Journal of Crystal Growth 223 (2001) 466 483
GaN decomposition in H2 and N2 at MOVPE temperatures
and pressures
D.D. Koleske*, A.E. Wickenden, R.L. Henry, J.C. Culbertson, M.E. Twigg
Laboratory for Advanced Material Synthesis, Code 6800, Electronics Science and Technology Division, Naval Research Laboratory,
Washington, DC 20375, USA
Received 31 July 2000; accepted 12 December 2000
Communicated by C.R. Abernathy
Abstract
GaN decomposition rates were measured in H2, N2, and mixed H2 and N2 flows for pressures and temperatures
typically encountered in metalorganic vapor phase epitaxy. The rates for GaN decomposition, Ga desorption, and Ga
droplet accumulation, were obtained from weight measurements before and after annealing the GaN films in a close-
spaced showerhead reactor. In H2 at constant temperature, the GaN decomposition rate is enhanced when the reactor
pressure is greater than 100 Torr. Unlike H2, the decomposition rate in N2 did not change as a function of pressure. The
enhanced GaN decomposition rate in H2 is not due to an increase in the Ga desorption rate, which is constant vs.
pressure, but instead is due to H2 dissociation on the surface followed by NH3 formation and desorption. NH3
formation is suggested by the cubic decrease in the GaN decomposition rate as N2 is substituted for H2. The measured
activation energies, EA, for the GaN decomposition range from 0.34 to 3.62 eV and depend strongly on the annealing
conditions. By comparing measured and literature values of the EA, four distinct groupings of the EA are observed. The
four distinct groupings of the EA imply that there are possibly four different reactions which limit the GaN
decomposition rate. Connections between the GaN decomposition and improved GaN growth are discussed. This
includes a discussion of changes that occur in the nucleation layer during the ramp from low to high temperature, as
well as increases in GaN grain size as the growth pressure is increased. Published by Elsevier Science B.V.
PACS: 81.05.Ea; 81.15.Gh; 82.60.Cx; 81.05.D; 68.55; 61.16.B; 61.72
Keywords: A1. Decomposition; A1. Desorption; A1. Grain size; A1. Growth; A3. Metalorganic vapor phase epitaxy; B1. Gallium nitride
1. Introduction these device successes, detailed understanding of
GaN growth is of great interest for further
Despite the lack of a lattice matched substrate, material and device improvements. Studies of
GaN of sufficient quality has been grown to GaN growth using molecular beam epitaxy
produce high brightness blue LEDs [1], lasers [2], (MBE) have provided far greater insight into the
and microwave power devices [3 5]. Because of growth details because the growth species are
simpler and in situ diagnostics are available [6 11].
However, a similar level of understanding has not
*Corresponding author. Fax: [phone].
been attained in metalorganic vapor phase epitaxy
E-mail address: [e-mail] (D.D. Koleske).
0022-0248/01/$ - see front matter Published by Elsevier Science B.V.
PII: S 0022- 0248( 01) 00617- 0
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 467
(MOVPE) [12 15],1 because of the more complex The principal aim of this paper is to demon-
chemical reactions and changes at the surface are strate how P; T, and gas flow (i.e. H2 or N2)
difficult to measure during growth.2 Optimal change the GaN decomposition rates and mechan-
MOVPE growth conditions vary from reactor to isms. The paper also unifies previous literature on
reactor, primarily reflecting the influence of GaN decomposition kinetics, since many of the
reactor design on the individual chemical reaction previously measured kinetic values are corrobo-
rates [16]. In addition, some of the chemical rated in the present study by varying the annealing
reactions, specifically NH3 decomposition, have conditions.
been suggested to be catalytic, relying on specific The paper is structured as follows: in Section 2,
surface sites for full activation [17 19]. results from previous GaN decomposition
It is well established that GaN material quality studies are discussed along with proposed decom-
depends heavily on the nucleation layer [20] and position mechanisms. In Section 3, the experi-
high temperature GaN layer [16] growth condi- mental details of the weight loss measurements
tions. Recently, we have reported the influence of and annealing conditions are presented. In Section
pressure, P, on GaN film quality [21,22]. Using 4, the experimental results on GaN decompo-
cross sectional TEM, the GaN grain size was sition, Ga desorption, and Ga surface droplet
shown to increase as the growth P increased accumulation are presented. Details of the Ga
[21,22]. Prior to this work we observed that GaN droplet coalescence are presented along with a
decomposition also increases as the reactor P comparison to a theoretical droplet coalescence
increases [23 25], implying a correlation between model. Also in Section 4, the GaN decomposition
enhanced decomposition at high P and the larger in mixed H2 and N2 flows will be presented,
grained GaN films [21]. These studies [21 25] along with an Arrhenius analysis of the GaN
illustrate a close link exists between GaN film decomposition kinetic parameters. In Section 5,
quality and the initial growth stages, specifically mechanisms for GaN decomposition under the
the competition between GaN growth and decom- various experimental conditions are discussed.
position. This will include a discussion of the four different
Although GaN decomposition has been exten- groups of kinetic parameters listed in Table 1.
sively studied, there are substantial differences in Also in Section 5, the implications of this
the reported kinetic parameters and decomposi- decomposition study to GaN growth are consid-
tion mechanisms. This is partly due to the ered, including a discussion of how the GaN
differences in P, temperature, T, and gas flows nucleation layer changes during the high tempera-
used to study GaN decomposition. For example, ture anneal and how the growth conditions
the onset of GaN decomposition has been affect GaN grain size. Section 6 contains conclu-
reported for T as low as 4008C [26] and as high sions from this study on decomposition and
as 10708C [27]. Additionally, activation energies, general implications for other GaN reactor con-
EA, as low as 0.4 eV [28] and as high as 3.93 eV [29] figurations.
have been reported. Literature values for EA and
the pre-exponential factor, A0, are listed in Table 1
[6,19,24,26,28,30 40]. The EA and A0 are sepa- 2. Previous studies of GaN decomposition
rated into groups labeled A D to reflect their
dependence on annealing conditions. Many early reports of GaN growth also
included observations and concern over GaN
decomposition, mostly because decomposition
competed with growth [41,42]. In 1932, GaN
1
For recent reviews of GaN growth, see Ref. [12].
synthesis and decomposition were reported by
2
While in situ reflectometry and optical pyrometery are
Johnson, Parsons, and Crew, who noticed that
useful for determining surface roughness and growth rate, they
GaN decomposes when the temperature, T, was
do not measure atomic level changes in the surface structure
like RHEED does. greater than 8008C [41]. In later papers, GaN
468 D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483
Table 1
Measured kinetic parameters for GaN decomposition (groupings A D) and Ga desorption (grouping E) gathered from the literature
and the present work. For each grouping label (i.e. A E) the conditions for pressure and temperature are listed. The first column lists
the measurement technique used for measuring the kinetic parameters. The second column lists the pre-exponential factor, A0, in units
of cm 2 s 1, while the third column lists the base 10 logarithm of A0. The fourth column lists the activation energy, EA, for each
measurement in eV. The last column lists the reference from the literature and the year the work was conducted in square brackets
Event and experimental conditions A0 (cm 2 s 1) Log10ðA0Þ (cm 2 s 1) EA (eV) Ref. [Year]
(A) GaN decomposition, P476 Torr, T> 9258C, N2 formation and desorption limited
(1) Thermogravimetric }} 2.7 30 [1956]
(2) Thermogravimetric 4 1029 29.60 3.1 31 [1965]
(3) Mass spectroscopy in vacuum 5 1028 28.70 3.1 32 [1974]
(4) Mass spectroscopy in vacuum 1.2 1031 31.08 3.93 33 [1996]
(5) Ga flux in vacuum 5.0 1029 29.70 3.45 34 [1998]
(6) Reflectivity data in vacuum 5.1 1031 31.7 1.2 3.7 0.3 35 [1999]
(7) Microbalance at 760 Torr 1.4 1029 29.1 3.2 36 [2000]
(8) Weight loss in H2 at 40 and 76 Torr 2.8 1029 29.4 1.8 3.4 0.2 This work
(9) Weight loss in N2 at 76 and 150 Torr 1.2 1029 29.1 1.6 3.62 0.14 This work
(B) GaN decomposition, P476 Torr, T59258C, N2H25x54 formation and desorption limited
(10) Surface photoadsorption N loss in H2 }} 0.91 37 [1998]
(11) Weight loss in H2 at 40 and 76 Torr 8.7 1019 19.94 0.40 0.98 0.07 This work
(C) GaN decomposition, P5150 Torr, T59008C, NH3 formation and desorption limited
(12) Redhead analysis of NH3 TPD peak }} 1.6 2.0 19 [1999]
(13) Gravimetric in H2 at 760 Torr }} 1.8 26 [1974]
(14) Reflectometry at 760 Torr 1 1025 25 1.87 28 [1999]
(15) Weight loss in H2 at 150 and 250 Torr, 7.9 1024 24.9 1.3 1.7 0.2 This work
(D) GaN decomposition P5150 Torr, T> 9008C, Ga diffusion limited
(16) Ga diffusion EA}theory calculation }} 0.4 38 [1999]
(17) Reflectometry at 760 Torr 3 1017 17.5 0.38 28 [1999]
(18) Microbalance at 760 Torr 1 1015 15 0.44 36 [2000]
(19) Weight loss in H2 at 150 and 250 Torr 3.5 1017 17.54 0.53 0.34 0.1 This work
(E) Ga desorption, in vacuum and at all pressures in N2, and H2
(20) Desorption from liquid Ga }} 2.8 39 [1969]
(21) RHEED study of hexagonal GaN }} 2.76 40 [1996]
(22) RHEED study, cubic GaN 1.0 1028 28.0 0.26 2.69 0.05 6 [1996]
(23) Weight loss in H2 at 40 250 Torr 6.5 1026 26.81 0.34 2.74 0.08 24 [1999]
(24) Weight loss in N2 at 76 and 150 Torr 5.4 1025 25.73 1.5 2.69 0.4 24 [1999]
decomposition was found to initiate at lower T 2GaNðsÞ !2GaðgÞ þN2ðgÞ: ð1Þ
(400 8008C) in H2 [26,27,41 44] compared inert
gases (N2 or Ar) or vacuum (>9008C) [26,27,30 2GaNðsÞ !2GaðlÞ þN2ðgÞ !2GaðgÞ: ð2Þ
32,43 48].
Early on, several different mechanisms were GaNðsÞ !GaNðgÞ or ½GaNŠxðgÞ: ð3Þ
proposed to explain GaN decomposition. These
included decomposition into gaseous Ga and Evidence for all three reactions has been observed
nitrogen (Eq. (1)) [31], liquid Ga and nitrogen experimentally. For example, N2 evolution has
(Eq. (2)) [26,31,32], and sublimation of GaN as a been observed using mass spectroscopy [27,32]
diatomic or polymeric product (Eq. (3)) [31]. along with GaN+ and Ga2N+ species [47,48].
2
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 469
Occasionally, Ga droplets 20 30 mm in size have tion for Ga (i.e. 2.83 eV) [56] and the measured EA
been observed on the GaN surface [27,30 32,43], suggests a simple Ga Ga bond cleavage for the
indicating that the GaN decomposition rate, kGaN, rate-limiting step.
can exceed the Ga desorption rate, kGa. Liquid Ga The N desorption kinetic parameters have been
and In droplets have been shown to catalyze GaN measured in a RHEED study on cubic GaN
decomposition [49 51]. [6]. Both the measured EA (6.1 eV) and A0
As mentioned above, the onset T for GaN (2 1044 cm 2 s 1) are large for N desorption [6].
decomposition is lower in H2 compared to inert Extrapolating the N desorption rate, kN, to
environments (i.e. N2, Ar, and vacuum). Hydrogen 10208C gives a kN that is 1000 times kGa [16].
could assist decomposition by the reverse GaN
synthesis reaction, i.e. the reformation of NH3 via
GaNðsÞ !3=2H2 ! GaðlÞ þNH3ðgÞð4Þ 3. Experimental details
To measure the NH3 formation rate, Thurmond
The experimental procedure is described in Ref.
and Logan cleverly titrated the furnace exhaust
[23]. The GaN films used in these studies were
during GaN decomposition in H2 into a column of
grown in a close-spaced showerhead (CSS) reactor
HCl solution containing methyl red indicator [42].
[3] and in a custom-designed vertical reactor [16].
The NH3 reformation reaction is significant for
Growth in both reactors have produced GaN films
GaN growth because it removes N from the
with specular morphology and Si doped electron
surface and provides a reversible pathway for
mobilities>600 cm2 V 1 s 1 [22,23].
bringing GaN synthesis closer to equilibrium.
The decomposition study was conducted in the
Previously measured values of the activation
CSS reactor.3 In the CSS reactor, the temperature,
energy, EA, and the pre-exponential factor, A0, for
T, measurement and control consists of a W/Re
GaN decomposition are listed in Table 1. For H2 thermocouple in direct contact with the backside
pressure, P, less than 76 Torr, in N2, or in vacuum
of the susceptor. The T was calibrated by
(Table 1A) the EA is 2.7 3.93 eV. Surprisingly, this
observing the melting of 0.00500 diameter Au wire
is lower than the EA of 4.7 eV for GaAs decom-
placed on the sapphire and correlating it with the
position [52,53], despite the stronger bond strength
T of the W/Re thermocouple. The T was
in GaN (4.1 eV) compared to GaAs (2.0 eV) [54].
reproducible to within 108C after 2 years of use.
When GaN films are annealed in H2 at higher P
For the decomposition study, pieces of the GaN
and lower T, the EA decreases (Table 1B D). For
grown on a-plane ð1120Þ sapphire were cleaved
example, Morimoto [26] and Rebey et al. [28] have
and weighed to within 0.1 mg using an analytical
measured EA of 1.8 eV and 1.87 eV, respectively.
balance [23]. The pieces were loaded into the
In H2 at higher P and higher T, two groups have
reactor and the process flows established. To
measured lower EA of 0.38 eV [28] and 0.44 eV
minimize the effects of decomposition during the
[36]. These lower EA for decomposition in H2 ramp from low to high T, the pieces were ramped
suggests hydrogen aids in N removal from the
at a controlled rate (258C/min) from 8008C to the
surface, possibly by NH3 formation (i.e. Eq. (4)).
final annealing T. After annealing for a set time,
Unlike the GaN decomposition kinetics, the
ranging from 3 to 180 min 1, the heater was shut
kinetic parameters for Ga desorption have only a
off. Each piece was re-weighed to determine the
weak dependence on pressure or gas flow [23,24].
mass loss.
The measured kinetic parameters listed in Table
In many instances, Ga droplets were observed
1E are similar to the parameters for Ga desorption
on the surface and were removed in dilute HNO3
from GaAs surfaces [55] and liquid Ga [39]. Good
acid. After removal of the Ga droplets, each piece
agreement has been observed in the measured A0
(ranging from 1 1028 to 6 1029 cm 2 s 1) and 3
Manufactured by Thomas Swan & Co., Ltd. The use of a
EA (ranging from 2.69 to 2.76 eV) [6,23,24,39,40].
commercial product does not imply endorsement by the U.S.
The close agreement between the heat of forma- Government.
470 D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483
was weighed again to determine the weight of the GaN pieces were annealed at 9928C for 10 min in
liquid Ga. The weight of the decomposed GaN 6 SLM of H2 at pressures of 76 (Fig. 1(a)) and
was calculated by subtracting the weight after 150 Torr (Fig. 1(b)). Note that the Ga droplets are
removal of the Ga droplets from initial weight. larger and more numerous in Fig. 1(b) (150 Torr)
The weight of desorbed Ga was calculated from compared to Fig. 1(a) (76 Torr). Generally, as the
the mass balance difference between the decom- P increased, the size of the Ga droplets increased.
posed GaN and the liquid Ga. Finally, the piece In pure N2, Ga droplets were not observed for
was annealed at 10808C until all traces of GaN T510008C in N2. However, Ga droplets were
were removed from the sapphire piece. The area, observed in N2 for T > 10008C. For similar T and
A, of the irregularly shaped pieces was determined P, the Ga droplet size was 10 times smaller in N2
from the bare sapphire weight using the formula, compared in H2.
A ź m=td, where m is the sapphire weight, t is the The Ga droplet size also increases as a function
sapphire thickness (typically 0.033 cm), and d is of time at fixed T and P. In Fig. 2, images of the
the sapphire density (3.98 g cm 3). After convert- GaN surface after (a) 3, (b) 10, (c) 20, and (d)
ing the weights to molar quantities, the kinetic 80 min of annealing in 6 SLM H2 at 150 Torr and
rates (atoms cm 2 s 1) were determined by divid- 8118C are shown. In Fig. 3, Ga droplet size
ing by the sapphire area and anneal time. During distributions, NS, vs. size are shown for annealing
the ramp from 8008C to the final T, some GaN times of (a) 10, (b) 20, (c) 30, and (d) 45 min. The
decomposition occurred. Direct comparison of NS were obtained from Nomarski transmission
258C/min and 508C/min ramp rates produced images using an analysis program to measure the
similar kinetic rates, therefore decomposition droplet area [52,53]. The droplet size was then
during the ramp was not corrected for in the calculated as the square root of the area. The NS
kinetic rates. for 20, 30, and 45 min are shifted vertically in
The surface morphology after annealing was Fig. 3. The solid line is a fit to NS based on power
viewed using a Nomarski microscopy in the phase law and bell shaped curve distributions [55]. Note
contrast and transmission mode. Transmission that the mean droplet size (i.e. peak of bell shaped
images were analyzed using NIH public domain distribution) in Fig. 3 is 8 9 mm after 10 min and
software4. The atomic-force-microscope (AFM) increases to 20 mm after 45 min. Comparison of
measurements were performed under atmospheric NS shown in Fig. 3 to the growth model of Family
conditions using a Park Scientific Instruments and Meakin [57] is discussed in Section 5.1.
system in the contact mode.5
4.2. Kinetic rates vs. pressure in H2
4. Results The total P had a strong influence on the GaN
decomposition rate, kGaN, in flowing H2. The kGaN
4.1. Ga droplet accumulation and growth is plotted in Fig. 4 vs. pressure for T of 811 (filled
diamonds), 902 (open squares), and 9928C (filled
After annealing in H2, the most notable change circles). It is clear from Fig. 4 that the kGaN
in the GaN surface morphology is the appearance increases as the H2 pressure increases at constant
of Ga droplets as shown in Fig. 1. In Fig. 1, the T. A more dramatic increase in the kGaN occurs at
lower T. For example, at 9028C the kGaN at
400 Torr is 9 times the kGaN at 40 Torr. Also, the P
4
Images were analyzed using NIH public domain software where the kGaN is enhanced shifts to higher P as
(NIH Image program developed at the U.S. National Institutes
the T decreases. This can be seen in Fig. 4 by
of Health and available on the Internet at http://rsb.info.nih.-
comparing the onset for enhanced decomposition
gov/nih-image/).
5 at 9928C, which occurs at a P of 76 Torr to the
Manufactured by Park Scientific Instruments. The use of a
onset at 9028C and 8118C, which occur at 120 and
commercial product does not imply endorsement by the U.S.
Government. 200 Torr, respectively. It is also clear from Fig. 4
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 471
Fig. 1. Nomarski phase contrast pictures of the GaN surface after heating GaN for 10 min at 9928Ci nH2 pressures of (a) 76 Torr and
(b) 150 Torr. The droplets are liquid Ga. The bar on (a) indicates a length of 100 mm.
Fig. 2. Nomarski phase contrast pictures of the GaN surface after annealing GaN films at 8118C for (a) 3, (b) 10, (c) 20, and (d) 80 min
in H2 at a pressure of 150 Torr. The bar on (d) indicates a length of 50 mm.
Fig. 3. The natural logarithm of the GaN droplet size
distribution is plotted for GaN surfaces annealed at 8118C for
(a) 10, (b) 20, (c) 30, and (d) 45 min in H2 at a pressure of Fig. 4. GaN decomposition rate measured is 6 SLM of H2
150 Torr. The solid line fit is combination of a power law plotted vs. pressure at temperatures of 9928C (filled circles),
dependence (at small sizes) and a bell shaped curve (at large 9028C (open squares), and 8118C (filled diamonds). The solid
sizes). and dashed lines are guides to the eye.
472 D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483
that the kGaN at higher P (i.e. >200 Torr) and
lower T (i.e. 9028C) can exceed the kGaN at lower P
(i.e. 5 40 Torr) and higher T (i.e. 9928C). The
strong influence of P on the kGaN suggests a
change in the GaN decomposition mechanism as P
increases.
For the same T and P shown in Fig. 4, the liquid
Ga accumulation rate, kS;Ga, on the surface is
plotted in Fig. 5. This rate was obtained from the
weight difference between the annealed GaN piece
(with Ga droplets) and the same piece after
removal of the Ga in dilute HNO3. The kS;Ga
closely coincide with the enhanced kGaN shown in
Fig. 4. Note that for P greater than 100 Torr, the
Fig. 6. Same as Fig. 4, except the Ga desorption rate is plotted.
kS;Ga at 9028C exceeds the kS;Ga at 9928C. This is
due to the increased kGaN at higher P and the
reduced Ga desorption rate at lower T.
The Ga desorption rate, kGa, is plotted in Fig. 6
for the same T and P shown in Figs. 4 and 5. Note
that in Fig. 6, the kGa are plotted on an
exponential scale. Unlike the kGaN and kS;Ga, the
kGa is relatively constant as a function of P at each
T, only increasing at P of 80 Torr at 9928C,
110 Torr at 9028C, and 120 Torr at 8118C. The
increase in the kGa at these P may be due to the
increased Ga surface area that occurs at onset of
droplet coalescence as described in Appendix A.
To achieve a measurable weight loss for the
rates shown in Figs. 4 6, the annealing time was
varied from 10 to 180 min. For Figs. 4 6, it is
assumed that the rates were constant in time (i.e.
Fig. 7. GaN decomposition (filled circles), Ga surface accumu-
lation (open squares), and Ga desorption (filled diamonds) rates
are measured in 6 SLM H2 for various annealing times. For this
plot the temperature was 8118C and the pressure was 150 Torr.
steady state). This assumption was checked by
measuring the three rates as a function of time. In
Fig. 7, the kGaN, kS;Ga, and kGa in 150 Torr of H2 at
8118C are plotted as a function of total annealing
time. Note that after 20 min all three rates become
constant and that the kGa is constant after 10 min.
At shorter times, the kGaN is larger and the kGaN
and the kS;Ga are equal for times 510 min. At
higher T the incubation time to achieve constant
rates (in time) will be shorter because the rates
Fig. 5. Same as Fig. 4, except the Ga surface accumulation rate
depend exponentially on T.
is plotted.
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 473
Fig. 8. Atomic force microscopy images of the GaN surface after annealing at 8118C for 60 min in 40 Torr of H2. The images are
(a) 40 40 mm2, (b) 10 10 mm2, and (c) 1 1 mm2.
The observation that the enhanced kGaN coin- N2 fractional flow (i.e. [N2]/[N2] +[H2]) at a T of
cides with the formation of liquid Ga on the 9928C and a P of 76 Torr. In pure N2, the kGaN
surface suggests that liquid Ga metal catalyzes the and kGa drop by a factor of 10 compared to the
decomposition of GaN [47,50,51]. To confirm this rates in pure H2. The kS;Ga decreases exponentially
observation, the GaN surface was predosed with as N2 is added, until in pure N2, Ga droplets are
trimethylgallium for 10 min at 6008C to deposit not observed.
liquid Ga on the surface. On the Ga predosed Unlike the kGaN measured in H2, no strong P
surface, the kGaN increased by 30% for annealing effect was observed for the kGaN in N2. This is
at 9928C in 6 SLM of H2 at 76 Torr. This increase shown in Fig. 10, where the kGaN in N2 are similar
in kGaN it is not as large as the 300% increase in at P of 76 and 150 Torr. Note in Fig. 10 that the
the kGaN as P increases at T ź 9928C as shown in kGaN at 76 (open squares) and 150 (solid circles)
Fig. 4. torr have the same curvature as N2 is substituted
Images of the surface morphology after etching for H2. The solid and dashed lines are cubic fits to
the Ga droplets in HNO3 are shown in Fig. 8. For kGaN vs. N2 fractional flow, with only the kGaN at
this sample the GaN film was annealed at 8118C 100% N2 not included in the fit. For T > 10008C,
for 60 min in 40 Torr of H2. The length scales are the kGaN in N2 was less uniform, typically
(a) 40 40 mm2, (b) 10 10 mm2, and (c) 1 1 mm2.
The raised features in Fig. 8(a) correlate well with
previous locations of Ga droplets when the images
of Ga droplets (see Fig. 1(a) of Ref. [22]) are
compared. The sunken features shown in Fig. 8(a)
Ú
extend 1000 A into the GaN film. In Fig. 8(a)
and (b) the measured step heights ranged from 50
Ú
to 200 A and the maximum height (dark to light)
Ú
relief in Fig. 8(a) was approximately 2000 A. In
Ú
Fig. 8(c), atomic step heights of 2.5 A are
measured. The AFM images indicate that the
terraces are atomically smooth, suggesting that
decomposition occurs in a step flow fashion.
4.3. Kinetic rates vs. P in N2 and mixed H2 and N2
flows
Fig. 9. GaN decomposition (filled circles), Ga surface accumu-
As N2 flow is substituted for H2 flow, the
lation (filled squares), and Ga desorption (open diamonds) rates
kGaN, kS;Ga, and kGa all decrease. This is shown in
are measured at 9928C and 76 Torr as a function of N2 fraction
Fig. 9, where the kinetic rates are plotted vs. the of the total N2 and H2 flow.
474 D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483
Fig. 11. Arrhenius plot of the GaN decomposition rate in H2 at
Fig. 10. GaN decomposition rate at 76 Torr (open squares) and
40 (filled circles) and 76 Torr (open circles) and in N2 at 76 Torr
150 Torr (filled circles) at 9928C as a function of N2 fraction of
(filled squares) and 150 Torr (open squares). Parameters for the
the total N2 and H2 flow.
exponential fits (lines 8 and 9) are listed in Table 1.
decomposing from the edges of the sapphire pieces
before the center.
4.4. Arrhenius parameters for GaN decomposition
The kGaN in pure H2 and pure N2 are plotted vs.
reciprocal T in Fig. 11. The kGaN was measured in
H2 at P of 40 Torr (filled circles) and 76 Torr (open
circles), and in N2 at P of 76 Torr (filled squares)
and 150 Torr (open squares). In H2 a pre-
exponential, A0, of 2.8 1029 cm 2 s 1 and an
activation energy, EA, of 3.4 eV are measured,
while in N2 a A0 of 1.2 1029 cm 2 s 1 and an EA
of 3.62 eV are measured. These values of A0 and
EA along with Log10ðA0Þ are listed on lines 8 (H2)
and 9 (N2) of Table 1.
Fig. 12. Arrhenius plot of the GaN decomposition rate in H2 at
As shown in Fig. 4, the kGaN in H2 shows a
40 (filled circles), 76 Torr (open circles), 150 Torr (filled
strong dependence on P. For this reason the kGaN diamonds), and 250 Torr (open diamonds). Parameters for the
were measured vs. T at P ranging from 40 to
exponential fits (lines 8, 11, 15, 19) are listed in Table 1.
250 Torr. An Arrhenius plot of the kGaN in H2 is
shown in Fig. 12 for P of 40 (filled circles), 76
59008C (line 11), (C) the P5150 Torr and T
(open circles), 150 (filled diamonds) and 250 Torr
59008C (line 15), and (D) the P5150 Torr and
(open diamonds). Plotting the kGaN this way,
T > 9008C (line 19).
shows that the two lower and two higher P have
nearly the same kGaN vs. 1=T. In addition, both
the low and high P data sets have a break in slope. 4.5. Decomposition rate vs. H2 flow rate
As a result, the kGaN in Fig. 12 were fit with four
exponentials: (A) the P476 Torr and T > 9008C The kGaN is plotted vs. H2 flow rate in Fig. 13 at
(line 8 of Table 1), (B) the P476 Torr and T a P of 76 Torr and a T of 9928C. As shown in
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 475
Fig. 14. GaN decomposition rate measured over several
months time. The GaN pieces were annealed at a temperature
Fig. 13. GaN decomposition (filled circles), Ga surface accu-
of 9928C and a pressure of 76 Torr in 6 SLM H2 for 10 min. The
mulation (open squares), and Ga desorption (filled diamonds)
filled circle is the measured decomposition rate reported in Ref.
rates at 9928C and 76 Torr as a function of H2 flow rate.
[19] the open squares were measured after cleaning the
showerhead, and the filled diamonds were measured after the
installation of a new showerhead.
Fig. 13, the kGaN increases by 25% as the H2 push
flow increases from 1 to 10 SLM. A much larger
increase in the kS;Ga is observed (4.5 times) as the
H2 push flow is increased. A similar increase in the
kGaN vs. the H2 push flow was previously reported
[28].
4.6. GaN decomposition measured vs. time
In Fig. 14, the kGaN is measured over 18 months
for 10 min anneals at 9928C in 6 SLM of H2 at
76 Torr. As shown in Fig. 14, the initial kGaN
measured in the Fall of 1997 [23] was larger than
measurements under identical annealing condi-
Fig. 15. Plots of (a) the GaN decomposition rate in 6 SLM of
tions in March through August of 1999 after
H2 (open circles) at Tź 9928C, (b) the GaN decomposition rate
cleaning the showerhead. In November of 1999 a
in 2 SLM of NH3 and 4 SLM of H2 (filled circles) at
new showerhead was installed on the CSS reactor.
Tź 10308C, (c) the GaN growth rate in 2 SLM of NH3,
After the showerhead change, the kGaN was
4 SLM of H2, and 32 mmol/min of TMGa (open squares) at
Tź 10308C, and (d) the calculated incorporation rate obtained
measured four times (see Fig. 14). The kGaN after
by adding curve b to curve c (filled diamonds) as a function of
the showerhead change are similar to those
reactor pressure.
measured in March through August of 1999.
4.7. Growth rate vs. growth pressure
Fig. 15(c) except that no TMGa was used. Note
that as the P increases the kGaN increases and the
The GaN growth rate at 10308C is plotted vs. P
GaN growth rate decreases. For comparison the
in Fig. 15(c). For the growth, 2 SLM NH3, 4 SLM
kGaN in pure H2 at 9928C vs. P (from Fig. 4) are
H2, and 32 mmol of TMGa were used. In Fig. 15(b)
plotted in Fig. 15(a). The kGaN in both H2
the kGaN is plotted for the same conditions used in
476 D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483
(Fig. 15(a)) and mixed H2 and NH3 (Fig 15(b)) can be seen in bottom left corner of Fig. 2(b). The
have a similar shape for P from 40 to 300 Torr. In close fit by the combination power law and bell
fact, if the kGaN in Fig. 15(b) are multiplied by shaped curve suggests good agreement between the
30, the kGaN in H2 (Fig. 15(a)) and in mixed H2 droplet size distributions and the theoretical model
and NH3 (Fig. 15(b)) overlap, suggesting a similar [57]. Further comparison between the droplets and
mechanism for GaN decomposition in both gas the model [57] is outside this papers scope.6
environments. Clearly, as the H2 pressure increases, both the
kGaN (Fig. 4) and the kS;Ga (Fig. 5) increase.
Despite this coincidence, the presence of droplets
5. Discussion are most likely the result of the increased kGaN at
higher P. Liquid Ga enhances GaN decomposition
In the previous section, changes in the GaN locally on the surface [51], because H2 is cracked
decomposition rate ðkGaNÞ, liquid Ga accumula- on liquid Ga surfaces [59,60]. However, the
tion rate ðkS;GaÞ, and Ga desorption rate ðkGaÞ primarily reason for the increases in kGaN and
were presented for a range of gas mixtures, P, and kS;Ga vs. H2 pressure is because the N removal rate
T. In this section we compare the measurements exceeds kGa as is discussed in Section 5.4.
presented in Section 4 to the measurements of
GaN decomposition reported in the literature and
5.2. Kinetics of Ga desorption
offer explanations as to how the GaN decomposi-
tion mechanism changes as the annealing condi-
Except for the increase near 80 120 Torr, the
tions change.
kGa does not depend on H2 pressure. As shown in
Table 1E, the EA for Ga desorption are very
5.1. Ga droplet accumulation and growth
similar. The EA measured for Ga desorption are
nearly identical to the EA of 2.8 eV measured for
Previous studies of GaAs decomposition by
Ga desorption from liquid Ga (line 20, Table 1)
Carlow et al. showed that Ga droplets form in the
[39] and the heat of vaporization for Ga which is
absence of an As overpressure when the substrate
2.8 eV [56]. This close agreement of EA suggests
is heated to 6608C [58]. Further heating of the
that the barrier to Ga desorption is the breaking of
GaAs surface causes the Ga droplets to grow in
a single Ga Ga bond. Despite the similarity in the
size and coalesce. Carlow et al. suggested that the
EA, the value of A0 is 12 times larger in H2
Ga droplet coalescence mechanism was consistent
compared to N2 (compare lines 23 and 24 in Table
with the growth model of Family and Meakin [57].
1). Possible origins for the difference in A0 are
To verify the Family and Meakin model, Carlow
discussed in Appendix B. An explanation for the
et al. used the combination of a power law and bell
increase in the kGa between 80 and 120 Torr in
shaped curve to fit the Ga droplet distribution.
Fig. 6 is suggested in Appendix A.
Likewise, this combination of fits was used in
Fig. 3.
5.3. Kinetic rates vs. time
The coalescence growth model of Family and
Meakin assumes that when two spherical droplets
When measuring kinetic rates it is important
coalesce they form a single droplet on the center of
that the rates approach a constant value in time,
mass of the original droplets [57]. The droplets
6
initiate from the Ga accumulation on the surface,
According to the coalescence model of Ref. [54], the change
because the kGaN > the kGa. When droplets begin in droplet size from 30 to 45 min should be larger than the
experimental Ga droplet distributions, where no significant
to coalesce, the droplet size increases [57]. Droplet
increase in the droplet size is observed between Figs. 2(c) and
formation is shown in Figs. 2(a) and (b), where the
2(d). The discrepancy between the model and the experimental
droplets are 1 mm. Over time these small
data may be due to a change in kGa, which depends on the
droplets coalesce and the droplet size increases as
surface area as described in Section 5.2 and is not included in
shown in Fig. 3. Evidence of droplet coalescence the coalescence model.
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 477
especially if time is varied to measure slow rates. In 5.5. Reaction order for NH3 production during
Fig. 7, the kinetic rates at a T of 8118C and P of GaN decomposition in H2
150 Torr are shown vs. annealing time. Despite a
short induction time, which is 10 min, all three Dissociated H on the GaN surface can recom-
rates become constant at longer times. Because the bine with surface N to form NH3. This reaction
rates become constant, the slower kinetic rates channel is shown in Fig. 10 by the decrease in the
could be measured by increasing the anneal time. kGaN as N2 is substituted for H2. The solid
Several factors may contribute to the observed (150 Torr) and dashed (76 Torr) lines are cubic fits
induction time, including an increase in the kGaN at to the data, showing that kGaN is proportional to
the surface due to the larger concentration of [H2]3. This is the expected number of H2 that need
subsurface H [61]. The larger surface H concentra- to be cracked on the surface to form NH3, if only
tion may enhance decomposition mechanisms with one H per H2 molecule attaches to a surface N
lower EA such as NH3 formation. Because the atom. Recently, Mayumi et al. measured the kGaN
kGaN is initially larger, the kS;Ga observed at 10 min to be proportional to partial P of H2 to the 2/3
in Fig. 7 is larger. Note that at higher T the power [36]. The difference in power dependence
induction time will be shorter than 10 min, due to between the data in Fig. 10 (third power) and in
the increased rates at higher T. Ref. [36] (2/3 power) is not currently understood.
This difference may be due to the lower T used in
Ref. [36], which would reduce the H2 desorption
5.4. Rate of atomic H production in CSS reactor rate from Ga atoms on the surface. Further work
is needed to understand the reactor influence on
For NH3 formation, H2 must be cracked on the the H2 dissociation rate, because it plays a large
GaN surface to form adsorbed H. Ga metal is role in GaN decomposition and growth.
known to dissociate H2 at high temperatures to
form Ga hydrides [59,60]. Recently, Bartram and 5.6. Grouping of GaN decomposition kinetic rates
Creighton have shown that H2 can be dissocia-
tively chemisorbed at elevated temperatures on the The dependence of the GaN decomposition
GaN surface [19]. Under an atomic H flux of kinetic parameters on process conditions is sum-
1018 cm 2 s 1, King et al. have calculated that at marized in Table 1. Unexpectedly, the measured
least 50% of the surface N bonds should be H kinetic parameters fall into one of four different
terminated at 8008C, while at higher atomic H groups, denoted as A D. The four distinct slopes
fluxes ( 1020 cm 2 s 1), 50% H termination in Fig. 12 imply that four different decomposition
occurs at 10008C [62]. reactions become rate limiting. Here, we propose
Although the degree of atomic H, Hat, how each of these reactions limit GaN decom-
production was not measured, an estimate can be position.
made from the NH3 production rate. In Fig. 10, In Table 1A, the Arrhenius parameters are listed
the kGaN is 1.6 1016 cm 2 s 1 at 150 Torr at for T> 9008C, in H2 at P of 40 and 76 Torr (line 8)
9928C. In order to produce NH3 at this rate, the and in N2 at P of 76 and 150 Torr (line 9). Under
steady state Hat flux must be greater than these conditions, an EA of 3.4 0.2 eV is measured
4.8 1016 cm 2 s 1 or 42 ML s 1. However, NH3 i nH2, while a slightly larger EA of 3.62 0.14 eV is
formation from surface N and H requires the measured in N2. The values for the EA compare
formation of three N H bonds in three sequential well to previous EA measured in vacuum (lines 1
reversible chemical reactions. This suggests that 7, Table 1), suggesting that under these conditions
the Hat flux generated at the GaN surface the GaN decomposition mechanism is similar. The
is 1 4 orders of magnitude larger than proposed rate limiting step for GaN decomposi-
4.8 1016 cm 2 s 1. Certainly, Hat is produced at tion is the formation and desorption of N2
a substantial rate in the MOVPE reactor during [26,31,32]. Forming N2 on the surface would
growth and decomposition. involve the diffusion of at least one N atom to
478 D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483
form N2, however N surface diffusion is unlikely partially hydrogenated N species bond to form a
because of a large kinetic barrier [6,16]. Alterna- hydrogenated dinitrogen molecule ðN2HxÞ, which
tively, a subsurface N could combine with a then desorbs from the surface. The formation of
surface N to form N2. This mechanism has this molecule would have a lower EA compared to
recently been observed by Bartram and Creighton, the EA for N2 formation and desorption. While the
15
where N14N was shown to desorb after dosing H surface coverage at lower P is larger at lower T,
15
the surface with NH3 [19]. Nitrogen formation the coverage is not large enough to form NH3.
from surface and subsurface N atoms would For T59008C and P5150 Torr the kGaN
follow first order desorption kinetics as previously increases compared to the kGaN at lower P (Table
observed [6]. The EA (Table 1A) for forming N2 1C). This can be seen in Fig. 12 by comparing line
would then equal to the energy required to remove 11 (lower P) and line 15 (higher P). Under these
a subsurface N atom, i.e. the energy required to annealing conditions an EA of 1.7 eV is calculated
create a N-vacancy. Estimates of the N-vacancy (line 15 of Table 1). Similar EA have been observed
formation energy range from 2.7 eV for a Ga-rich by Morimoto [26] (line 13, Table 1) and Rebey
surface to 4.6 eV for a N-rich surface [63]. Note et al. [28] (line 14, Table 1) at similar T and P. The
that the N-vacancy formation energy is within the rate limiting step for GaN decomposition under
range of measured EA in Table 1A. The increased these conditions (i.e. T59008C and P 5150 Torr)
A0 in H2 vs. N2 is most likely related to the is NH3 desorption from the surface. Recently,
increased A0 also observed for Ga desorption in Bartram and Creighton found that the NH3
H2 vs. N2.7 desorption rate from the GaN surface peaks at a
For T59258Cat P476 Torr, the measured EA T of 4478C [19]. Using Tź 4478C, a Redhead
and A0 decrease for the kGaN (see Table 1B). The analysis gives an EA of 1.6 2.0 eV if an A0 of 1013
decrease in EA and A0 is seen in Fig. 12, where for 1020 cm 2 s 1 is assumed [64]. The range of 1.6
T59258C the kGaN is larger than the expected 2.0 eV agrees well with the EA listed on lines 13 15
kGaN based on extrapolation of line 8. A fit to kGaN of Table 1, suggesting that NH3 desorption limits
for T59258C gives an EA of 0.98 eV (line 11 of GaN decomposition at lower T and higher P.
Table 1). This EA is close to the EA (0.91 eV) Finally, as the T is increased at higher P, the
measured Kobayashi and Kobayashi in 76 Torr of kGaN is relatively constant as shown by line 19 in
H2 (line 10 of Table 1) [37]. The decrease in EA Fig. 12 (see Table 1D). An EA of 0.34 eV is
from 3.5 to 1 eV implies a change in the rate calculated from line 19. This value of EA agrees
limiting step for decomposition. At lower T, the H well with the EA of 0.38 eV measured by Rebey
coverage will increase due to the decreased H et al. (line 17, Table 1) [28] and the EA of 0.44 eV
desorption rate from N sites [37], resulting in measured by Mayumi et al. (line 18, Table 1) [36].
increased stability of  NH and  NH2 species.8 At these T and P, Ga droplets are prevalent
Because N H bonds are strong (i.e. 4.5 eV [56]), because kGaN exceeds kGa. The change in slope
the  NH and  NH2 species will be more weakly near 9008C suggests that NH3 desorption no
bound to the surface. We propose that two longer limits the kGaN. Instead the lower EA is
suggestive of a kinetic barrier for metallic surface
diffusion [65]. Recently, the Ga surface diffusion
7
The increase in the pre-exponential factor, A0, for GaN
barrier on a Ga terminated surface was calculated
decomposition, kGaN, i nH2 vs. N2 is similar to the increased A0
to be 0.4 eV (line 16 in Table 1) [38]. This value
for Ga desorption in H2 vs. N2 described in Ref. [23]. Note that
for the Ga diffusion barrier agrees well with the
in both cases the kGaN and kGa are larger in H2 than in N2. In
Appendix B, it was proposed that the increase A0 for Ga experimentally measured EA listed in Table 1D. As
desorption is due to an increased entropy of more volatile Ga
a result, the limiting factor for GaN decomposi-
H species. Increased desorption of Ga would increase the
tion is the rate of Ga diffusion so that the next
number of N exposed at the surface, which in turn would
layer of N atoms are exposed for removal via the
increase the number of N2 desorbing per second.
8
formation of NH3. Note that the kS;Ga plotted in
An increased surface and subsurface H coverage were
measured in MOVPE grown GaN films in Ref. [61]. Fig. 5 drops dramatically for T59008C and
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 479
P > 100 Torr, in agreement with the change in the increase GaN decomposition should accelerate the
rate limiting step from Ga diffusion limited (Table rate of the GaN NL evolution.
1D) to NH3 desorption limited (Table 1C).
It is apparent from the A D grouping in Table 1 5.8. Relationship of GaN decomposition to the
that GaN decomposition strongly depends on quality of GaN growth
P, T, and gas type (i.e. H2 or N2). One key to
understanding the changes in the kGaN is the rate While there is not an obvious link between the
of Hat production. While the production rate of high T GaN growth and GaN decomposition, we
Hat on the GaN surface may be reactor dependent, have grown better quality GaN growth in the CSS
in general it should increase as P increases. In fact, reactor at P > 100 Torr [21,22]. When the GaN
one of the major distinguishing characteristics epitaxial layer is grown at 150 Torr, we find a near
between atmospheric and low P GaN growth may doubling of the electronic mobility (m > 500 cm2/
be the Hat production rate and its subsequent V s) compared to films growth at 76 Torr on
influence on GaN growth and decomposition. identically grown NL [21]. For films grown above
100 Torr, the GaN grain size increased from 51to
2 5 mm, which may be directly responsible for the
5.7. GaN nucleation layer decomposition and increased mobility [21,22]. Other groups using CSS
evolution or high speed rotating disk reactors have also
reported improved electric properties when GaN is
Several groups have shown that the nucleation grown at P > 100 Torr [29,72]. Watanabe and
layer, NL, evolves during the ramp from low to coworkers have shown that the dislocation density
high T used for the main layer growth [21,29,66 decreases, as the growth P is increases [73].
70]. As shown by X-ray diffraction, annealing of Previously, we speculated that having some degree
the GaN NL at high T increases its crystallinity of GaN decomposition during GaN growth is
[66]. Recently Sugiura et al. have shown that a T important for removing more weakly bound Ga
of 850 9008C(P ź 760 Torr) produces an optimal and N atoms, which would increase ordering
NL for high quality films, while above 9508C during growth [16]. From a thermodynamic
reevaporation of the NL can occur [67]. Annealing perspective, lower activation barriers for decom-
the NL at higher P increases the NL roughness position bring the GaN growth closer to equili-
and the GaN electron mobility when compared to brium.
annealing the NL at lower P [29]. Similar results Theoretically, growth rate equals the incorpora-
have been reported by Wickenden et al., where tion rate minus the decomposition rate [16]. At a
higher m and larger GaN grain sizes were produced constant growth flux, the incorporation rate
on GaN NL annealed at 150 Torr compared to should not change; therefore any increase in the
76 Torr [21]. The grain size in the NL also decomposition rate will decrease the growth rate.
increases during annealing at 10808C with the This is shown in Fig. 15(d), where the incorpora-
tops of the GaN NL grains becoming flatter [68 tion rates (solid diamonds) are calculated vs. P by
70]. The flattening occurs because the NL grains adding the kGaN (Fig. 15(b)) to the GaN growth
decomposition during the high T anneal and the rates (Fig. 15(c)). Below 150 Torr, the incorpora-
Ga atoms reform GaN by reaction with NH3. If tion rate is constant at 1.7 1015 cm 2 s 1,
the N loss from the surface is not compensated by however at 300 Torr the incorporation rate de-
enough NH3 during NL annealing, the entire GaN creases. This decrease in the incorporation rate
NL decomposes. This has been observed by may be due to gas phase pre-reaction [74] or site
Kobayashi et al., who noted the loss of the GaN blocking at the GaN surface [75]. At equilibrium,
NL when heated to 10208C in NH3 (0.25 SLM) the decomposition and incorporation rates are
and H2 [71]. Because the kGaN is enhanced at equal and no growth occurs. Note that at
higher P, the NL may evolve at a faster rate when 300 Torr, the kGaN and the incorporation rate are
annealed at high P. Therefore, conditions that the closest for all P, suggesting that at this P the
480 D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483
GaN growth is closest to equilibrium. The kGaN in decreased incorporation rate at 300 Torr in
pure H2 (Fig. 15(a) and in mixed H2 and NH3 Fig. 15(d). Recently, Briot and coworkers pro-
(Fig. 15(b)) increase similarly as P increases posed that the GaN growth rate decreases at high
suggesting that the same decomposition mechan- NH3 flow because the excess NH3 blocks surface
ism, i.e. H2 dissociation and removal of surface N sites necessary for growth [75]. When large
as NH3, occurs in both cases as P increases. amounts of Hat are generated on the GaN surface,
Other groups have also observed decreases in the H coverage may become large enough to block
growth rate when the growth P is increased surface sites, thereby reducing the incorporation
[28,33,76,77]. For example, Khan et al. observed and subsequent growth rates. Further study is
a factor of 2 decrease in the GaN growth rate as P necessary to determine the relative contributions
was increased from 40 100 Torr [76]. Reductions of site blocking and kGaN to limiting the GaN
in growth rate have also been observed when H2 is growth.
substituted for N2 as the carrier gas [33]. The use Finally, a benefit of large surface H coverages
of higher H2 push flows during growth has might be an increase in the Ga diffusion length [16]
previously been shown to lead to discontinuous through the formation of more volatile and mobile
growth followed by no GaN film growth [77]. Ga H species. For example, Morishita et al.
Sasaki also showed that the transition between measured an increased Ga diffusion length in H2
continuous and discontinuous growth occurred at and Hat. The increased diffusion length was
lower H2 push flows when the growth P was speculated to be due to the formation of Ga H
increased from 70 to 780 Torr [77]. The observa- species [81]. In addition, Okamoto and coworkers
tions of Sasaki are consistent with the increase in have recently showed a suppression of 3D growth
kGaN shown in Fig. 14 and in Ref. [28]. For each morphology when Hat is used during MBE growth
case the likely cause for the decreased GaN growth of GaN [82], implying increased surface mobility
rate is an increase in the kGaN. of the Ga H species. Note that the A0 for the kGa
Besides affecting the GaN growth rate, the GaN are larger in H2 (line 23) than in N2 (line 24) as
crystal quality is also improved when pure H2 is shown in Table 1. As proposed in Appendix B,
used instead of pure N2. In a comparison study of Ga H species that desorb may redeposit at a
GaN growth in H2 vs. N2, Kistenmacher et al. different position along the GaN surface, conse-
showed that the FWHM of the GaN films grown quently increasing the Ga diffusion length. High
in H2 had narrower X-ray rocking curve line- levels of surface H may also scavenge and remove
widths and were better aligned compared (i.e. graphitic C from the surface. Growth parameters
smaller mosaic dispersion) to GaN films grown in that enhance kGaN should also decrease the surface
only N2 [78]. Schon and coworkers find smoother and bulk C concentrations.
¨
morphologies and better electrical properties when
growth is conducted in H2 compared to N2 [79].
Tadatomo and coworkers find that the lateral 6. Conclusions
growth rate in N2 increases compared to growth in
H2, however, there is a greater degree of mis- This paper has shown how the kinetics rates of
orientation in the overgrown region and a smooth GaN decomposition, Ga surface accumulation,
surface cannot be attained [80]. These studies and Ga desorption change under different anneal-
indirectly suggest higher quality growth in H2 ing conditions. We have shown that the GaN
compared to growth in N2. We speculate that an decomposition kinetic parameters can be arranged
increase in the kGaN when H2 is used as the carrier into four distinct groups. We have also suggested
gas plays a role in the initial grain formation and possible chemical reactions, which ultimately limit
hence the resulting GaN quality. the decomposition. The reformation of ammonia
It is not clear to what extent the surface H atoms is suggested when GaN is decomposed in H2 at
etch the GaN surface or block sites necessary for higher P. The observation of the NH3 reformation
GaN growth. Site blocking may account for the reaction is important for growth because it
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 481
suggests that at higher P GaN growth is closer to Appendix A
equilibrium [16]. Unlike GaN decomposition, the
Ga desorption kinetics are similar in H2 and N2 In this section, we explain the peak in the Ga
environments with only a difference in the pre- desorption rate between 80 and 120 Torr in Fig. 6.
exponential factor. This peak is probably due to an increase in the
The occurrence of GaN decomposition during surface area, S, of the Ga droplets that occurs in
GaN growth has several implications. Clearly, this pressure regime. An increase in S is expected
GaN decomposition during growth affects the from the model of Family and Meakin as the
quality of the GaN films. Annealing the NL at droplets begin to coalesce [57]. If the droplet size,
higher temperatures and pressures allows the NL R, has a combination power law, i.e. R x, and bell-
to evolve to a greater extent for the formation of shaped distribution, i.e. expð dðR R0Þ2Þ, the
large grained GaN [21]. Limiting GaN grain droplet size distribution, NðRÞ, is given by
formation at higher P (because of the increased N0R x expð dðR R0Þ2Þ, where N0 is the num-
GaN decomposition) also contributes to the ber density as R! 0; x is the power law depen-
increased GaN grain size observed in the full dence, d is a constant proportional to the width of
GaN films [21,22]. From Fig. 14 it appears that the the bell shaped distribution, and R0 is the peak
decomposition rate is nearly constant; this even droplet size of the bell shaped distribution. The S
after cleaning the showerhead and replacement of a spherical droplet on a flat surface, is given by
with a new showerhead. pR2ð1 þ tan2ðy=2ÞÞ, where y is the contact angle
The extent of GaN decomposition that occurs [83]. To estimate the total S of the Ga droplets
during growth is condition (see Fig. 15(b)) as well requires summing S of each droplet over the
as reactor dependent [16]. Here, we have shown droplet distribution, NðRÞ. Assuming that the
that the H2 dissociation rate plays a critical role in contact angle between the Ga droplets and
the decomposition rates. The kinetic rates for the GaN surface is constant, the surface area
many of the GaN growth reactions such as the distribution, SðRÞ Åºconst ðNðRÞÞ2 ź const ½N0
cracking rate of H2 may explain, at least partly, the R xexpð dðR R0Þ2ÞŠ2. Integration of SðRÞ over
reason why GaN growth conditions are not R would provide the total S. Therefore, the change
directly transferable from reactor to reactor. in SðRÞ can be estimated from the change in NðRÞ
Finally, because of the differences in reactor de- as the droplets initially form, begin to coalesce,
signs (horizontal, vertical, two flow, etc.) and flow and grow in size.
dynamics (gas composition, preheating, residence Based on Family and Meakin s theory, the
time, etc.), the measured GaN decomposition rates power law term, R x, is nearly constant and only
in other types of reactors may differ from those the bell shaped term, expð dðR R0Þ2Þ, changes
presented here. However, the general trends and as the droplets coalesce [57]. This is observed in the
rate limiting decomposition mechanisms presented fits in Figs. 3 and 4 of Ref. [57]. Initially, when the
here are likely present in all GaN MOVPE Ga droplets first appear, the expð dðR R0Þ2Þ
reactors. Clearly, more research is necessary to term is negligible, SðRÞ /ðN0R xÞ2, and integra-
measure kinetic factors that control GaN growth. tion of SðRÞ over R will be a constant. As the
droplets begin to coalesce, there will be a large
intensity of droplets at a small value of R0,
Acknowledgements increasing the expð dðR R0Þ2Þ term, which will
increase the total S. As the droplets further
We thank V.A. Shamamian, V.M. Bermudez, coalesce, the expð dðR R0Þ2Þ term will decrease
R.J. Gorman, J.A. Freitas, Jr., and M. Fatemi for as R0 increases and as a result the total S will
their assistance in this work. We thank J.E. Butler decrease. Eventually, as the droplets become
for the use of a phase contrast microscope. This larger, the contribution to the total S from the
work was supported by the Office of Naval bell shaped distribution will again become negli-
Research. gible, SðRÞ /ðN0R xÞ2, and S will again be a
482 D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483
constant. Therefore, the Ga droplet surface area is [2] S. Nakamura, M. Senoh, S. Nagahama, N. Iwasa, T.
Matushita, T. Mukai, MRS Internet J. Nitride Semicond.
initially constant before coalescence, increases at
Res. 4S1 (1999) G1.1.
the onset of coalescence, and then decreases back
[3] S.N. Mohammad, A.A. Salvador, H. Morkoc, Proc. IEEE
to the initial value as the Ga droplets grow in size.
83 (1995) 1306.
This behavior is qualitatively observed in Fig. 6, as
[4] S. Strite, H. Morkoc, J. Vac. Sci. Technol. B 10 (1992)
the H2 pressure increases at 9928C and the Ga
1237.
[5] R.F. Davis, Proc. IEEE 79 (1991) 702.
droplets initially form (10 40 Torr), begin to
[6] O. Brandt, H. Yang, K.H. Ploog, Phys. Rev. B 54 (1996)
coalesce (76 150 Torr), and increase in size (250
4432.
700 Torr).
[7] R. Held, B.E. Ishaug, A. Parkhomovsky, A.M. Dabiran,
P.I. Cohen, J. Appl. Phys. 87 (2000) 1219.
[8] D. Crawford, R. Held, A.M. Johnston, A.M. Dabiran, P.I.
Cohen, MRS Internet J. Nitride Semicond. Res. 1 (1996)
Appendix B
12.
[9] E.J. Tarsa, B. Heying, X.H. Wu, P. Fini, S.P. DenBaars,
The larger pre-exponential factor, A0, for Ga
J.S. Speck, J. Appl. Phys. 82 (1997) 5472.
desorption in H2 compared to N2 may be due to an
[10] C.R. Jones, T. Lei, R. Kaspi, K.R. Evans, Mater. Res.
increase in the frequency factor, v, or the reaction
Soc. Symp. Proc. 395 (1996) 141.
entropy, S, if a standard thermodynamic inter- [11] R. Held, D.E. Crawford, A.M. Johnston, A.M. Dabiran,
P.I. Cohen, Surf. Rev. Lett. 5 (1998) 913.
pretation of A0 is used, i.e. A0 ź veðS=kÞ, where k is
[12] S.P. DenBaars, S. Keller, Semicond. Semimet 50 (1998) 11.
the Boltzman constant [84]. Previously, we pro-
[13] O. Ambacher, J. Phys. D 31 (1998) 2653 2710.
posed that a reduced v for Ga desorption in N2 vs.
[14] O. Briot, J.P. Alexis, M. Tchounkeu, R.L. Aulombard,
H2 might be possible because of the more efficient
Mater. Sci. Eng. B 43 (1997) 147.
heat transfer between N2 and the hot surface [24]. [15] R.D. Dupuis, J. Crystal Growth 178 (1997) 56.
[16] D.D. Koleske, A.E. Wickenden, R.L. Henry, W.J.
Additionally, increases in v and S might be
DeSisto, R.J. Gorman, J. Appl. Phys. 84 (1998) 1998.
possible if the Ga adatoms become hydrogenated
[17] S.S. Liu, D.A. Stevenson, J. Electrochem. Soc. 125 (1978)
in H2. Increases have been observed in the A0 for
1161.
Ga surface diffusion when the GaAs surface is
[18] M. Kamp, M. Mayer, A. Pelzmann, K.J. Ebeling, Mater.
exposed to H2 and atomic H [81]. In this study, the Res. Soc. Symp. Proc. 449 (1997) 161.
[19] M.E. Bartram, J. R. Creighton, MRS Internet J. Nitride
increase in A0 for Ga surface diffusion was
Semicond. Res. 4S1 (1999) G3.68.
attributed to the formation of GaH species, which
[20] H. Amano, N. Sawaki, I. Akasaki, Y. Toyoda, Appl. Phys.
were more mobile on the surface [81]. Recent
Lett. 48 (1986) 353.
thermodynamic calculations show that when GaN
[21] A.E. Wickenden, D.D. Koleske, R.L. Henry, R.J. Gor-
is heated in H2 near 10008C GaH species are more
man, J.C. Culbertson, M.E. Twigg, J. Electron. Mater. 28
(1999) 299.
abundant in the gas phase than Ga atoms [50].
[22] A.E. Wickenden, D.D. Koleske, R.L. Henry, R.J. Gor-
Another calculation shows that Ga desorption in
man, M.E. Twigg, M. Fatemi, J.A. Freitas Jr., W.J.
mixed H2 and NH3 environments is in the form of
Moore, J. Electron. Mater. 29 (2000) 21.
Ga hydrides rather than Ga atoms [85]. Finally,
[23] D.D. Koleske, A.E. Wickenden, R.L. Henry, M.E. Twigg,
the gas phase entropy for GaH is larger than for
J.C. Culbertson, R.J. Gorman, Appl. Phys. Lett. 73 (1998)
2018; Erratum Appl. Phys. Lett. 75 (1999) 1646.
Ga atoms [56]. Therefore, the larger A0 measured
[24] D.D. Koleske, A.E. Wickenden, R.L. Henry, M.E. Twigg,
in H2 vs. N2 could be the result of forming Ga H
J.C. Culbertson, R.J. Gorman, MRS Internet J. Nitride
species, which have larger entropy and/or the
Semicond. Res. 4S1 (1999) G3.70.
reduced heat transfer between H2 and the surface,
[25] D.D. Koleske, A.E. Wickenden, R.L. Henry, MRS
discussed in Ref. [24].
Internet J. Nitride Semicond. Res. 5S1 (2000) W3.64.
[26] Y. Morimoto, J. Electrochem. Soc. 121 (1974) 1383.
[27] M. Furado, G. Jacob, J. Crystal Growth 64 (1983) 257.
[28] A. Rebey, T. Boufaden, B. El Jani, J. Crystal Growth 203
References
(1999) 12.
[29] J. Han, T.-B. Ng, R.M. Biefeld, M.H. Crawford, D.M.
[1] S. Nakamura, M. Senoh, T. Mukai, Appl. Phys. Lett. 64 Follstaedt, Appl. Phys. Lett. 71 (1997) 3114.
(1994) 1687. [30] R.J. Sime, J.L. Margrave, J. Phys. Chem. 60 (1956) 810.
D.D. Koleske et al. / Journal of Crystal Growth 223 (2001) 466 483 483
[31] Z.A. Munir, A.W. Searcy, J. Chem. Phys. 42 (1965) 4223. [59] H. Remy, Treatise on Inorganic Chemistry, Elsevier, New
[32] R. Groh, G. Gerey, L. Bartha, J.I. Pankove, Phys. Status
York, 1960, p. 18.
Solidi A 26 (1974) 353.
[60] W.R.S. Garton, Proc. Phys. Soc. A 64 (1951) 509.
[33] O. Ambacher, M.S. Brandt, R. Dimitrov, T. Metzger, M.
[61] M.M. Sung, J. Ahn, V. Bykov, J.W. Rabalais, D.D.
Stutzmann, R.A. Fischer, A. Miehr, A. Bergmaier, G.
Koleske, A.E. Wickenden, Phys. Rev. B 54 (1996) 14652.
Dollinger, J. Vac. Sci. Technol. B 14 (1996) 3532.
[62] S.W. King, E.P. Carlson, R.J. Therrien, J.A. Christman,
[34] R. Held, D.E. Crawford, A.M. Johnston, A.M. Dabiran,
R.J. Nemanich, R.F. Davis, J. Appl. Phys. 86 (1999) 5584.
P.I. Cohen, Surf. Rev. Lett. 5 (1998) 913.
[63] I. Gorczyca, A. Svane, N.E. Christensen, Phys. Rev. B 60
[35] Grandjean, J. Massies, F. Semond, S.Yu. Karpov, R.A.
(1999) 8147.
Talalaev, Appl. Phys. Lett. 74 (1999) 1854.
[64] P.A. Redhead, Vacuum 12 (1962) 203.
[36] M. Mayumi, F. Satoh, Y. Kumagai, K. Takemoto, A.
[65] E.G. Seebauer, C.E. Allen, Prog. Surf. Sci. 49 (1995) 265.
Koukitu, Jpn. J. Appl. Phys. 39 (2000) L707.
[66] A.E. Wickenden, D.K. Wickenden, T.J. Kistenmacher, J.
[37] Y. Kobayashi, N. Kobayashi, J. Crystal Growth 189/190
Appl. Phys. 75 (1994) 5367.
(1998) 301.
[67] L. Sugiura, K. Itaya, J. Nishio, H. Fujimoto, Y. Kokubun,
[38] T. Zywietz, J. Neugebauer, M. Scheffler, Appl. Phys. Lett.
J. Appl. Phys. 82 (1997) 4877.
73 (1998) 487.
[68] X.H. Wu, D. Kapolnek, E.J. Tarsa, B. Heying, S. Keller,
[39] R.E. Honig, D.A. Kramer, RCA Rev. 30 (1969) 285.
B.P. Keller, U.K. Mishra, S.P. DenBaars, J.S. Speck,
[40] P. Hacke, G. Feuillet, H. Okumura, S. Yoshida, Appl.
Appl. Phys. Lett. 68 (1996) 1371.
Phys. Lett. 69 (1996) 2507.
[69] S.D. Hersee, J. Ramer, K. Zheng, C. Kranenberg, K.
[41] W.C. Johnson, J.B. Parsons, M.C. Crew, J. Phys. Chem.
Malloy, M. Banas, M. Goorsky, J. Electron. Mater. 24
36 (1932) 2651.
(1995) 1519.
[42] C.D. Thurmond, R.A. Logan, J. Electrochem. Soc. 119
[70] K. Lorenz, M. Gonsalves, Wook Kim, V. Narayanan, S.
(1972) 622; R.A. Logan, C.D. Thurmond, J. Electrochem.
Mahajan, Appl. Phys. Lett. 77 (2000) 3391.
Soc. 119 (1972) 1727.
[71] Y. Kobayashi, T. Akasaka, N. Kobayashi, Jpn. J. Appl.
[43] G. Jacob, R. Madar, J. Hallais, Mater. Res. Bull. 11 (1975)
Phys. 37 (1998) L1208.
445.
[72] B.T. McDermott, R. Pittman, E.R. Gertner, J. Krueger, C.
[44] M.R. Lorenz, B.B. Binkowski, J. Electochem. Soc. 109
(1962) 24. Kisielowski, Z. Lilienthal-Weber, and E. Weber, Talk D2.2
[45] M. Kuball, F. Demangeot, J. Frandon, M.A. Renucci, J. at Fall Material Research Society Meeting, Boston, 1997.
Massies, N. Grandjean, R.L. Aulombard, O. Briot, Appl. [73] A. Watanabe, H. Takahashi, T. Tanaka, H. Ota, K.
Phys. Lett. 73 (1998) 960.
Chikuma, H. Amano, T. Kashima, R. Nakamura, I.
[46] C.J. Sun, P. Kung, A. Saxler, H. Ohsato, E. Bigan, M.
Akasaki, Jpn. J. Appl. Phys. 38 (1999) L1159.
Razeghi, D.K. Gaskill, J. Appl. Phys. 76 (1994) 236.
[74] A. Thon, T.F. Kuech, Appl. Phys. Lett. 69 (1996) 55.
[47] A.S. Bolgar, S.P. Gordienko, E.A. Ryklis, V.V. Fesenko,
[75] O. Briot, S. Clur, R.L. Aulombard, Appl. Phys. Lett. 71
in: G.V. Samsonov (Ed.), Chemistry and Physics of the
(1997) 1990.
Nitrides, Naukova, Dumka, Kiev, 1968, p. 151 (in
[76] M.A. Khan, R.A. Skogman, R.G. Schulze, and M.
Russian).
Gershenzon, Appl. Phys. Lett. (1983) 43492.
[48] I.G. Pichugin, D.A. Yas kov, Izv. Akad. Nauk SSSR,
[77] T. Sasaki, J. Crystal Growth 129 (1993) 81.
Neorganicheskie Mater. 6 (1970) 1973.
[78] T.J. Kistenmacher, D.K. Wickenden, M.E. Hawley, R.P.
[49] R.C. Schoonmaker, A. Buhl, J. Lemley, J. Phys. Chem. 69
Leavitt, Mater. Res. Soc. Symp. Proc. 395 (1996) 261.
(1965) 3455.
[79] O. Schon, B. Schineller, M. Heuken, R. Beccard, J. Crystal
¨
[50] A. Davydov, private communication, 1998.
Growth 189/190 (1998) 335.
[51] A. Pisch, R. Schmid-Fetzer, J. Crystal Growth 187 (1998)
[80] K. Tadatomo, Y. Ohuchi, H. Okagawa, H. Itoh, H.
329.
Miyake, K. Hiramatsu, MRS Internet J. Nitride Semi-
[52] T. Kojima, N. Kawai, T. Nakagawa, K. Ohta, T.
cond. Res. 4S1 (1999) G3.1.
Sakamoto, M. Sawashima, Appl. Phys. Lett. 47 (1985)
[81] Y. Morishita, Y. Nomura, S. Goto, Y. Katayama, Appl.
286.
Phys. Lett. 67 (1995) 2500.
[53] M. Van Hove, P.I. Cohen, Appl. Phys. Lett. 47 (1985) 726.
[82] Y. Okamoto, S. Hashiguchi, Y. Okada, M. Kawabe, Jpn.
[54] Y. Seki, H. Watanabe, J. Matsui, J. Appl. Phys. 49 (1978)
J. Appl. Phys. 37 (1998) L1109.
822.
[83] J. Israelachvili, Intermolecular and Surface Forces, 2nd
[55] J.R. Arthur Jr., J. Appl. Phys. 39 (1968) 4032.
Edition, Academic Press, London, 1991, p. 149.
[56] R.C. Weast (Ed.), CRC Handbook of Chemistry and
[84] S.W. Benson, The Foundations of Chemical Kinetics,
Physics, 55th Edition, CRC Press, Cleveland, 1974.
McGraw-Hill, New York, 1960, p. 72.
[57] F. Family, P. Meakin, Phys. Rev. Lett. 61 (1988) 428.
[85] Beaumont, P. Gibard, J.P. Faurie, J. Crystal Growth 156
[58] G.R. Carlow, R.J. Barel, M. Zinke-Allmang, Phys. Rev. B
56 (1997) 12519. (1995) 140.