Parasitic Al_xO_yN_z surface defects on high-temperature annealed AlN and their role in hillock formation

During the last few years, the importance of disinfection and sterilization was emphasized by the COVID-19 pandemic. Thus, UV-C radiation, which can be used for disinfection/sterilization purposes, gained a lot of interest. Since microorganisms exhibit the largest germicidal efficacy around an emission wavelength of 265 nm, the respective UV-C LED development focusses on this wavelength range.¹ Unfortunately, LEDs emitting in the UV-C regime exhibit inferior performance compared to their blue and UV-A pendants. This is because of physical restrictions as well as technological challenges that have to be surpassed.^2–4 To name a few, higher vulnerability of radiative recombination to defects, both, extended threading dislocations as well as point defects, low transparency of the AlN growth substrates due to intrinsic point defects, light outcoupling due to changes in polarization of the emitted photons, doping limitations, as well as processing challenges due to a stronger wafer bow caused by the higher coefficient of thermal expansion (CTE) mismatch between AlN and the commonly used sapphire substrates.^2–7 

The high-temperature annealing (HTA) of thin AlN layers on sapphire can tackle many of the problems listed above. Due to their reduced dislocation density and high transparency, HTA AlN templates are a well-suited candidate for the growth of UV-C LEDs.^8,9

In contrast to HTA templates, MOVPE AlN templates usually have a higher screw dislocation density. However, in this case, the observed spiral growth mode of AlGaN on MOVPE AlN templates is not a problem for the fabrication of UV LEDs. Due to a vast number of screw-dislocations, the regions affected by an individual screw dislocation overlap, which then leads to a relatively flat surface, suited for growth of the active area and further processing of the epitaxial wafer.¹⁰ In HTA AlN, however, the situation is different. The low screw dislocation density that is typical for HTA AlN can lead to huge singular spirals in the AlGaN layer, since each individual spiral grows faster compared to the predominantly screw-dislocation-free surface that grows in step-flow mode.^10,11 These huge singular spirals tend to incorporate more gallium compared to the spiral-free surface area.^12,13 The higher gallium incorporation leads to increased strain, which, in turn, leads to the commonly observed fringing of the spirals after exceeding a certain AlGaN thickness. The fringed bottom part of the spiral results in a self-enforcing gallium incorporation, due to the preferred sticking of gallium at macrosteps. Gradually, the former spiral develops into a large surface defect, either with an open or closed core. The open-core version was found to carry leakage currents in final UV emitting devices.¹⁴ In addition, a light-emitting device processed out of an epitaxial wafer containing such surface defects shows a broadened emission peak due to the different levels of gallium incorporation across the wafer (spiral vs flat area).¹¹ Screw dislocation and the spiral-growth mode of AlGaN are inherent properties of this material system (as long as growth does not happen under extreme conditions regarding supersaturation or on wafers with an uncommonly high off-cut). Therefore, these defects are mostly inevitable. Typical densities of the described surface defects in the AlGaN layer are in the 10⁵–10⁶ cm⁻² range, corresponding to the total screw-dislocation density of the HTA AlN template,¹⁵ but were also reported to be as low as mid 10³ cm⁻².¹¹ 

In addition, hillock defects can be observed in AlN and AlGaN films. Here, we want to distinguish between the explained spirals that are also often denoted as hillocks in the literature and hillocks that stem most likely from surface contaminations and nucleation on top of these, which will be shown in the following sections. These defects are also known from growth on native AlN substrates.¹⁶ They appear as flat-top mesas with differing heights. It is not easy to separate between defects that are spiral-related and hillocks in later stages of a growth process. Differentiation using an AFM or SEM is only possible after some hundred nanometers of growth, when the screw-related defects still retain their spiral shape, but the contamination related hillock defects already appear as hexagonal mesas. Indeed, depending on the origin of the hillock, it can already be observed during AlN growth stages, whereas singular spirals are only appearing as large surface irregularities as soon as gallium is introduced. A distinct identification is possible by transmission electron microscopy (TEM). Screw dislocations originating from the sapphire/AlN interface in the case of HTA AlN, or from inside the AlN bulk material, can be seen as an inherent property of the material, whereas defects related to contaminations are avoidable and denoted as hillocks in this publication.

Nominally 350 nm thick commercial AlN samples grown by physical vapor deposition (PVD) on c-oriented sapphire substrates were used for the experiments. PVD-grown AlN on sapphire typically grows with columnar morphology along the [0002]-direction. The as-grown Al-polar samples exhibit (0002) and (10–12) XRD reflexes with FWHMs of ∼200 and ∼2500 arc sec, respectively.

The HTA process was carried out inside a cold-wall furnace in N₂ atmosphere at ambient pressure. The AlN samples were annealed in face-to-face configuration for 5 h in 1 slm of N₂ to suppress thermal decomposition, as introduced by Miyake et al. in 2016 for AlN.^17,18 More details about the annealing setup were published elsewhere.^6,19 The annealed AlN/sapphire templates were used for AlN regrowth inside a Thomas Swan 3 × 2 in. CCS MOVPE reactor to recover a step-flow morphology suited for AlGaN growth. For this, two similar HTA templates were used with different bake conditions, either a bake in pure H₂ atmosphere (125 mbar) at 1100 °C for 5 min (sample A) or a bake in H₂/NH₃ (14.5 and 0.5 slm, respectively) atmosphere (50 mbar) at 1150 °C for 10 min (sample B). Subsequently, 500 nm AlN were grown with a growth rate of approximately 1.25 μm/h on top of the HTA AlN at 1130 °C and 50 mbar with a V/III ratio of 50. After regrowth, the surfaces of samples A and B were analyzed using PeakForce tapping-mode with a Bruker Dimension Icon atomic force microscope (AFM). Additionally, cross-sectional transmission electron microscopy (TEM) lamellae of sample A including three different hillocks were prepared in a ThermoFisher Helios G4 dual beam focused ion beam (FIB) using 2 kV gallium ions for the final milling step. The lamellae were investigated in an image-corrected FEI Titan 80–300 operated at 300 kV by means of selected-area electron diffraction (SAED), weak-beam dark field (WBDF) imaging, energy dispersive x-ray (EDX) spectroscopy, and electron energy-loss spectroscopy (EELS). Images and diffraction patterns were recorded with a Gatan UltraScan 1000XP CCD camera. EDX spectra were obtained with an Oxford Instruments X-max 80 mm² detector, and EELS data were collected with a Gatan Quantum 965 ER detector. Lamella thicknesses were determined via the log-ratio method using a mean free path of 173 nm for AlN for the given experimental parameters (semi-convergence angle of 10 mrad and semi-collection angle of 23 mrad), a density of 3.26 g/cm³; and the parametrization is given in Ref. 20. Elemental quantification was performed with the EELS plugin in Gatan's Digital Micrograph (version 3.42.3048.0) using power-law background subtraction and Hartree-Slater modeling of the core-loss ionization continuum.

The two different AlN regrowth samples were examined using AFM. Depending on the applied bake conditions inside the MOVPE system prior to growth, the AlN surface morphology changed drastically. In the case of the H₂ bake (sample A), AFM scans reveal a surface that is significantly impaired by hexagonal hillocks (defect type 1, T1) and faceted macrosteps (defect type 2, T2). The defect types have a density of 1.5 × 10⁶ cm⁻² (T1) and 0.8 × 10⁶ cm⁻² (T2) extrapolated from ∼1600 μm² wafer area, respectively. In between the defects of sample A, the RMS roughness is <0.4 nm. In comparison, the harsher NH₃-containing bake (sample B) leads to a clean, defect-free surface with zero measured defects on ∼6400 μm², which would result in a surface defect density of better than 1.5 × 10⁴ cm⁻². Sample B has a similar RMS roughness of < 0.4 nm. A NH₃-containing bake was also found to be favorable by Shojiki et al., due to reorganization of the HTA AlN surface morphology without pit formation as it was observed using a pure H₂ bake at 1200 °C.²¹ Since the bake conditions have a strong influence on the outcome of the surface, we want to postulate a surface contamination on top of the HTA AlN template that is a consequence of the annealing process but can be removed by appropriate bake conditions. An exemplary overview scan for both samples is shown in Figs. 1(a)/1(b). More detailed scans on sample A reveal the nature of the two observed T1 defects, as it is shown in Figs. 2(a)/2(b): The scans exhibit an even number of threading dislocations with screw component and seemingly oppositely oriented Burgers-vectors for each defect. The AFM results alone do not allow differentiation between pure screw- or mixed-type dislocations with screw components. Since the pairs of oppositely oriented spirals did not annihilate during MOVPE regrowth, it is likely that the dislocations are actually mixed-type and the sum of their Burgers-vectors does not allow annihilation. The resulting spirals around the dislocation core wind in opposite directions. If the spirals' distance is small, the collision of the two growth fronts of each spiral leads to the emission of concentric circular surface steps.²² With increasing distance of the spiral pair, the emission rate of concentric circular surface steps is not high enough to excel the superimposed step-flow growth of the AlN; thus, only step-pinning and a resulting formation of facetted hexagonal macrosteps occurs (cf. Fig. 3).²² The hillock in Fig. 2(a) does not yet show a fully developed facet on the upper part of the scan, but the onset of bunching. In comparison, the hillock in Fig. 2(b) reveals already emerged facets in all directions. Therefore, depending on the distance and the amount of spirals and the off-cut of the used sapphire wafer (thus, the difference between growth velocity of the concentric circles and the superimposed step-flow growth), either a hillock defect like T1 or a facetted macrostep like T2 emerges, as shown in Fig. 3.²² The AFM scan shown in Fig. 3 exhibits the same type of intertwining spirals, but the core of the spirals is not visible, since the formed faceted macrostep is pinned at each dislocation core. This effect is well-described by Uwaha in Ref. 22. On top of the emerging plateau, a new nucleation site is visible, leading to enlargement of the macrostep by Volmer–Weber growth. Consequently, we expect the origin of T1 and T2 defects to be the same, which is counterintuitive at first due to their different appearance in the AFM scans.

As suggested above, a surface contamination as the source of these defects due to the annealing process is likely. These contaminations were removed by using the harsher NH₃/H₂ bake. One possible contamination is aluminum oxynitride (AlON). The strong affinity between oxygen and aluminum or AlN is well-known. AlN tends to fully oxidize to Al₂O₃ in the presence of oxygen already at temperatures below 1200 °C.²³ There are also reports in the literature about the strong interdiffusion of oxygen into the AlN lattice during HTA.²⁴ Hagedorn et al. published TEM and HRTEM measurements of HTA AlN, which revealed either amorphous AlON defects originating from the AlN/sapphire interface leading to the so-called volcano defects or epitaxial (111)-oriented γ-AlON islands at the AlN/air interface after HTA.^25,26

The nature of the interfacial defects in our samples is also revealed by TEM results. As shown in the WBDF images in Fig. 4(a), several dislocations are found underneath all three investigated hillocks, which converge at 80–100 nm wide platelets at the regrowth interface. According to the $g →⋅ b →≠0$ visibility criterion, mainly edge-type dislocations are observed; however, mixed-type threading dislocations are found in the case of hillock 2 as well, which may explain why no annihilation of the oppositely twisting spiral pairs is observed. Please note that hillock 1 was investigated using the [1–100] zone axis and hillocks 2 as well as 3 using the [11–20] zone axis, respectively, leading to different in-plane two-beam conditions. Furthermore, the bright surface layer is caused by the platinum deposition during FIB preparation, which was performed at 2 kV accelerating voltage for hillock 1 and 5 kV for hillocks 2 and 3 explaining the observed layer thickness differences. The EDX spectra presented in Fig. 4(b) clearly show the presence of oxygen at the regrowth interface in all three cases proving the hypothesis of surface contamination after annealing as a cause for the defect formation. Interestingly, the highest oxygen-to-nitrogen signal ratio is observed for hillock 2, indicating that it is centered best in the TEM lamella, which might explain the absence of dislocations with screw component reaching the surface of hillocks 1 and 3 in the WBDF images shown in Fig. 4(a).

To investigate the phase of the platelets, hillock 3 was investigated after further thinning of the lamella down to approximately 90 nm thickness. Figure 5(a) shows an SAED pattern with the AlN matrix tilted to the [1–100] zone axis. The additional principal Bragg reflections due to the platelet are marked as $g precipitate 1$ and $g precipitate 2$ [green arrows in Fig. 5(a)] and all remaining spots are consistent with linear combinations of these and the AlN Bragg vectors [blue and red markings in Fig. 5(a)], i.e., with dynamical diffraction conditions. From the SAED pattern, we find $| g precipitate 1|=0.554(2)⋅ 11 − 20 AlN$ and $| g precipitate 2 = 0.543 ( 2 ) ⋅ | 0002 AlN$⁠. These values are very consistent with (111)-oriented epitaxial $γ$-AlON with [2–20]_AlON ǁ [11–20]_AlN in-plane relation. In fact, using the bulk lattice constants, i.e., omitting strain effects, one expects $| 2 − 20 AlON , bulk|=0.553⋅ 11 − 20 AlN , bulk$ and $| 111 AlON , bulk|=0.542⋅ 0002 AlN , bulk$ according to Refs. 27 and 28. This finding is further supported by the oxygen anion fraction obtained by EELS as presented in Figs. 5(b)/5(c). In fact, the plateau in Fig. 5(c) at around 84 at. % suggests the chemical composition Al₂₃O₂₇N₅, which is also referred to as γ-AlON in the constant anion model and consistent with ab initio calculations given in Ref. 29. Certainly, the presented combination of EELS and SAED data is a strong for Al₂₃O₂₇N₅. However, since an underestimated oxygen fraction due to remaining contributions of the AlN matrix in the lamella cannot be fully ruled out, constant cation γ-AlON (Al₂₄O₃₀N₄)²⁹ and even γ-alumina (Al₂O₃)³⁰ are possible phases as well, while the nitrogen-rich phases of ideal γ-AlON (Al₂₄O₂₄N₈), 27R (Al₉O₃N₇), 21R (Al₇O₃N₅), and 12H (Al₆O₃N₄) discussed in Refs. 29 and 31 can be excluded.

The TEM results emphasize the epitaxial nature of the interfacial defect below the measured hillocks, and thus, we want to attribute the hillock formation during AlN MOVPE regrowth to nucleation on top of epitaxial γ-AlON islands. Due to the measured lattice mismatch between γ-AlON and re-nucleated AlN, the bunched edge-type dislocations beneath the hillocks may be identified as threading arms originating from misfit dislocations forming at the γ-AlON/AlN interface. The γ-AlON islands evolve during the HTA process and are triggered by the high oxygen concentration inside the layer. As discussed by Cancellara et al., oxygen diffusion into the AlN and simultaneous formation of compensating vacancy complexes enable the dislocation climb during HTA.²⁴ At the same time, the oxygen impurity concentration inside the HTA AlN almost reaches atomic percentages, presumably leading to segregation of AlON phases at the top surface of the HTA AlN template. The segregation might be enhanced by a change in oxygen solubility during temperature ramps of the HTA process. Since hillocks 1 and 3 do not show dislocations below the γ-AlON platelets, it seems like the γ-AlON islands nucleate or segregate randomly during HTA on the AlN top surface and their appearance is not connected to an enhanced oxygen diffusivity (and thus, locally increased concentration) by means of pipe diffusion along a dislocation core [cf. Fig. 4(a)]. However, since the EDX results in Fig. 4(b) indicate that hillock 2 is centered best inside the lamella, threading dislocations below the γ-AlON platelets of hillocks 1 and 3 may be cut out of the lamella. Finally, a pipe diffusion mechanism cannot be ruled out completely.

The influence of different surface pretreatments prior to the AlN regrowth on HTA AlN templates was discussed. An NH₃-containing harsh bake is suitable to remove surface contaminations, stemming from the HTA process, and thereby reduce the density of hillock defects in MOVPE grown AlN. Furthermore, the nature of this contamination was studied by using different TEM-related measurement techniques. The phase of the segregation was found to be most likely Al₂₃O₂₇N₅, which is also referred to as γ-AlON in the constant anion model. These findings are consistent with earlier reports and thermodynamical considerations of the Al₂O₃–AlN system. Finally, an appropriate pre-conditioning of HTA AlN templates allows to fabricate atomically smooth (RMS roughness of 0.4 nm), hillock-free AlN, which is ready for subsequent AlGaN epitaxy.

The help of Volker Radisch during TEM sample preparation and the equipment of the “Collaborative Laboratory and User Facility for Electron Microscopy” (CLUE, Göttingen) is highly acknowledged.

This work was partly funded by Deutsche Forschungsgemeinschaft (DFG) in the framework of the SPP 2312 (Energieeffiziente Leistungselektronik “GaNius”), Project No. 462737320 (Aluminium Nitrid für die vertikale Leistungselektronik).