MOVPE growth of GaN on patterned 6-inch Si wafer

Three different 6-inch Si (111) substrates for GaN growth were investigated: a bulk silicon substrate (non-patterned) and patterned silicon substrate with 10 μm (sample A) or 40 μm (samples B) deep trenches between mesas. Mesas were square shape with sizes of L: 500, 1500, 3000, 5000 μm, and trenches between mesas, W: 20, 50, 150 μm as shown on a figure 1. All substrates were p-type silicon (111) with total thickness 1000 ± 25 μm.

Zoom In Zoom Out Reset image size

In order to determine effect of corner stresses, zones 2, 4 and 8 (figure 1) had round shape corners, zones 1, 3 and 5–7—sharp corners, the difference shown on figure 2. The substrates were manufactured by Okmetic Oyj.

Zoom In Zoom Out Reset image size

GaN (0001) layers were grown on Si wafers using the standard step graded Al_xGa_1−xN and AlN approach by the metalorganic vapour-phase epitaxy (MOVPE). The reactor had a 1 × 6-inch close coupled showerhead configuration. Thimethylaluminium (TMAl), thimethylgallium (TMGa) and ammonia (NH₃) were used as precursors for aluminium, gallium and nitrogen, respectively. Hydrogen was used as the carrier gas, more details can be found in [5].

Figure 3 presents cross-sectional SEM image of the structures grown on non-patterned Si substrates. The growth process was started by in-situ annealing to remove the surface native oxide. After annealing, a 300 nm AlN buffer was deposited. The buffer consists of low temperature and high temperature AlN grown at 980 °C and 1085 °C nominal substrate surface temperature, respectively. Next, a step graded Al_xGa_1−xN buffer was grown at 1060 °C with three different compositions. These were 260 nm of Al₈₀Ga₂₀N, 280 nm of Al₅₀Ga₅₀N and 470 nm of Al₂₀Ga₈₀N. Finally, 900 nm of GaN was grown at 1040 °C. The sample parameters are listed in table 1.

Zoom In Zoom Out Reset image size

The reactor pressure during growth was 100 mbar, except for the last GaN which was grown under 400 mbar pressure. The higher growth pressure increases the crystalline quality of GaN while material grown at 100 mbar pressure has a higher carbon concentration and forms a semi-insulating layer [14]. This type of semi-insulating layer is typically used for device insulation, for example, in GaN high electron mobility transistors (HEMTs). Device layers, such as HEMT or light emitting diode (LED), could then be grown on these epitaxial templates.

An optical in-situ measurement system was used to assess the growth rate and wafer surface temperature profile. A more detailed measuring mechanism description is explained in detail in [5]. Besides an optical in-situ measurements, the thickness measurements were verified with cross-sectional scanning electron microscope (SEM). Moreover, a behaviour of crack distribution was studied by SEM. The surface roughness of the GaN layers was determined by atomic force microscopy (AFM).

Confocal Raman spectroscopy is a nondestructive and sensitive optical method for quantitative characterization of stress distribution in GaN layers via observing Raman peak shifts [12, 15–17]. Modern Raman tools operate close to the optical diffraction limit and are capable of detecting the local stresses with lateral spatial resolution of only 0.5 μm [18]. However, due to the tiny magnitude of Raman peak shifts, it is noteworthy that the Raman 1-D, 2-D and 3-D measurements are easily influenced by various potential error sources that can distort the scan results if not accounted for [19–21]. For accurate measurements, a laser induced heating and stress should be considered due to their effect on the Raman peak shift [22]. Another major error source can be a lack of proper spectrometer calibration, often affected by the environment in real time, because it is used to linearize the spectral axis and thus correct the Raman peak shifts [17]. In order to estimate absolute stress values, both of these effects were accounted for in this work. The essential point to remember is that the collected spectra are to be interpreted with care due to the fact that a multi-layered high refractive index film is being measured as depicted in figure 3. In other words, although the laser beam was carefully focused on the air-GaN interface, its collection volume extends through all the grown layers and results in a weighted sum of each layer spectrum.

To evaluate the local residual stress differences in the grown GaN layers, a confocal Raman microscope, WITec alpha300 RA+ was used in the backscattering configuration to estimate the absolute biaxial stress, σ_xx from the Raman peak position, ω of GaN's main optical phonon mode, ${{\rm{E}}}_{2}^{H}$. When the biaxial stress component is dominating, their linear relationship can be written as ${\omega }_{0}-\omega =K\cdot {\sigma }_{{xx}}$, where, in case of the Raman peak ${{\rm{E}}}_{2}^{H},{\omega }_{0}\approx$ 568 rel. cm⁻¹ is a still-debated stress-free value [12, 15] and K = 4.2 cm⁻¹ GPa⁻¹ is an experimentally measured phonon deformation potential [12, 23]. The Raman scattering was excited by a 488nm wavelength laser with 10 mW power, focused through an intermediate magnification/numerical aperture×20/0.4 dry objective lens. A MATLAB toolbox wit_io was used to boost the data analysis [24].

As a practical example of biaxial stress analysis, the GaN layer grown on non-patterned Si substrate was line scanned ∼1 cm from its center and edge, as demonstrated in figure 4. The center and the edge regions of the wafer were found to be spatially uniform with tensile biaxial stresses of σ_xx = 468 ± 33 MPa and σ_xx = 528 ± 41 MPa, respectively. Here plus-minus sign represents standard deviation from the mean value. Averaged tensile biaxial stress is then found to be σ_xx = 498 ± 47 MPa (total of 102 data points from total of 2 line scans). Observed anomalous deviations from the mean in figure 4 (b)–(c) and (f)–(g) are attributed to some local changes within the multilayer film structure, for instance, to a local defect, but detailed analysis is beyond the scope of this work. The top layer surface uniformity was verified by AFM measurement with a root-mean-square (RMS) roughness of 1.46 nm for 10 × 10 μm image scan.

Zoom In Zoom Out Reset image size

The obtained biaxial stresses are not high and are comparable to the previous research [12]. The small stress difference could be explained by the different temperature in the center and edge of the substrate during GaN growth.

In order to account for the instrument spectral axis nonlinearity, a highly spectrally stable bulk ammonothermally grown GaN reference sample spectrum was also taken with each measurement. The reference sample spectrum was once meticulously linearized using dozens of neon atomic lines, explained in detail in this [17]. Each collected spectra was then Raman shifted together with the corresponding reference spectrum in a way that the peak of interest, the main Raman peak of GaN, ${{\rm{E}}}_{2}^{H}$ recovers its absolute position. It is noted here that possible but small laser-induced heating effect to the peak positions was assumed to be sufficiently equivalent for each sample and thus is handled by the previously described methodology.

The same MOVPE process recipe used for the GaN growth on non-patterned Si was then utilized for a patterned Si wafer, denoted as Sample A. Figure 5 presents emissivity-corrected in-situ substrate surface temperature and 633 nm wavelength reflectance during a growth run on (a) non-patterned Si and (b) patterned Si substrate (Sample A). It can be seen that the patterned substrate does not significantly alter the growth. The growth rates are 0.5 μm h⁻¹, 1.1 μm h⁻¹, and 2.3 μm h⁻¹ for AlN, AlGaN and GaN layers, respectively for the growth on both substrates. The similar growth rate of high pressure GaN in both cases (1.2 μm h⁻¹) shows that epitaxy is not affected by substrate patterning. Nevertheless, the amplitude is slightly smaller on a patterned substrate due to the fact that measured point covers both the mesa and trench. The effect of mild layer degradation during AlGaN buffer deposition (5000–9000 s) is due to surface roughening, this can be optimized in future.

Zoom In Zoom Out Reset image size

Figure 6 demonstrates microscope top view images of grown GaN layers on 20/500 μm mesas. It can be seen that GaN grown on both mesas with sharp corners (a) and round corners (b) is crack-free. However, the GaN grown in the areas outside mesa patterns are cracked from trenches with sharp corners, as is shown by arrows in figure 6(c). For round corners in (d), they remained crack-free. In fact, sharp corner trenches generate GaN cracks between mesas while round corner trenches prevent GaN from cracks. The high-resolution stitching microscope image of the whole crack-free pattern is demonstrated in figure 7.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 8 shows a cross-section edge of the 20/1500 μm square patterned zone of Sample A. The thickness difference of the grown layer in mesa and trench areas is mostly due to a higher growth rate and earlier coalescence of GaN crystallites on top of the mesas. The GaN growth direction in flat mesas and trenches bears a similarity to a standard c-plane (0001) growth on (111) Si. The growth direction in side walls is supposedly (11-20) a-plane, but proof of this requires additional x-ray diffraction measurements.

Zoom In Zoom Out Reset image size

Figure 9 displays a GaN stress distribution comparison between the sharp (zone 3, figure 1) and round (zone 8, figure 1) cornered mesas. The perceptually uniform color map of matplotlib [25] was chosen for representing the stress maps to aid in the visual inspection in the least biased way. This allows us to see in figure 9 that GaN in the round corner mesa (b) is slightly more tensile stressed than in the sharp corner mesa (a). The cross-sections (figure 9(e)) indicate to an excellent stress distribution uniformity in vertical (V) and horizontal directions (H). In contrast to previous reports on the circle shaped mesas, where compressive strain is maximum at the center and relaxes towards all the edges [12, 16], here stresses distribute uniformly for the whole square mesa surface. However, the slight asymmetry in figure 9(e) could be caused by an unintentional artifact of accumulated instrumental errors [17], since biaxial stress distributions in figures 9(a) and (b) tend their highest values towards the bottom-right corners. From the comparison of figures 9(c) and (d) it is easy to note that the round corners seem to alleviate the stress more quickly than the sharp corners. The same effect is presented in the cross-section measurements (f). Here semi-transparent thick trend lines are the 1st order polynomial fit lines and were added to demonstrate that the round corner mesa reaches the stress equilibrium slightly faster than that of the sharp corner mesa, also seen in (e).

Zoom In Zoom Out Reset image size

The line-averaged tensile biaxial stresses calculated in table 2 supports Raman map results. Based on table 2, patterned Si substrate has some effect on the GaN stress values. GaN on non-patterned Si substrate has slightly lower tensile biaxial stresses. This is probably due to GaN thickness difference between samples. The GaN layer in round corner mesas is slightly more stressed than in sharp ones. This is probably due to stress difference between the edge and center of wafer similarly to non-patterned Si.

The effect of substrate patterning on thicker GaN films was investigated with sample B, where time for Al₈₀Ga₂₀N, Al₅₀Ga₅₀N and GaN^a scaled by factors of 2, while other layers kept similar to sample A. For sample B a patterned substrate with trench depth of 40 μm was used. In-situ reflectance data of sample B is shown in figure 10. As can be seen, reflectivity trends same behavior with sample A. A small decrease in reflectance amplitude, most probably due to a different trenches depth in measurement area, otherwise reflectivity is similar to the sample A/non-patterned Si. Nevertheless, the slight drop in the reflectivity average during Al_xGa_1−xN buffer growth indicates to increasing surface roughness, that require further optimization.

Zoom In Zoom Out Reset image size

Figure 11 shows microscope image of Sample B 20/500 μm mesa surface, no cracks are seen. The total thickness of grown layers is supported by the cross-sectional SEM image (figure 11 (b)). In case of L ≥3000 μm, the GaN grown on the mesa often exhibited cracks. This effect of mesa size is probably due to a substrate bowing and cooling effects [26]. Although the growth process was developed for non-patterned Si substrates, it could be further optimized for growth on patterned wafers. As a result, with simply increasing growth time its possible to obtain crack-free thick GaN on patterned substrate.

Zoom In Zoom Out Reset image size

One of the crack-free GaN 20/500 μm mesas was 2-D scanned with confocal Raman spectroscope to show its stress distribution for visual comparison purposes as illustrated in figure 12. Averaged tensile biaxial stress is then found to be σ_xx = 761 ± 24 MPa (total of 65 data points from total of 2 line scans) which is more tensile stressed than sample A, but still has no cracks. In contrast to the sample A (Figure 9(a)), where stresses are distributed more evenly on the mesa surface, here the stress maximum is concentrated in the center of the mesa (figure 12 (a)). The color scaling was set same in all figures for comparison purposes. The cross-sections for normalized biaxial stress values (Figure 12 (b)) also demonstrates reaching the equilibrium point significantly slower than sample A (figure 9(e)). The effect of higher GaN strain and different stress distribution is probably due to the larger contribution of the low pressure GaN^a to the summarizing GaN layer. For further growth optimizations, the thickness ratio of GaN^a/GaN^b should be considered.

Zoom In Zoom Out Reset image size