AlN formation by an Al/GaN substitution reaction

Aluminium nitride (AlN) is a promising semiconductor material for use as a substrate in high-power, high-frequency electronic and deep-ultraviolet optoelectronic devices. We study the feasibility of a novel AlN fabrication technique by using the Al/GaN substitution reaction method. The substitution method we propose here consists of an Al deposition process on a GaN substrate by a sputtering technique and heat treatment process. The substitution reaction (Al + GaN = AlN + Ga) is proceeded by heat treatment of the Al/GaN sample, which provides a low temperature, simple and easy process. C-axis-oriented AlN layers are formed at the Al/GaN interface after heat treatment of the Al/GaN samples at some conditions of 1473–1573 K for 0–3 h. A longer holding time leads to an increase in the thickness of the AlN layer. The growth rate of the AlN layer is controlled by the interdiffusion in the AlN layer.

To determine a novel technique for growing AlN to increase the possibility of further developments, here, we introduce an Al/GaN substitution reaction method. In this method, an Al layer deposited on a GaN substrate is used as a precursor, and AlN is obtained by the interfacial reaction between Al and GaN. The details are given in the next section. There are two purposes of this study. The first is to develop a novel technique to potentially grow a bulk AlN crystal by a substitution reaction of Al and GaN. The second is to fabricate an AlN film/GaN substrate structure that is also useful in some devices, for example, as a substrate in AlGaN/AlN/GaN highelectron-mobility transistors 28,29 or as an insulated gate in AlN/GaN heterostructure field-effect transistors 30,31 .
To the best of our knowledge, there have only been a few reports on fabricating AlN by heat treatment of Al on a GaN substrate. Luther et al. obtained a 2-3-nm-thick AlN layer by the heat treatment of Ti/Al and Pd/Al on GaN under Ar atmosphere at 600 °C for 15 and 30 h 32 . Moreover, Wang et al. demonstrated the advantage of Al buffer layers to obtain high-quality and stress-free GaN epitaxial films on Si substrates. In their study, an Al buffer layer with a thickness of 20-40 nm was grown on a Si substrate and, subsequently, a 250-nm-thick GaN film was grown on the Al buffer layer at 1123 K by PLD. The AlGaN peak was observed in the X-ray diffractometer (XRD) 2θ − ω scan profiles. However, AlN was not obtained 33 .
This paper focuses on the investigation of AlN fabrication by the substitution method at relatively low temperatures (around 1573 K). In this method, GaN was used as a starting material because it has the same wurtzite structure as AlN and has close lattice constants with those of AlN 34,35 . Currently, GaN is commercially available from several companies and institutions. There has been a tremendous amount of research on GaN. It has already been grown and investigated by the HVPE method 36,37 , ammonothermal method 38,39 , Na flux method [40][41][42][43] and high-pressure solution growth method as reported in a review by Amano 44 . Although the GaN substrate is still expensive, large size GaN substrates will hopefully be available at a reasonable cost in the near future based on the ongoing intensive studies 45 . The substitution method we propose here consists of only Al deposition on a GaN substrate by a sputtering technique and heat treatment process, which provides some benefits such as much lower growth temperature, compared with the sublimation method (2473 K), and a simple and easy process. This study introduces the details of the Al/GaN substitution method including some fundamental results and discussion on the growth mechanism.
principle of Al/Gan substitution method Figure 1 shows a schematic diagram of the Al/GaN substitution method. This process starts with an Al layer deposited on a GaN bulk crystal, as shown in Fig. 1 (left). AlN is thermodynamically more stable than GaN. Ga atoms in the GaN can be substituted by Al atoms during heating the sample in an inert gas atmosphere. Thus, an AlN layer forms at the Al/GaN interface by the substitution reaction (1). This process proceeds with time by atomic diffusion through the AlN layer. The driving force of the reaction and mass transport can be controlled by selecting the temperature.
From the thermodynamic point of view, the standard Gibbs energy of reaction (1)  Results thermal analysis. Prior to the Al/GaN substitution reaction experiment, the thermal stabilities of metallic Al, single crystalline GaN and an Al layer deposited on GaN (Al/GaN) were studied by thermogravimetry-differential scanning calorimetry (TG-DSC). Figure 2 shows the TG-DSC profiles of these materials in an Ar atmosphere. Figure 2a shows that the profile of metallic Al was almost parallel to the profile of an empty cell (as a baseline). This implied that vaporization of Al is not significant up to 1673 K. However, the GaN profile started to exhibit a mass reduction from its baseline at 1473 K (as shown by the red dashed line in Fig. 2b). This implied that the GaN started to dissociate into Ga and nitrogen gas at 1473 K according the following reaction:  Fig. 2c), 150 K lower than that for the GaN dissociation. If the substitution reaction (1) takes place alone, no mass reduction would occur. However, the GaN dissociation reaction (3) can take place together with reaction (1) at a lower temperature, because the Ga activity is greatly reduced by mixing Ga with Al.
Al/Gan substitution reaction. Cross-sectional SEM image. Figure 3 shows the cross-sectional scanning electron microscopy (SEM) image of the AlN layer formed on the GaN substrate by the substitution reaction: A 7.6-μm-thick Al layer deposited on Ga-polar GaN substrate was annealed in an Ar atmosphere for 3 h at 1573 K. The AlN layer had the same crystal orientation with that of the GaN substrate, which will be explained by the in-plane crystallographic relationship described later. Figure 4 shows the bird's-eye view SEM-EDS images of the Al on GaN substrate heat treated at 1573 K for 3 h. A metallic Ga droplet was observed on the Al/GaN sample. The Ga was formed by the substitution reaction (1) at the Al/GaN interface, and somehow it moved up to the surface of the Al/GaN sample. This is evidence of Ga formation by the substitution reaction. The AlN layer was hardly observed at this scale because the AlN thickness was only around 1.5 µm.

Bird's-eye view SEM-EDS images.
Crystalline orientation. Figure 5a shows the XRD 2θ − ω scan profiles of the bare GaN substrate and Al/GaN samples with and without heat-treatment at 1573 K for 3 h. The profiles show that a c-axis-oriented AlN layer was obtained after heat treatment of the Al/GaN sample. The peak position of the AlN (0002) at a 2θ value of 36.0° reflections is also shown by the dashed line in Fig. 5a as a reference. AlN (0002) and GaN (0002) peaks were obtained for the heat-treated Al/GaN sample. Figure 5b shows the ϕ-scans of AlN {10-12} and GaN {10-12} for the heat-treated Al/GaN substrate. Both AlN {10-12} and GaN {10-12} exhibited 6 peaks, and they agreed with each other. From Fig. 5a,b, the in-plane crystallographic relationship between the AlN layer and GaN substrate is:   www.nature.com/scientificreports/ Ga peaks were not observed, which implied that the amount of formed Ga was too small to be detected by XRD. A certain amount of GaN may dissolve in AlN forming an Al x Ga 1−x N solution, which would cause the blunt peak of AlN (0002). The formation of Al x Ga 1−x N will be described in the cross-sectional transmission electron microscope (TEM) observation section.
Lattice constant. Figure 6 shows the lattice constant of the AlN layers obtained after heat treatment of Al/GaN samples with various heat treatment temperatures and holding times. The lattice constants c of the AlN layers are almost the same with that of bulk AlN, but the lattice constants a are slightly larger than that of bulk AlN, and they approach that of bulk AlN with holding time.
Residual stress. Figure 7 shows the residual stresses of the AlN layers evaluated from the lattice constants presented in Fig. 6. The residual stresses along c-axis are almost zero. However, the residual tensile stresses along a-axes exist in the AlN layers and they approach zero with holding time. The thermal expansion coefficient along a-axis of GaN (6.2 × 10 -6 K −1 ) 48 is smaller than that of AlN (6.9 × 10 -6 K −1 ) 49 , which may generate tensile stress along a-axis in the AlN layer near the AlN/GaN interface during cooling.      Table 1.
Cross-sectional TEM observation. Figure 9a shows the cross-sectional TEM image of the AlN layer obtained after heat treatment of an Al/GaN sample at 1573 K for 3 h with an incident beam along GaN . Thus, the AlN layer and GaN substrate could be clearly seen and distinguished from each other. It was observed that   Figure 9b shows the EDX spectra at points a, b, c and d designated in Fig. 9a. Al and N peaks were detected at point a. However, Al and N peaks were detected in addition to a Ga peak at points b and c. Thus, Ga nonuniformly distributed in the AlN layer. The N peak intensities at points b, c, and d were lower than that at point a. This may imply that nitrogen atoms exited in the form of N 2 gas, which resulted in some voids. Thus, the formation of N 2 gas may cause the deviation of the TG curve of the Al/GaN sample from the baseline observed in Fig. 2c. The oxygen peak appearing at point b may have originated from contamination during the sputtering process of Al layer, and the oxygen was trapped in the AlN layer during the heat treatment. Figure 9c

Discussion
Growth model of the substitution reaction method. The growth model of the Al/GaN substitution reaction method is proposed as follows. Initially, Al reacts directly with GaN forming an AlN layer at the Al/ GaN interface. A subsequent reaction occurs through the mass diffusion in the AlN layer. Figure 10 shows the growth model of an AlN layer in the Al/AlN/GaN structure. There are two interfaces: the Al/AlN and AlN/GaN interfaces. At the Al/AlN interface, metallic Al is oxidized to be Al 3+ , then it diffuses in the AlN layer towards the AlN/GaN interface. At the AlN/GaN interface, the Al 3+ substitutionally reacts with GaN forming AlN and Ga 3+ . The Ga 3+ then diffuses towards the Al/AlN interface, and it is reduced to be metallic Ga by the reaction with three electrons. The growth model is summarized as follows, At the Al/AlN interface: The total reaction at the Al/AlN interface is given by.

At the AlN/GaN interface:
The overall reaction is given by the sum of reactions (6)- (9) The growth rate of AlN can be controlled by either interfacial reactions or interdiffusion. Assuming the interfacial reaction rates are much faster than interdiffusion, the growth rate is controlled by the diffusion.  Figure 11a shows the holding time dependence of the AlN thicknesses after heat treatment of the Al/GaN samples at 1473-1673 K. Here, the AlN thickness was measured from the cross-sectional SEM images. There was no AlN layer formed at zero holding time at 1473 and 1573 K. This implied that the substitution reaction proceeds at a slow rate and needs time to form the AlN below 1673 K. The temperature effect on the thickness of AlN is difficult to observe because GaN decomposition is more aggressive in high temperatures (1623 and 1673 K) and affects the AlN thickness. The longer the holding time leads to the thicker AlN film. Assuming the parabolic rate law, Fig. 11a was revised as Fig. 11b. The parabolic rate constant k p ' (μm 2 /h) is given by the following equation, here x (μm) is the AlN thickness and t (h) is the holding time. The Arrhenius plot is shown in Fig. 11c. The activation energy was calculated from the slope of the Arrhenius plot to be 121 ± 66 kJ/mol. The uncertainty is large owing to non-uniform AlN thickness after heat treatment of Al/GaN at temperatures of 1623 and 1673 K. This value has the same order with diffusion-controlled mechanism of some high-temperature oxidation studies of AlN. For instance, the activation energies for the oxidation of AlN obeying the parabolic rate law have been reported as 160 kJ/mol for the CaC 2 -doped AlN bulk at temperatures above 1523 K 50

Methods
Sample preparation. Al films were deposited on Ga-polar GaN substrates using magnetron pulsed DC sputtering (Shimadzu, HSR552). An Al target (High Purity Chemicals, diameter: 101.6 mm, purity: 99.999 mass%) was used. A pulsed DC power of 600 W (Advanced Energy, Pinnacle Plus + 10 kW) was used with a frequency of the pulse of 100 kHz and a duty cycle of 60%. The square-wave pulse type was chosen. The distance between the target and the GaN substrate was 60 mm. The Ar gas (99.9999% purity) equipped with an oxygen filter (Nanochem Purifilter; Matheson PF-25 Serial number P02241) was introduced into the chamber with a flow rate of 1.7 × 10 −4 L/s (10 sccm) and the total pressure was maintained at 0.6 Pa during sputtering. The oxygen filter removed NO x , SO x , H 2 S, < 0.1 ppb of H 2 O, O 2 , CO 2 , < 1 ppb of CO, and < 0.1 ppb of non-methane hydrocarbons from the argon gas. The growth temperature was fixed at 298 K. The sputtering time was 27 min that corresponded to 7.6-μm-thick Al on Ga-polar GaN substrates (Suzhou Nanowin Science and Technology Co. Ltd., size: 10 × 10.5 mm 2 , thickness: 350 ± 25 μm, crystal orientation: c-plane (0001), off-angle toward m-axis: 0.35° ± 0.15°, resistivity at 300 K: < 0.1Ω · cm, surface roughness of the front surface: Ra < 0.2 nm).  Substitution reaction experiment. Figure 12 shows a schematic diagram of the experimental setup for the Al/GaN substitution reaction experiment. The Al/GaN sample was placed upside-down. The samples were heated to the temperature range of 1473-1673 K in an Ar gas atmosphere with a flow rate of 30 mL/min at 293 K and cooled to room temperature after reaching each heat treatment temperature. From the TG-DSC result of Al/ GaN in Fig. 2c, the starting temperature of Al/GaN substitution reaction was 1323 K. However, from Fig. 11, the AlN thickness at 1473 K even for 3 h was very small. Therefore, we selected 1473 K as the lowest experimental temperature. On the other hand, the GaN decomposition became more aggressive with increasing temperature as shown in Fig. 2b. Thus, the maximum temperature was selected at 1673 K, but it was applied only for the short-duration experiments less than 1 h. The heating and cooling rates were held constant at 10 K/min. The total pressure was kept at 106.5 kPa in the chamber. The holding time was varied from 0 to 3 h for 1473 and 1573 K, but only 0 and 1 h for 1623 and 1673 K.
Sample characterization. The thickness, crystalline quality and cross-sectional image of the AlN layers formed between the Al layer and GaN substrate were evaluated around the middle part of the substrates. The interface morphology and the bird's-eye view of the AlN layers were examined using a SEM (JEOL JCM-5700). The AlN thickness was evaluated from these images. The 2θ − ω scan profile, where 2θ is the diffraction angle between the incident X-ray and the detector, and ω is the incident angle between the incident X-ray beam and the sample surface, and the X-ray rocking curve (XRC) profile were obtained using an XRD (Bruker, D8 Discover MR). An X-ray source of Cu-Kα radiation was selected. The XRD system was equipped with two Ge (400) crystals in its monochromator and a single-bounce Ge (220) in the analyser. The voltage and current in the X-ray cylinder during the XRD measurement were 40 kV and 40 mA. The 2θ − ω scan was conducted with a step size of 0.05°. The ϕ-scan was performed with a 0.1° step size where ϕ is a rotational axis normal to the sample surface. A TEM (Hitachi High Technologies, H-9000NAR), with an acceleration voltage of 200 kV and a magnification accuracy of ± 10%, was used to acquire the TEM images and electron diffraction patterns. An EDX equipped to the TEM system (Hitachi High-Technologies HD-2700) was used to carry out the elemental analysis at some points of the sample. The beam diameter was approximately 0.2 nm. Before the sample was measured by TEM and EDX, the remaining Al on the AlN was removed by wet etching using a 0.1 mol/L HCl aqueous solution at 353 K for 3 h, and then, the sample was pre-treated with a thinning process by focused ion beam (FIB) apparatus using the μ-sampling method.