results in several different
reconstructions. Phases with (2×7) and (2×5) at a low flux of disilane and a (2×2) at high flux were distinguished that consist of
only two basic elements: a low-H-coverage double-row structure and the gradual insertion of high-H-coverage (2×2)-like heavy
domain walls. Above H coverages resulting in the (2×5) larger domains of (2×2) separate, for which a model is suggested. Besides
an anisotropic shape of the Si islands in all stages of growth, a strong influence of the H-induced reconstructions on the silicon
growth mode is found. Whereas step flow dominates for the lower H-coverage structure (2×7) and (2×5), the formation of the
(
2×2) leads to a massive nucleation of Si islands arranged in a quasi-periodic way. This difference in nucleation behaviour results
in a mechanism for a layer-by-layer-like growth mode, even on a surface with a very high hydrogen coverage in contrast to Si(111).
1998 Elsevier Science B.V. All rights reserved.
©
Keywords: Chemical vapor deposition; Epitaxy; Growth; High index single crystal surfaces; Scanning tunneling microscopy; Silane;
Silicon; Surface relaxation and reconstruction; Surface structure, morphology, roughness, and topography
Although silicon-based surface technology is still
dominated by (001) and (111) surfaces, high
indexed surfaces have also attracted increasing
interest in the last years. In particular, the thermal
stability of Si(113) has made it a possible candidate
for future technological applications.
The structure of the clean Si(113) surface was
investigated intensively by scanning tunnelling
microscopy in recent years [1–3]. A (3×2) recon-
struction with a high density of translational
domain walls was found to be the typical recon-
struction at room temperature. A model by
Dabrowski et al. [4] explains this reconstruction
and the observed transition between the (3×2)
and a (3×1) phase by a unit cell containing a
subsurface self-interstitial. The surface energy that
they calculated is small and close to the value for
a hypothetical Si(001)-p(2×2) [4]. In X-ray scat-
tering studies [5], however, only a (3×1)- and no
evidence of a (3×2) reconstruction was found.
Probably because of the small differences between
both reconstructions, the X-ray structure factor of
the (3×2) periodicity seems to be small compared
to that resulting from the (3×1) periodicity. The
calculated unit cell dimensions of this reconstruc-
*
Corresponding author. Fax: (+49) 234 7094 173;
e-mail: volker.dorna@ep4-of.ruhr.uni
Present address: Department of Applied Physics, Hunan
University, People’s Republic of China.
1
0
039-6028/98/$19.00 © 1998 Elsevier Science B.V. All rights reserved.
PII S0039-6028 ( 98 ) 00071-5

LETTERS
SURFACE SCIENCE
L376
V. Dorna et al. / Surface Science 401 ( 1998) L375–L382
tion are 1.152 nm along [ 1: 10] and 0.637 nm along
33 2: ].
Whereas the structure of the clean Si(113) seems
pumped UHV chamber. Disilane was supplied
[
through a standard UHV leak valve. The Si(113)
wafers were outgased at 600°C for 8 h and then
flashed several times up to 1250°C. After the last
flash, the samples were cooled down at a rate of
10°C s−1.
to be resolved, little is known about the growth
behaviour on this surface, even in homoepitaxy.
In this context, the behaviour of hydrogen plays
an important role because gas source MBE often
involves H-containing precursors. However, in
growth regimes where facetting on other Si surfaces
occurs, quite often (113)-type faces appear as
stable facets, i.e. in the case of the formation of
The time-resolved STM measurements were per-
formed in situ by taking snapshots of a selected
surface area during epitaxial growth at elevated
temperature. It has to be considered that the STM
tip, depending on its geometrical form, may more
or less shade the observed area and thus lower the
effective pressure of the precursor in this area. For
disilane, this effect is – for tips that are not too
blunt – negligible. Because of the low sticking
coefficient of about 1×10−3 ([16] on Si(111)) the
molecules can diffuse underneath the tip using
multiple collisions. Nevertheless, the features seen
in in-situ STM observations were compared with
images that were obtained without the presence of
a tip during growth to exclude tip-induced effects
on the growth process. All STM images shown in
this paper were obtained in the constant current
3d islands of Ge on Si [6] and especially during
homoepitaxial growth under the influence of
hydrogen [7]. The adsorption of hydrogen on
Si(113) was found to proceed in two steps investi-
gated by LEED, AES, ARUPS and HREELS [8–
11]. During the first step, only Si–H bonds are
formed, and the (3×2) turns into a (3×1). In the
second step, additionally, H–Si–H bonds are
formed, and the reconstruction changes into
(
1×1) at a higher dosis. In contrast to these data,
Wilson et al. [12] found, using STM studies, a
mixture of (2×2) and (2×3) after exposure to
atomic hydrogen at room temperature with no
signs of (3×1).
mode at 0.5 nA and a sample bias U of −2 V.
B
After preparation in UHV, the clean Si(113)
Although hydrogen adsorption on Si(113) was
intensively studied, little has been published in
context with epitaxial growth. On a low-index
silicon surface, the detailed mechanism of the
decomposition of disilane (Si H ) used as a precur-
surface shows a (3×2) reconstruction with a high
density of anti-phase boundaries at room temper-
ature (Fig. 1). The average miscut of the wafer
2
6
sor gas is known. On Si(111), disilane decomposes
above 420°C into 2SiH and 2H (g) on the surface
2
[
13]. Above 500°C, two additional reaction paths,
where Si and H are deposited as reaction products,
have been identified [13,14]. Also, the morphology
of the growing surface is known in the case of
Si(111) and Si(001) [15].
The maximum of the hydrogen desorption peak
on silicon is around 550°C. Therefore, we per-
formed experiments on Si(113), especially between
420 and 550°C, to determine the correlation
between the hydrogen-induced structure and the
growth mode on this surface. In this temperature
range, the competition between H supply by
adsorption of disilane and thermal desorption
leads to different hydrogen coverages, depending
on the temperature and growth rate.
Fig. 1. Clean (3×2)-reconstructed Si(113) surface at room tem-
perature with the typical high density of anti-phase boundaries
The experiments were performed in an ion-
walls (U =−2 V, 17×15 nm2).
B

SURFACE SCIENCE
LETTERS
V. Dorna et al. / Surface Science 401 ( 1998) L375–L382
L377
used as a substrate was 0.2°, resulting in a typical
terrace width of 100 nm.
surface structures become increasingly anisotropic.
As Fig. 2c shows, a further increase in disilane
pressure (8.5×10−6) intensifies this effect. At this
moderate flux of disilane, the effective diffusion
length of the adsorbed species along [ 1: 10] is still
in the order of the terrace width of the investigated
samples, and step flow is dominant, although some
string-shaped islands along the [ 1: 10] direction with
a length of 100 nm and a width of several nanome-
tres are seen. A surface reconstruction, visible as
streaks along the [ 1: 10] direction, which could not
be observed at lower pressures, is accompanied by
this growth mode. This will be investigated in
further detail below.
If the samples are exposed to disilane at an
elevated temperature and very low pressures in the
range of 5×10−9 mbar, a (3×1) reconstruction
becomes visible, which vanishes with increasing
flux of disilane. This effect can be reversed by
lowering the partial pressure. In contrast to the
static adsorption of H at room temperature [8,9],
the pressure dependence enables us to dynamically
adjust the hydrogen coverage of the surface.
As on Si(111), no continuous growth could be
observed below 420°C on (113). For an apprecia-
ble growth rate, the temperature has to be some
1
0°C above this threshold. For all temperatures
The size of the surface structures decreases fur-
ther with increasing disilane pressure, and finally
at pressures above 2×10−5 mbar, a massive forma-
tion of islands takes place, which are also charac-
terized by their strong anisotropy. At
3.7×10−5 mbar (Fig. 2d ), the typical length of
these islands along the [ 1: 10] direction is 10 nm,
between 450 and 550°C, we observed the same
growth modes – in this range of temperature, the
limiting process for the growth rate is the thermal
desorption of hydrogen [17] as on Si(111) and
Si(001). To concentrate on the disilane flux depen-
dence in this paper, only in-situ measurements at
4
80°C are presented.
For the investigation of dynamic processes, time-
whereas their width is only several angstroms. A
quasi-periodical arrangement of these islands per-
pendicular to the [ 1: 10] direction can be observed
resolved continuous scans of the same surface area
were performed to obtain information on changes
in surface morphology. In the course of an experi-
ment, the disilane pressure was increased succes-
sively. Fig. 2 shows four selected images of such a
sequence of STM scans, in which an overall view
of a surface area can be seen. The time interval
between the first and last frame is 44 min, the
scanned area is 140×140 nm2.
In Fig. 2a, several terraces separated by mono-
or diatomic steps are visible. The image was taken
at a partial pressure of 1.2×10−8 mbar. At such a
low flux of disilane, diffusion of adsorbed species
leads to a final accommodation of silicon only at
the step edges. Step flow perpendicular to the step
edges is predominant, but already a restructuring
of the steps takes place. The step edges consist of
smooth sections parallel and rough sections per-
pendicular to the [ 1: 10] direction. The [ 1: 10] and
(see below).
A 20-min exposure at a disilane pressure of
2×10−4 reduces the length of the nucleated
islands, but no roughening of the surface was
observed. This may be a sign that the growth is in
a state of equilibrium, and no facetting – in
contrast to the Si(111) surface [7,13] – takes place.
After quenching the disilane partial pressure, the
islands dissolve again, and the obtained images
are similar to Fig. 2a.
To better resolve the different atomic arrange-
ments in the different stages of growth, ‘‘zoomed-in
scans’’ with an atomic resolution area of
15×15 nm2 are shown in Fig. 3. A structure with
a (2×7) reconstruction is obtained at a pressure
of 6.5×10−6 mbar (Fig. 3a) that contains two
structural elements, one part consisting of three
rows along [ 1: 10], each row with a width of two
[
33 2: ] directions, which are parallel and perpendicu-
lar to the dimers in the (3×1) reconstruction, are
marked in Fig. 2a.
Fig. 2b shows the same surface area, after the
partial pressure was increased to 4.0×10−6 mbar.
Although growth still proceeds at step edges, the
unit cells and a ‘‘double chain’’-like structure
(marked with #) and another row with clearly
visible protrusions (marked with 6). The twofold
periodicity of both parts along [ 1: 10] in the unit
cell can be understood as being a consequence of
the dimerization within the (001)-like Si-sites of

LETTERS
SURFACE SCIENCE
L378
V. Dorna et al. / Surface Science 401 ( 1998) L375–L382
Fig. 2. Selected frames of a time-lapsed sequence of STM images (140×140 nm2) giving an overview of the morphology during
growth. The scans show the same surface area during deposition of silicon from disilane at 480°C. With increasing flux of disilane,
the step flow turns into an asymmetric restructuring of step edges (b, c) and the nucleation of quasi-periodically arranged islands
(d ). The corresponding disilane partial pressures are (a) 1.2×10−8, (b) 4.0×10−6, (c) 8.5×10−6, and (d ) 3.7×10−5 mbar.
the bulk truncated surface consisting of (111)- and
subsequent rows are shifted with respect to each
other resulting in a non-rectangular unit cell with
an angle of 73° (see Fig. 4). Whereas our model
reflects the correct symmetry of the unit cell, the
protrusions visible in the STM images (e.g. Fig. 3d)
are stretched along [33 2: ], which means perpendicu-
(
001)-like sites alternating along [33 2: ] (see below).
Let us first concentrate on the second structure
element (6) that was found to dominate at a
higher disilane flux forming a (2×2) reconstruc-
tion with a unit cell of 0.68×1.25 nm2, indicated
in Fig. 3d at
a
corresponding pressure of
lar to the direction of the (001)-like dimers.
Further STM measurements and calculated STM
images of the structural model together with the
information about the locations of hydrogen are
needed to determine which structural elements are
visible in the filled state images. We did not observe
an ordered structure different from (2×2) even at
a higher flux, and we especially did not observe a
(1×1). In these terms, the (2×2) structure is the
upper limit in hydrogen coverage during Si homoe-
pitaxy with disilane as a precursor.
1
.8×10−5 mbar.
We propose a structural model for this recon-
struction as shown in Fig. 4. The (2×2) unit cell
consists of a threefold coordinated ((111)-like) site
and a tetramer containing a (001)-like dimer along
[
1: 10] and has eight dangling bonds for possible
hydrogen adsorption. The structural elements are
the same as in the model proposed by Wilson et al.
[
12], obtained from a STM study of H adsorption
at 380°C; only the symmetry of the unit cell is
different. Whereas Wilson et al. propose a rectan-
gular unit cell, our data show that the dimers in
Returning to Fig. 3a, the surface structure can
then be assigned as a reconstruction with a lower

SURFACE SCIENCE
LETTERS
V. Dorna et al. / Surface Science 401 ( 1998) L375–L382
L379
Fig. 3. Scans at increasing disilane flux with atomic resolution (15×15 nm2, U =−2 V). (a, b) Low-coverage phase with (2×7)-
B
and (2×5) reconstructions at 6.5×10−6 and 7.5×10−6 mbar disilane partial pressure, (c) transition from (2×5) to (2×2) reconstruc-
tion (9.5×10−6 mbar) with nucleated islands, and (d) high-coverage phase with (2×2) reconstruction at 1.8×10−5 mbar. The images
are high-pass filtered.
hydrogen coverage that consists of two coexisting
phases, one row (half a unit cell) of the (2×2)-
like high-coverage phase (6) and the ‘‘double
chain’’ structure interpreted as a low-H-coverage
structure (#). Other in-situ and ex-situ measure-
ments (not shown here) indicate that the single
chains in this ‘‘double chain’’ structure are not
equivalent. One row appears to be higher and is
shifted against the other along the [ 1: 10] direction
and both have a poorly visible twofold periodicity
along [ 1: 10]. However, these data are not sufficient
for a structure model as yet.
Back to the course of the experiment, where we
gradually increase the flux of disilane. In addition
to the atomic details visible in Fig. 3, the develop-
ment of the overall morphology of the domain
structure is shown in Fig. 5.
Fig. 4. Structural model for the high-coverage phase. The unit-
cell of the (2×2) reconstruction is marked (see text).
A carefully increased disilane flux (Fig. 3b)

LETTERS
SURFACE SCIENCE
L380
V. Dorna et al. / Surface Science 401 ( 1998) L375–L382
Fig. 5. Rearrangement of the surface reconstruction on a larger scale with increasing disilane flux (40×40 nm2). (a) Restructuring of
the surface at low flux, (b) insertion of high-coverage domain walls with (2×2) resulting in a (2×5) reconstruction, (c) (2×2)
domains in coexistence with the low-H-coverage phase, (d) high-H-coverage phase with (2×2) reconstruction with a small fraction
of nucleated islands. The corresponding disilane partial pressures are (a) 2.0×10−6, (b) 8.2×10−6, (c) 9.4×10−6, and (d )
3
.0×10−5 mbar.
changes the (2×7) into a (2×5) surface recon-
struction with two, instead of three, low-coverage
a high-coverage phase with decreasing distances
between the heavy domain walls takes place. In
particular, no (2×3) superstructure was observed
in contrast to the case of hydrogen adsorption
reported in Ref. [12]. This means that a single
‘‘double chain’’ is not stable. Instead, increasing
the disilane flux further leads to the formation of
extended (2×2) reconstructed surface areas,
whereas the remaining part of the surface still
shows the (2×5) – both phases coexist at the same
flux (see Fig. 5c). Further increased flux leads to
a mostly (2×2) reconstructed surface with only
occasionally visible low-coverage ‘‘double chains’’
that then can be considered as light domain walls
in a dense H-(2×2) (see the upper part of Fig. 3c).
Finally, at the highest flux, the surface is com-
‘‘double chains’’ in between the high-coverage
rows. The alternation between high- and low-
hydrogen-coverage structures can be interpreted
as a domain wall structure with a striped symme-
try, where the (2×2)-like rows depict heavy
domain walls. At a higher flux of Si H and
2
6
therefore increasing H coverage, the separation of
these walls shrinks. On a larger scale, a statistical
mixture of (2×5) and (2×7) is present (Fig. 5b),
the average separation shifts continuously with
increasing flux of disilane. At step edges already
at very early stages, domains with local (2×2)
appear.
Beyond the (2×5), no continuous transition to

SURFACE SCIENCE
LETTERS
V. Dorna et al. / Surface Science 401 ( 1998) L375–L382
L381
pletely covered with the (2×2) (see Figs. 3d and
d).
A strong correlation between the hydrogen-
by-layer growth mode is strongly favoured, even
at a very high growth rate, where the surface is
largely hydrogen-covered.
5
induced reconstruction and the growth mode
exists. The formation of separated Si islands during
growth does not take place below a flux of about
The interplay of hydrogen-induced reconstruc-
tions and nucleation behaviour of new islands
therefore prevents a roughening of the (113) sur-
face at a high flux of disilane in contrast to Si(111).
This high stability of Si(113) during growth may
make this surface orientation an interesting candi-
date for technological applications. Heteroepitaxy
on Si(113) and the influence of typical surfactants
like Sb [18] are presently investigated and will be
of future interest.
9
×10−6 mbar. In Fig. 3b, small clusters nucleate
at the high-coverage domain walls, but they are
only temporary stable and dissolve again. Only at
a higher flux are island chains created (Fig. 3c),
which again nucleate on the high-H-coverage
(
2×2) regions of the surface.
The preferential nucleation on areas covered by
the high-coverage phase – the site selectivity – is
probably the reason for the abrupt transition from
step flow to island nucleation, if the surface recon-
struction changes in this pressure range as seen in
Fig. 2c–d.
On top of the nucleated islands, the reconstruc-
tion differs from the underlying (2×2) and has
initially the same ‘‘double chain’’ structure as
found in the low-H-coverage regions at a lower
flux in Fig. 3a,b. The width of the islands depends
dynamically on the actual flux; increasing the
disilane partial pressure leads to narrower islands.
Finally, the islands consist of only one ‘‘double
chain’’ of the low H-coverage structure. On a
perfect (2×2)-reconstructed surface, the observed
islands show a quasi-periodical arrangement with
a fourfold periodicity along [33 2: ], which means
In conclusion, it was possible to show that the
morphology and the growth mode of gas source
MBE-grown Si on Si(113) are strongly driven by
the observed H-induced surface reconstructions,
which are different to those obtained by adsorption
of atomic hydrogen. Using high-temperature, time-
lapsed STM ‘‘movies’’, real space information on
growth processes can be obtained in a very efficient
and reliable manner.
Acknowledgements
We like to thank F. Meyer zu Heringdorf and
H.J. G u¨ nter (Universit a¨ t Hannover) for helpful
discussions and the Volkswagenstiftung for finan-
cial support.
that every second row of the (2×2) is occupied
by an island (see bottom right part of Fig. 5d).
The mode of continuous growth at a high disi-
lane flux is dominated by the difference in nucle-
ation rate on the (2×2) and on the islands showing
the low-H-coverage structure. The growth of a
new layer can be divided into two steps: on a
completely (2×2)-reconstructed surface, islands
nucleate with the low-H-coverage structure on top.
Because the nucleation on these islands is strongly
hindered, half a monolayer is formed in a near-
perfect layer-by-layer growth mode until a periodic
arrangement of islands is formed (see Fig. 2d).
Only in the second step, when the islands coalesce,
does the reconstruction gradually change to high-
H-coverage (2×2). In this way, the nucleation of
the next layer is suppressed until approximately
References
[
[
[
[
[
1] J. Knall, J.B. Pethica, J.D. Todd, J.H. Wilson, Phys. Rev.
Lett. 66 (1991) 1733.
2] W. Arabcyk, S. Hinrich, H.J. M u¨ ssig, Vacuum 46 (1995)
473.
3] H. Sakama, D. Kunimatsu et al., Phys. Rev. B 53 (1996)
6
927.
4] J. Dabrowski, H.J. M u¨ ssig, G. Wolff, Phys. Rev. Lett. 73
1994) 1660.
5] S. Song, S.G.J. Mochrie, Phys. Rev. B 51 (1995) 10068.
(
[6] U. K o¨ hler, O. Jasko, G. Pietsch, B. M u¨ ller, M. Henzler,
Surf. Sci. 248 (1991) 321.
[
7] U. K o¨ hler, L. Andersohn, H. Bethge, Phys. Stat. Sol. (a)
59 (1997) 39.
8] U. Myler, P. Althainz, K. Jacobi, Surf. Sci. 251252
1991) 607.
[9] K. Jacobi, U. Myler, Surf. Sci. 284 (1993) 223.
1
[
(
0.75 of a complete layer has grown, and a layer-

LETTERS
SURFACE SCIENCE
L382
V. Dorna et al. / Surface Science 401 ( 1998) L375–L382
[
10] H. Huang, S.Y. Tong, U. Myler, K. Jacobi, Surf. Rev. [15] R.J. Hamers, U.K. K o¨ hler, J.E. Demuth, J. Vac. Sci. Tech-
Lett. 1 (2)3) (1994) 221.
nol. A 8 (1990) 195.
[
[
11] S.M. Scholz, K. Jacobi, Phys. Rev. B 52 (1995) 5795.
12] J.H. Wilson, P.D. Scott, J.B. Pethica, J. Knall, J. Phys.
Cond. Mat. 3 (1991) 133.
[16] L. Andersohn, T. Berke, U. K o¨ hler, B. Voigtl a¨ nder, J. Vac.
Sci. Technol. A 14 (1996) 312.
[17] F. Meyer zu Heringdorf, to be published.
[18] H.J. M u¨ ssig, J. Dabrowski et al., J. Vac. Sci. Technol. B
14 (1996) 982.
[
[
13] U. K o¨ hler, L. Andersohn, B. Dahlheimer, Appl. Phys. A
5
7 (1993) 491.
14] S.K. Kulkarni, S.M. Gates et al., J. Vac. Sci. Technol. A
(1990) 2956.
8