The high-temperature epitaxial growth of gallium nitride (GaN) and related device structures directly on [00.1]-oriented silicon carbide (SiC) substrates is significantly hampered by the mismatches in the lattice parameters and the coefficients of thermal expansion between these materials. In the use of metalorganic vapor phase epitaxy (MOVPE) it is common practice to deposit an intermediate buffer/nucleation layer of aluminum nitride (AlN) or aluminum gallium nitride (AlGaN) to facilitate nucleation of the subsequently deposited GaN film. Buffer layers on SiC(00.1) have not been as intensively investigated as those on similarly oriented sapphire. In the latter case, a buffer layer deposited at low temperatures promotes the nucleation and growth of GaN islands at high temperatures. Tanaka et al. have shown that AlN deposited on 3.5° off-axis 6H-SiC(00.1) substrates at ∼1100°C by gas source molecular beam epitaxy grows via the Stranski–Krastanov (S–K) mode. A coherent AlN layer covers the SiC surface to a thickness of ∼4 nm at which point the total strain energy in the film exceeds the surface energy associated with the formation of AlN islands which nucleate and subsequently coalesce. Complete AlN films are commonly grown via MOVPE at ∼1100°C using on-axis SiC(00.1) substrates.
Buffer layers and subsequent GaN layers contain a high density of threading dislocations when grown on the basal planes of both sapphire and SiC. The source of threading dislocations was originally postulated to result from the tilt and twist in low-angle boundaries which form during coalescence of GaN islands in the initial stages of growth. However, results from Mahajan and co-workers have indicated that this hypothesis is not correct for GaN films deposited on sapphire. These investigators assert that in this material assembly, the source of the threading dislocations are either climb segments of dissociated Shockley partials (pure edge dislocations) or coalescence of Frank faults (pure screw dislocations), both of which are located in the highly faulted regions of the low-temperature GaN nucleation layer. Stacking faults are rarely observed in AlN(00.1) and GaN(00.1) films grown on 6H-SiC(00.1) and AlN/SiC(00.1) substrates, respectively, via MOVPE; thus, the source(s) of threading dislocations in these structures have not been definitively established and may well be a function of island coalescence in the initial AlN and GaN layers. Evidence to support this conjecture is presented herein.
A significant percentage of the dislocations in the AlN deposited on the SiC(00.1) continues to thread into the subsequently grown GaN. However, as GaN deposited via MOVPE at 1010°C has been shown by Einfeldt et al. to also grow on coalesced AlN films via the S–K mechanism, new threading dislocations are likely formed during the coalescence of the islands of the GaN. Related research by the same team of investigators has recently determined that Al Ga N films grown on SiC(00.1) initially phase separates into GaN- and AlN-rich islands. The former grows vertically and laterally more rapidly than the latter, and the Al-rich islands are eventually enveloped by the GaN-rich phase. These defects are predominantly edge in character (a-type), are oriented normal to the surface of the film and have a density of approximately 1×10 cm within the GaN. They are now presenting significant barriers to the advancement and commercialization of selected III-Nitride-based devices as they cause (1) carrier scattering and recombination, (2) high leakage currents at device junctions, (3) lower than expected breakdown fields and (4) decreased internal efficiencies and higher threshold current densities in InGaN/GaN laser diodes.
A commonly quoted structural property used to relate to the character of a particular thin film is the full-width at half-maximum (FWHM) value of the X-ray rocking curve. Broadening of a rocking curve is related to a marked increase in the dislocation content. Multiple peaks in this curve are indicative of multiple crystal domains of different orientations in the sampling area. On-axis rocking curves of III-Nitride films grown on SiC(00.1) are typically acquired from the (00.2), (00.4) or (00.6) reflections, and their FWHM are influenced by the density of screw dislocations. Off-axis rocking curves are commonly acquired from several peaks including the (10.3), (10.5) or (20.1) reflections, and their FWHM are influenced by the densities of dislocations having screw, edge and mixed characters. Direct measurement of the direction of pure edge dislocations is not possible in on-axis (00.1) substrates, such as those used in this research, as the incident beam direction would have to be 90°, i.e. from the side of the wafer. However, techniques to estimate the density of pure edge dislocations are described below and employed for that purpose in this research.
It is important to note that domains and domain tilting in the SiC substrates can have a pronounced effect on the measured FWHM of the rocking curves of subsequently deposited nitride films. Typical 6H-SiC(00.1) substrates contain numerous domains with an approximate average diameter of a millimeter that are misaligned to each other . In GaN/AlN/6H-SiC(00.1) assemblies the substrate variability causes sufficiently large changes in the FWHM of these curves to hide improvements in the films due to a reduction in dislocation density. The research presented below is concerned with (1) the characterization routes and the analysis necessary to determine the domain structure and defect substructure in SiC wafers, (2) the procedures and analyses needed to remove the influence of the domain structure on the FWHM values of X-ray rocking curves of GaN, (3) the dislocation densities in GaN layers as a function of thickness and (4) the efficacy of (a) the surface microstructure of AlN buffer layers deposited at different temperatures and (b) of AlGa N buffer layers in reducing the FWHM of the rocking curves in subsequently deposited GaN films.
2. Experimental procedure
The GaN films investigated in this research were grown in situ on as-received, 2 in diameter, research grade, on-axis, 6H-SiC(00.1) wafers on which a 100 nm AlN or a 100 nm Al angle change between the peaks was neglected. It was therefore important to choose film and substrate peaks close to each other in reciprocal space to minimize geometry variations.Ga N buffer layer had been previously deposited. The growths were conducted in a vertical flow, pancake style, metalorganic vapor-phase epitaxy system at approximately 1020°C and 20 Torr. High-resolution X-ray diffraction was conducted using a Philips X’Pert Materials Research Diffractometer (MRD) with a copper X-ray source at a power setting of 40 kV and 45 mA. Approximately 100( /2 ) scans were conducted over the entire surface of each SiC wafer using a four-bounce Ge 220 crystal monochromater on the incident beam side and an open slit on the detector side. This arrangement allowed both the acquisition of the GaN and SiC rocking curve peaks in the same scan and the same spot illumination for both peaks, if the slight spot size variation due to the 2
Synchrotron white-beam X-ray topography (SWBXT) images were acquired at the Stony Brook Synchrotron Topography Facility at the National Synchrotron Light Source located at the Brookhaven National Laboratory. The double-sided polished 6H-SiC(00.1) wafers provided maximum image clarity. Images were taken in back-reflection geometry with the reciprocal wave vector as =00.24, a wavelength of =1.24 Å, and specimen to film distance of 20 cm.
3. Results and discussion
The domains present in typical research grade 6H-SiC wafers are quite variable in size and in the degree of tilt. As such, characterization of numerous substrates was conducted to determine if trends existed among wafers.
Fig. 1. Plot of normalized frequency of occurrence of the FWHM values of X-ray rocking curves determined for SiC(00.6) vs. selected values of these FWHM measured within the total range found in 26 6H-SiC(00.1) wafers.
Fig. 2. Four composite sets of X-ray rocking curves acquired from different areas of a GaN(00.2) film and the associated 6H-SiC(00.1) substrate that show that the (00.2) spectra of the GaN mimic the width and principal features of the underlying (00.6) spectra of the SiC.
The variability of the FWHM is particularly obvious when the values are arranged in a contour plot format such as the 2.5 mm resolution map of a typical wafer in this study as shown in
Fig. 3. Contour map of the FWHM of X-ray rocking curves acquired from a typical 6H-SiC(00.1) wafer illustrating the variability within a single wafer. The scale at right shows the ranges of values of the FWHM used in the map and the gray-scale value associated with each.
Fig. 4. Synchrotron back reflection topography image of the same wafer shown in Fig. 3. Note the correlation between the darker, more defective areas in the topography image and the higher values of the FWHM shown in Fig. 3.
Investigations concerning whether or not the results presented above were due to surface and/or subsurface damage caused by polishing were also conducted. -axis of the boule during growth. Therefore polishing or etching the surface, including hydrogen etching at high temperatures, would not effect the domain tilting observed in the substrates.
Fig. 5. Maps of X-ray rocking curve FWHM for four SiC(00.1) wafers from a single boule. Defective areas of the wafers propagate vertically through the entire boule resulting in similar patterns in the four wafers. The scale at right shows the ranges of values of the FWHM used in the maps and the gray-scale value associated with each.
3.2. GaN X-ray measurement technique
The tilt among the domains measured within the GaN thin films grown on the AlN/6H-SiC(00.1) substrates mimics the tilt among the domains within the underlying wafer, as shown in
Fig. 6. FWHM values of X-ray rocking curves of GaN and SiC plotted with respect to each other. The vertical dashed line marks the transition from where the FWHM is limited by GaN dislocation broadening (left side of line) to where the FWHM measurement limited by dislocation broadening and domain tilt (right side of line): (a) on-axis measurements and (b) off-axis measurements.
The FWHM values for the on-axis X-ray rocking curves of the [00.2] direction in GaN that occur to the left of the knee in the upper curve in
3.3. GaN thickness results
As described in the Introduction, direct X-ray measurement of pure edge dislocation rocking curve broadening is not possible because of the sample geometry. The X-ray inclination angle technique discussed by Srikant et al. and others allows an estimation of the density of the pure edge dislocations in a film by making X-ray measurements of the dislocation broadening starting with planes oriented parallel to the normal surface (0°) and subsequently tilting the sample towards the pure edge geometry at 90° from the normal surface. Using this technique it is possible to quantify the dislocation populations in the film, if the exact character and distributions of the dislocations in the film are known. Defect populations and distributions of GaN films on AlN/SiC substrates grown via MOVPE have not been established by prior TEM examinations, so it is not possible to make quantitative calculations for the current samples. However, it is possible to make a meaningful comparison between samples, if one assumes that similar growth conditions will result in similar dislocation character and distribution.
A good test case for this hypothesis is the comparison of GaN thin films with different thicknesses. Increasing the thickness of these films is known to decrease the density of the threading dislocations through dislocation annihilation mechanisms ; thus, a 0.5 um film should have a greater density of threading dislocations than a 2.5 um film.
Fig. 7. Plots of the FWHM of X-ray rocking curves obtained at six inclination angles in GaN films of various thicknesses. All data were acquired from the portions of the curves similar to those shown in controlled only by dislocation broadening. All films had AlN buffer layers. Line fits are a guide to the eye.
A point of note with regard to the X-ray measurement sampling volume is that each Bragg peak will have a different spot size due to the projection of the beam cross-section onto the tilted sample. Peaks acquired using larger spot sizes will represent a larger number of substrate domains and thus appear to have more scatter and higher FWHM values if a poor area of the film has been interrogated. This effect is particularly noticeable in some higher-order planes such as the (10.1) and the (20.1), which are analyzed using spot sizes that are roughly double of that used for the (00.2) peak. Alternatively, the (00.4) and (00.6) peaks have spot sizes that are respectively 50% and 33% of the size of the (00.2) peak.
3.4. Buffer layer results
SEM analyses of our AlN buffer layers grown to the same thickness on 6H-SiC(00.1) wafers at 1010°C (the growth temperatures for our GaN films), 1130°C and 1220°C reveal an extremely pitted, a less pitted and a very smooth microstructure, respectively, as shown in
Fig. 8. SEM images of the surface microstructure of AlN films after 100 nm of growth on 6H-SiC at 1010°C (left), 1120°C (center), and 1220°C (right).
Fig. 9. Plots of the FWHM of X-ray rocking curves obtained at three inclination angles in GaN films deposited at 1010°C on (a) AlN buffer layers grown at different temperatures and (b) buffer layers of different chemical composition. All data were acquired from the portions of the curves similar to those shown in Fig. 6 controlled only by dislocation broadening. All buffer layers are 100 nm thick. Line fits are a guide to the eye.
An Al Ga N buffer layer is employed to both reduce the lattice misfit that is present at the GaN/buffer interface and to allow doping of this alloy for vertical device structures. The roughness of the surface of the Al Ga N layer grown on SiC(00.1) surfaces decreases with increasing growth temperature and influences the initial nucleation and the growth of the GaN film. Recent research has shown that GaN does not form islands on an Al Ga N buffer layer but continues the terrace structure of this latter layer via step-flow. The steps of the GaN also gradually become straighter and more regularly spaced, which indicates that the number of defects at the steps that pin growth fronts are decreasing. As shown in
The difference in the edge dislocation densities in the GaN films grown on the AlGaN and the AlN buffer layers indicates two possibilities that influence the formation of edge dislocations in GaN. (1) Island coalescence increases the edge dislocation density of the final GaN films. GaN/AlN/SiC and GaN/AlGaN/SiC assemblies have two and one island nucleation and coalescence processes, respectively; for as implied above, GaN does not re-nucleate on AlGaN. (2) Misfit at the GaN/buffer layer interface may contribute to the formation of edge dislocations. Growth of GaN on an AlGa N buffer layer reduces the lattice mismatch between these layers by approximately 80% relative to growth on AlN.
Wafers of 6H-SiC possess a varied domain structure that strongly influences the values of the FWHM of the rocking curves of this material. The magnitude of this influence makes it necessary to account for this variation when investigating via X-ray methods the structure and microstructure of, e.g., AlN and GaN films that may be grown on these wafers. In this research, numerous FWHM data points were simultaneously acquired for GaN(00.1) films and the underlying SiC(00.1) substrates using on- and several off-axis diffraction directions. These data were used to obtain estimates of the values of the FWHM of the GaN rocking curves derived only from GaN dislocation broadening while avoiding the influence of the domain tilting in the SiC substrates. These values, acquired at particular angles of lattice plane inclination, showed that the influence of the edge dislocation component on the FWHM of the rocking curves increased with increasing inclination. This method also showed, as expected, that the density of edge dislocations decreases as a function of GaN thickness due to dislocation annihilation when these films are grown on AlN/SiC substrates. It was assumed that samples grown under similar conditions have similar dislocation population distributions.
The influence of the buffer layer on dislocation density in GaN films was investigated via growth of this material on both AlN and AlGaN buffer layers. A GaN film deposited on an AlN buffer layer grown at 1130°C possessed lower FWHM values in the rocking curves for the edge dislocation component of the film than did similar films deposited on AlN grown to the same thickness at 1010°C or 1220°C. GaN films deposited on AlGa N buffer layers grow via a step-flow mechanism which measurably reduces the density of threading edge dislocations because of the combination of reduced misfit between the GaN and the AlGaN layers and the absence of boundaries of coalesced islands.