On the polarity determination and polarity inversion in nitrogen-polar group III-nitride layers grown on SiC

We investigate the interfaces and polarity domains at the atomic scale in epitaxial AlN and GaN/AlN grown by hot-wall metal organic chemical vapor epitaxy on the carbon face of SiC. X-ray diffraction, potassium hydroxide (KOH) wet chemical etching, and scanning transmission electron microscopy combined provide an in-depth understanding of polarity evolution with the film thickness, which is crucial to optimize growth. The AlN grown in a 3D mode is found to exhibit N-polar pyramid-type structures at the AlN – SiC interface. However, a mixed N-polar and Al-polar region with Al-polarity domination along with inverted pyramid-type structures evolve with increasing film thickness. We identify inclined inversion domain boundaries and propose that incorporation of oxygen on the h 40 – 41 i facets of the N-polar pyramids causes the polarity inversion. We find that mixed-polar AlN is common and easily etched and remains undetected by solely relying on KOH etching. Atomic scale electron microscopy is, therefore, needed to accurately determine the polarity. The polarity of GaN grown on mixed-polar AlN is further shown to undergo complex evolution with the film thickness, which is discussed in the light of growth mechanisms and polarity determination methods.


INTRODUCTION
][3][4] As a consequence of their wurtzite crystal structure, which lacks an inversion symmetry along the c-axis, c-plane oriented III-nitride layers can have either metal or nitrogen polarity.The (0001) surface is defined as metal-polar (e.g., Al-, In-, or Ga-polar), and the (000 1) surface is referred to as nitrogen-polar (N-polar).The polarity has an important effect on material properties such as impurity incorporation, surface roughness, defects, and recombination efficiency.The polarity can be exploited to tune the device performance and design novel device architectures.7][18][19][20][21][22] However, challenges remain for metal organic chemical vapor deposition (MOCVD) growth of N-polar III-nitrides. 11In particular, polarity inversion domains (IDs) with various shapes are among the most often encountered types of defects in N-polar epitaxial layers and device heterostructures.
The cause of IDs is typically attributed to impurity incorporation, such as oxygen 23,24 or metal ad-layers in combination with strain effects. 257][28][29][30][31] These defects interfere negatively with the desired p-type conductivity, which can be reduced or even altered to n-type.Effectively, local polarity IDs also deteriorate the crystal quality and, thus, device performance.Therefore, it is critical to reliably determine and control the polarity of III-nitride epilayers and device heterostructures.
6][37] This technique is a simple and rapid method to verify the polarity through monitoring the surface morphology since the etching rate and surface roughness depend on the chemical nature of the surface.For example, KOH etching of an N-polar surface results in a rough surface with hexagonal hillocks of varying sizes.On the other hand, KOH etched Al-polar or Ga-polar surfaces are not affected and remain smooth.It was shown that in contrast to N-polar surfaces, the etching speed of Al-polar or Ga-polar surfaces is negligible. 38Furthermore, it was shown that the etching speed of N-polar AlN is higher than that of N-polar GaN. 34Recently, we proposed to combine high resolution x-ray diffraction (HRXRD) with KOH etching and scanning electron microscopy (SEM) for polarity determination of AlN nucleation layers. 226][37] The method has also the potential to verify metal-polar inclusions in mixed-polar layers, which could remain after the N-polar material is etched away and hence detected by XRD.However, depending on thickness, specific IDs shapes and distribution conclusive determination of polarity in mixed-polar III-nitride epilayers and heterostructures by KOH etching even in combination with HRXRD may be rather challenging.
In this work, we employ scanning transmission electron microscopy (STEM) to investigate the polarity of epitaxial AlN and GaN/AlN layers grown on C-face 4H-SiC by hot-wall MOCVD.Polarity IDs and general atomic interface arrangements of the ID boundaries (IDBs) are revealed and discussed.A new evidence showing limitations of KOH and XRD techniques in determining the polarity of mixed-polar III-nitrides is presented.Furthermore, a comprehensive picture of polarity evolution from the AlN nucleation layer into the GaN and extending to the surface is presented.Polarity inversion mechanisms are discussed, and strategy for growth optimization of targeted polarity in III-nitrides is suggested.

EXPERIMENTAL DETAILS
The AlN nucleation layers (NLs) and GaN are grown by hot-wall MOCVD with a horizontal reactor design on carbon face on-axis semi-insulating (SI) 4H-SiC (000 1).All growth experiments were performed at a low pressure of 50 mbar.Trimethylaluminum (TMAl), trimethylgallium (TMGa), and ammonia (NH 3 ) were used as the source precursors for Al, Ga, and N, respectively.Epi-ready C-face SiC substrates are precleaned using a standard routine as described in Ref. 22. Before growth, the cleaned substrates were annealed and etched with hydrogen at 1360 C in the MOCVD reactor.After a pre-flow of NH 3 , AlN NLs growth takes place with a mixture of N 2 and H 2 as the carrier gas at 1050 C and a V/III ratio of 1257.GaN was grown on top of AlN NLs at a temperature of 1000 C and a V/III ratio of 15 000.The surface morphology of the AlN NLs and GaN layers were investigated by SEM and atomic force microscopy (AFM).Zeiss SEM LEO1550 and Bruker AFM Dimension 3100 in a contact mode are used, respectively.The polarity of the AlN NLs is examined by KOH etching with a KOH concentration of 0.18 mol/l at 55 C for 5 min, while for GaN, a KOH concentration of 8.91 mol/l and a etching time of 15 min at 55 C is applied. 34Panalytical HRXRD (Empyrean) is used to characterize the crystallographic orientation and verify the polarity of AlN NLs and GaN.2θ-ω scans around the AlN and GaN (0002) peaks were performed on as-grown and the respective KOH etched layers using the same acquisition parameters.
STEM was employed to investigate the polarity of epitaxial GaN/AlN/C-face 4H-SiC.TEM samples were thinned to electron transparency by using standard mechanical polishing and Ar ion milling procedures.Characterization was carried out with the image and probe aberration corrected Linköping Titan 3 60-300 (S) TEM equipped with a high brightness Schottky field emission gun operated at 300 kV, providing 0.6 Å resolution.Energy dispersive x-ray (EDX) was performed using a Super-X EDX spectrometer, and STEM images were acquired with an annular dark field (ADF) detector.Electron energy-loss spectroscopy (EELS) was carried out using a Gatan Quantum GIF spectrometer.Simulations of ADF STEM images were performed using Dr. Probe software [X], rendering supercells of AlN and adopting standard structure parameters (a = 3.111 97 Å, c = 4.980 89 Å, u iso = 0.3866). 39Microscopy parameters used for the simulations were measured prior to imaging.The parameters include acceleration 300 kV, 21.5 C, convergence semi-angle 21.5 mrad, energy spread 1.3 eV, probe size 0.6 Å, probe current 10 pA, twofold astigmatism ,2 nm, spherical aberration 200 nm, chromatic aberration 1.5 mm, threefold astigmatism ,30 nm, axial coma ,30 nm, fourfold astigmatism 300 nm, star aberration 500 nm, and ADF angular range 80-200 mrad.A frozen phonon model was included utilizing 25 modes to account for vibrational effects at room temperature.Pixel dwelltime was set to 8 μs.Image processing was carried out using the commercially available software Gatan Microscopy Suite 3.4 with built-in routines.

Journal of Applied Physics
conclude that the contrast variations in ADF STEM images arise from different polarity regions and their inversion boundaries (IDBs).The pyramidal polarity region onset after the first few layers of AlN growth is an indication of preferential structure change along high angle facets.Al-polar regions appear to grow faster confining the N-polar regions to steep pyramidal structures.This is an important observation because it shows that large area overview STEM images can be used to determine the polarity inside and outside the pyramids only by considering whether inverted or regular pyramids are observed.Previously, similar approach has been proposed for Mg pyramid inversion domains in GaN. 30,41For example, it was found 42 that pyramidal N-polar regions were identified in a predominantly Ga-polar GaN cap layer, while others 24,43 reported V-shape domains with Al-polar regions inside the V-shape and N-polar outside.
The origin of the transformation from N-polar to Al-polar is important to address.The observed inclined ID boundary (IDB) [Fig.2(c)] corresponds to the (40 41)/(04 41) planes and their mirrors.The IDB represents a partial overlap of two AlN lattices with Al-and N-polarity, respectively.Figure 3 shows an illustration of an atomic model of the observed inclined IDB in AlN and the corresponding simulated and experimentally observed ADF STEM image.Partial occupancy was assumed at overlap positions of Al from N-polar (blue) and Al-polar (red), whereas the N positions (black) remained constant [Fig.3(a)].Such a scenario could arise when N-polar and Al-polar domains have nucleated in 3D fashion and coalesce generating the observed IDB.However, we could not find evidence for Al-polar domains that nucleate at the interface with the SiC and, therefore, excluded this scenario as a probable mechanism.Oxygen is ubiquitous impurity in N-polar III-Nitrides and is incorporated on the N site.We note that the O levels in our AlN NL were found to be in the order of 10 20 cm À3 by secondary ion mass spectroscopy.For comparison, the O levels in the GaN layer on top of the AlN NLs are at least two orders of magnitude lower even though GaN is gown at lower temperature.The reason for the high O levels in AlN is related to the reaction between the TMAl precursor and oxygen in the gas stream, which results in the formation of volatile Al(CH 3 ) 2 CH 3 OH product.This product is believed to be responsible for the incorporation of oxygen. 44educing residual oxygen incorporation in AlN is further complicated by the fact that the maximum H 2 O/H 2 ratio allowed over Al-containing surface before oxygen incorporation occurs is minute, e.g., in the order of 0.01 ppb for AlGaAs. 44Al acts as a gatherer of O and there is always O present to some extent in the MOCVD environment.Additional source of O close to the interface with the substrate may be potentially associated with the native SiO 2 on SiC.SiO 2 is removed before growth by appropriate substrate preparation and conditioning (see experimental section) but O may remain in the gas system and get incorporated during the initial stages of growth.Indeed, O incorporation in the AlN NL/SiC is always higher in the defective region close to the substrate as compared to AlN grown further way from the interface.We recall that AlN NL is necessary for the growth of the GaN-based structure on SiC to accommodate the lattice mismatch.
Furthermore, it is well known that N-polar and semi-polar/ nonpolar surfaces incorporate higher levels of O (impurities in general) 45,46 compared to the metal-polar case.Hot-wall MOCVD grown Al-polar AlN further away from the interface has typical concentrations of the order of 10 18 cm À3 . 47For comparison using the same growth reactor, we routinely achieve O levels of (2-5)Â 10 15 cm À3 in Ga-polar GaN, which is at the detection limit of the SIMS measurements employed.In such structures, the O in the respective Al-polar AlN NLs is still of the order of (5-6)Â 10 19 cm À3 .
It has also been shown that a higher angle facet is more prone to bind oxygen for the case of GaN. 48It is, therefore, plausible to suggest that the inclined facets of the N-polar 3D islands are likely to incorporate residual oxygen from the MOCVD environment.In such instances, the N atoms will be partly replaced by O. Previously, it has been shown that the formation of Al x O y N z during the early stages of the nitridation process causes a planar IDB and is responsible for the inversion from N-to Al-polarity in AlN grown on sapphire. 23Such an Al O y N z formed at the growing facets of the N-polar AlN islands could provide an explanation for the observed inclined IDB [Fig.2(c)] and the polarity inversion in our case.This is also consistent with previous reports showing that annealing in oxygen can result in change from N-polar to Al-polar AlN. 24Hence, we suggest that oxygen is likely responsible for polarity inversion in AlN grown on SiC and in the case of inclined IDBs as well.This second scenario is illustrated in the atomic arrangement model in Fig. 4(a The h40-41i N-polar pyramid geometry is illustrated in Fig. 4(e).Note that it is still the same lattice overlap that occurs as in Fig. 3(a).However, in the latter, when the IDB is closer to the SiC interface, the appearance is of an inverted Al-polar pyramid in a dominant N-polar background.Furthermore, no local variations of oxygen at the pyramid boundaries are observed by EELS.We believe that this is related to the observed overlapping of pyramids

Journal of Applied Physics
of random sizes (see Fig. 2) that causes delocalization of O-K signal (see Fig. S1 in the supplementary material).Future works employing a higher sensitivity setup using single electron detection may help resolve the oxygen incorporation profile in such instances.
In general, increasing V/III ratio can reduce the incorporation of O in N-polar III-nitrides but in the MOCVD environment there will be always residual levels sufficient to trigger polarity inversion.Although O may case the formation of IDBs, it is the availability of the semi-polar surfaces of the AlN islands during the 3D growth mode that provide highly susceptible sites for O incorporation.Therefore, another approach to suppress polarity inversion is to transition from 3D growth to step-flow growth mode in order to dispense with the inclined island facets.Indeed, when grown in a step-flow mode we find that AlN preserves N-polarity throughout the entire film thickness and no inversion domains could be observed (see Fig. S2 in the supplementary material) even though the O levels are also of the order of 10 20 cm À3 .Detailed study on inversion domain suppression and effect of growth conditions on the polarity will be reported elsewhere.
The STEM results clearly show that the AlN layer has mixed polarity and that in these instances KOH etching could not provide an adequate polarity assignment.Arguably, the latter may serve only as an indirect indication for the dominating polarity.The KOH etching method should be used with caution as it may give similar results for purely N-polar and mixed-polarity layers.However, different growth strategies would be required for these two distinctive cases to control polarity of subsequently grown layers.It is likely that GaN layers grown on mixed-polarity AlN such as the one shown in Fig. 2 exhibit mixed-polarity.Hence, we further investigated the polarity and atomic arrangements at the interface of epitaxial GaN grown on the mixed-polarity AlN discussed above.

GaN on AlN/SiC
Figure 5 summarizes the AFM, SEM, and XRD results before and after KOH etching of a 50-nm-thick GaN layer grown on the AlN NL described above.The surface of the as-grown GaN is rougher as compared to the AlN NL with an RMS surface roughness of 5.9 nm [Fig.5(a)].The surface morphology reveals large islands with characteristic sizes of 1-3 μm and height of 5-10 nm.This surface morphology is indicative of the Volmer-Weber growth mode.A closer look at the islands shows that they  constitute of hexagonal plates that are sequentially arranged on top of each other and rotated by some degrees.Therefore, a spiral growth around a screw dislocation may occur.The SEM topography of the as-grown GaN in Fig. 5(b) is consistent with the AFM surface morphology.After KOH etching, a significant part of the layer is etched away as revealed by SEM [Fig.5(c)].The remaining material encompasses entirely etched surfaces with hexagonal shapes as indicated in the magnified picture in the inset of Fig. 5(c).Hence, one can speculate that the hexagonal-like islands observed in the as-grown GaN have N-polarity and are, thus, etched away.The remaining part of the layer is most likely Ga-polar on a microscopic scale.Figure 5(d) shows the GaN (0002) 2θ-ω scans before and after KOH etching, respectively.It is seen that parts of both GaN and AlN layers remain after KOH etching as evidenced by the non-zero intensity of the GaN and AlN (0002) diffraction peaks.Thus, the approach combining KOH etching and XRD indicates that the GaN is mixed-polar.
Cross-sectional STEM was used to further discern the polarity evolution and atomic arrangements in order to get insight into the specific growth mechanisms at the nanoscale.Figure 6(a 6 consists of single crystalline AlN, while the more grayish pyramid regions show lattice overlap appearance.The latter is a result of coexisting N-polar and Al-polar domains in the volume of the TEM lamella, which in projection appear as overlapping lattices.We hereafter refer to such regions of coexisting opposite polarities as mixed-polar.The GaN is positioned symmetrically on top of the AlN pyramid structure forming an archway pillar appearance, highlighted by white dashed lines.Figures 6(d) and 6(e) provide a detailed view of the atomic structure at the interfaces of typical peaks and valleys in the archway structures, respectively.Also, Fig. 6(d) verifies that AlN archway regions, that appear dark, are Al-polar.The atop grown GaN, that appears bright, is Ga-polar.Figure 6(e), on the other hand, shows a mixed-polar pyramid regions where coexisting N-polar and metal-polar domains are present in the imaged volume.The mixed-polar AlN intersects with a mixed-polar inverted GaN pyramid.The coexisting of GaN domains with opposite polarities is concluded by comparing simulated ADF STEM images of overlapping N-polar and metal-polar lattices.Based on our STEM results, we propose that if the N-polar AlN pyramids enveloped with Al-polar material extend up to the surface, the atop grown GaN forms an archway type of structure, which replicates the underlying polarity, i.e., Ga-polar GaN forms on Al-polar AlN and N-polar GaN forms on N-polar AlN.Since the growth is faster along the metal-polar direction as compared to the N-polar case, 49 the resulting mixed-polar GaN contains larger Ga-polar regions with smaller N-polar domains embedded within.The latter when imaged by STEM appear only as overlapped lattices with the overall Ga-polar matrix.
It is seen from Fig. 6(c) that the GaN forms deep into the valleys between the AlN V-shape domains.The GaN islands are proposed to be nucleated on the AlN surface through the

Journal of Applied Physics
Volmer-Weber growth mode during the initial deposition stage. 50he regions where Ga-polar and N-polar domain coexist is expected to have a lower growth rate as compared to the neighboring purely Ga-polar GaN.The competitive growth with Ga-polar growing faster than N-polar material results in two types of domains with Ga-and mix-polarity as illustrated in Fig. 7.Such a scenario provides an explanation for the observed two distinct topographies in the AFM and SEM images of the GaN surface .When switching to GaN growth, the V/III ratio is increased from 1257 to 15 000 in order to maintain the N-polarity.The step also includes a 5-min nitridation at 1000 C before the GaN growth is initiated.The high HN 3 during the nitridation may cause surface roughening due to etching and be responsible for the observed rough interface [Fig.6(c)].To verify this hypothesis, we investigated the surface morphology of an AlN NL which was exposed to the nitridation step without growing the subsequent GaN (Fig. S4 in the supplementary material).The results show that the RMS surface roughness increases from 1.9 nm for the as-grown sample [Fig.1(a)] to 3.7 nm after nitridation.In comparison, the V-shape ondulations at the GaN/AlN interface observed in Figs.6(a) and 6(c) have characteristic size of 5-7 nm.This indicates that additional factors besides the nitridation contribute to the roughening.We note that the roughening can be completely overcome by tuning the growth mode of the AlN NL to stepflow, as exemplified in Fig. S2 in the supplementary material.In such instances, we find that variation, for example, in the V/III ratio from 250 to 15 000 results in pure N-polar GaN and no ID could be observed.Further details on the effect of different growth conditions on the polarity of III-nitride layers and heterostructures grown by hot-wall MOCVD will be reported elsewhere.

Limitations in polarity determination by KOH
By STEM investigation, we demonstrate that when AlN grows in a 3D fashion steep pyramid-like structures with N-polarity formed and which become enveloped by Al-polar regions as the film thickness increases (see Fig. 2).This type of mix-polar AlN films with coexisting N-polar and Al-polar domains is entirely etched away even with a low molarity of KOH solution at 50 C.Previously, Hussey et al. 43 have reported that a combination of low molarity of KOH etching solution and SEM techniques revealed a very low density of IDs in N-polar AlN, which could be easily over-etched and, thus, rendered undetected.Based on our results, we suggest that this is due to the steep N-polar pyramid that extends to the surface, which served as an initial site and pathway for the etching in depth.A similar scenario is proposed also to take place in the case of GaN on AlN/SiC where most of the layer is etched away [see Fig. 5(c)] despite the predominant Ga-polarity.
Figure 8 illustrates the proposed model for reverse etching in the case of GaN grown on AlN/SiC.Initially, the N-polar regions are etched, which results in inverse pyramidal pillars of Al-polar AlN and Ga-polar GaN, see Figs.

Journal of Applied Physics
metal-polar domains and allow for their reverse etching.The process of further advancing the removal of metal-polar domains by reverse etching is depicted schematically in Fig. 8(c).This stage represents the experimentally observed situation in GaN on AlN/SiC after KOH etching [Fig.5(c)].In the case of the AlN NL, the reverse etching process advances faster since there are larger N-polar domains, as can be inferred from the STEM results, where pure N-polar regions could be observed (see Fig. 2).Consequently, the entire AlN NL is etched.In contrast, in GaN, we could only observe mix-polar regions in the STEM images, i.e., the N-polar GaN domain sizes are smaller and they only appear as overlapped lattices [Figs.6(c)-6(e)].As a result, the reverse etching process of GaN is slower since in general GaN is etched slower than AlN, and it will take longer time to reach the N-polar AlN regions and initiate the reverse etching of Al-polar AlN and the respective Ga-polar atop.One might speculate that given sufficient time the entire GaN/AlN stack could be etched away.However, we could not completely remove GaN even if the etching is performed for significantly longer times.This suggests that the reverse etching process may be self-limited governed by the specific ratio of N-polar to metal-polar domain sizes, the ID boundaries type and the etching conditions.
The immediate limitation of KOH and SEM/AFM interpretation of polarity becomes apparent when considering that the AlN is mixed-polar with predominant Al-polarity but all is etched away, which is typically considered an indication of N-polarity. 22On the other hand, if mixed polarity is manifested with well separated laterally regions with different polarities, the KOH method may provide more conclusive picture.This is the case with the GaN/AlN presented here, where the Ga-polar domains in the film remain unaffected by the etching, while the mix-polar or predominantly N-polar GaN domains are etched away.

CONCLUSION
The applicability of standard methods for determination of polarity in III-nitride epitaxial layers and heterostructures, such as KOH wet chemical etching in combination with XRD and SEM, is critically evaluated.Exemplary polarity situations are presented for AlN epitaxial layers and GaN/AlN heterostructures grown on C-face 4H-SiC (000 1) substrates by hot-wall MOCVD via detailed STEM investigations.We show that there are limitations in solely relying on KOH etching for polarity determination for mixed-polar layers consisting of metal-polar and N-polar domains.STEM images reveal steep pyramidal N-polar structures in AlN films grown in the 3D growth mode that extends to the surface and are completely enveloped in Al-polar regions.We have found that this type of mixed-polar AlN could be easily etched and remain undetected by KOH etching, as N-polar regions are undermining the Al-polar regions as well as extending to the surface allowing for complete removal of the Al-polar material via reverse etching.We hypothesize that this is a universal situation applicable to all substrate types.Inclined inversion domain boundaries were identified in AlN on SiC and it is suggested that incorporation of oxygen on the h40-41i facets of the growing islands is likely responsible for polarity inversion in similarity to lateral IDB in AlN on sapphire.
GaN exhibits a rough V-shape interface with the AlN NLs with the 3D growth mode.This is only partly attributed to roughening of the AlN island-like surface during nitridation.Additional roughening may be caused by the initial growth of GaN with mixed polarity.The lower growth rate of the N-polar region competes with the higher growth rate of the Al-polar region and forms steep pyramid structures that extend up to the surface.The atop grown GaN forms an archway type of structure with Ga-polar and mixed-polar regions.Thus, the mixed type of Ga-polar and N-polar GaN layers formed under competitive growth could be associated with two different types of surface topography, easily detected by AFM and SEM.In such instances, KOH etching could provide adequate indication of polarity.Nonetheless, atomic scale electron microscopy proves necessary to accurately determine the polarity.

SUPPLEMENTARY MATERIAL
See the supplementary material for additional details on the elemental quantification profile across the GaN/AlN/SiC interfaces, on the atomically resolved N-polar GaN/AlN/SiC structure with flat interfaces, and on the surface morphology of AlN after NH 3 pretreatment.

Figure 1
Figure 1 summarizes the AFM, SEM, and XRD results for a representative 35-nm-thick AlN NL.The AFM and SEM images of

FIG. 2 .
FIG. 2. (a) ADF STEM cross-sectional overview image showing the AlN nucleation layer grown on SiC.(b) STEM-EDX maps showing Al-K (blue) and Si-K (yellow) signals.(c) Atomically resolved ADF STEM image demonstrating steep pyramid structures in the AlN with boundaries highlighted by dashed lines.
) with the corresponding simulated ADF STEM image [Figs.4(b) and 4(c)] and experimentally observed STEM image [Fig.4(e)].A representative experimental STEM image of the IDB further away from the interface with the SiC is chosen in order to better reveal the edges of the steep pyramids representing the IDB.In such instances, the N-polar pyramid appears embedded in an Al-polar background [Figs.4(b) and 4(c)].

FIG. 3 .
FIG. 3. (a) An illustration of the atomic model of an inclined IDB in AlN, where partial occupancy was assumed at overlap positions of Al from N-polar (blue) and Al-polar (red), whereas the N positions (black) remained constant.(b) The corresponding simulated ADF STEM image and (c) STEM image of the inclined IDB presented in (a) and (b).

FIG. 4 .
FIG. 4. (a) An illustration of the atomic model of the inclined IDB for an embedded N-polar AlN h40-41i pyramid in Al-polar AlN presented in the [11-20] direction.Partial O occupancy (50%) was accounted for at h40-41i surfaces.(b) Shows the corresponding simulated ADF STEM image.Al in Al-polar (red), Al in N-polar (blue), N (black), and N/O (black/green).(c) Atomic model overlaid on the ADF STEM image.(d) Illustration of the h40-41i N-polar pyramid geometry.(e) STEM image of the inclined IDB presented in (a) and (b).

FIG. 5 .
FIG. 5. (a) AFM and (b) SEM images of the as-grown GaN, (c) an SEM image after KOH etching and (d) 2θ-ω scans around the GaN and AlN (0002) diffraction peaks before (solid black line) and after (red dashed line) KOH etching.The respective RMS roughness of the as-grown GaN layer is indicated in (a).The inset in (c) shows a magnified region after KOH etching that reveals the etched hexagonal areas surrounded by non-etched material.
) presents a structural overview of the AlN and GaN layers in cross section, characterized using ADF STEM imaging.The AlN/GaN interface shows clear contrasted regions with steep pyramids in the lower part intersecting with wide inverted pyramids above (shown in bright intensity).

Figure 6 (
b) shows the EDX map of Ga-K and Al-K signals confirming a rough structural transition judging by the Ga and Al signals across the interface with inverted GaN pyramids extending down in the AlN layer.More details can be seen in Fig. S3 in the supplementary material exemplifying the EDX spectra and Fig. S1 in the supplementary material where C-K, N-K, and O-K EELS maps show no compositional variation across the interface.Transition AlGaN regions can be identified at the boundaries of the archway-pillar interface between GaN and AlN in the STEM-EDX maps [Fig.6(b)].Hence, the observed shoulders between 34.5 and 35.5 in the XRD scan in Fig. 5(d) could be ascribed to the AlGaN (0002) peak.Figure 6(c) presents an atomically resolved ADF STEM image of the GaN-AlN interface acquired along a [11 20] of SiC.A closer inspection reveals that the darker regions in the AlN in Fig.

FIG. 6 .
FIG. 6.(a) HAADF STEM cross-sectional overview image of the GaN-AlN interface.(b) STEM-EDX map showing Ga-K (green) and Al-K (red) signals.(c) Atomically resolved HAADF STEM image of the GaN-AlN pyramid structures (white dashed lines) with two regions of interest (white dashed rectangles) showing an Al-polar to Ga-polar interface (d) and a mixed-polar AlN to mix-polar GaN region (e).
[Figs.5(a) and 5(b)].The rough GaN/AlN interface with pyramid-like structures [Figs.6(a) and 6(c)] may cause destructive interference, which could explain the reduced intensity of the AlN (0002) 2θ-ω peak observed in the XRD experiments [Fig.5(d)] FIG. 7. (a) Cross-sectional ADF STEM image showing the GaN surface structure.A gray dashed line separates a mixed-polar region from a Ga-polar region.(b) Magnified view of the interface of the mix-polar and Ga-polar GaN.A Ga-polar GaN atomic model is presented in red.A color scheme is applied to highlight the intensity differences of the Ga and N positions.

FIG. 8 .
FIG. 8. Proposed model for revers KOH etching of metal-polar domains in N-polar pillar-archway AlN-GaN interfaces.(a) Schematic showing the polarity regions observed, Ga-polar in green, N-polar GaN in yellow, Al-polar in red, N-polar AlN in blue, and SiC in black.(b) Initial pathway for KOH etching through N-polar regions providing channels for KOH to reverse etch the metal-polar domains.(c) Pillars of metal-polar regions remain with angled facets that enable reverse etching.