Impact of Si doping on dislocation behavior in MOVPE-grown AlN on high-temperature annealed AlN buffer layers

AlGaN-based light-emitting diodes (LEDs) operating in the ultraviolet (UV) wavelength range require transparent AlN template layers with low threading dislocation density (TDD).^1,2 One of the most efficient possibilities to produce such templates is to anneal defect-rich AlN layers at high temperatures in an N₂ ambient, the so-called high-temperature annealing (HTA) technique.^3,4 HTA-AlN layers on sapphire show not only strongly reduced TDDs⁵ but also a decrease of the in-plane lattice constant.^4,6 As a result, subsequent homoepitaxial metalorganic vapor phase epitaxy (MOVPE) growth on HTA-AlN leads to compressive strain and an increased wafer curvature.⁷ This becomes even more critical when AlGaN-based LEDs are grown on the HTA-AlN templates with low TDDs, but high compressive strain.⁸ In this case, strain relaxation takes place by the formation of new defects drastically degrading the quality of the layers and making them inappropriate for use in LEDs. Huang et al.⁹ have shown that compressive strain can be released by the introduction of Si-doping into the MOVPE-grown AlN layers. In particular, a Si concentration of 2 × 10¹⁹ cm⁻³ led to the relaxation of compressive strain during MOVPE growth of AlN on HTA-AlN templates while the low TDD could be preserved. The authors suggested that the strain relaxation was achieved by surface-mediated dislocation climb.^9,10

Our study visualizes and compares for the first time the defect behavior in the MOVPE-AlN layers grown with and without intentional Si-doping on HTA-AlN/sapphire layers to confirm these observations. In particular, scanning transmission electron microscopy (STEM) was applied for this analysis. We will show that homoepitaxial AlN growth on HTA-AlN results in strain relaxation by the formation of dislocation half-loops, most probably originating from the layer surface and moving down to the homoepitaxial MOVPE-AlN/HTA-AlN interface during the increase of the MOVPE-AlN layer thickness. Similar half-loops of an irregular shape were observed in thick AlN layers obtained by epitaxial lateral overgrowth (ELO)¹¹ as well as in AlGaN layers grown on AlN with a relatively low TDD.⁸ For subsequent AlGaN growth, this defect formation mechanism becomes disadvantageous, especially when the defect formation takes place close to the active zone of devices. In contrast, we will prove that Si-doping leads to a stronger threading dislocation (TD) line inclination in MOVPE-AlN layers, which gives an alternative route for strain relaxation. This allows us to avoid the formation of unfavorable dislocation half-loops.

To obtain the HTA-AlN templates, 350 nm thick AlN layers were sputtered onto sapphire substrates in an Evatec Clusterline™ 200. C-plane sapphire with a nominal off-cut of 0.2° toward the m-plane was used as substrate material. Sputter process parameters as described in Ref. 12 were applied to deposit AlN.

Subsequently, the HTA process was performed in a sintering oven in a N₂ atmosphere at a total pressure of 1000 mbar and a temperature of 1680 °C for 2 h in a face-to-face configuration of the samples. Heating was performed with a rate of 5.5 °C/min. The temperature was controlled by a thermocouple close to the heater.

After that, two 1.8 μm thick AlN layers were grown on identical HTA-AlN templates using an AIX2400G3HT MOVPE planetary reactor with standard precursors (TMAl, NH₃, Si₂H₆). The following growth parameters were applied: growth pressure of 50 mbar, growth temperature of 1158 °C, input group-V/III ratio of 30, and a NH₃ flow of 0.011 mol/min. One of the AlN layers was not intentionally doped (in the following referred to as AlN:uid), whereas the other one was doped with Si by introduction of a Si₂H₆ flow of 0.13 μmol/min. The Si concentration in the doped AlN layer (in the following referred to as AlN:Si) amounts to about 2 × 10¹⁹ cm⁻³ [measured by secondary ion mass spectrometry (SIMS)], while in the undoped sample, the Si background is below 10¹⁶ cm⁻³ (see Fig. S1 in the supplementary material).

In an additional experiment, a 100 nm thick AlN:uid layer was grown by MOVPE on 350 nm thick HTA-AlN at the same growth conditions. This sample was grown for comparison of its defect structure with that of the 1.8 μm thick AlN:uid layer. The 405 nm reflectance from the sample surface and the wafer curvature during MOVPE-AlN growth were monitored by a LayTec EpiCurveTT in situ metrology system.

The defect distribution in the layers was analyzed in cross-sectional (in both [1−100] and [11−20] zone axes) as well as in plan-view by STEM using an annular dark-field detector (ADF STEM). Additionally, x-ray rocking curve (XRC) measurements were carried out to compare the layer quality and strain state. AFM analysis was used to reveal the surface morphology of the layers.

Figure 1 shows exemplary cross-sectional ADF STEM images of about 1.8 μm thick AlN:uid and AlN:Si layers, which allow for comparison of the dislocation distribution in these layers. The relaxation process in the AlN:uid layer takes place by two mechanisms: (i) a slight inclination of TD lines as they propagate from the HTA-AlN into the MOVPE-grown layer [see the TD lines marked by the black arrows in Figs. 1(a) and 1(b)]; (ii) accumulation of dislocation lines of an irregular shape (including horizontal line segments) at the MOVPE-AlN/HTA-AlN interface [see the dislocation half-loops marked by the white arrows in Figs. 1(a) and 1(b)]. In case of the AlN:Si layer, such irregular dislocation lines were not observed [Figs. 1(c) and 1(d)]. The relaxation process takes place only by inclination of TD lines.

Looking at the shape of the irregular dislocations [Figs. 1(a) and 2(a)], we suggest that in AlN:uid dislocation half-loops with horizontal line segments nucleate most probably at the layer surface and then propagate from the surface down to the homoepitaxial MOVPE-AlN/HTA-AlN interface. This relaxation mechanism is well known for a number of materials,^13–15 especially for layer systems with low defect densities. In the classical model, after nucleation at the layer surface, the half-loops start expanding under the influence of the film stress and glide deep into the layer leaving two threading dislocations behind [see the sketch in Fig. 2(b)]. However, to our knowledge, this relaxation mechanism was rarely observed for group-III nitrides so far. This was connected to the lack of shear stresses acting in potential glide planes of the wurtzite structure.¹⁶ Nevertheless, dislocation gliding has come into focus recently as a possible cause of dislocation motion under temperature-induced stress during HTA thermal cycle annealing of AlN.¹⁷ Furthermore, strain relaxation through dislocation half-loop formation has been previously observed for InGaN layers¹⁸ as well as AlGaN layers grown on GaN.¹⁹ This process involved glide of mixed-type dislocations with the Burgers vector 1/3⟨11−23⟩ on pyramidal {11−22} planes and resulted in the formation of misfit dislocation arrays. Similar to the classical model sketched in Fig. 2(b), the authors observed half-loops with straight TD lines propagating through the whole layer and connected by the horizontal misfit segments in the heterointerfacial plane.

In contrast to the classical model, in our experiment, we do not observe any drastic increase in TDD of the AlN:uid layers due to dislocation half-loop nucleation compared to the TDD of the underlying HTA-AlN template. This can be explained by the fact that in contrast to classical semiconductors, where the initial TDD is extremely low and the half-loop formation and propagation take place undisturbed, group-III nitride layers mostly exhibit rather high initial TDDs. Consequently, the fact that the TDD is not drastically increased may be attributed to a strong interaction of pre-existing TDs (stemming from HTA-AlN) and the newly formed half-loop defects, which can result in a dynamic process of dislocation line redirection and annihilation of some dislocation segments. Furthermore, for high half-loop densities, an additional strong interaction between the individual half-loops can take place as sketched in Fig. 2(c). Such interaction processes explain the irregular shape of the defects observed in our experiment.

Diffraction contrast imaging was used to prove if the half-loop defects can relieve strain, i.e., if their Burgers vector has an in-plane component. Figure 3 shows weak-beam diffraction contrast images obtained at different two-beam conditions. The images in Fig. 3 show one and the same sample area tilted out of the [11−20] zone axis to the corresponding two-beam conditions. The defects with the horizontal line segments are visible when 1−100 and 1−210 reflections are strongly excited for the imaging [Figs. 3(a)–3(c)]. For the 0002 reflection, no defects are visible [Fig. 3(b)]. Bright dots visible within the HTA-AlN layer in Fig. 3(b) correspond to small voids, which are formed during the HTA process.²⁰ This analysis confirms that the Burgers vector of the dislocation half-loops $b→$ lies in the (0001) growth plane. Consequently, these dislocations are able to compensate the mismatch between the HTA-AlN and MOVPE-AlN. Note that in the zone axis images obtained in the [11−20] projection [Figs. 1(a) and 2(a)], the horizontal segments of the dislocation half-loops are arranged parallel to the [1−100] direction. This might lead to the wrong interpretation that the corresponding dislocation lines are parallel to the [1−100] direction. It should be kept in mind that these images represent only a projection of the defects on the (11−20) imaging plane, and the horizontal dislocation lines can be arranged along any other direction in the (0001) plane. Further plan-view analysis is required to clarify the exact arrangement of the dislocation lines, their exact Burgers vector, and the geometry of their movement plane.

According to our knowledge, such a relaxation mechanism has not yet been observed for homoepitaxial GaN. However, we have observed similar dislocation loops of irregular shape in AlN layers with low TDDs, for example, in ELO-AlN,¹¹ or in AlN-based layer systems with increasing strain, e.g., in AlN/AlGaN layers.⁸ Obviously, when the number of threading dislocations (TDs) decreases, their contribution to strain relaxation (e.g., through TD line inclination) becomes strongly limited and alternative relaxation routes, like the formation of dislocation half-loops, become possible. We suggest that high growth temperatures applied for the AlN growth promote the movement of the dislocation half-loops by point defect diffusion (i.e., climb processes take place). According to our STEM studies of AlGaN/AlN layer systems, for TDDs ≤ 10⁹ cm⁻², we often observe electron beam induced movement of the dislocations half-loops from the top layers down to the bottom (see Fig. S2 in the supplementary material). In this case, the electron beam of high energy produces point defects in the analyzed material that initiates the climb process. High temperatures can act as a similar trigger. This observation allows us to assume that also in the observed homoepitaxial AlN layers, the formation of irregular dislocation lines takes place at the top surface of the layers, and they subsequently move down to the strained MOVPE-AlN/HTA-AlN interface. Obviously, point defect diffusion is required for the dislocation climb process. It is well known that high-temperature annealed AlN layers exhibit high densities of point defects causing strong defect emission at a wavelength of about 400 nm in their luminescence spectra. MOVPE-AlN layers grown on HTA-AlN also feature an increased defect luminescence compared to the MOVPE-AlN grown without any HTA template underneath (Fig. S3 in the supplementary material). The defect luminescence may be attributed to a number of defects, like Al or N vacancies and/or vacancy-oxygen defect complexes.^21,22 Vacancy diffusion through dislocation cores was also considered to be predominant mechanism facilitating threading dislocation climb during the HTA process.²³ The fact that the half-loops stop at the MOVPE/HTA interface despite the generally higher point defect density in the HTA-AlN layers suggests that the dislocation movement is energetically favorable only as long as there is a sufficient strain in the system. Thus, they predominantly stop when the lattice constant difference vanishes. Note that although the majority of these defects accumulates at the MOVPE-AlN/HTA-AlN interface, some of them are “frozen” within the MOVPE-grown layer [Figs. 1(a) and 2(a)] during the cooling down procedure at the end of the layer growth. These dislocations could not move anymore due to the temperature decrease and the corresponding decrease in the point defect mobility. Still to completely exclude the glide process as the half-loop movement mechanism, one needs plan-view investigations for analysis of the exact dislocation line directions as well as the movement planes. The formation of dislocation half-loops takes place during the growth at high temperatures as the layer thickness increases and the accommodated strain energy increases simultaneously. Thus, to prove the suggested relaxation mechanism, an AlN:uid layer with a smaller thickness of about 100 nm was additionally grown and analyzed in cross section. No irregular dislocation lines were observed in this sample (Fig. 4), whereas the TDs from the HTA-AlN propagate into the MOVPE-AlN:uid with a certain line inclination. Obviously, when the layer thickness is low, the slight inclination of the already existing TD lines stemming from the HTA-AlN is sufficient to partially relax the compressive strain occurring due to the difference in the lattice constants of MOVPE-AlN and HTA-AlN. However, with increasing layer thickness, the strain energy stored in the AlN film will increase, reaching a critical value and initiating relaxation through half-loop formation and their movement through the layer. When such defects are present close to the active region of an LED,¹² they will deteriorate optical device properties. Consequently, suppression of this relaxation mechanism becomes extremely important.

Huang et al. showed that the introduction of Si-doping allows overcoming the excessive compressive strain in the AlGaN layers grown on HTA-AlN.⁹ Our analysis of AlN:Si layers on HTA-AlN shows that also a suppression of half-loop formation becomes possible when using Si-doping. Figures 1(c) and 1(d) show that in the AlN:Si layer, the inclination of pre-existing TD lines is stronger compared to that observed in AlN:uid [Figs. 1(a) and 1(b)]. TD line inclination in Si-doped layers, especially in GaN, has been widely discussed in the literature.^24,25 It is believed that this dislocation movement takes place through surface-mediated dislocation climb.¹⁰ Weinrich et al. showed that in GaN:Si layers, the TD line inclination shortens the extra half-planes of edge and mixed-type dislocations, resulting in an increase of tensile strain in the layers.²⁶ Similarly, the larger inclination of the TD lines in AlN:Si on HTA-AlN introduces a higher tensile strain component reducing the compressive strain arising due to the difference in lattice constants of the MOVPE-grown and HTA-treated materials. Another important consequence of the strain reduction is that there is no additional formation of any other defects in the 1.8 μm thick AlN:Si layers on HTA-AlN. Thus, the formation and propagation of dislocation half-loops can be suppressed using this approach.

According to plan-view ADF STEM analysis of over 700 TDs in the AlN:Si layers, about 98% of them incline to one of the crystallographically equivalent ⟨1−100⟩ AlN directions [Figs. 5(a) and 5(b)]. Consequently, the true inclination angle of the TD lines can be measured in the [11−20] cross-section, since in the [1−100] cross-section, a part of TDs will appear vertical due to the inclination into and out of the viewing direction, whereas inclination into the other four ⟨1−100⟩ directions will only show projected angles. In contrast, when viewing dislocation inclination angles in the [11−20] zone axis, the true inclination angle can be measured for the dislocations inclined into the [1−100] and [−1100] directions [see the blue-colored directions and TD lines in Figs. 5(b) and 5(c)], whereas inclination into the other four equivalent directions will be viewed in projection [see red-colored directions and TD lines in Figs. 5(b) and 5(c)]. Figure 6 shows inclination angles of about 400 TD lines measured in the [11−20] cross-section for both AlN:uid and AlN:Si layers. The positive and negative inclination angles indicate the opposite inclination directions. Note that according to the sketch in Fig. 5(c), there should be two groups of inclination angles for each sample: (1) the group of maximum inclination angles, which correspond to the true inclination angle α_true into the [1−100] and [−1100] directions; (2) the group of smaller inclination angles corresponding to the projected inclination angle α_proj. When all six crystallographically equivalent inclination directions ⟨1−100⟩ are represented uniformly, then the number of dislocations with α_proj should be two times larger than the number of dislocations with the maximum inclination angle α_true.

The results in Fig. 6 show that for AlN:uid layers, the majority of inclination angles lies in the region between 5° and 10°, whereas the inclination angles in AlN:Si are between 20° and 30°. Using the approach of Romanov and Speck,²⁷ the total plastic relaxation by TD line inclination at the surface of the 1.8 μm thick AlN layer was calculated for both samples according to the formula

where b is the length of the Burgers vector of edge dislocations in AlN (b = 0.3111 nm), h is the layer thickness, ρ_TDD is the threading dislocation density, and α is the inclination angle of TD lines. This calculation is based on the fact that the projection of inclined TD lines acts as a misfit component enabling strain relaxation. According to this model, a slight inclination of TD lines in the undoped AlN:uid layer leads to a strain relaxation of 0.06%. In contrast, the inclination of about 29° results in a relaxation of 0.25%. For comparison, Table I summarizes the data obtained on the two samples and compares these relaxation values with those measured by high resolution x-ray diffraction (HRXRD)²⁸ as well as by in situ sample curvature measurements during MOVPE-AlN growth²⁹ (Fig. S6 in the supplementary material). The total TDD in the layers was calculated from XRC measurements according to Ref. 30 as well as from plan-view ADF STEM images. Compared to XRC results, the STEM analysis shows slightly higher values (obtained from a total specimen area of 45 μm²).

In the following, we will shortly discuss the main results demonstrated in Fig. 6. In the AlN:uid layer, the majority of dislocations exhibit negative inclination angles [Fig. 6(a), also see Fig. S4 in the supplementary material], which means that they incline preferentially to one direction with respect to the vertical dislocation line. This finding can be attributed to the off-cut of the sapphire substrate, which might influence the inclination direction and lead to the non-uniform distribution of inclination angles between initially crystallographically equivalent directions. In contrast, in the AlN:Si layer, the distribution of inclination directions becomes more uniform. However, the inclination angles in the opposite directions are different [compare the positive and negative α_true inclination angle values in Fig. 6(b)]. This suggests that the contribution of TD inclination into the relaxation process is not equivalent in the six ⟨1−100⟩ directions. This may be caused not only by the initial off-cut direction but also by the growth and annealing conditions of the HTA-AlN template as well as the spatial TD distribution and TD clustering.

The black squares in Fig. 6(b) represent all inclination angles measured in the AlN:Si layer, whereas the violet circles in Fig. 6(a) show all inclination angles observed in AlN:uid. According to our previous explanations [Figs. 5(b) and 5(c)], the inclination angles in AlN:Si can be divided into two groups showing projected and true inclination angles [Fig. 6(b)] according to the inclination into the six ⟨1−100⟩ directions. In contrast, in the undoped AlN:uid layer, the inclination angles do not show this clear separation into two groups [Fig. 6(a)]. This allows us to conclude that in our experiment, the inclination model presented in Figs. 5(b) and 5(c) as well as the calculations based on Romanov and Speck²⁷ can be applied to estimate the strain relaxation by TD inclination in the doped AlN:Si layers but cannot be applied for the undoped AlN:uid layer. The reason for that may be the simultaneously present inclination into directions different from the ⟨1−100⟩ AlN as well as the predominant influence of the substrate off-cut onto the inclination direction. For example, using plan-view TEM analysis, Weinrich et al. showed that in undoped GaN:uid layers, dislocation line inclination takes place predominantly into a-directions, whereas in Si-doped GaN:Si layers, the inclination directions are close to m-directions.²⁶ The similar situation might appear in the AlN:uid layers used in our experiment. In the case of the more or less random as well as inhomogeneous TD inclination directions, the simple geometric model of Romanov and Speck,²⁷ which considers the formation of periodically arranged and homogeneously distributed “net” of inclined TD line segments, is suggested to overestimate the impact of the TD inclination angle on the strain relaxation. Unfortunately, our numerous attempts to prepare plan-view samples of MOVPE-grown AlN:uid layers on HTA-AlN failed because the samples fractionized during ion milling most probably due to high stresses and areas with a high defect concentration (e.g., at MOVPE-AlN/HTA-AlN interface) where crack nucleation may appear preferentially. In contrast, plan-view preparation of AlN:Si layers was successful, which allowed us to analyze the TD inclination directions in this layer precisely [Fig. 5(a)].

Table I shows summarized data on strain relaxation values calculated using different experimental methods: ɛ_XRD calculated from high resolution omega/2theta scans of the 004 reflection,²⁸ ɛ_plastic calculated according to the model of TD inclination of Romanov and Speck,²⁷ and ɛ_{in situ} calculated from in situ curvature measurements²⁹ (see Fig. S6 in the supplementary material). The in situ measurement of the sample curvature at the initial stage of growth shows an initial mismatch between the HTA-AlN and MOVPE-AlN:uid layers of 0.08% (see Fig. S6 in the supplementary material), which is only slightly higher than the value of plastic relaxation of 0.06% calculated from the TD inclination angles in AlN:uid. However, STEM analysis shows the formation of a high number of dislocation half-loops in this sample, suggesting that only the observed TD line inclination alone is insufficient to compensate this mismatch. These data support the above suggestion based on the untypical distribution of the inclination angles in AlN:uid (Fig. 6) that the model of Romanov and Speck²⁷ cannot be applied for the AlN:uid layer when the TD inclination directions have not been analyzed precisely.

The reason for the proposed different TD behavior and the possible differences in the inclination directions may lay in the different driving forces for the inclination in the AlN:uid and AlN:Si layers. We suggest that in the undoped AlN:uid layer, the TD line inclination takes place initially as a strain relaxation mechanism, which is however limited by dislocation mobility at a certain temperature and at a certain point defect concentration. In this case, the inclination direction can be strongly influenced by local TD distribution and local strain and can deviate from the ⟨1−100⟩ directions. Additionally, in the undoped AlN:uid layer, there is a stronger influence of the substrate off-cut onto the TD inclination direction (see Fig. S4 in the supplementary material). In contrast, in Si-doped layers, the tensile stress increases even in the initially tensilely strained layers (e.g., in GaN²⁵), which clearly shows that the TD line inclination in the Si-doped layers appears independent of the strain state of the layers and should not necessarily lead to strain relaxation. Moram et al.³¹ suggested that there might be the formation of Si impurity atmospheres at dislocation cores, which limits their movement during the growth resulting in a higher tensile stress compared to undoped GaN. Using microphotoluminescence, Forghani et al.²⁴ measured an asymmetry in the strain dipoles around the TD cores, suggesting that Si atoms accumulate at the compressively stressed side of an edge dislocation, reducing the compressive strain in this region and causing the formation of strain monopoles. Weinrich et al.²⁶ proved that in GaN:Si, the TD line inclination takes place with the shortening of the extra half-plane of edge and mixed TDs, which leads to the generation of tensile strain. There are two inclination models supporting this finding: an SiN_x masking of dislocation cores³² and an enhanced formation of Ga vacancies due to the Fermi level shift toward the conduction band of n-doped material resulting in the vacancies incorporation at the TD cores.³³ We suggest that similar to the GaN:Si study of Weinrich et al.,²⁶ the observed stronger TD inclination in AlN:Si [Figs. 1(c) and 1(d)] and the associated increase of the tensile stress in this layer (see curvature measurements in Fig. S5 in the supplementary material) are not connected with the initial necessity of the system to relax the compressive strain due to the lattice constant difference between MOVPE-AlN and HTA-AlN but appears as an effect of Si-doping, which leads to the shortening of the TD extra half-planes.

It has been shown that homoepitaxy of MOVPE-AlN layers on HTA-treated sputtered AlN templates on sapphire substrates results in strong relaxation mostly by the formation of newly generated dislocation half-loops of irregular shape showing horizontal line segments. We propose that these defects are formed with increasing layer thickness at the surface of the growing layer, and then move down, probably by climb processes, until they finally accumulate at the HTA-AlN/MOVPE-AlN interface. The introduction of Si-doping during AlN homoepitaxy on HTA-AlN leads to a higher TD line inclination introducing a tensile strain component, which allows us to overcome the compressive strain generally present in these layers due to the difference in lattice constant of HTA-treated AlN and MOVPE-grown AlN. Furthermore, the formation of half-loops can be completely suppressed. This approach allows the fabrication of AlN templates showing a better crystalline quality and a modified strain state to further improve subsequent AlGaN epitaxy.

The supplementary material includes SIMS profiles showing Si distribution in the AlN:uid and AlN:Si layers (Fig. S1), a series of images showing dislocation loop movement and its interaction with threading dislocations under electron beam irradiation (Fig. S2), photoluminescence spectra showing defect luminescence in HTA-AlN and MOVPE-AlN layers (Fig. S3), a bar diagram visualizing distribution of inclination angles of threading dislocations in the analyzed layers (Fig. S4), in situ measured reflectance and wafer curvature of the Si-doped AlN layer (Fig. S5), and comparison of wafer curvature and strain values estimated from the curvature measurements for AlN:uid and AlN:Si layers (Fig. S6).

This work was partially supported by the German Federal Ministry of Education and Research (BMBF) through the consortium “Advanced UV for Life” under Project Contract No. 03ZZ0134B.

The authors have no conflicts of interest to disclose.

The data that support the findings of this study are available within the article and its supplementary material.