Strategies for defect reduction for heteroepitaxially grown GaN
As described in the previous section, heteroepitaxial layers grown on planar foreign substrates exhibit very-high-defect densities. Many different attempts were undertaken to reduce the density of BSFs and dislocations. A BSF density of less than ~104 and dislocation densities below 108 cm-2 would be desirable, since for such values optically pumped green lasing was observed (Strittmatter et al., 2010). The observation in c-plane GaN that the threading dislocation density decreases with thickness by dislocation annihilation applies to stacking faults and partial dislocations in non-polar GaN (Tavernier et al., 2004). In principle, the partial dislocations could bend and annihilate together with the stacking fault. However, climb or glide on the c-plane of these dislocations seems to be hindered by the presence of the stacking faults, making it impossible to use the growth of thick layers for defect reduction. In semipolar GaN a dislocation and BSF reduction was observed, leading to sufficiently low-defect densities (Usikov et al., 2010). The mechanism is unresolved as yet, but it could be caused by glide of partial dislocations in the c-plane due to shear stress that is not present in non-polar GaN (Romanov et al., 2006). Another defect-reduction approach successfully employed for c-plane GaN is three-dimensional growth. This idea was applied to non- and semipolar GaN as well (Hollander et al., 2008; Sun et al., 2009a; Sun et al., 2009b). A reduction of the BSF and dislocation density was achieved, but the defect densities are not yet at the desired levels.
Also, in situ and ex situ nanomasking with SiN (Chakraborty et al., 2006b; Johnston et al., 2009) and ScN (Johnston et al., 2009) or TiN (Tavernier et al., 2004) has been examined, and typically a strong reduction of threading dislocation density could be observed, but no or very little reduction of the BSF density.
The only successful approach to reduce the BSF density to a level at or below ~ 104 cm-1 based on a maskless in situ technique was to exploit the strain relaxation for BSF reduction. This can be accomplished by introducing low- temperature AlN layers (Dadgar et al., 2011). The SF terminate at the (Al,Ga)N interface by the formation of partial misfit dislocations. However, this mechanism works only for (10ll)-type planes (Schulz et al., 2013). After a strong BSF reduction at the interface, the dislocation density can be reduced further by dislocation annihilation after growth of thick layers down to a level near 108 cm-2 (Usikov et al., 2010).
Another approach also known from c-plane GaN is lateral epitaxial overgrowth (ELOG) (Beaumont et al., 2001). For this technique, GaN is partly covered by a lithographically patterned dielectric mask (e.g., SiO2 or SiNx). During overgrowth, GaN does not grow on the mask, but adatoms diffuse to the GaN surface and are incorporated in the overgrown GaN. This way, GaN can grow vertically and laterally to form a coalesced layer (see Fig. 8.9). Defects from the underlying GaN layer are terminated at the mask, and the laterally grown region above the mask is mostly defect-free. This technique was successfully applied to a-plane (Wu et al., 2003), m-plane (Haskell et al., 2005) and (1122) GaN (Bougrioua et al., 2007). Usually, stripe masks are employed which fit well to the surface symmetry of semipolar and non-polar GaN. Due to the low symmetry of the surface, the stripe orientation is very crucial (Craven et al., 2002b) as it strongly influences the defect reduction as well as the formation of facets, the lateral growth rate, and the coalescence of the layers (Netzel et al., 2008). Here we will discuss briefly ELOG of a-plane GaN with the low-index stripe orientations of [1l00], , and  stripe orientations (90°, 45° and 0° to the c-direction). As depicted in Fig. 8.10 for  stripes, the overgrown crystal develops 10l0 m-plane facets that lead to some dislocation reduction (as can be seen from CL), but due to the equivalency of the 10l0 surfaces the lateral growth rate is limited. Also, the inclined top facets impede coalescence. The monochromatic CL image at 3.42 eV also show that BSF can be found in the coherently grown region (window) as well as in the laterally grown region (wing). Since the BSF are perpendicular to the stripe orientation they can enter the laterally overgrown region. The  stripes develop (1011), (1120), (0111), and (1102) facets which exhibit very different growth rates. The fast lateral growth of the (0111) and (1102) facets causes a strong dislocation reduction. The inclined
Fig. 8.9. Working priciple of epitaxial lateral overgrowth.
(1011) facets are overgrown during the coalescence, leading to a reduction of threading dislocation density even in the wing regions (Wernicke et al., 2009). The fast lateral growth and the (1120) facet also support coalescence of the layers. But also for this stripe orientation, BSF can enter the wing region, since they are inclined to the stripe direction. [1l00] stripes exhibit a rectangular profile with (0001), (1120), and (0001) facets, again with very anisotropic growth rates. The CL mapping shows a very strong reduction of threading dislocation and BSF density in the -grown wing (Wu et al., 2008; Bastek et al., 2008). This shows that BSF can be reduced by lateral overgrowth only if the growth direction is perpendicular to the basal plane. By side-wall epitaxy (Imer et al., 2006) a further reduction of the defect densities could be realized. These studies show why many attempts to reduce the BSF density do not produce sufficiently low BSF densities if the crystallographic nature of this defect is not taken into account. In all cases the BSF easily penetrate overgrown regions, as long as the growth is not in the  direction.
Although the defect densities obtained by the ELOG approach are very good, growth on stripe-patterned substrates can achieve even lower defect densities and can be employed for arbitrary surface orientation. The basic idea is to nucleate (0001) oriented GaN on a tilted side facet of a foreign substrate and to achieve a semipolar surface by coalescence. Growth on Si <111> etched into (001) Si by KOH was reported first (Honda et al., 2001). Using differently oriented Si substrates (10l1) (Honda et al., 2002), (1122) (Tanikawa et al., 2008a), (1120) (Tanikawa et al., 2008b) and (1010) (Ni et al., 2010), oriented GaN was grown.
Fig. 8.10. SEM and CL images of a-plane ELOG with , , and  stripe orientation (Netzel et al., 2008; Bastek et al., 2010). Images are courtesy of U. Zeimer (FBH) and B. Bastek.
Also, sapphire can be structured so that (0001) c-plane GaN nucleates on either (0001) c-plane or (1120) a-plane-like surfaces. With this technique, (1010) (Okada et al., 2008), (1120) (Okada et al., 2011a), (1011) (Schwaiger et al., 2010), (1122) (Schwaiger et al., 2011) and (2021) (Okada et al., 20116) GaN layers have been fabricated. The key for such layers is to control nucleation sites by masking with SiO2 and by the nucleation conditions. In addition, achieving coalescence is a major issue requiring optimization of substrate miscut and growth conditions. For structured substrates the c-plane nucleation allows rather low threading dislocation densities which are further reduced by the subsequent growth, and dislocation densities as low as 1 x 105 cm-2 have been achieved (Murase et al., 2011). Also, the formation of stacking faults during nucleation can be prevented in this way. However, the layers still exhibit a low density of stacking faults (the density of BSF bundles is typically 2 x 103 cm-1 (Schwaiger et al., 2011)) due to an exposed (0001) facet during coalescence. But approaches for eliminating these residual stacking faults were proposed (Schwaiger et al., 2011). Heteroepitaxial growth of low-defect-density non- and semipolar GaN is much more challenging than c-plane growth, especially due to the formation of basal-plane stacking faults and associated defects. So far, the most promising approach appears to be the use of c-plane nucleation on side facets of structured sapphire or silicon substrates.