Thickness inhomogenities in the organometallic chemical vapor deposition of GaP

Article abstract of DOI:10.1063/1.3029742

We analyze exponential lateral-thickness variations observed in the growth of GaP on (001) GaAs, thermally generated Si O₂, (001) Si, and nanoscopically roughened Si surfaces by organometallic chemical vapor deposition, using as a reference the polycrystalline GaP deposited on the Mo susceptor surrounding the 2 in. wafers. We find these variations to be due to differences in the chemical reactivities of the various surfaces toward the generation of a precursor, probably a H-P=Ga-C H₃ dimer adduct, by heterogeneous catalysis followed by desorption and diffusion through the gas phase.

Full text of DOI:10.1063/1.3029742

Thickness inhomogenities in the organometallic chemical vapor deposition of GaP
X. Liu and D. E. Aspnes
Citation: Applied Physics Letters 93, 203104 (2008); doi: 10.1063/1.3029742
View online: http://dx.doi.org/10.1063/1.3029742
View Table of Contents: http://scitation.aip.org/content/aip/journal/apl/93/20?ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
Lattice-engineered Si1-xGex-buffer on Si(001) for GaP integration
J. Appl. Phys. 115, 103501 (2014); 10.1063/1.4864777
Structural and optical properties of GaInP grown on germanium by metal-organic chemical vapor deposition
Appl. Phys. Lett. 97, 121909 (2010); 10.1063/1.3492854
Thickness inhomogeneities and growth mechanisms of GaP heteroepitaxy by organometallic chemical vapor
deposition
J. Vac. Sci. Technol. A 28, 583 (2010); 10.1116/1.3442805
Erratum: “Elimination of nonuniformities in thick GaN films using metalorganic chemical vapor deposited GaN
templates” [J. Appl. Phys. 90, 6011 (2001)]
J. Appl. Phys. 91, 6778 (2002); 10.1063/1.1471940
Elimination of nonuniformities in thick GaN films using metalorganic chemical vapor deposited GaN templates
J. Appl. Phys. 90, 6011 (2001); 10.1063/1.1415363
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.189.170.231 On: Tue, 23 Dec 2014 11:59:25

APPLIED PHYSICS LETTERS 93, 203104 ͑2008͒
Thickness inhomogenities in the organometallic chemical vapor
deposition of GaP
X. Liua͒ and D. E. Aspnes
Department of Physics, North Carolina State University, Raleigh, North Carolina 27695-8202, USA
͑Received 24 October 2008; accepted 30 October 2008; published online 18 November 2008͒
We analyze exponential lateral-thickness variations observed in the growth of GaP on ͑001͒ GaAs,
thermally generated SiO2, ͑001͒ Si, and nanoscopically roughened Si surfaces by organometallic
chemical vapor deposition, using as a reference the polycrystalline GaP deposited on the Mo
susceptor surrounding the 2 in. wafers. We find these variations to be due to differences in the
chemical reactivities of the various surfaces toward the generation of a precursor, probably a
H–P=Ga–CH3 dimer adduct, by heterogeneous catalysis followed by desorption and diffusion
through the gas phase. © 2008 American Institute of Physics. ͓DOI: 10.1063/1.3029742͔
Lateral thickness inhomogenities are commonly ob-
served for films deposited by organometallic chemical vapor
deposition ͑OMCVD͒ for a wide range of reactor con-
figurations such as vertical,1–4 pancake,5 and horizontal.6–12
Extensive experimental studies1,4,6,7,9,12 and numerical
simulations2–6,8,10–12 of this topic exist for the epitaxial
growth of GaAs,1,2,6–9,11 GaN,4,12 and Si.3,5,10 These inho-
mogenities are generally ascribed to depletion of gas-phase
reactants,6,10,12 parasitic deposition,1,9,11 variation of the
thicknesses of diffusion-boundary layers,1,5,7 gas-flow
characteristics,2,8 thermal effects,3,10 and rate limiting by
mass transport as opposed to diffusion1 or surface kinetics.5
More interesting are the systematic variations that occur
near sample peripheries. Hemmingsson et al.4 and Dam
et al.12 found the growth rate at the edge of wafers in GaN
homoepitaxy to be higher than that at the center. They attrib-
uted this to lateral back diffusion of GaCl from the area
surrounding the substrate, assuming lower depletion by the
polycrystalline GaN deposited there. Similar effects have
been reported in the selective-area growth of GaN,13–15
where a higher edge growth rate was again ascribed to back
diffusion of reactants from the mask regions. This diffusion
process was verified by Mitchell et al.15 to occur primarily in
the gas phase.
ated SiO2, ͑001͒Si, nanoscopically roughened ͑nr͒ Si, and the
polycrystalline GaP deposited on the molybdenum ͑Mo͒ sus-
ceptor surrounding the 2 in. diameter wafers, showing that
information about growth mechanisms can be obtained as
well. We describe analytically the exponential lateral varia-
tions that we observe by solving a one-dimensional ͑1D͒
diffusion equation. The data are consistent with a growth
process where a precursor, probably a H–P=Ga–CH3 dimer
adduct, is formed by heterogeneous catalysis, mainly des-
orbs, then—consistent with the studies mentioned above—
diffuses in the gas phase.
The OMCVD reactor is a modified Emcore model
GS3300 of shower-head, cold-wall, and rotating-disk design.
Growth was done with trimethylgallium ͑TMG͒ and phos-
phine ͑PH3͒ sources at a chamber pressure of 4 Torr to en-
hance diffusion. The 2 in. diameter substrates were placed in
the slightly recessed central region of a 3 in. diameter Mo
susceptor, so the samples are surrounded by a Mo ring that is
0.5 in. wide. Prior to loading, the ͑001͒GaAs substrates were
given a 10% HCl rinse. The as-polished ͑001͒Si substrates
were given a standard RCA clean plus a final 10% HF rinse.
The SiO2 substrates of root mean square ͑rms͒ roughness
less than 0.5 nm were generated in situ by first heating the
Si substrates to 800 °C under H2 flow, then exposing the
surface six times to O2 bursts of 2 s each. Two types of
nrSi substrates were used. The first, denoted nrSi͑NH4F͒,
Here, we report comparative studies of edge effects for
GaP deposited by OMCVD on ͑001͒GaAs, thermally gener-
TABLE I. Growth conditions and thickness variations for the representative depositions.
␻
FTMG
͑sccm͒
GC
͑nm/s͒
c
Run
Subs.
Gas
͑rpm͒
T͑°C͒
V/III
TE ͑nm͒
A
B
C
D
E
F
GaAs
GaAs
H2
H2
H2
N2
N2
N2
N2
1200
0
570
580
570
600
550
680
600
3300
530
0.52
0.38
0.25
0.52a
0.63
0.34
0.91
0.15
0.06
0.09
0.17
N/A
N/A
0.21b
75–0.3ex/5.5
134–0.5ex/6.0
255–2.8ϫ10−9 ex/1.1
302–0.07x
N/A
SiO2
1200
1200
0
400a
6200a
6300
2400
2700
SiO2
Si
nrSi͑NH4F͒
nrSi͑1:1:1͒
0
N/A
107+0.3ex/5.0
G
0
aStabilized value.
bDetermined from TE divided over time rather than real-time spectroscopic polarimetry.
cUnits of x are mm.
a͒
Author to whom correspondence should be addressed. Electronic mail: xliu.ncsu@gmail.com.
0003-6951/2008/93͑20͒/203104/3/$23.00
93, 203104-1
© 2008 American Institute of Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.189.170.231 On: Tue, 23 Dec 2014 11:59:25

203104-2
X. Liu and D. E. Aspnes
Appl. Phys. Lett. 93, 203104 ͑2008͒
Conditions for the seven representative depositions ana-
lyzed in detail here are summarized in Table I. Kinematics
calculations show that in all cases, growth occurred under
diffusion-limited conditions. Thickness profiles were deter-
mined after growth by four-phase-model analysis of 20
pseudodielectric-function ͗␧͘ spectra taken per sample by
spectroscopic ellipsometry at evenly spaced intervals across
the 1 in. radius. Results are expressed in Fig. 1 as an
effective thickness TE=͑bulk thicknessϫbulk concentration
+top thicknessϫtop concentration͒ and a nominal thickness
TN. Figure 1 also show the best-fit exponentials for deposi-
tions A–D and G. Here, x=0 and 25.4 mm represent the
center and edge of the wafer, respectively. Table I also pro-
vides the numerical forms of TE and the growth rates at the
film centers for all depositions except E and F, which were
done on substrates that favor growth of three-dimensional
islands. For those that can be analyzed, the films are either
thinner ͑A–C͒ or thicker ͑G͒ at the edges than at the center,
but all show an exponential dependence on radius.
We model the thickness inhomogenities as follows. We
suppose that the inhomogenities are due to lateral diffusion
and derive the analytic form of a general concentration n͑x͒
from the 1D continuity equation
ץn
ٌ · J +
= gn − rn.
͑1͒
ץt
ꢀ
Here J=−D͑ץn/ץx͒x is the mass-transport current and gn and
rn are the generation and removal rates of the active species,
respectively. In steady state ץn/ץt=0. A 1D treatment is ac-
ceptable if the characteristic length of the thickness variation
is small compared to the substrate radius, as is the case here.
In our system, diffusion can only occur in the gas phase since
a gap between the sample and the outer ring of the susceptor
prevents diffusion over the surface. We assume that the ac-
tive species is removed by decomposition, deposition, and/or
diffusion normal to the surface, which together act to yield
Poisson statistics with time constant ␶. With these assump-
tions, Eq. ͑1͒ becomes
ץ2n
ץx2
n
− D
= − + gn.
͑2͒
␶
This can be solved using the Green-function approach. We
obtain
n͑x͒ = dx G͑x,x ͒g ,
͑3͒
͵
0
0
n
where G͑x,x0͒=L/2De−͉x−x ͉/L with a characteristic diffusion
0
2
ͱ
length L= D␶. If gn is a constant n0, then n͑x͒=n0L /D is
also a constant. In this case, deposition is uniform, as in the
central regions of the substrates. However, if gn=n0 only
over a limited range, then Eqs. ͑2͒ and ͑3͒ give
FIG. 1. ͑Color online͒ ͑a͒–͑e͒ Nominal ͑blue͒ and effective ͑red͒ lateral
thicknesses together with best-fit exponential dependences for depositions A
͑a͒, B ͑b͒, C ͑c͒, D ͑d͒, and G ͑e͒, which are defined in Table I.
n0L2 n0L2
͑x−a͒/L
n͑x͒ =
−
e
for xϽa,
͑4a͒
was generated by etching in a 10% NH4F solution at 80 °C
for 2 to 7 min. This process is very slow, being anisotropic
and reaction-rate limited,16 leaving a surface that is fairly
rough ͑10 nm rms͒ but with a low density of kink sites. The
second, denoted nrSi͑1:1:1͒, was generated by etching with
HF:HNO3:H2O=1:1:1 at room temperature for 10 to 30 s.
For this composition, the process is rapid, isotropic, and dif-
fusion limited,17 leaving a surface of rms roughness of about
D
2D
and
n0L2
n͑x͒ =
e−͑x−a͒/L for xϾa.
͑4b͒
2D
The two solutions join smoothly at x=a. Thus exponential
behavior occurs near the boundaries, with the exponentials
5 nm with a high density of kink sites.
increasing or decreasing as the boundary is approached de-
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.189.170.231 On: Tue, 23 Dec 2014 11:59:25

203104-3
X. Liu and D. E. Aspnes
Appl. Phys. Lett. 93, 203104 ͑2008͒
pending on which side the concentration is higher. Generali-
zation to the case of different generation rates and/or mul-
tiple regions is straightforward. Equations ͑4a͒ and ͑4b͒ with
adjustable scaling factors for the exponentials were used to
calculate the best-fit curves in Fig. 1.
ductors as noted above. Although our measurements were
made at a relatively low pressure, higher pressures are ex-
pected to affect mainly diffusion lengths and not basic
mechanisms. The results also show that different parts of the
surface, including the susceptor, are in constant contact with
each other during growth.
Since the above theory coincides well with the data, we
interpret the origin of the thickness inhomogenities in terms
of these expressions. The assumption of gas-phase diffusion
appears to contradict the conclusion indicated by the calcu-
lated Péclet numbers, which are larger than 1 but smaller
than 10. These values imply mass transport primarily by ad-
vection. In fact this is what occurs for TMG and PH3 and
their homogeneous decomposition products. Hence, we con-
clude that these precursors alone cannot be directly respon-
sible for deposition. Instead, deposition must result from an-
other species, one that involves both Ga and P, is formed by
heterogeneous catalysis, largely desorbs, and communicates
with different regions via gas-phase diffusion. From the data,
it follows that the generation rates of this species are highest
for GaAs and the high-quality epitaxial GaP on GaAs, inter-
mediate on SiO2 and the polycrystalline GaP deposited on
SiO2 and Mo, and lowest on Si, nrSi, and the rough GaP
deposited on nrSi.
A primary candidate is the H–P=Ga–CH3 dimer ad-
duct. The formation of similar adducts has been suggested
for OMCVD growth of InP from trimethylindium and PH3
sources,18–20 and of GaAs from TMG and arsine ͑AsH3͒.21,22
This adduct would be generated through a reaction between
decomposition products of TMG and PH3. It would also be
likely to decompose in the gas phase, contributing to ␶ in Eq.
͑2͒. Since at our growth temperatures PH3 decomposition is
very inefficient in the gas phase and depends largely on the
catalytic activity of the substrate,18–20 we conclude that it is
the rate-limiting step for adduct generation. This is consistent
with the relative bond weakening effects exerted on PH3 de-
composition by various substrates, which from activation-
energy measurements18,23 occur in the sequence III-V ͑InP͒
͑Ref. 18͒ ϾSiO2 ͑Refs. 18 and 23͒ ϾSi.23
This work is supported by the Army Research Labora-
tory under Grant No. W911NF-04-2-0035.
1G. Costrini and J. J. Coleman, J. Appl. Phys. 57, 2249 ͑1985͒.
2Y. N. Makarov and A. I. Zhmakin, J. Cryst. Growth 94, 537 ͑1989͒.
3C. R. Kleijn, Thin Solid Films 365, 294 ͑2000͒.
4C. Hemmingsson, G. Pozina, M. Heuken, B. Schineller, and B. Monemar,
J. Cryst. Growth 310, 906 ͑2008͒.
5I. Oh, C. G. Takoudis, and G. W. Neudeck, J. Electrochem. Soc. 138, 554
͑1991͒.
6J. van de Ven, G. M. J. Rutten, M. J. Raaijmakers, and L. J. Giling, J.
Cryst. Growth 76, 352 ͑1986͒.
7M. Sato and M. Suzuki, J. Electrochem. Soc. 134, 1540 ͑1987͒.
8R. J. Field, J. Cryst. Growth 97, 739 ͑1989͒.
9Y. Li and L. J. Giling, J. Cryst. Growth 156, 177 ͑1995͒.
10H. Habuka, J. Cryst. Growth 223, 145 ͑2001͒.
11S. Mazumder and S. A. Lowry, J. Cryst. Growth 224, 165 ͑2001͒.
12C. E. C. Dam, A. P. Grzegorczyk, P. R. Hageman, R. Dorsman, C. R.
Kleijn, and P. K. Larsen, J. Cryst. Growth 271, 192 ͑2004͒.
13Y. Kato, S. Kitamura, K. Hiramatsu, and N. Sawaki, J. Cryst. Growth 144,
133 ͑1994͒.
14X. Li, A. M. Jones, S. D. Roh, D. A. Turnbull, S. G. Bishop, and J. J.
Coleman, J. Electron. Mater. 26, 306 ͑1997͒.
15C. C. Mitchell, M. E. Coltrin, and J. Han, J. Cryst. Growth 222, 144
͑2001͒.
16M. A. Hines, Int. Rev. Phys. Chem. 20, 645 ͑2001͒.
17A. Schwartz and H. Robbins, J. Electrochem. Soc. 123, 1903 ͑1976͒.
18C. A. Larsen, N. I. Buchan, and G. B. Stringfellow, J. Cryst. Growth 85,
148 ͑1987͒.
19N. I. Buchan, C. A. Larsen, and G. B. Stringfellow, Appl. Phys. Lett. 51,
1024 ͑1987͒.
20N. I. Buchan, C. A. Larsen, and G. B. Stringfellow, J. Cryst. Growth 92,
605 ͑1988͒.
21C. A. Larsen, N. I. Buchan, and G. B. Stringfellow, Appl. Phys. Lett. 52,
480 ͑1988͒.
22C. A. Larsen, S. H. Li, N. I. Buchan, G. B. Stringfellow, and D. W. Brown,
J. Cryst. Growth 102, 126 ͑1990͒.
The heterogeneous-catalysis/desorption/diffusion pro-
cess is probably common since similar precursors have also
been invoked to explain OMCVD growth of other semicon-
23G. G. Devyatykh, V. M. Kedyarkin, and A. D. Zorin, Russ. J. Inorg.
Chem. 14, 1055 ͑1969͒.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.189.170.231 On: Tue, 23 Dec 2014 11:59:25