Low-temperature atomic layer epitaxy of AlN ultrathin films by layer-by-layer, in-situ atomic layer annealing

Low-temperature epitaxial growth of AlN ultrathin films was realized by atomic layer deposition (ALD) together with the layer-by-layer, in-situ atomic layer annealing (ALA), instead of a high growth temperature which is needed in conventional epitaxial growth techniques. By applying the ALA with the Ar plasma treatment in each ALD cycle, the AlN thin film was converted dramatically from the amorphous phase to a single-crystalline epitaxial layer, at a low deposition temperature of 300 °C. The energy transferred from plasma not only provides the crystallization energy but also enhances the migration of adatoms and the removal of ligands, which significantly improve the crystallinity of the epitaxial layer. The X-ray diffraction reveals that the full width at half-maximum of the AlN (0002) rocking curve is only 144 arcsec in the AlN ultrathin epilayer with a thickness of only a few tens of nm. The high-resolution transmission electron microscopy also indicates the high-quality single-crystal hexagonal phase of the AlN epitaxial layer on the sapphire substrate. The result opens a window for further extension of the ALD applications from amorphous thin films to the high-quality low-temperature atomic layer epitaxy, which can be exploited in a variety of fields and applications in the near future.

Since ALD is a layer-by-layer process, the in-situ plasma treatment can be applied in each ALD cycle. Hence, the surface temperature of the sample might increase as a result of the plasma treatment, leading to the annealing effect during the layer-by-layer ALD growth. Accordingly, the in-situ, layer-by-layer plasma treatment in each ALD cycle is proposed in this paper and denoted terminologically as "atomic layer annealing" (ALA). The ALA treatment may enhance the adatom movement and migration on the surface, which is favorable for the improved crystallization of the deposited films. Another effect caused by ALA is to facilitate the removal of the ligands of the chemisorbed precursors. Consequently, the physical or chemical properties of film surface can be tailored by the ALA treatment, leading to the promotion of crystallization of the films at relatively low deposition temperatures.
Aluminum nitride (AlN) has attracted a lot of attention due to its distinguished material characteristics including wide direct band gap, high piezoelectric response, and good stability at high temperature. Hence AlN has been widely applied in ultraviolet light-emitting diodes 13 , resonators 14 , and sensors 15 . For example, the two-dimensional electron gas (2DEG) near the Al(Ga)N/GaN interface has been recently exploited in high-electron-mobility transistors (HEMTs) 16 . In general, high-quality AlN epilayers were conventionally grown by metal-organic chemical vapor deposition (MOCVD). However, a very high temperature (> 900 °C) is required to grow high-quality AlN epilayers using MOVCD. Besides, a large thickness up to ~600 nm is needed to achieve a fully coalesced, continuous MOCVD AlN hetero-epitaxial layer on the sapphire substrate 17,18 . The high growth temperature and fully coalesced thickness result in practical limitations on applications of MOCVD AlN epilayers.
Because of the distinguished advantages of ALD, a lower deposition temperature and a smaller coalesced thickness could be expected in the AlN thin films prepared by ALD. Thus the deposition and characteristics of the AlN thin films prepared by ALD have been intensively investigated recently [19][20][21][22][23][24] . However, owing to the low deposition temperature, the ALD-deposited AlN thin films are generally amorphous or polycrystalline, resulting in a low polar field which deteriorates the performance of HEMTs. Therefore, the improvement of crystallinity of the AlN layer prepared by ALD is highly desirable.
In this study, we make a proposal of a novel technique and concept of ALA in the ALD process to achieve high-quality epitaxial growth of AlN at a low deposition temperature of 300 °C. By introducing an in-situ, layer-by-layer Ar plasma treatment into each ALD cycle, the crystallinity of ALD-deposited AlN films was improved significantly, and a high-quality AlN epilayer with a thickness of only ~30 nm was achieved. The mechanism of ALA was also experimentally investigated and discussed, indicating that the crystallization takes place due to the surface heating caused by the in-situ ALA. Finally, the high-quality AlN layer prepared with the ALA treatment was grown on a bulk GaN epilayer, revealing the 2DEG characteristics near the AlN/GaN heterojunction with the significant improvement of mobility and sheet electron concentration by the in-situ ALA.

Result and Discussion
Atomic layer annealing (ALA). The scheme of the ALD cycles for growing the AlN epilayers using the in-situ ALA is plotted in Fig. 1. The standard ALD cycle consisted of the following steps: (1) TMA pulse, (2) Ar purge, (3) remote N 2 /H 2 plasma, and (4) Ar purge. As for the ALD cycle including the in-situ ALA, an additional step (5) of the Ar plasma treatment was added after the step (4) in each ALD cycle. In this way, each as-deposited monolayer formed by the standard ALD cycle (1)-(4) was modulated by the subsequent Ar plasma treatment (5). Hereafter the "standard" AlN layer is an AlN layer that was prepared without the additional step (5) for the ALA treatment.
The growth rate, uniformity, roughness, and surface morphology of the ALA-treated AlN layer. Figure S1 in the Supplementary Information shows the thickness of the ALA-treated AlN layer as a function of the applied ALD cycles, revealing a linear dependence between the film thickness and the applied ALD cycles including the additional step (5) of the Ar plasma treatment. The result indicates that the thickness of films treated with in-situ ALA can be precisely and digitally controlled by the number of the applied ALD cycles. The growth rate of the ALA-treated AlN layer deduced from Figure S1 was ~0.9 Å per ALD cycle. Figure S2 in the Supplementary Information illustrates the uniformity of the film thickness of the ALA-treated AlN layer on the sapphire substrate with an effective area corresponding to a 6" diameter. The percent non-uniformity of the film thickness is only ± 1.1%, indicating high uniformity of the AlN layer prepared with the ALA treatment. The surface roughness of the standard AlN and ALA-treated AlN layers characterized by AFM is shown in Figure S3 in the Supplementary Information. The root mean square (RMS) of the surface roughness of the ALA-treated AlN layer is 0.663 nm, which is slightly higher than that (0.223 nm) of the standard AlN layer but still in the same order of magnitude. The increase of surface roughness could be explained by the enhancement of grain size in the AlN layer treated with in-situ ALA as a result of the improvement of crystallinity, which is mentioned in the following section. The SEM image of the ALA-treated AlN layer is shown in Figure S4 in the Supplementary Information. It can be observed that the surface of the ALA-treated AlN layer is very smooth without any obvious pits or pinholes, which might be attributed to the layer-by-layer growth of ALD. In addition, the high conformality of the films prepared by ALD results in the filling of surface pits and pinholes by the following deposited layers, which is beneficial to reducing the density of pits or pinholes on the surface. Furthermore, the thermal stress induced by the difference in the thermal expansion coefficient between AlN and sapphire may lead to the formation of pits and cracks 25 . Since only a few layers of atoms near the surface were heated in the case of in-situ ALA, the pits and cracks caused by the thermal stress were much suppressed. Therefore, the ALA-treated AlN layer exhibited a very smooth and uniform surface which is nearly free of pits and cracks.
Crystallinity of the ALA-treated AlN layer. The θ-2θ XRD patterns of the standard AlN and ALA-treated AlN layers are shown in Fig. 2(a). In order to clarify the effect of ALA, the XRD pattern of the AlN layer prepared with the post-deposition Ar plasma treatment was also shown in Fig. 2(a) for comparison. The plasma power and total treatment time of the post-deposition Ar plasma treatment (300 W, 4000 sec) were controlled to be identical to those of ALA. Significant reflection at ~36.0° associated with the (0002) hexagonal AlN in Fig. 2(a) indicates that the ALA-treated AlN layer was well-crystallized with a wurtzite structure. As compared with the standard AlN layer and the AlN layer prepared with the post-deposition Ar plasma treatment, the remarkable enhancement of the XRD peak intensity indicates that the ALA treatment greatly improves the crystallinity of the AlN ultrathin film. Notice that multiple satellite peaks were also clearly observed around the (0002) main peak of the ALA-treated AlN layer. These satellite peaks originates from the interference due to an ultrathin hetero-epitaxial layer 26 , which provides further evidence of high crystallinity and smooth surface of the ALA-treated AlN layer. The significant difference in the crystallinity between the AlN layers prepared with the ALA and the post-deposition Ar plasma treatments can be attributed to the enhanced adatom migration on the surface by the layer-by-layer, in-situ ALA treatment. During the ALA treatment in each ALD cycle, the loosely bound adatoms on the surface are prone to migrate, leading to high crystallinity of the ALA-treated AlN layer. Figure 2(b) shows the ω-scan rocking curve of the ALA-treated AlN layer with a thickness of only ~30 nm. The full width at half-maximum (FWHM) of XRD ω-scan rocking curves is only 144 arcsec, which is close to the typical value of a thick (~1 μ m) AlN epilayer grown by MOCVD at a high temperature of 900~1100 °C 27 . Hence, Fig. 2 clearly reveals that the crystallinity of AlN is improved dramatically by the ALA treatment, and high-quality AlN epilayers with a thickness of a few tens of nm can be achieved at a low growth temperature as low as 300 °C. Figure 3 shows the θ-2θ XRD patterns of the AlN layers treated with different Ar plasma power and time during the in-situ ALA in each ALD cycle. When the Ar plasma power was kept at 100 W ( Fig. 3(a)), the XRD peak intensity increased with the plasma treatment time from 10 sec to 40 sec, indicating the improvement in crystallinity of the AlN layer with the increase of the plasma treatment time. As the Ar plasma power was raised to 300 W (Fig. 3(b)), the XRD peak intensity increased significantly with the increment of the plasma treatment time from 10 sec to 20 sec, and then saturated when the plasma treatment time was greater than 20 sec. Notice that the FWHM of all the (0002) AlN XRD peaks shown in Fig. 3(a,b) is almost the same. Since the vertical crystallite size is limited by the thickness of the film with high crystallinity, the approximately identical thickness of the AlN layers leads to indistinguishable difference between the FWHM of the (0002) AlN XRD peaks. As for the saturation of the crystallinity of the AlN layer when the plasma treatment time was greater than 20 sec (Fig. 3(b)), it can be understood that the thermal crystallization would reach saturation as the annealing time is sufficient 28 . The result shown in Fig. 3 indicates that the increase in the dose of plasma treatment, including the plasma power and time, is capable of enhancing the effect of in-situ ALA.
In order to confirm the effect of in-situ ALA, the duration between the Ar plasma treatment (5) and the following TMA pulse (1) (see Fig. 1), which is defined as the delay time, was changed from 0 to 20 sec in each ALD cycle. Figure 4(a) shows the θ-2θ XRD patterns of the ALA-treated (300 W, 20 s) AlN layers with different delay times. The AlN (0002) XRD peak intensity as a function of the delay time is plotted in Fig. 4(b). It is seen that the XRD peak intensity decreases significantly with the delay time, but is still higher than that of the standard AlN layer without the ALA treatment as the delay time is 20 sec. Hence it can be deduced that two mechanisms take place for the crystallinity improvement during the ALA treatment: one depends on the delay time and the other is independent of the delay time. During the in-situ ALA, the AlN layer gets the energy from the incident Ar ions and radicals, which promotes the migration of adatoms on the surface so as to bind with the atoms on the underlayer and the nearest neighboring adatoms. Thus the crystallization occurs in the deposited atomic layer during the Ar plasma treatment. This process termed as "surface crystallization" is supposed to be independent to the delay time. On the other hand, the energy transferred from the Ar plasma effectively results in the increase of the surface temperature. This "surface heating" effect will enhance the reactivity of the following TMA precursors adsorbed on the surface, leading to the higher effectiveness for ligands desorption from the precursors as well as the adatom migration 29 . Thus the effect of surface heating on the following adsorbed TMA precursor strongly depends on the delay time, and eventually decays with the increase of the delay time as a result of heat dissipation into the substrate and the surroundings. Therefore, the XRD peak intensity decays quickly with the delay time. Because the surface crystallization takes place during the Ar plasma treatment, the crystallinity of the ALA-treated AlN layer is still higher than that of the standard AlN layer without the ALA treatment, even with a very long delay time of 20 sec between the Ar plasma treatment and the following TMA pulse.  HRTEM was performed on the ALA-treated AlN layer to further investigate the microstructure of the film and the characteristics of the interface. Figure 5(a) shows the cross-sectional HRTEM image at the AlN/sapphire interface, revealing a sharp interface and a well-defined crystal lattice of AlN. The fast Fourier transform (FFT) diffractograms shown in Fig. 5(b,c) refer to the areas enclosed in the region of the AlN layer and the sapphire substrate, respectively. The diffractogram in Fig. 5(b)  Hence the growth of the ALA-treated AlN layer proceeds epitaxially layer-by-layer on the underlying surface from just the first layer on the substrate. Because of the lattice mismatch, the misfit dislocations can be formed at the interface between the sapphire substrate and the AlN epitaxial layer, and then extend as threading dislocations into the upper layer. The disarray of the AlN lattice fringes as indicated by the dotted lines in Fig. 5(a) may be ascribed to threading dislocations.
Chemical composition and depth profile analysis of the ALA-treated AlN layer. The Auger survey spectrum of the ALA-treated AlN layer (as show in Figure S5 in the Supplementary Information) exhibited four elements: carbon, nitrogen, oxygen, and aluminum. Figure 6 shows the Auger depth profile of each chemical composition in the ALA-treated AlN layer. Signals of contamination such as carbon and oxygen in the region I (as indicated in Fig. 6) decrease sharply from the surface and approach to the detection limit of AES when the etching time is greater than 1 min, indicating that the carbon and oxygen in the ALA-treated AlN layer are mainly ascribed to surface contamination and oxidation caused by the ambient atmosphere after the film deposition. In  the region II, the nitrogen and aluminum signals remain almost identical down to the interface between the AlN layer and the sapphire substrate (region III), suggesting that the composition of AlN is quite uniform across the layer. The slightly higher nitrogen content than aluminum (Al:N~1:1.2) in the region II might result from the plasma enhanced dissociation of nitrogen during the AlN deposition 30,31 . Refractive index of the ALA-treated AlN layer. The refractive index of the ALA-treated AlN layer was extracted from the spectroscopic ellipsometer (SE). Figure 7 shows the dispersion of the refractive index of the ALA-treated AlN layer in the wavelength range between 256 and 640 nm. The SE data was fitted by the quantum mechanical model, which is widely used for semiconductor materials. The refractive index of the ALA-treated AlN layer at λ = 633 nm is 2.04 and the bandgap energy given by the quantum mechanical model is 6.01 eV, which is close to the reported values of high-quality epitaxial AlN films 32,33 . Electrical properties of the ALA-treated AlN layer. In order to investigate the electrical properties of the AlN ultrathin films prepared with the ALA treatment, 10 nm-thick layers of standard AlN and ALA-treated AlN (300 W) with different plasma treatment time were deposited on 4 μ m-thick undoped-GaN epilayers which were grown by MOCVD. Figure 8 shows the sheet electron concentration (n s ) and mobility of the AlN/GaN heterojunction. As compared with the standard AlN/GaN heterojunction with n s of only 1.6 × 10 12 cm −2 , the n s was greatly increased by one order of magnitude to 1.4 × 10 13 cm −2 in the ALA-treated AlN/GaN heterojunction as the plasma treatment time was greater than 10 sec. The mobility also increased from 53.3 to 214.1 cm 2 V −1 s −1 with the ALA treatment on the AlN layer. The remarkable enhancement of the AlN crystallinity by the ALA treatment is responsible for the significant increase of the n s and mobility. The high crystalline quality of the AlN layer will induce the spontaneous polarization and strain-induced piezoelectric field 34 . Hence the enhancement of AlN (0002) preferred orientation by the ALA treatment improves the polarization field in the [0001] direction, leading to the increment of the n s and mobility. Therefore the significant increase of the sheet electron concentration and mobility could be attributed to the onset of the 2DEG near the AlN/GaN heterojunction prepared with the ALA treatment. Note that the increase of the n s and mobility saturates when the plasma treatment time is greater than 20 sec, which is in good agreement with the saturation of the AlN crystallinity as shown in Fig. 3(b).

Conclusion
In summary, a novel concept and approach termed "atomic layer annealing" (ALA) was proposed to realize the low-temperature atomic layer epitaxy of ultrathin films. ALA is the layer-by-layer, in-situ Ar plasma treatment applied in each ALD cycle to significantly improve the crystal quality of AlN ultrathin films, converting the structure from the amorphous phase to single-crystalline epitaxial layer. This is attributed to the surface crystallization caused by moving or migration of adatoms on the surface, so as to bind with the atoms on the underlayer and the nearest neighboring adatoms as the crystal. In addition, the surface heating effect that enhances the reactivity of the following precursors also occurs for the improvement of film crystallinity during the ALA treatment. Thus a high-quality AlN epilayer with a thickness of a few tens of nm was achieved by the ALA at a low deposition temperature of 300 °C. The XRD and HRTEM characterization clearly revealed a high-quality single-crystal hexagonal AlN epilayer. The 2DEG characteristics of the AlN/GaN heterojubction was also accomplished by the ALA treatment on the AlN layer. The result demonstrates that the layer-by-layer, in-situ ALA in each ALD cycle is a promising technique to achieve high-quality atomic layer epitaxy of nanoscale ultrathin films at a low temperature, which extends the applications of ALD tools from the conventional thin film deposition to high-quality epitaxial growth.

Method
ALD process. The AlN epilayers were grown on (0001) sapphire substrates by plasma-enhanced ALD.
Trimethylaluminum (TMA) and remote N 2 /H 2 plasma were utilized as the precursors for aluminum and nitrogen, respectively, which were delivered into the reaction chamber by the Ar carrier gas. The deposition was conducted at a low temperature of 300 °C.

Measurement.
The thickness and refractive index of the AlN layers were measured by the spectroscopic ellipsometer (SE, Elli-SE, Ellipso Technology). The crystalline phase and quality of the AlN thin films were characterized with the high-resolution transmission electron microscopy (HRTEM, FEI Tecnai G2 F20) and X-ray diffraction (XRD, PANalytical X'Pert Pro and Bruker D8) in θ-2θ and ω-scan (rocking curve) modes. Atomic force microscopy (AFM, NT-MDT) was used to evaluate the surface roughness of the AlN layers. The chemical composition and depth profile of the ALA-treated AlN layer were measured by the Auger electron spectroscopy (AES, JAMP-9510F Field Emission Auger Microprobe, JEOL). The acceleration voltage, electron current, and incident angle were 5 kV, 4 nA, and 60°, respectively. The sample was etched by 3 keV Ar ions and the duration of each etching cycle in the AES measurement was 3 sec. Scanning electron microscope (SEM, FEI Helios Nanolab 600i dual beam system) was used to examine the surface morphology of the ALA-treated AlN layer. The electrical properties of 2DEG were measured by the Ecopia HMS-3000 Hall-effect measurement system in the van der Pauw configuration at room temperature.