ScAlMgO4: an Oxide Substrate for GaN Epitaxy
The availability of an appropriate substrate for epitaxial growth is crucial to
the practical application of new semiconductor materials. The factors
determining appropriateness include crystallographic, physical, chemical, and
economic parameters. For the wide band-gap semiconductor GaN, the most
appropriate substrate for practical applications such as light-emitting-diodes
(LED's) has proven to be c-plane sapphire.
[1]
This fact might be surprising in view of some inappropriate crystallographic
parameters, but not so surprising considering its physical, chemical and
economic parameters. Sapphire is strong, hard, and inert. It is available in
large, inexpensive, high quality wafers from a number of vendors. However, for
potential use of GaN in blue lasers, these virtues may prove to be insufficient
to counteract the one drawback of sapphire: its huge lattice mismatch with GaN.
The -13% misfit [a] results in a very large
dislocation density in GaN epitaxial films on sapphire
[2]
.
It is at present not understood why LED's work as well as they do in the
presence of such a large dislocation density. For epitaxial growth, 13% is
almost the largest misfit that can be tolerated while still getting a
well-aligned film. The surprisingly high quality of GaN films that have been
obtained on sapphire may be due to an interfacial AlN reaction layer.
[3, [3]
A variety of alternatives to sapphire have been investigated over the years,
[4]
including Si, GaAs, NaCl, GaP, InP, W, TiO2, SiC, ZnO,
MgAl2O4 (spinel), and MgO. Only the last four can be
considered possible improvements over sapphire.
The mismatch is only -3.1% to GaN and -1.0% to AlN. SiC has the additional
advantage, for many applications, of being conductive. Although excellent
results have been obtained with epitaxial GaN films on 6H-SiC,
[5]
the 6-H SiC substrates are very expensive.
ZnO has the wurtzite structure, and is only mismatched by +2.3% to GaN.
Positive misfits can be compensated by the addition of In to make the (GaIn)N
alloy.
[6]
However, ZnO crystals are not very easy to make, and ZnO is not as thermally
stable as would be desired. Zn is considered a troublesome contaminant for many
processing techniques, including MBE.
The (111) face of MgO is mismatched by -6.4% to GaN. The main difficulty with
MgO is that the (100) face has a much lower surface energy than the (111)
face, both in terms of cleavage and crystal growth, which makes the (111) face
extremely difficult to prepare. For growth of zincblende GaN, MgO remains an
attractive possibility,
[7]
although substrates larger than 1'' may not be practical.
Spinel is one of the most interesting of the substrates that have been used
previously.
[8]
The (111) face is the low surface energy face, so it is relatively easy to
prepare. The oxygen lattice is face-centered cubic, which presents a similar
stacking layer along (111) as found in the wurtzite structure. Best of all, the
spinel structure and spinel related structures are found in a large number of
oxides. Our group has studied two different spinel type materials,
Fe2NiO4, a -7.1% mismatch and
LiGa5O8, a -8% mismatch.
[9]
Despite the diversity of the spinels, there are very few that have really good
lattice matches to GaN. The oxides are mostly too small and the sulfides are
all too large. The oxides that come closest are Na2MoO4,
Na2WO4, and In2CdO4, with
mismatches +1.3%, +1.5%, and +1.3% respectively.
As the cations in spinels get larger, the spinel structure becomes unstable
with respect to other structures. One such structure-type is the
YbFe2O4 structure.
[10]
This structure-type is a rhombohedral layered structure, with hexagonal
a-lattice constants between 3.236 (ScMgAlO4)
[11]
and 3.489Å (YFeZnO4).
[11]
The structure can be considered to be a superlattice of rock-salt-like layers
and wurtzite-like layers. The oxygen lattice is near to close-packed. The
smallest of the known YbFe2O4-type materials,
ScAlMgO4 is well matched (+1.8%) to the hexagonal face of
wurtzite-structure GaN. The structure of ScAlMgO4 and its relation
to that of GaN is depicted in figure 1.
In the rest of this paper, we will describe the preparation of
ScAlMgO4 crystals, and the epitaxial growth and characterization of
GaN films on these crystals.
Two types of ScAlMgO4 substrate crystals were prepared and used for
epitaxial growth.
Platelets as large as 1 cm across were grown by slow cooling of a
stoichiometric melt. A 50g batch prepared from stoichiometric amounts of MgO,
Sc2O3 and Al2O3 was placed in an Ir
crucible. The sample was melted using RF induction heating under an
N2 atmosphere. The sample was then cooled slowly to yield crystals
approaching 1 cm diameter and 1 mm thick, although many of these had low angle
grain boundaries. Flat flakes were peeled from the micaceous crystal mass and
soldered with indium to molybdenum heater blocks for epitaxial growth.
A compelling reason to consider ScAlMgO4 as a potentially practical
substrate for GaN epitaxial films is that it can be grown by the Czochralski
method. Thus, the techniques needed to produce large diameter, high quality
crystals on an industrial scale are already well established. A melt of
stoichiometric composition was prepared by mixing 44.01g MgO and 75.31g
Sc2O3, forming the mixture into a pellet, placing the
pellet into a conventional iridium crucible together with 55.68g
Al2O3, and heating the charged crucible under
N2 in a conventional RF induction furnace. The starting materials
were commercially available, and of at least 99.99% purity.
The charge was completely molten at about 1900°C. At this point, a thin
iridium rod was dipped into the liquid and a button of polycrystalline
ScAlMgO4 was formed on the tip of the rod. The rod with solidified
ScAlMgO4 thereon was slowly raised (initially at 2.5 mm/hr) and
rotated at 15 rpm. After about 1.5 hours the pull rate was gradually increased
to 4 mm/hr. Pull conditions were regulated under computer control by
maintaining weight gain to yield a boule having a neck (about 7mm in diameter,
about 60mm long) that blended smoothly into the main body of the boule, about
60mm long. This was done to favor the growth of a limited number of relatively
large crystals in the boule. Away from the seed end of the boule, the
crystallites were large enough to separate with a razor blade into 20mm
diameter near-single crystal slices. The slices were then mounted on a
polishing block with black wax and polished with emery paper until they were
flat on a cleavage plane. The slices were flipped and the process repeated on
the reverse side, followed by polishing of the intended growth surface with
LINDE A® and LINDE B® polishing compound. The
surface chips rather easily, so considerable care is required during this
stage, so as not to gouge the surface. After a final polish using
Syton® on polishing paper, the slices were soldered to heater
blocks with indium.
The substrates are somewhat flexible, but will break into flakes rather than
cleave, as expected from a micaceous material. They are clear and insulating.
Low angle grain boundaries were visible in some substrates
A critical parameter for the compatibility of a substrate and an epitaxial film
is the thermal expansion match. If the thermal expansion coefficient of the
substrate is too small, the film is liable to crack, while if it is too large,
the film can peel or blister on cooling. GaN is a rather hard material and has
a relatively small thermal expansion coefficient, 5.6x10-
6/°C
(in the a direction), which is between that of sapphire (7.5x10-
6/°C)
and that of silicon (3.6x10-
6/°C).
Most oxides tend to have larger thermal expansion coefficients. Micaceous
materials such as ScAlMgO4 tend to have smaller thermal expansion
coefficients parallel to the cleavage planes and larger thermal expansion
coefficients normal to the cleavage planes.
The temperature dependence of the unit cell parameters of ScAlMgO4
was measured using a DISPLEX cryostat with beryllium windows on a
diffractometer equipped with graphite monochromated Cu K
radiation.
Powder patterns were taken at 50K, with subsequent increments of 50K up to
300K. Lattice parameters were refined using least squares. The data is shown in
Figure 2. The refined cell parameters are available in Data File 1. Thermal
expansion or contraction of the cryostat resulted in non-uniform shifts that
could not be fully accounted for due to the limited 2-
range of the
data. These shifts particularly affected the refined a-lattice
parameters. The cell parameters at 300K, a=3.2405 and c=25.106
are quite close to the published cell parameters, a=3.236 and
c=25.15; the small differences may be due to slight differences in the
stoichiometry of the melt-cooled ScAlMgO4. Using the refined unit
cell parameters for 200 and 250K, we obtain thermal expansion coefficients of
6.2x10-
6/°C
for the a direction and 12.2x10-
6/°C
for the c direction, roughly what we expected.
Nitride growth was done in a Riber molecular beam epitaxy (MBE) system with a
custom nitrogen plasma source. The plasma source is described in Reference
[12]
The substrates were heated to 700°C in vacuum, then exposed to the
nitrogen plasma at a temperature between 600 and 650°C briefly before
starting the GaN deposition. The nitrogen plasma is excited with 10-20 watts of
RF power in nitrogen at 65-85 mTorr. The RHEED (reflection high energy electron
diffraction) pattern was observed through the whole process. Figure 3a shows
the RHEED pattern of a ScAlMgO4 substrate before growth. Although
somewhat blurry, it shows the streaks indicative of a surface with atomically
smooth areas. Exposure to nitrogen resulted in a marked improvement in the
RHEED pattern of the substrate, as seen in figure 3b. The streaks are still
blurry, but they are significantly brighter. The patterns in 3a and 3b are too
similar to determine whether the surface has been nitridized, or merely cleaned
of contaminants. The substrate was then exposed to a Ga beam by opening the Ga
furnace shutter, starting the growth. The Ga furnace temperature was set to
obtain GaN growth rates between 350 and 5000 Å/hr. The RHEED pattern
remains bright at the start of GaN growth, in contrast to growth on
Al2O3, for which the RHEED pattern is sharply reduced in
intensity at the start of growth. Figure 3c shows the pattern after growth of
GaN for 50 seconds at 350Å/hr. Almost no qualitative change in the RHEED
pattern is observed, instead, the pattern grows brighter and more distinct.
This is characteristic of coherent epitaxy. For higher growth rates, the
streaky pattern is sustained for only a short time, while for low growth rates,
the streaks become brighter and sharper to the end of the growth, as seen in
Figure 3d. Figure 3e shows the RHEED pattern we observe in films of high
quality after cooling to 250°C in vacuum. This 3x3 reconstruction, which
we have confirmed by low energy electron diffraction (LEED) and also observe in
the best of our films on sapphire, may be a result of nitrogen loss at the
surface. We have not been able to find reports of the 3x3 reconstruction on GaN
in the literature; the (3x3) reconstructions known for InSb (
)
[13]
and 6H-SiC (0 0 0 1)
[14]
surfaces could be related.
After the films are removed from the growth chamber they are un-soldered from
the heater block. The films are smooth as seen under a Nomarski microscope,
although under some growth conditions gallium droplets can form. No peeling is
observed, although our first film had some surface chipping remaining from the
surface preparation. The films are further characterized by x-ray diffraction,
optical transmittance, and by photoluminescence.
The film crystallinity was examined using a 4-circle x-ray diffractometer using
monochromated Cu K radiation. In addition to the usual -2
and 
scans, scans on in-plane diffraction peaks of the film and
substrate were used to measure the azimuthal order. This measurement is a
particularly relevant to large misfit epitaxial systems, as discussed in
Reference
[15]
.
The -2 scan for a 0.22 µm thick film on ScAlMgO4 is
shown in figure 4. Prominent are the substrate (0 0 3l) peaks and the GaN
(0 0 2l) peaks. The diffractometer resolution was insufficient to
resolve the -2 peak widths. (rocking) scans on the
substrate (0 0 18) peak revealed two crystallites in the measured region with a
0.3° difference in orientation, each with peak widths of
0.5°. [b] The (0 0 4) rocking
curve for the film showed 0.8° wide peaks. These numbers are to be
compared with the 0.37° rocking curve width measured for GaN grown
concurrently on sapphire, which had an peak width limited
instrumentally to 0.22°. The 
scans of substrate and GaN azimuthal
peaks indicated only the 3 peaks spaced at 120° intervals for the
rhombohedral sapphire and ScAlMgO4, and the 6 expected peaks in the
hexagonal GaN. Figure 5 shows the peaks in detail. The ScAlMgO4
substrate again reveals two crystallites, misoriented in by 0.4degree
the sapphire peak again exhibits the instrumental resolution. The azimuthal
broadening of the GaN peaks is clearly apparent, and is similar for growth on
sapphire and growth on ScAlMgO4. The Matthews theory of island
rotations for large misfit systems
[16]
predicts that the width on sapphire should be a factor of 3 larger than
that on ScAlMgO4, which we do not see in the data. This suggests
that island nucleation is not the principle source of the azimuthal broadening
in this film.
One of the easiest ways to measure the GaN film thickness for the thicker films
is to measure the optical transmittance. The Fabry-Perot fringes can be counted
to determine the film thickness, and the position of the absorption edge can be
verified. The amplitude of the fringes can be used to give a rough estimate of
the substrate index of refraction. It is roughly 0.3 lower than GaN over the
visible range, or about 2.
Figure 6 shows the comparison of the photoluminescence (PL) spectra of
0.16µm thick GaN films grown on ScAlMgO4 and on sapphire in the
same growth run. A low growth rate and a low plasma power were used for this
run. A He-Cd laser was used as an excitation source; the emitted light was
dispersed by a 0.5 m monochromator and detected by a CCD camera. The emission
spectra of both samples look very similar for both temperatures and the
intensity is roughly the same. The spectra measured at 5 K show a rather small
donor bound exciton peak with a slightly red shifted
[17]
maximum at 3.43 eV (Fig. 6a). These spectra are dominated by the donor-acceptor
pair transition at 3.26 eV and its phonon replicas. The broad band in the
yellow spectral region with maximum at 2.25 eV is commonly believed to result
from defects.
[18]
The intensity of this band is low in these samples, a sign of good crystal
quality. At room temperature we observe a band edge peak at 3.37 eV although
the main emission features are broad bands with their maximums around 2.0 and
1.7 eV.
The PL results show that the band edge emission from thin GaN films grown on
ScAlMgO4 and on sapphire is roughly comparable. Films of superior
quality have been grown on sapphire by MOCVD; these films are typically an
order of magnitude thicker than our MBE films. We find that, between sapphire
and ScAlMgO4, the luminescence intensity is much more dependent on
growth conditions than on the substrate type. Considering the crystal quality
of the substrate, which has to be improved, the luminescent properties of GaN
films grown on ScAlMgO4 look very promising.
What should be clear from our results is that ScAlMgO4 is a very
promising substrate material for epitaxial growth. What should also be clear is
that ScAlMgO4 substrates are not the magic solution to all problems
facing the realization of practical high performance GaN based devices.
Although we can expect further improvement in film properties when truly single
crystal ScAlMgO4 substrates are available, we do not observe the
dramatic improvements that we might naively expect from an order of magnitude
reduction of misfit. This could indicate either that misfit dislocations are
not the primary limitation to, for example, luminescence properties in our
films, or that new types of defects are introduced by the substrate.
Since the c-axis lattice constant of ScAlMgO4 is 4 1/2 times
that of GaN, we expect stacking faults to occur at surface steps on
ScAlMgO4 (which should be c/3 (0.75nm) high) A similar
situation holds for the sapphire (0 0 0 1) surface. If this sort
of stacking fault is important in GaN, then film quality should be sensitive to
the surface mis-orientation. Stacking faults of this type should not occur for
growth on isostructural substrates, such as ZnO. On the other hand, the
insensitivity of the photoluminescence results to the choice of substrate
suggests that misfit per se is not limiting the film quality of films on
sapphire.
We note that since a small fraction of InN will increase the film lattice
constant, substrates with positive misfits to GaN can be perfectly lattice
matched to alloys in the (In,Ga,Al)N system. Thus the +1.8% misfit for
ScAlMgO4 should be perfectly lattice matched to
In0.16Ga0.84N and In0.30Al.0.70N,
while the -3% misfit to 6H-SiC cannot be eliminated in this manner.
The initial results reported here indicate that ScAlMgO4 is a
promising substrate material for GaN epitaxial growth. Although growth
conditions for GaN on ScAlMgO4 may have to be re-optimized to
benefit from the order-of-magnitude reduction in the misfit, we anticipate that
when large, high quality single crystals are available, growth on
ScAlMgO4 will result in epitaxial films surpassing the quality of
those currently produced on sapphire. This may be a crucial step towards the
realization of high-performance GaN-based opto-electronic devices.
The authors would like to thank A. J. Valentino for assistance in the substrate growth.
T a c
12 3.2372 25.0349
100 3.2382 25.0381
150 3.2394 25.0486
200 3.2373 25.0636
250 3.2383 25.0789
298 3.2405 25.1061
last updated Wednesday, September 9, 1998 2:48:50 AM.© 1996-1998 The Materials Research Society
