Two different forms of GaN, AlN and InN can exist. They are Wurtzite
(hexagonal) structures, or they can exist as cubic forms. The
most common forms of these materials are the Wurtzite structures.
Cubic nitrides are more difficult to synthesize as a practical
matter because lattice matched cubic substrates do not exist.
It can be seen that the bandgap of hexagonal GaN corresponds
to a wavelength of approximately 380nm in the UV. Alloys of AlGaN
move the bandgap to higher energies where emission wavelengths
closer to 300nm appear possible.
|
|
| Figure 1 Bandgaps for Wurtzite and cubic III-nitrides |
There is a significant difference in the manner of change in bandgaps
of the III-nitrides compared to the more conventional III-arsenide
semiconductors. AlN is significantly lattice mismatched to GaN
while AlAs is nearly lattice matched to GaAs. As a result, it
is not possible to grow very thick clad layers of AlN on GaN without
generating lattice defects to accommodate the difference in lattice
constants. It is very difficult to make lateral cavity lasers
with sufficient optical confinement for good efficiency when constrained
to the thicknesses presented in the AlGaN materials system.
The majority of nitrogen will enter the chamber as N2.
Since it cannot be used for growth of GaN, it
is thought that it must be rapidly
removed to avoid degrading the desired growth process. A 2400
l/sec turbomolecular pump efficiently removes unreacted N2
from the chamber and keeps the pressure in the range of low 10-5T
while N2 is flowing in at a rate of 2-3 sccm. The
drawbacks to using a turbomolecular pump include oil backstreaming
from oil-based roughing pumps, and the possibility of catastrophic
failure of internal seals which can rupture and fill the chamber
with turbo pump oil. This mode of failure did occur during November
1995. The problem was repaired and no contamination of GaAs films
was subsequently detected. It is presumed that GaN was similarly
unaffected by the failure, but the possibility remains that the
results of this study have been affected in ways that are not
evident at the present time.
 |
An additional concern when employing source gas in a MBE machine
is the possibility of contamination of the growth environment
through source gas impurities. Oxygen is normally to be avoided
in MBE growth of semiconductors. The starting point for high
purity nitrogen gas used in this facility is employment of boil-off
gas from a liquid nitrogen tank as a nitrogen source. This is
one of the highest purity starting gases available. Following
standard ultra-high purity practices, the gas enters the MBE gas
cabinet where it is further purified through the use of commercial
resin filters. An additional point of use resin filter is used
just before gas enters the RF plasma source.
The Kaufman ion source was never used extensively. It produced only very poor GaN films during our brief efforts. The Mg and Si dopants will be discussed later in this section.
|
|
| Figure 3 Optical signal corresponding to atomic nitrogen for the Oxford Cars-25 is shown as a function of RF power and MBE chamber pressure. |
The curve does not show the limited range in pressure over which
a plasma can be struck, or ignited, at each power level. The
plasma will only turn on within a narrow band of pressure somewhere
within the center of each curve shown. The band is widest for
highest RF power and narrower for low power. If the plasma is
operated outside of this strike range and should go out, it will
not restart without intervention to change pressure back to the
strike range.
Another aspect of plasma operation is source temperature. The Oxford source is designed to be cooled by a large flow of cold N2 gas. The source of gas is LN2 which is warmed through a heating coil to enter the source at a temperature of approximately -20°C. The amount of LN2 consumed dictates that the source must be fed from a large outdoor tank to be operational for periods on the order of a day or more without interruption. In practice, it is very difficult to regulate the flow of nitrogen gas to the source because the vacuum jacketed delivery system is shared by the LN2 phase separator used to cool the MBE machine. The pressure variations in this system are so great that a solenoid in the gas line to the RF source must be controlled by a thermocouple in the source. Although this system works for the majority of conditions encountered, there are times when pressure transients cause the source to become too hot for safe operation, or so cold that the plasma is extinguished. Between 25 and 50% of all growth starts were ruined when the plasma went out. Some improvement could be seen by being careful to keep the plasma pressure within the strike range, but reliable operation remains a topic of development. For this reason, this commercial plasma source would not be recommended for new programs involving the MBE growth of GaN unless extensive measures are taken to solidify plasma stability.
A near catastrophic failure of the Oxford CARS-25 source occured approximately 18 months after installation. The gas feed line broke into two pieces during operation. The effect of this failure was to bring the MBE chamber up to atmospheric pressure instantly. Fortunately, the gas which leaked into the chamber was N2 from the cooling gas which surrounds the gas line inside of the cooling chamber portion of the source. The source furnaces do not appear to have been damaged from the instant atmospheric pressure contamination. The growths since this time have been mostly from the Uni-bulb source.
The cause of the failure would appear to be a result of coupling stray RF from
the feed going to the plasma RF coil into the gas line. The precaution which
is designed to avoid this problem is to keep the gas line isolated from ground
at both ends by ceramics. At the discharge tube side, the gas line is
insulated through its metal to ceramic seal. At the gas entrance side, the gas
line is isolated through a series ceramic connection. It is our best guess
that contamination of the cooling gas components carried enough of some sort of
conducting deposits (metal filings, etc) onto the ceramic spacer to cause it to
short out and increase its coupling of RF energy. A current maximum in the
standing wave was then sufficient to melt the stainless gas feedline.
|
|
| Figure 4 Comparison of substrate thermocouple temperature and substrate temperature measured by a pyrometer. |
The excellent thermal contact offered by indium solder has been
helpful towards minimizing one problem in the control of substrate
temperature in MBE. The substrate is very close in temperature
to the temperature of the moly block. This helps to identify
the actual temperature of the substrate itself. The block is
not in direct contact with the substrate heater station thermocouple.
The thermocouple is radiatively heated by the back of the moly
block. Unfortunately, the thermocouple receives some heat from
the heater filaments which raises the thermocouple temperature
above that of the moly block. A comparison between the thermocouple
temperature and the moly block /substrate temperature measured
by a pyrometer is seen in Figure 4. It can be seen that the thermocouple
temperature is approximately 200°C hotter than the actual
substrate temperature. Although all reports of temperature in
this effort are those measured by the thermocouple, it can be
seen that actual temperatures are lower. The difference between
the actual and the control temperature can vary widely between
different laboratories. These differences make it difficult to
utilize conditions reported by other labs without undergoing the
expensive process of formulating parameter variation curves to
establish similar relationships.
Although indium mounting provides excellent heat transfer from
the moly block to the substrate, it has a disadvantage for the
growth of GaN that is often not seen for GaAs growths. GaN grows
at slow growth rates (0.1µm to 0.2µm per hour) compared
to GaAs (1.0 µm per hour). As a result, typical growths
take 5 to 10 times longer - sometimes lasting for more than one
day. During these long times, the substrate is held at temperatures
of approximately 700°C and an initial substrate nitridization
occurs at approximately 800°C. At these long times and high
temperatures, significantly more indium solder evaporates away
from between the substrate and moly block than occurs for GaAs.
The effects of this evaporation are very harmful to GaN growth.
 |
When indium evaporates, the thermal contact between the moly block
and substrates becomes smaller and the substrate surface temperature
falls while the moly block temperature is held constant by the
control thermocouple. Substrate temperature is an important parameter
in controlling the quality of GaN as shown later in this report.
As a result of falling substrate temperature, the growth conditions
change so much during a long growth that the material quality
varies with depth and is poorest at the surface. It becomes very
difficult to develop techniques to optimize material quality under
these conditions. In addition, approximately one wafer out of
every four grown will fall off of the moly block because too much
indium solder evaporates.
|
|
| Figure 6 Room temperature mobility of GaN doped at approximately 1x1018 cm-3 as a function of thermocouple temperature. |
These difficulties took several months to uncover, however the
solution which was reached has been very valuable to reproducibility
of GaN growth. The new method of mounting substrates is seen
in Figure 5. The
Uni-block
mounting system is now used.
The wafer is held by clips between a backing plate
and front plate. The backing plate has a cutout of the same shape
and size as the substrate. This permits heat from the heaters
to directly radiate onto the substrate.
The new substrate mounting technique presents a new problem.
The blackbody radiation from the heaters largely passes through
the transparent sapphire substrate. In order to raise the actual
substrate temperature to a given value, the heaters must be set
to much higher power than was used for the indium substrate mounting.
This is undesirable for a number of reasons. The higher power
will produce more outgassing of unintentional background gases
from the hotter substrate heater mechanism. These gases can contaminate
the purity of the growing GaN wafer. The higher power will also
shorten the lifetime of the substrate heater filament, and will
establish a smaller upper limit to maximum substrate temperatures
that can be reached. For these reasons, the substrates are metallized
with tungsten, and prepared with roughened backsides to increase
radiative heat absorption. If the substrate is not roughened
prior to metallization, the metal will resemble a mirror and reflect
significant amounts of heat back to the heaters rather then transfer
the heat to the substrate. Using this technique, it is possible
to obtain pyrometer measurements of substrate temperature that
are very close to those of indium bonded substrates.
Most of the wafers for this study were grown before the change to non-indium bonded substrates. In another study, a new calibration of GaN wafer properties as a function of substrate temperature was performed. One of the most important properties of GaN for application to devices is mobility. High mobilities are desired to permit low resistance electrical conduction between ohmic contacts and the laser active region.
Another sapphire substrate issue surfaced during June 1997. It has been found that the highest reproducible mobilities can only be grown on 2 inch diameter substrates. Attempts to obtain mobilities above 200 cm2/Vsec at doping of 1x1018 cm-3 on 1/4 wafers have only succeeded once, while this figure of merit has been routinely obtained on 2 inch substrates using the Uni-bulb source.
|
|
| Figure 7 RHEED reconstruction for a sapphire substrate prior to growth of GaN. |
An important element of GaN growth by MBE is the ability to monitor the growing surface using reflection high energy electron diffraction (RHEED). The RHEED reconstruction seen on a phosphor screen can give indications about the relative quality of the growing surface. It has been useful to identify the type of GaN crystal lattice also. It is clear from RHEED when GaN is a Wurtzite form, or a cubic form. Some cubic GaN was grown for this effort on cubic SiC which clearly showed cubic symmetry. When GaN was grown on GaAs, the RHEED pattern did not show only cubic symmetry, but instead a combination of cubic and Wurtzite phases were observed.
This is an important observation because X-Ray diffraction of
GaN grown on GaAs often is more sensitive to the cubic phase of
the material and has lead to many false reports of cubic GaN on
GaAs.
RHEED observations of GaN growth begin with observation of the
sapphire substrate. The RHEED reconstruction from a typical sapphire
substrate is seen in Figure 7. The horizontal lines indicate
that the surface consists of a regular pattern of atoms. The
RHEED reconstruction can be thought of in terms of the reciprocal
of the surface atom arrangement. A regular flat array of atoms
will be transformed in appearance to an array of lines while a
non-flat array will give rise to spots. A spotty RHEED pattern
indicates that there is surface roughness on more than an atomic
scale. A crystal that grows under this condition for extended
thicknesses will likely exhibit defects in crystalline perfection
which will be observed with X-Ray diffraction, or electrical and
optical properties.
Roughened surfaces with increased layer thicknesses are often
seen for lattice mismatched growth. Rough surfaces have been
seen in most GaN growth performed under this effort. Typical
RHEED reconstruction at different points in the growth of GaN
is seen in Figure 8.
Growth initially remains flat as a thin buffer is grown at low
temperatures. The low temperature layer is necessary to initiate
nucleation. If higher thermocouple temperatures, such as 800°C
are used to grow on the bare substrate, no growth will take place,
even after several hours of exposure to Ga and atomic nitrogen.
Not shown is the RHEED pattern during this initial 700°C
growth. It would show that the surface appears to be amorphous
after approximately 5-10 monolayers (ML)s. When the growth is
interrupted for 2 minutes, the pattern anneals to the form seen
in Figure 8. With this very thin layer of growth, the surface
seems to be planar and regular. It becomes apparent with further
growth that the surface becomes rough and somewhat irregular.
After just a limited amount of additional growth, the RHEED pattern
becomes spotty and remains this way for any thickness which follows.
The spotty growth is presumed to result from rough surfaces caused
by the lattice mismatched growth.
|
|
|
|
|
|
| |
| Figure 8 Changes in RHEED pattern during growth of GaN on sapphire using 600W RF power and 1.1x10-5T chamber pressure (wafer GS0091) | ||
This rough surface growth is almost always seen. The exception is when growth conditions are selected which provide insufficient nitrogen overpresure for stoichiometric growth. When the Ga flux is raised to too high of a growth rate, excellent RHEED patterns are seen. The RHEED is streaky and bright which would seem to indicate very good crystal quality. The films, however show large densities of Ga droplets all over the surface. It is not hard to understand how insufficient nitrogen would lead to Ga droplets - excess Ga atoms that migrate on the growing surface would migrate into balls which would continue to grow. It is difficult to understand why the RHEED pattern would look so good. Presumably it indicates that the small GaN islands between the Ga droplets might be of very good quality. There may be some 3D growth taking place free of lattice mismatch dislocations. If so, some form of control of this condition might lead to a better quality crystal. GaN grown with insufficient nitrogen also usually has much stronger photoluminescence quality than films grown without Ga droplets. This characteristic would be desirable for laser application. The density of the droplets, however, is much too high to permit laser device fabrication.
Since the time when these RHEED observations were first seen, progress in nucleation layer growth has produced growth conditions where RHEED images similiar to those of GaAs are seen. For up-to-date details, see the Reports for the MURI program.
Some AFM images from early growths are seen in Figure 9. In the
top image, the white areas are Ga droplets which are higher than
the surface of the GaN. At higher magnification in the image
at the bottom, islands of GaN are seen at this early stage of
growth. In some areas the boundaries of the islands can be seen.
These boundaries that appear are the interface defects which
exist between the early GaN island nucleation and continue to
exist throughout the entire vertical layer growth of an observed
thickness. The electrical and optical properties of these defects
are discussed later in this report.
|
|
| Figure 9 Atomic force microscopy (AFM) views of GaN grown on sapphire |
|
|
| Figure 10 X-ray diffraction patterns comparing a sapphire substrate and GaN grown on top. The center locations have been adjusted for overlap. |
 |
The mobilities follow an inverse square root relationship with
doping (carrier) concentration typical of other semiconductors.
Mobilities are very high for the highest doped GaN grown at Cornell.
These samples doped beyond 1x1020 cm-3
have the highest conductivity reported in GaN to date. Although
the mobilities are approximately 20-40 times lower than those
for highly doped GaAs, the doping concentrations are significantly
higher than those that can be obtained in GaAs. The product of
doping and mobility is therefore slightly higher for GaN than
for GaAs. This material should produce very low parasitic resistances
in device applications.
 |
| Figure 12 Carrier concentration in n-type GaN as a function of reciprocal temperature of the Si doping furnace |
The mobilities obtained at Cornell at lower doping concentrations
are lower than reported in other labs and are a matter of concern.
The low mobilities are thought to result from our choice of relatively
thin structures for evaluation of doping and mobility studies.
The data from this effort shown above are from layers that are
1 micron thick. It is known that mobility in GaN gets higher
as thicker layers are grown. Presumably this is the result of
defects associated with lattice mismatched growth which are greatest
in density near the substrate. A single attempt to grow a thicker
layer of 3 microns did not have significantly higher mobility
than the 1 micron thick reference. It is not known whether this
result is a single anomaly, or is representative of all thicker
layers. It is the only thick wafer growth attempted to date.
Another possible explanation for unexpectedly low mobility is
scattering from impurities or defects. It is unlikely that ionized
impurities exist in significant concentration for these wafers.
Shown in Figure 12 is the Si doping calibration data for GaN
obtained for 600W plasma power and 0.1 µm/hour growth rate.
The curve contains two interesting features. First the curve
is linear over a wide range, including at lower doping concentrations.
This would not be the case if there were a significant concentration
of undoped impurities. In addition, undoped material has higher
resistivity and does not exhibit significant concentrations of
ionized impurities. Thus, large compensation densities as a cause
of low mobility is unlikely. Second the very high doping concentrations
that are shown are surprising in this wide bandgap semiconductor.
These levels are usually only seen in small bandgap semiconductors,
and not even materials such as GaAs exhibit such good doping behavior.
In contrast to the n-type doping, p doping has proved to be more
troublesome. Although a p-type doping curve for Mg in GaAs has
been established during this study, it cannot be sure that the
few p-type data points for GaN are unambiguous. P-doping of GaN
remains a difficult problem which must be solved for fabrication
of lasers and HBTs. Present lateral GaN lasers suffer significant
IR drops due to poor p-type GaN doping concentrations.
PL from GaN grown at 600W RF plasma power, 2x10-5T
chamber nitrogen pressure, 840°C substrate temperature (using
indium bonded samples) and Si doped to the levels indicated, is
shown in Figure 13. The incident laser intensity is very low
compared to PL commonly performed on semiconductors. The 20mW
laser power is spread over a beam width of approximately 5mm diameter
at the surface of the sample. This corresponds to a power density
of approximately .01W/cm2.
The low incident power density provides an opportunity to closely examine PL from deep centers within GaN. Since these centers would be expected to saturate and leave only observation of the bound exciton at the band edge at high power densities, low excitation power density permits easy observation of the relationship between deep level filling and band edge luminescence.
|
|
| Figure 13 Room temperature photoluminescence of GaN as a function of n-type doping concentration. Above is linear intensity scale and below is log intensity scale. |
It can be seen that at low doping concentrations, luminescence
at yellow wavelengths dominates the PL spectra.
As doping concentration is raised, the yellow
luminescence falls and band edge luminescence rises. Increased
band edge PL with increased doping is not unexpected as the electrons
provided by the donors increase the probability of recombination
with free holes generated by the incident photons. The decrease
in deep level luminescence is not expected for increased doping.
Decreased recombination from the deep level at increased doping
concentration can be explained by a model.
The deep level is modeled to be 2.2eV above the
valence band of GaN. The deep level luminescence is from the
transition from the deep level to the valence band. The physical
origin of the deep level is presumed to be from defects arising
from lattice mismatch. These defects are often seen to rise perpendicularly
from the substrate interface to the surface. A cross section
might appear as simple vertical lines. A schematic showing the
band diagram as a function of position is seen in Figure 14.
|
|
| Figure 14 Model of band diagram surrounding vertically oriented defect in GaN which shows Fermi-level pinning at the defect level and band bending away from the defect. |
The defect level pins the Fermi-level near mid-gap. The area
surrounding the defect is depleted of carriers. If the combination
of low doping density and high defect density exists, the material
will exhibit very high resistivity. The undoped samples that
we have grown at 600W do exhibit high resistivity.
|
|
| Figure 15 Model of defect level in the presence of two different doping concentrations illustrating different depletion regions surrounding the defect. |
Figure 15 shows the effect of doping on the depletion regions
surrounding the defect. Higher doping will result in a much greater
fraction of material which is ìflat-bandî and much
less material will be depleted surrounding the defect. The probability
of band edge luminescence will be highest for the highest doped
material, and the probability of luminescence from the defect
will be lowest. In addition, the highest doped material will
move the Fermi-level above the defect level where it will be completely
occupied.
Both AlGaN and InN were grown. InN was grown as an initial step
towards growth of GaInN quantum wells. The InN growths were unsuccessful.
Presumably the substrate temperature used was too high for this
material. The nucleation of InN on sapphire did not occur similar
to lack of nucleation of GaN when high temperatures were used.
The lowest temperature attempted was 500°C. Lower temperatures
may prove successful.
AlGaN was grown with ease. Al mole fractions were predicted based
on calibrations from AlGaAs growths in the same machine. X-ray
was used to determine Al content of AlGaN layers. The properties
of AlGaN resemble those of GaN with simple shifts in energies
corresponding to changes in bandgap. For example, PL (shown in
Figure 16) of AlGaN is dominated by deep level emission which
is shifted towards green wavelengths from yellow wavelengths seen
for GaN. Si doping of AlGaN also resembles GaN characteristics.
Single data points have been collected for Si doped AlGaN at
concentrations of 1x1019 cm-3 and 5x1019
with comparable mobilities to those for GaN.
The photoluminescence of AlGaN is affected similarly to GaN by
doping. High n-type doping results in reduced defect emission
and increased band edge emission. This observations support the
deep level defect model
|
|
| Figure 16 Room temperature PL from undoped AlGaN for two different compositions. No band edge luminescence is seen (the high energy signal is from the UV laser). |
Some effort was directed towards the growth of cubic GaN on cubic
SiC. Since cubic GaN is a closer lattice match to cubic SiC,
lower defect densities might be expected. There would be possible
advantages for laser applications using the cubic forms of GaN
and AlGaN. The wavelength ranges available for this material
system shown in Figure 1 provide a slightly different band of
emission possibilities. There is also some indication that electron
mobilities might be higher in cubic GaN which would lead to even
lower parasitic resistances.
X-ray diffraction of cubic GaN grown on SiC is seen in Figure 17.
The cubic peaks for cubic (002) and (004) GaN are clearly visible
near the cubic SiC peaks. This result is very encouraging as
it shows GaN full width half-maximum values similar to those of
the SiC substrate (which is itself grown on Si).
|
|
| Figure 17 X-ray diffraction of 0.2µm thick cubic GaN grown on SiC at Cornell at 600W RF power |
Much progress has been made in learning about the effects of growth
conditions on electrical and optical properties of GaN. This
first effort has been successful in identifying the differences
in approach required between growth of GaN and previous programs
based on GaAs. Some fundamental techniques including indium-free
mounting of metallized sapphire substrates, low temperature nucleation
buffers, and even improving the reliability of the plasma RF source
have been developed.
Electrical characterization of GaN shows the demonstration of
exceptional n-type doping characteristics. N-type carrier concentrations
of 6x1020 cm-3 have been easily obtained
for application to low resistance laser contacts. P-doping has
been largely unsuccessful and will require future development.
Undoped material with low conductivity has also been demonstrated.
The dominant materials difficulty which has been encountered has
been the existence of significant defect densities arising from
lattice mismatched growth of GaN on sapphire substrates. A model
to understand the electrical and optical nature of the deep level
arising from these defects has been developed. The only obvious
solution to this problem is to employ GaN substrates which still
have not been developed.
AlGaN appears to have similar properties to GaN and these materials
should be able to be combined to form device structures providing
careful attention is paid to critical thickness constraints.
Some benefit might come from using cubic GaN and early growths
of this material during this effort has been successful.
Access
times since 12 Feb 1997.
September 26, 1997
