Growth of GaN epitaxial films on polycrystalline diamond by metal-organic vapor phase epitaxy

Our earlier work had indicated that GaN epitaxial layers grown on PD have nitrogen polarity (N-polarity) [14]. Since a different procedure for growing the AlN nucleation layer was used in this work, the polarity-dependence of the etch rate of GaN in KOH was used to identify whether or not the epitaxial growth had proceeded along the (0 0 0 1) or (0 0 0 $\bar{1}$) direction [15]. The layers produced in this work were confirmed to be N-polar.

To demonstrate the role of the Si_xC layer to the epitaxial growth of III-nitride hetero-epitaxy on PD substrates, EDX spectroscopy (data not shown) was performed after removing the N-polar III-nitride heteroepitaxy in KOH. Strong Si and C signals were observed, demonstrating the resilience of the only significant sources of residual Si on this surface, namely the Si_xC and possibly some SiO₂ as revealed in the XPS spectrum in figure 1. Further, it was found that the substrate could be re-used and a successive AlN/GaN heterostructure was successfully grown on it.

A fragment of a composite Si(1 1 1) (2 µm)/PD(60 µm) substrate (sample 1 in table 1) was used to identify the epitaxial relationship between the nitrides and PD. In this case the Si(1 1 1) layer acts as a reference for characterising the crystallographic relationship. Whilst the shape of the Si(1 1 1)/PD substrate fragment was irregular, it contained a straight edge aligned to a principle cleavage plane of Si and this was used as a reference direction for the XRD analysis of the GaN film. An AlN nucleation layer (~80 nm) and a GaN layer (~200 nm) was grown on the Si(1 1 1)/PD fragment to form a GaN(200 nm)/AlN(80 nm)/Si(2µm)/PD(60µm) structure (sample 1 in table 1). This was loaded into the diffractometer with the cleaved edge parallel to the x-ray beam to measure the sample rotation angle (ϕ) at which the GaN(10–13) crystallographic direction could be identified. Afterwards, sample 1 was immersed into a KOH solution (~2 M) at ~95 °C to remove both the nitride layers and the 2 µm Si(1 1 1) layer to leave just the PDA substrate topped with its thin residual Si_xC layer. This substrate was then used to grow sample 2, which comprised of a 20 nm thick AlGaN followed by a  ∼200 nm thick unintentionally doped GaN layer, both grown using the conditions described above. The procedure used to identify the direction of GaN(10 $\bar{1}$ 3) relative to the Si cleavage plane of sample 1 was then repeated for sample 2.

When the straight edge of the wafer was parallel with the x-ray beam, the GaN(10–13) diffraction of both samples 1 and 2 were located at ω ~ 31.52°, tilt angle κ ~ 32.52° and 2θ ~ 64.43°. Based on the epitaxial relationship between GaN and Si(1 1 1) established by Krost et al [16], these results identify the cleaved edge of the Si/PD composite substrate of sample 1 as Si[−1 1 0], which is a preferred cleavage plane of Si and GaN [11$\bar{2}$ 0]. This also indicates that the epitaxial relationship between a GaN epitaxial layer and the PDA growth substrate of sample 2 is the same as that between a GaN layer and the starting Si(1 1 1) substrate on which the PD was grown. Therefore, the epitaxial relationship between GaN and the PD can be represented as GaN[0 0 0 1] parallel to Si[1 1 1] or equivalently GaN [11$\bar{2}$ 0] parallel to Si[−1 1 0].

Samples 3 and 4 were designed to identify the growth procedures needed to realize planar GaN layers suitable for eventual device applications. Both were grown on PDB substrates using an AlGaN (AlGaN) grown by the 3D island process, with the GaN layers grown by the two-step process, but with the duration of the reduced temperature 3D growth varied. sample 3 was grown using a prolonged 3D growth process of ~30 min to result in a total GaN thickness of ~1100 nm, of which  ∼900 nm is 3D GaN, whereas in sample 4 the final GaN layer thickness was ~300 nm and was grown over a thinner AlGaN formed using a 3D growth process of just ~7 min duration.

As a consequence of the likely partially-ordered nature of the Si_xC layer, it is expected that an AlN or AlGaN grown on a PD substrate will contain more defects than an equivalent one grown on a Si substrate. This is because edge dislocation arrays [17], which form between the nucleation layer and the Si_xC layer for efficient relaxation of mismatch stress, will degrade the structural periodicity. Figure 3 compares the evolution of the reflectivity of the growing surface in samples 3 and 4. The surface of sample 4 roughens during growth, causing the observed decay in the oscillations in the intensity of the reflected light. On the other hand, the oscillations in the intensity of the reflected signal during the growth of the GaN layer of sample 3 do not decay, indicating that a planarized layer is forming. The SEM images in figure 4 confirm that hexagonal hillocks have formed on the surface of sample 4 (figure 4(a)), whereas the much longer 3D growth step of sample 3 (∼25 min) results in an almost complete elimination of the hillocks and a planar surface (figure 4(b)).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The elimination of hillocks by growing a thick 3D GaN layer bears a marked similarity to the dislocation annihilation and bending process observed in GaN layers grown on PD substrates by epitaxial lateral overgrowth (ELOG) [14]. Specifically, a high density of extended defects in the nucleation layer will penetrate a growing GaN layer and lead to the formation of grains with significant tilts and/or twists, which result in the occurrence of facets, such as GaN(1 0 –1h) (h  =  0, 1, 2, 3.....). These facets grow at different rates and consequently the surface becomes rough, as seen in figure 4(a). A prolonged 3D growth process will eliminate many of these facets caused the expansion of the slow growing crystallographic facets, at which point the strain can be released locally. The resulting strain gradient can then cause dislocations to bend, interacting to annihilate each other, to result in a more ordered epitaxial film.

This can be seen in figure 5 which compares the x-ray rocking curves of samples 3 and 4 for (a) GaN(10 $\bar{1}$ 3) and (b) GaN(0 0 0 2) diffraction. The twist in sample 3 is smaller than that in sample 4 (figure 5(a)) as a result of its thicker 3D GaN layer. Similarly the tilt in sample 3 is smaller than in sample 4, reduced to 0.35°compared with 0.71° tilt in sample 4 (figure 5(b)).

Zoom In Zoom Out Reset image size

As a result of the high CTE of GaN compared to that of diamond, a mismatch of ~27% [18], a high tensile stress in the GaN layer will develop on cooling from the growth temperature to room temperature. Stress control techniques, such as inserting an AlGaN multilayer stack between an AlN nucleation layer and a GaN layer, or using an intermediate AlN layer between two neighbouring GaN layers, are useful for the growth of crack-free, light–emitting diodes on a Si(1 1 1) substrate, because the much-reduced stiffness of Si [19] at temperatures above 1000 °C allows the Si substrate to bow, according to the Stoney equation [20, 21]. Such an approach may not be appropriate for the growth of GaN layers on PD since the stiffness of the PD barely changes with temperature [22], while the effect that pseudomorphic compressive strain deliberately introduced during epitaxial growth will not cause the substrate to bow in a way that compensates for the tensile stress induced by the CTE mismatch during cool-down. Instead, strain relief by forming misfit dislocations, or worse, cracking, is more likely. Further, as the thermal conductivity of AlGaN is small, about 1% of that of diamond [23], any AlGaN layer introduced into the buffer layer of a device will degrade the thermal resistance between it and the substrate [24].

Dadgar et al have reported that a 3D growth process can also be used to generate a compressive stress by increasing the GaN grain size through growing it on a Si_xN-mask formed in situ on a low-temperature AlN nucleation layer [25]. The procedures used in growing samples 3 and 4, with their initial 3D growth step at  ∼800 °C, were, in effect, a modified version of the approach of Dadgar et al, but without the Si_xN layer, which was omitted in order to eliminate its contribution to thermal resistance between the 2D GaN layer and the PD substrate. Further, it has been shown that the wavenumber of the E2^H mode of the Raman scattering spectrum of GaN provides a sensitive measure of strain in the layer structure [26]. This technique has been used to assess in samples 3 and 4 the impact of their different duration 3D growth steps, hence GaN grain size, on reducing CTE mismatch strain in GaN epitaxial layers grown in PD substrates.

Figure 6 shows the Raman scattering spectra in the region of the E2^H phonon mode of the two samples. The GaN layer of both samples is under tensile stress, but the greater thickness (∼900 nm) of the 3D GaN layer in sample 3 has caused a shift of ~1.5 cm⁻¹ of the E2^H phonon peak towards the zero strain position of 568 cm⁻¹, compared with its position for sample 4. The rate of shift of the E2^H peak with biaxial strain in GaN epitaxial films has been reported to lie between a theoretical value of 2.4 cm⁻¹ GPa⁻¹ and measured values in the range 2.7–6.2 cm⁻¹ GPa⁻¹ (for a review, see 26). Tripathy et al reported a coefficient of 4.3 cm⁻¹ GPa⁻¹ for the strain-induced shift of the E2^H phonon peak in GaN and Al_xGa_1−xN/GaN heterostructures grown on Si(1 1 1) [27]. Using the latter value on the basis that it was obtained from hetero-epitaxial structures under tensile stress as our data were, the reduction in tensile stress resulting from the more extensive 3D growth is estimated to be  ∼350 MPa.

Zoom In Zoom Out Reset image size

The tensile stress in an as-grown GaN layer can also be reduced by utilising the inherent bow of PD substrates. The origin of this curvature of the PD is residual stress developed in the PD during its growth on the silicon substrate [28]. Because of the high thermal conductivity of diamond, the curvature is not expected to lead to a significant temperature variation across a PD substrate during GaN epitaxial growth, particularly when an appropriately curved substrate holder is used. The convex bow of PDA type substrates is preferred because this will tend to compensate the higher CTE of GaN on cool-down from its growth temperature. Figure 7 compares the Raman scattering spectra near the E2^H peak of samples 5 and 6 which were on PDB and PDA, i.e. substrates of opposite curvature, respectively. Apart from the different sign of the substrate bow, the growth parameters of these samples were the same as those of sample 3, except that the thickness of the AlN nucleation layer grown at ~800 °C was increased to 40 nm.

Zoom In Zoom Out Reset image size

Despite samples 5 and 6 both having concave curvatures after growing the III-nitride heteroepitaxy, the E2^H peak in the Raman scattering spectrum of samples 6 is ~1.8 cm⁻¹ closer to its unstrained wavenumber, a shift (based on the data in [27]) that corresponds to a tensile stress ~420 MPa lower than that in sample 5. Further, no cracks were observed in sample 6, whilst some scattered cracks appeared in sample 5. The reduced tensile stress in sample 6 is related to its curvature change from convex to concave during cooling down. The curvature flip, from convex to concave before and after growing the III-nitride heteroepitaxy, occurs because the PDA substrates are only 50–60 µm thick (as were the PDB substrates) and therefore less rigid than thick diamond substrates (>0.2 mm). As such, they were thin enough to be deformed on cool-down from the GaN growth temperature. This use of flipping the pre-existing wafer bow provides a promising method for mitigating CTE induced tensile stress in GaN epitaxy grown of PD substrates, but needs detailed characterisation and modelling.