You are on page 1of 5

Thin Solid Films 427 (2003) 1115

Surface diffusion of SiH3 radicals and growth mechanism of a-Si:H and


microcrystalline Si
R. Dewarrat, J. Robertson*
Engineering Department, Cambridge University, Cambridge CB2 1PZ, UK

Abstract
Existing growth mechanisms of hydrogenated amorphous silicon (a-Si:H) and microcrystalline silicon (mc-Si) assume that the
growth species SiH3 can diffuse over the hydrogen-saturated Si surface. However, recent calculations suggest that this could not
happen. We have carried out local density formalism pseudopotential calculations of the binding of SiH3 to hydrogen-terminated
(1 1 1)Si surfaces. The bound site is not the three-centre SiHSi bridging site previously assumed. It has a direct SiSi bond
between the SiH3 and the surface Si, and the surface hydrogen is displaced to a bond centre of a surface SiSi bond. A bound
site validates conventional models of growth of a-Si:H and mc-Si, in which a mobile growth species creates smooth surfaces.
2002 Elsevier Science B.V. All rights reserved.
Keywords: a-Si:H; Microcrystalline silicon; Hydrogen

1. Introduction
Hydrogenated amorphous silicon (a-Si:H) is the key
large area electronic material, used in solar cells and
thin film transistors in active matrix liquid crystal
displays and sensors. There is now a trend to use
microcrystalline Si (mc-Si) for these applications,
because of its improved mobility and electrical stability.
Both materials are grown by plasma enhanced chemical
vapour deposition, so an understanding of the growth
process might hep improve the material quality.
The growth species of a-Si:H and mc-Si is generally
believed to be the SiH3 radical w15x. The a-Si:H is
fully terminated by hydrogen, so it does not possess
sites for the SiH3 radical to immediately form a direct
SiSi bond. However, some form of binding is needed
to allow surface diffusion. It is thought that growth
occurs by the weak physisorption of SiH3 onto the
hydrogen-terminated a-Si:H surface, the SiH3 diffuses
over it, and abstracts a hydrogen from a surface SiH
bond, to create a Si dangling bond (DB). A second
SiH3 radical then also physisorbs on the surface, diffuses
and adds to this DB, to give growth w16x.
The key factors are that SiH3 is a mono-radical, and
so it cannot by itself insert directly into a normal SiSi
*Corresponding author. Tel.: q44-1223-33-2689; fax: q44-122333-2662.
E-mail address: jr@eng.cam.ac.uk (J. Robertson).

bond to initiate growth. The second factor is that surface


diffusion is required to give the observed smooth a-Si:H
surfaces. If growth occurred by sticking without diffusion, this would give the physical vapour deposition
conditions, which was ruled out some years ago by the
surface coverage experiments of Tsai et al. w7x.
There are three possibilities for a growth process. The
growth species could stick where it lands on the surface.
This is random deposition. It gives a rough surface in
which the surface roughness varies with film thickness
h as Wsh 0.5. The second possibility is that the growth
species sticks only if it lands at certain energetically
preferred sites, such as step-edges or valley bottoms,
where it makes more bonds to the existing film. This
gives a smoother surface than case 1, but the sticking
coefficient, s, would need to be very much less than 1
(under 0.1) for this, which is not observed for a-Si:H
(ss0.3). The third possibility is that the species forms
a weakly bound state on the surface, which allows the
radical to diffuse over the surface to find a preferred
site. This produces a smooth surface, combined with a
higher sticking coefficient. The surface roughness W in
general varies as Wsahb, with a roughness exponent b
w810x.
The observed exponent b is plotted as function of
temperature in Fig. 1 w1114x. b is seen to decrease
from nearly 0.5 at room temperature, to approximately
0.25 at typical deposition temperatures of 250 8C, and

0040-6090/03/$ - see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 0 4 0 - 6 0 9 0 0 2 . 0 1 1 7 3 - 2

12

R. Dewarrat, J. Robertson / Thin Solid Films 427 (2003) 1115

Fig. 1. Variation of the roughness exponent b with deposition temperature. Data from Smets et al. w13x, Tanenbaum et al. w11x, Flewitt
et al. w12x and Bray and Parsons w14x.

finally reaches nearly 0 at high temperatures. A b of


0.5 corresponds to random deposition. The b value of
approximately 0.25 seen at typical deposition temperatures of 200300 8C requires surface diffusion. The
decrease of b from 0.25 to nearly 0 corresponds to a
crossover from the universality class of random deposition at low temperature to diffusive deposition at higher

temperature, as analysed by Das Sarma and Tamborenea


w15,16x.
The physisorption of SiH3 on the Si:H surface was
proposed to occur by the formation of a three-centre
SiHSi bond between the Si of the SiH3 and a surface
HSi group, as shown in Fig. 2c. This was based on the
analogy to bond-centred (BC) hydrogen site in crystalline Si, which is the stable configuration of H0 and Hq
w1719x. However, recent calculations have suggested
that the SiH3 radical does not bind to the Si:H surface
w2022x. This leaves the whole growth mechanism in
doubt. Either the calculations are wrong, or the growth
mechanism involves additional factors. This led to the
suggestion that a-Si:H could grow by the direct insertion
of SiH3 into weak SiSi bonds w22,23x. However, this
does not resolve the problem, because it is an Eley
Rideal process with no surface diffusion, so it still leads
to a rough surface, which is not seen. Smets et al. w13x
proposed instead that the surface DB, created by abstraction, could diffuse across the surface (instead of the
SiH3). This is equivalent to a surface H diffusing in the
opposite direction. This would give a smooth surface,
but it is not likely because DB diffusion requires the
breaking of a surface SiH bond, which costs 23 eV
w6x. We have therefore tried to resolve these issues by
carrying out detailed local density formalism pseudopotential calculations on the interaction of SiH3 with the

Fig. 2. Configuration of SiH3 next to the Si:H(1 1 1) surface. (a) Initial configuration; (b) silane molecular after H abstraction; (c) physisorbed
site with SiHSi bridge bond; (d) physisorbed site with direct SiSi bond and H displaced to AB site; (e) state with direct SiSi bond and
subsurface BC hydrogen. Large spheres, Si; small spheres, H.

R. Dewarrat, J. Robertson / Thin Solid Films 427 (2003) 1115

Fig. 3. Energy surface (in eV) of the SiH3 interaction with the
Si:H(1 1 1) surface, allowing only SiHSi bridging site, as a function
of SiH distances.

Si:H surface in a variety of configurations. The previous


calculations used clusters to represent the Si and a local
orbital basis, which may have limited their accuracy.
2. Method
The calculations are carried out on a 2=63 super-cell
with a slab of four (1 1 1) bilayers of Si, with the
surfaces terminated by hydrogen. The cell is 3.5 nm
high, so that the slabs are vertically separated by 2.23
nm of vacuum. The calculations are carried out with the
CASTEP code w24x, with the atoms represented by
ultrasoft pseudopotentials, a plane wave basis set with
a cut-off of 350 eV and the Schrodinger equation solved
in spin-restricted local density approximation with the
generalised gradient correction GGA-PW91 of the
exchange-correlation potential. The bond length and
bond angles of the SiH3 and of the first two bilayers of
the Si slab were allowed to optimise during the
calculations.

13

The resulting energy surface is shown in Fig. 3. The


SiH3 approaches the surface from the top of the figure
down the valley at x;0.15 nm, and then it exits along
the channel at large x and y;0.15 nm to become the
SiH4 molecule. There is a plateau at approximately xs
0.15 nm, ys0.25 nm, but there are no barriers which
would create a bound state. The barrier for abstraction
of H from the surface SiH group is under 0.1 eV, and
the gain energy of abstraction to form SiH4 is 0.8 eV,
compared to a free SiH3 placed at the centre of the
vacuum layer in our super-cell. Note that the abstraction
energy is slightly less than given by us earlier w25x. This
is because our energy for unbound SiH3 was not fully
converged, because we had not left enough vacuum
space between layers. The abstraction energy is slightly
higher than that found by Ramalingam et al. w20x and
Parsons w21x.
The absence of a bound state in the simple configuration led us to check other configurations. A bound
state is found for the configuration shown in Fig. 2d, in
which the Si of SiH3 forms a direct SiSi bond to the
surface Si atom, and the surface H atom is displaced
side-ways. The surface Si is now fivefold co-ordinated.
The displaced H atom now lies at the anti-bonding (AB)
site of a lateral surface SiSi bond. The binding energy
in this configuration is 0.2 eV compared to the free
SiH3.
A further unconstrained geometry optimisation found
that the displaced hydrogen atom rotates further until it
lies essentially at the bond centre of a lateral surface
SiSi bond, as shown in Fig. 2e. This configuration is
the most stable, and is bound by 0.74 eV compared to
SiH3. The hydrogen lies equidistant between the two
silicons, but slightly above the bond centre. The energy

3. Results
In the conventional picture, SiH3 approaches the Si:H
surface in the Si-down orientation. The interaction is
represented in terms of two distances, the SiH bond
length of the surface H, and the SiH distance between
the approaching SiH3 and the surface H, as in Fig. 2a.
The SiH bonds of SiH3 are staggered with respect to
the SiSi bonds of the Si slab. The SiSi bond length
in bulk Si is 0.235 nm, and the SiH bond length in
SiH4 is 0.154 nm, Fig. 2b. The total energy of the
system was found as a function of these two distances.
The energies were calculated on a xy mesh and then
interpolated by a cubic spline fit.

Fig. 4. Energy surface (in eV) of SiH3 interaction with Si:H(1 1 1)


surface, allowing the surface H to displace to AB site, as a function
of SiSi and surface SiH distances.

14

R. Dewarrat, J. Robertson / Thin Solid Films 427 (2003) 1115

contour of SiH3 for this configuration is shown in Fig.


4. The contours are plotted in terms of the SiSi
separation and the length of the surface SiH bond. In
this plot, the SiH3 approaches from the top along the y
direction, and the SiH4 leaves at 458. A clear bound
state is seen in the plot.
4. Discussion
Let us consider the stability and bonding of these
various configurations. It is clear that the three configurations in Fig. 2 correspond to the BC and the AB
sites of interstitial hydrogen at the sorption SiSi bond
or a lateral SiSi bond. The AB site is sometimes called
tetrahedral Td but this is not strictly correct as the SiH
bond is shorter than the SiSi bond. In bulk Si, the
neutral hydrogen interstitial H0 is marginally more stable
at BC than AB by 0.1 eV according to our calculations,
as found by van de Walle w17,18x whereas Chadi w19x
found the opposite. The positive H is much more stable
at BC, whereas the negative H is stable at the AB site.
H0 has an unpaired spin. At BC, the unpaired electron
occupies an orbital antisymmetric about the H, which is
an AB combination of Si states with no contribution
from the H s orbital itself. The stability of H0 at BC
arises because the energy cost of dilating the SiSi bond
to accommodate the H is exceeded by the energy gain
of delocalising the unpaired electron over the Si backbonds. This makes the H0 1 eV more stable than a free
H0 atom in the vacuum. At the AB site, the SiSi bonds
do not need to dilate, but there is less delocalisation
gain of the unpaired electron which now remains approximately 50% on the hydrogen itself.
For an interstitial H at the sorption SiSi bond, Fig.
2b, the BC site no longer costs a dilation energy because
the SiHSi is free to relax. However, there is also less
energy gain from delocalisation because the SiH bonds
of the SiH3 are less polarisable than SiSi bonds. On
the other hand, the AB site of Fig. 2c surrounds the H
atom with four SiSi bonds, which increases its delocalisation. This makes the AB more stable than the BC
on the sorption bond. Again there is no dilation energy
for an AB site. The subsurface BC site, Fig. 2c, is the
most stable because the H0 is surrounded by five SiSi
bonds, so the energy gain from delocalisation is most.
The new physisorbed state of SiH3 resolves the
problems with the growth mechanism. State 2(c) can
diffuse by a rolling processes from one Si to the next.
The H on the next surface Si is pushed into a Td site,
while the existing H at Td bounces back to become a
normal terminal HSi bond. The Si of SiH3 becomes
fivefold co-ordinated in the transition state. It may be
that the surface Td configuration was observed in a
molecular dynamics simulation of growth w26x.
Let use summarise surface processes on Si:H. The
abstraction and addition reactions of hydrogen with

Si:H(1 1 1) are EleyRideal, without surface diffusion


w27,28x. Additional hydrogen can be incorporated subsurface, but this is a separate matter. There is no fast
diffusing hot surface hydrogen site, as proposed by
Widdra w29x. This data is not relevant to the present
situation. On the other hand, Si species can diffuse on
the surface. This would be a simple matter if the Si
surface was not fully H-terminated, as there would be
numerous unsatisfied Si valences (DBs or reconstructed
bonds) to allow this. This is indeed what occurs during
the CVD of poly-Si above 550 8C. The growth species
there is the SiH4 molecule. The incoming SiH4 breaks
up into H and SiH3 on hitting the bare Si surface
w30,31x. The SiH3 can now diffuse around, until it finds
a kink or step-edge site, at which it rests and gives
growth.
The main questions come with Si:H surfaces which
have no DBs and are closed shell systems. A monoradical SiH3 cannot form a direct SiSi bond to it. It
can only form a weak physisorption bond of some form.
The present work has identified the nature of this bond.
It is not as first thought, it involves a direct SiSi bond
and the side-ways displacement of the surface H. The
more fundamental point is that this binding is needed to
give any surface diffusion, and thereby a smooth surface.
If the growth mechanism was by direct insertion into
weak SiSi bonds, it would still require binding and
surface diffusion, otherwise we would not get a smooth
surface.
The new physisorption site with H at a bond centre
of a lateral SiSi is similar to the configuration found
by Vittadini et al. w32x in the hydrogen etching of the
Si surface. They studied how H reacts with a Si:H
surface to evolve SiH4, and showed how the H must
attack the subsurface SiSi bonds to desorb a SiH4
species. An H can rearrange within a number of nearly
degenerate positions in the surface layer of Si, in order
to release the SiH4. Physisorption of SiH3 and growth
is the reverse process of etching, and so similar configurations will be used. Thus, growth and etching of Si
involves an Si atom arriving at or leaving the Si surface,
with the H configuration adjusting at each stage of the
process among various nearly degenerate positions. This
is equivalent to the hydrogen glass model w33x of
bonding in a-Si:H.
The migration energy of SiH3 on Si:H has been
calculated for various configurations. We distinguished
those with second neighbour SiH groups as on
Si(1 1 1):H, and those with first neighbour SiH groups
such as Si(1 1 0):H or the reconstructed 2=1
Si(1 0 0):H surface. The migration energy was found to
be 0.7 eV when SiH groups are second neighbours,
and 0.4 eV when they are first neighbours. This is a
critical difference, which has not been previously noted.
The SiH3 species can diffuse easily within a cluster of
nearest neighbour sites, but it is blocked by the larger

R. Dewarrat, J. Robertson / Thin Solid Films 427 (2003) 1115

barrier to second neighbour sites. Diffusion depends on


nearest neighbour sites forming a percolation path across
the surface. This is being studied in more detail. The
binding energy (0.7 eV) and diffusion migration energy
(0.4 eV) of SiH3 on the Si:H surface are reasonably
close to those used previously in empirical models of aSi:H growth w5,6x. This supports the existing ideas of
the growth mechanism of a-Si:H.
Acknowledgments
The authors are grateful to Paul Peacock for help
with computing. RD thanks the Swiss National Science
Foundation for funding, and JR thanks EPSRC and
Philips for funding.
References
w1x A. Gallagher, J. Appl. Phys. 63 (1988) 2406.
w2x A. Matsuda, K. Tanaka, J. Non-Cryst. Solids 97 (1987) 1367.
w3x A. Matsuda, K. Nomoto, Y. Takeuchi, A. Suzuki, A. Yuuki, J.
Perrin, Surf. Sci. 227 (1990) 50.
w4x J. Perrin, J. Non-Cryst. Solids 137 (1991) 639.
w5x J. Perrin, M. Shiratani, P. Kae-Nune, H. Videlot, J. Jolly, J.
Guillon, J. Vac. Sci. Technol. A 16 (1998) 278.
w6x J. Robertson, J. Appl. Phys. 87 (2000) 2608.
w7x C.C. Tsai, J.C. Knight, G. Chang, B. Wacker, J. Appl. Phys.
59 (1986) 2998.
w8x F. Family, T. Vicsek, J. Phys. A 16 (1986) L441.
w9x M. Kardar, G. Paradisi, Y.C. Zhang, Phys. Rev. Lett. 56 (1986)
889.
w10x A.L. Barabasi, H.E. Stanley, Fractal Concepts in Surface
Growth, Cambridge University Press, 1995.

15

w11x D.M. Tanenbaum, A.L. Laracuente, A.C. Gallagher, Phys. Rev.


B 56 (1997) 4243.
w12x A.J. Flewitt, J. Robertson, W.I. Milne, J. Appl. Phys. 85 (1999)
8032.
w13x A.H.M. Smets, D.C. Scram, M.C.M. van de Sanden, Mat. Res.
Soc. Symp. Proc. 609 (2000) A7.6.
w14x K.R. Bray, G.N. Parsons, Phys. Rev. B 65 (2002) 35311.
w15x S. Das Sarma, P. Tamborenea, Phys. Rev. Lett. 66 (1992) 325.
w16x P. Tamborenea, S. Das Sarma, Phys. Rev. B 48 (1993) 2575.
w17x C.G. van de Walle, Y. Bar-Yam, S.T. Pantelides, Phys. Rev.
Lett. 60 (1988) 2761.
w18x C.G. van de Walle, Phys. Rev. B 49 (1994) 4579.
w19x K.J. Chang, J.D. Chadi, Phys. Rev. B 40 (1990) 11644.
w20x S. Ramalingam, D. Maroudas, E.S. Aydil, S.P. Walch, Surf.
Sci. 418 (1998) L8.
w21x G.N. Parsons, J. Non-Cryst. Solids 266 (2000) 23.
w22x A. Gupta, H. Yang, G.N. Parsons, Surf. Sci. 496 (2001) 307.
w23x A. von Keudell, J.R. Abelson, Phys. Rev. B 59 (1999) 5791.
w24x M.C. Payne, M.P. Teter, D.C. Allan, T.A. Arias, J.D. Joannopoulos, Rev. Mod. Phys. 64 (1992) 1045.
w25x R. Dewarrat, J. Robertson, J. Non-Cryst. Solids 299 (2002)
48.
w26x S. Ramalingham, S. Sriraman, E.S. Aydil, D. Maroudas, Appl.
Phys. Lett. 78 (2001) 2685.
w27x D.D. Koleske, S.M. Gates, B. Jackson, J. Chem. Phys. 101
(1994) 3301.
w28x E. Srinivasan, H. Yang, G.N. Parsons, J. Chem. Phys. 105
(1996) 5467.
w29x W. Widdra, S.I. Yi, R. Maboudian, G.A.D. Briggs, W.H.
Weinberg, Phys. Rev. Lett. 74 (1995) 2074.
w30x T.I. Kamins, Polycrystalline Silicon for Integrated Circuit
Applications, Kluwer, 1988.
w31x J.H. Comfort, R. Reif, J. Electrochem. Soc. 136 (1989)
23862398.
w32x A. Vittadini, A. Selloni, R. Car, M. Cassarin, Phys. Rev. B 46
(1992) 4348.
w33x R.A. Street, J. Kakalios, C.C. Tsai, T.M. Hayes, Phys. Rev. B
35 (1987) 1316.

You might also like