Polarity Control of an All-Sputtered Epitaxial GaN/AlN/Al Film on a Si(111) Substrate by Intermediate Oxidization.

The ability to control the polarity of an all-sputtered epitaxial GaN/AlN/Al film on a Si(111) substrate via intermediate oxidization was investigated. A stable surface of GaN on a Si substrate is a N-terminated surface (-c surface); hence, for electric device applications, the Ga-terminated surface (+c surface) is preferable. The GaN/AlN/Al film on Si(111) showed a -c surface, as confirmed by time-of-flight low-energy atom scattering spectroscopy (TOFLAS) and X-ray photoelectron spectroscopy (XPS). The AlN layer was intentionally oxidized via air exposure during film growth. The GaN surface subjected to the oxidization process had the +c surface. Secondary-ion mass spectrometry measurements indicated a high oxygen concentration after the intentional oxidization. However, the intentional oxidization degraded the crystallinity of the GaN/AlN layer. By changing the oxidization point and repeating the GaN/AlN growth, the crystallinity of GaN was recovered. Such polarity control of GaN on Si grown by sputtering shows strong potential for the fabrication of large-diameter +c-GaN template substrates at low cost.

Introduction

GaN exhibits noteworthy optical and electric properties and can potentially be used for new power electronics devices.[1−3] In practical applications such as light-emitting diodes, a polar-plane GaN crystal with hexagonal symmetry on a c-plane sapphire or a Si(111) substrate has been used.[4−7] A Si substrate is preferable for cost efficiency. The direct growth of GaN introduces the issue of melt-back etching, which arises from the alloying of Si and Ga at the interface. At high temperatures, especially those above 1000 °C, melt-back etching is enhanced and degrades the interface and each layer.[7−9] The deposition of AlN is one approach to eliminating this problem, as previously reported.[7,8] The insertion of an AlN layer has an additional function of providing a buffer layer that converts the tensile strain due to the lattice mismatch between GaN and Si (−16.9%) to compressive strain due to the lattice mismatch between GaN and AlN (+2.5%), which improves the crystallinity of nitride semiconductors on Si.[10,11]

The growth technique typically used for nitride semiconductors is metal organic chemical vapor deposition (MOCVD). MOCVD requires a high growth temperature and a large chemical waste treatment system for ammonia gas. By contrast, sputtering is a preferable method in terms of low-temperature growth, large-area deposition, and cost efficiency. While the use of sputtered GaN as a bulk substrate has crystallinity issues, its use as a template substrate for MOCVD is expected to reduce cost by thinning the stress relief layer. For AlN insertion, some groups have reported using sputtered AlN as a template layer for MOCVD-grown GaN films.[12,13] In addition, interest in AlN films on Si substrates has recently been increasing because of the piezoelectricity of AlN itself.[14,15] Sc-doped AlN exhibits ferroelectricity.[16] These applications also require a large-scale AlN film on a Si substrate. For the growth of AlN by sputtering, a previous report suggested that Al termination improved the AlN film surface and reduced defects.[17] We also demonstrated the deposition of an all-sputtered GaN/AlN film onto a Si(111) substrate at a substrate temperature below 650 °C.[18] To increase the value of GaN on AlN/Si substrates, it is necessary to achieve the same polarity control as that achieved for GaN on other template substrates such as c-sapphire.

For practical device applications, the polarity of GaN plays an important role in controlling band alignment to form a two-dimensional electron gas. Polar-plane GaN has two terminations along the c-axis: the Ga-terminated plane (+c) and the N-terminated plane (−c). Typically, a GaN film grown on a c-plane sapphire substrate is +c-GaN, which has been widely used in practical devices. In the case of −c-GaN, Matsuoka et al. recently proposed a −c-GaN-based high-electron-mobility transistor device for use at higher frequencies.[19] They also reported that prenitridation of the c-sapphire substrate is important for obtaining −c-GaN. However, because of the difference in chemical stability between +c- and −c-GaN and the better process compatibility of +c-GaN, +c-GaN-based devices remain in the mainstream. In this regard, the sputtered GaN film is −c-GaN. The use of an Al buffer layer does not change this tendency. In this regard, the polarity inversion of sputtered AlN on c-plane sapphire by a postannealing process, as reported by Xiao et al.,[20] provides guidance for achieving polarity inversion in the sputtering of GaN onto Si. Xiao et al.[20] showed that high-temperature postannealing of the AlN/sapphire structure at temperatures greater than 1300 °C induced oxygen diffusion from the sapphire substrate to the AlN layer and formed a thin oxidized interlayer. The polarity was inverted at the boundary of the oxide layer. This process requires high temperatures that preclude the use of Si substrates. In addition, the initial oxidization of the Si substrate forms an amorphous SiO2 layer that prevents the crystallization of AlN. Therefore, in the present work, we attempted to control the polarity of the initial AlN layer by oxidizing it to enable the growth of all-sputtered +c-GaN/AlN films on Si substrates using a high-quality GaN ceramic target developed by Tosoh Corporation.[21,22]

Experimental Procedure

A Si(111) substrate was cleaned with an organic solvent solution and deionized water, followed by cleaning with HF solution. AlN films were deposited by DC-magnetron sputtering onto the substrate under a base pressure of 6.7 × 10–6 Pa. Figure shows the sample structures. On the substrate, a 1.5 nm thick initial Al layer (hereafter, 1st-Al) was deposited. An Al metal plate was used as a sputtering target (Kojundo Chemical Laboratory, 5N grade). The substrate temperature and gas pressure were set at 400 °C and 0.5 Pa, respectively. The AlN film (hereafter, 2nd-AlN) was deposited onto the 1st-Al. For the 2nd-AlN with a film thickness of 10 nm, the substrate temperature, sputtering gas pressure, and DC power were set at 650 °C, 0.5 Pa, and 150 W, respectively. The sputtering gas ratio for the N2-to-Ar-based gas mixture was set at 20% on the basis of our previous work.[23] The intentional oxidization process was performed during the growth of the 2nd-AlN at two different thicknesses of 3 and 10 nm (hereafter 3-Ox-AlN and 10-Ox-AlN, respectively). When the target film thickness was reached, the sputtering process was terminated to allow the samples to be exposed to O2. The substrate was moved to the transfer chamber when the substrate temperature reached 100 °C or lower. The transfer chamber was purged with high-purity N2 gas, and the sample was held in air for 1 min. The chamber was then evacuated, and the sample was moved to the growth chamber from the transfer chamber, where the substrate was heated under high vacuum and film deposition was resumed. The deposition of 3-Ox-AlN and 10-Ox-AlN was followed by the sequential deposition of AlN or GaN. These samples were compared to the unoxidized AlN (hereafter As-AlN). On the 2nd-AlN, the GaN film was deposited by radio frequency (RF) sputtering. We used a GaN ceramic target with a density of 4.2 g/cm3 and an oxygen content of less than 0.4 atom % (Tosoh Corporation).[21,22] The substrate temperature, N2-to-Ar sputtering gas ratio, and RF power were set at 650 °C, 20%, and 150 W, respectively. For some samples, another AlN and GaN film deposition sequence was performed but without the oxidization process. The film thickness was checked using a stylus step profiler (DekTak 6M); however, some variations in film thickness of up to 15% due to the oxidization process were observed. In the region thinner than 40 nm, where the measurement accuracy of the stylus step profiler is degraded, the average deposition rate was multiplied by the deposition time for films with a thickness of 50 nm or more. The average deposition rate for both AlN and GaN was 3.3 nm/min.

Figure 1

Schematics of sample structures. After the initial Al layer (1st-AlN) was deposited, 3 or 10 nm thick AlN films (2nd-AlN) were deposited. Some samples were exposed to the atmosphere after the substrate temperature was lowered, and AlN and/or GaN films were subsequently deposited.

The crystal structure was analyzed by X-ray diffraction (XRD; Bruker AXS, D8 Discover with a General Area Detector Diffraction System). To investigate the orientation and epitaxial relationship, a two-dimensional X-ray diffraction (2D-XRD) method was employed. The 2D-XRD method enabled part of the Debye–Scherrer ring to be two-dimensionally detected. The 2θ and χ angles could be simultaneously detected. The full-width at half-maximum (fwhm) of the χ angle for a film indicates its mosaicity.[24] The detailed crystallinity was investigated via ω rocking-curve X-ray diffraction (RC-XRD) of the (002) reflection of the AlN and GaN layers using a conventional scintillation counter. The surface morphology was observed by atomic force microscopy (AFM; Hitachi Science and Technology, AFM5300E). For the surface termination analysis, time-of-flight low-energy atomic scattering spectroscopy (TOFLAS, Pascal, TOFLAS-3000, see ref (25)) was employed. TOFLAS is a surface-scattering analysis method similar to coaxial impact-collision ion scattering spectroscopy[26] but uses a neutral atomic beam instead of ions to avoid charging the surface of highly resistive samples such as AlN. A shape of the neutral atomic beam was a rectangle of 2 mm × 1 mm. The polarity and interface structure were investigated by X-ray photoelectron spectroscopy (XPS). A chemical bonding analysis was conducted by XPS using a monochromated Al Kα X-ray source (hν = 1486.6 eV) with a measuring spot diameter of 400 μm (Thermo Scientific, Sigma Probe). The total energy resolution was set at 700 meV. To investigate the polarity dependence of valence band structure, a takeoff angle (TOA: θ) was set to 9.5° to the c-axis direction integrated with an acceptance angle of ±7.5°, which is minimum TOA in our measurement setup. The sample was connected to the electrical ground level of the XPS system, and the binding energy was calibrated against the Au 4f7/2 peak (84.0 eV) and the Fermi level for a Au plate on the system. The XPS data were fitted by a Voigt function after Shirley-type background subtraction.[27] The corresponding inelastic mean free path (IMFP, λ) of Al 2p core-level photoemission excited by X-rays was calculated by the Tanuma–Powell–Penn-2M to be λ = 3.1 nm.[28,29] The detection depth was approximately three times the IMFP (∼9 nm). Secondary-ion mass spectrometry (SIMS) analysis was carried out by Eurofins EAG Laboratories using point by point-correction SIMS (PCOR-SIMS),[30,31] which was optimized for thin stacking structures by considering the matrix effect of conventional SIMS measurements.

Results and Discussion

+c-GaN Growth

Figure shows 2D-XRD patterns and AFM images for GaN on 10-Ox-, 3-Ox-, and As-AlN. In the patterns for all the samples, reflections corresponding to AlN (002) and GaN (002) were confirmed at 2θ values of 36.0° and 34.4°, respectively.[32,33] However, the width of the χ angle for the GaN on 3-Ox-AlN was substantially smaller than those for the GaN on the other samples. The fwhm values for AlN (002) obtained by RC-XRD were 1.13°, 1.68°, and 0.96° for the 10-Ox-, 3-Ox-, and As-AlN layers, respectively. The nonoxidized sample exhibited the highest crystallinity, and the sample prepared with the oxidization process performed during AlN film formation exhibited the lowest crystallinity. This crystallinity affected the growth of GaN. The fwhm values for GaN (002) were 1.00°, 1.45°, and 0.83° for GaN films on 10-Ox-, 3-Ox-, and As-AlN layers, respectively. However, this difference in crystallinity could be eliminated by increasing the number of stacked structures and the thickness of the AlN and AlGaN buffer layers, as has been used in MOCVD.[34,35] With increasing thickness of the buffer layers, the difference in crystallinity between the two stacked structures disappeared to a large extent (Supporting Information, Figure S1). In addition, changes in grain growth were observed, with AFM showing that samples with better crystallinity exhibited larger grain growth. In contrast to the crystallinity, the trend of grain growth was maintained in the 2-cycle-grown samples (Figure S1).

Figure 2

2D-XRD and AFM images of GaN films on As-, 10-Ox-, and 3-Ox-AlN/Si(111) structures.

The surface polarity was investigated by TOFLAS and XPS. Figure shows experimentally obtained (Figure a–c) and theoretically simulated TOFLAS images (Figure d–f). As shown in the simulated images, the difference in polarity of the hexagonal structure is characterized by the distance and overlap of the hexagram with the black spot at the top (Figure d and e). The simulated image of +c-GaN shows a large difference in the distance between the spots of the two hexagram stars, and the 6-fold symmetry is clearly distinguishable; by contrast, the image of −c-GaN shows a similar distance between the spots and 12-fold symmetry is observed. A comparison of the TOFLAS images of 10-Ox- and 3-Ox-AlN with the corresponding simulated images indicates that the wurtzite-structured GaN was terminated with Ga. The pattern for GaN is more clearly observed in the image of 10-Ox-AlN than in that of 3-Ox-AlN. In addition, the image of the GaN/AlN/GaN stack structure (hereafter 2-cycle-deposited GaN) on 10-Ox-AlN shows a clearer pattern than the images of the other samples (Supporting Information, Figure S1). By contrast, for As-AlN, the TOFLAS image matches the image of the surface of the zincblende structure, which has regular 6-fold symmetry (Figure f) without the 12-fold symmetry structure seen at −c-GaN and the inner 6-fold symmetry structure seen at +c-GaN, and not a hexagonal structure. This result is not consistent with the 2D-XRD image. This difference might be attributable to defective GaN surfaces, which have been reported to rarely exhibit a zincblende structure.[36,37] In addition, first principle calculation also suggested that the N rich coordination easily changes to the Ga rich coordination at the GaN surface.[38] In fact, a hexagonal structure with an N-terminated surface has been confirmed on the surface of 2-cycle-deposited GaN with improved crystallinity (Supporting Information, Figure S1).

Figure 3

TOFLAS pole-figure images for GaN films on (a) 10-Ox-, (b) 3-Ox-, and (c) As-AlN/Si(111) structures. The bottom images are the simulated TOFLAS pole-figure images for the (d) (001) and (e) (00–1) planes of wurtzite-structured GaN and the (f) (1–11) plane of zincblende-structured GaN.

To confirm the surface polarity, XPS analysis, which has a slightly deeper probe depth than TOFLAS, was also performed. Figure shows the valence-band spectra of GaN films on As-, 3-Ox-, and 10-Ox-AlN. The relative intensity of the valence band at the lower binding energy side (∼5 eV) in the spectrum of the GaN film on As-AlN (this electronic state is denoted as P1) is greater than that of the electronic state at the higher binding energy side (∼9.5 eV). The P1 peak in the spectrum of the GaN film on As-AlN was not clear compared with the corresponding peaks in the spectra of the GaN films on 3-Ox- and 10-Ox-AlN. A similar polarity dependence of the valence band structure has been reported for GaN, InN, and ZnO with a polar hexagonal structure, which is strongly related to the valence-band structure.[39−42] Ohsawa et al. noted that, for GaN, the P1 corresponds to the N 2p and Ga 4p states, whereas the P2 corresponds to the N 2p and Ga 4s states.[42] In +c-GaN, the P1 peak top is larger than the P2 peak top in relative intensity and the shoulder shape of the peak around 5 eV is more pronounced (Figure S2). This feature is highly angle-dependent and is attenuated in structures with a high mixing ratio. The valence band spectra of the GaN films on 3-Ox- and 10-Ox-AlN showed a slightly lower relative intensity of the P1 compared to single crystals (Figure S2), which may be due to attenuation derived from the mixed layer or surface roughness. Although quantitative analysis is difficult, in the direction along the C-axis of the oxidized thin film sample, most of the +c-GaN is considered to be present in the GaN films on 3-Ox- and 10-Ox-AlN are +c-GaN. The relationship between the valence-band structure and the crystal structure, and the XPS detection depth being deeper than that for TOFLAS, means that the TOFLAS analysis of the GaN film on As-AlN corresponds to just the surface structure; the bulk region exhibits a hexagonal-type electronic structure.

Figure 4

Valence band spectra for GaN films on 10-Ox-, 3-Ox-, and As-AlN/Si(111) structures.

Polarity inversion was confirmed in the samples subjected to the oxidization process. As mentioned, polarity inversion due to oxidization induced by a high-temperature heat treatment has been reported for AlN on sapphire substrates.[20] In this previous report, oxygen was found to diffuse from the sapphire side of the substrate to form a highly oxidized AlON layer that was abundant in the AlN film to a depth of a few tens of nanometers from the interface; above the interface, an inversion layer was formed. In this structure, oxygen did not diffuse throughout the entire film; it was distributed unevenly in the oxidized area. To confirm the presence of an oxidized layer in our structure, we conducted PCOR-SIMS measurements (Figure ). In the GaN film on As-AlN, the oxygen content near the interface varies because of the matrix effect due to the thin layered structure of AlN and Al between GaN and Si. The oxygen content, including unintentional oxidization effects due to atmospheric exposure, will be the background. By contrast, the oxidized sample shows a change in signal intensity greater than 1 order of magnitude. The peak of the oxygen signal intensity is located at the Si interface side for 3-Ox-AlN and at the GaN interface side for 10-Ox-AlN.

Figure 5

Depth profile of the SIMS signals of Al, Ga, Si, and O for 10-Ox-, 3-Ox-, and As-AlN/Si(111) structures.

The polarity inversion is likely induced by the oxidization layer in the AlN film. However, according to previous reports, the formation of an inversion layer requires annealing at a temperature greater than 1000 °C. Mohn et al. emphasized that metal polarity was established through AlON formation during nitridation of Al2O3 and that no AlON layer was observed after high-temperature growth of the N-polar film.[43] To attain metal polarity, they grew an AlN buffer layer at 650 °C. During the nitridation process for subsequent AlN and GaN deposition at high temperature, an AlON layer was formed, which had a mixed polarity and induced polarity inversion. Top epitaxial layers were then deposited at temperatures greater than 1000 °C. The high-temperature process induced oxygen migration and polarity inversion in the AlN layer. However, our growth temperature was the same as the growth temperature used by Mohn et al. to grow the AlN buffer layer. As described in the next section, to understand the oxidization process and the migration of oxygen atoms, we investigated samples of AlN/Si and GaN/AlN/Si with various film thicknesses.

Polar Inversion at the GaN/AlN Interface

First, we evaluated samples in which only the thickness of the AlN layer was varied. Figure a shows the valence-band spectra of 46, 23, and 13 nm thick AlN on 10-Ox-AlN. The spectra of the AlN films indicate that the surface polarities were unclear. Note that, compared with the spectra of GaN, those of AlN show smaller differences among the P1/P2 intensity ratios; even our results for AlN on a sapphire substrate show only a weak peak at P1 (Supporting Information, Figure S3).[44] In the present results, no clear P1 peak is observed. The pole-figure TOFLAS images are also unclear (Figure b) although the surface roughness is lower than that of the GaN films (Figure c), and this trend is similar for other film thicknesses (not shown). However, the Al 2p spectra for the AlN films with various film thicknesses show a clear difference between the 13 and 23 nm thick films (Figure ). The Al 2p spectra for the 46 nm thick AlN films show two bonding states: AlN (∼73.8 eV) and Al–N–O (∼74.6 eV). The Al–N–O bonding state corresponds to defective AlN, which was oxidized unintentionally upon air exposure. For As-AlN, the spectra of all the samples are similar. By contrast, for AlN films on 10-Ox-AlN with thicknesses of 13 and 23 nm, an additional bonding state was confirmed at ∼76.5 eV. This binding energy is consistent with the Al–O bonding state of Al2O3 and is attributed to highly oxidized defective AlN (hereafter referred to as the Al–O bonding state). As the thickness of the AlN film was increased, the area intensity of the Al–O peak decreased. This result indicates the presence of highly concentrated oxygen near the interface, which is consistent with the SIMS results. A comparison of the results for As- and 10-Ox-AlN indicates that this layer containing high concentrations of oxygen is not formed simply by air exposure but by heating after the air exposure. However, these results imply that Al–O bond formation was not sufficient to invert the polarity. The 46 nm thick AlN film, which consisted of 36 nm thick AlN on 10-Ox-AlN, did not exhibit the +c structure.

Figure 6

(a) Valence-band spectra for AlN films of various thickness on 10-Ox-AlN. (b) TOFLAS pole-figure and (c) AFM images for a 46 nm thick AlN film on 10-Ox-AlN.

Figure 7

Al 2p core-level spectra for AlN films of various thickness on (a) As-AlN and (b) 10-Ox-AlN.

We also analyzed the GaN/AlN stack structures. Figure shows valence band spectra of GaN films with various film thickness deposited onto 10-Ox-AlN films. When the total thickness of the film was increased from 13 to 17 nm, the intensity of the P1 band clearly increased and the structure was +c-GaN. By contrast, in the spectrum of the 13 nm thick film, the spectral shape is AlN-type and the P2 peak is located at the lower binding energy side. According to a previous report, AlGa1–N with x > 0.77 exhibits an AlN-type valence-band structure.[45] In addition, Figure shows the Ga 2p and Al 2p core-level spectra, which consist of three and two bonding states, respectively. For Ga 2p, the highest bonding state corresponds to the metallic Ga and/or Al–Ga bonding state. The formation of AlGa1–N is possible. At the interface, numerous defects exist according to the metallic Ga bonding state. On the basis of electronegativity, Ga should attract electrons and the Ga 2p peak should shift to a higher binding energy. Collectively, the valence band spectra indicate that, for the 13 nm thick GaN/10-Ox-AlN/Si sample, the Ga 2p peak at ∼1119 eV corresponds to AlGa1–N, not GaN. By contrast, for the two bonding states in the Al 2p core-level spectra, the bonding states of AlGaN and AlN are difficult to distinguish. However, the peak corresponding to the Al–O bonding state in the spectrum of the 13 nm thick AlN/Si clearly disappeared, indicating a reduction of the highly oxidized AlN. These results suggest that the deposition of GaN onto AlN, in conjunction with substrate heating process, induced oxygen migration at the interface. In addition, the +c-GaN film was obtained after GaN film formation and subsequent air exposure when the AlN film was relatively thick (Supporting Information, Figure S4). The key factor governing the polarity inversion should be the GaN/AlN interface layer.

Figure 8

Valence band spectra for GaN films of various thickness on 10-Ox-AlN.

Figure 9

(a) Al 2p and (b) Ga 2p3/2 core-level spectra for GaN films of various thickness on 10-Ox-AlN.

Figure shows a summary of the role of the intentional oxidization layer at the GaN/AlN interface. The mechanism of polar inversion is similar to that for AlN on Al2O3. The high oxygen content of the formed AlN layer is important. However, for AlN, forming the inversion layer by inducing oxygen migration requires a high-temperature process, which is not compatible with Si substrates. To enhance the oxygen migration, the formation of an AlGaNO layer is critical. The crystallization temperature for GaN is lower than that for AlN. Ga ions related to defective GaN can form AlGaNO easily. This reaction process absorbs the AlN layer containing a high concentration of oxygen, and polarity reversal occurs. With respect to the thickness of AlN, in our initial investigation, a thinner initial AlN layer was correlated with better crystallinity (Supporting Information, Figure S4). However, oxidization of the Si substrate should be minimized. Note that the 3-Ox- and 10-Ox-AlN substrates showed the formation of SiO2 and SiON (Supporting Information, Figure S5). In this study, the initial AlN film thickness was optimized in consideration of the grain density, which can lead to deterioration of the crystallinity; room exists for further experimentation related to the growth rate and temperature range. Furthermore, the oxidation process is strongly influenced by surface topography and grain boundaries. Although the oxidation process was saturated under the conditions described in this paper, further investigation is needed for structures with improved crystallinity and planarity.

Figure 10

Schematic of the role of intermediate oxidization.

Conclusions

Polarity control of GaN/AlN films sputtered onto a Si structure was demonstrated. After intentional oxidization of the GaN/AlN interface, TOFLAS and XPS revealed the polarity inversion of GaN from −c to +c. The oxidization formed the Al–O–N bonding state. Subsequent deposition of another GaN layer led to the formation of AlO and the AlGaNO intermediate layer. During this process, oxygen migrated at the interface and formed the +c-GaN.

Hence, for applications in electronic devices, the Ga-terminated surface (+c) is preferable. The previously reported polarity inversion processes for GaN on a sapphire substrate are not suitable for Si substrates. The temperature used in our process is substantially lower than those used in previously reported processes, making our method suitable for use with Si. The ability to control the polarity of all-sputtered epitaxial GaN/AlN/Al films on Si(111) substrates via intentional intermediate oxidization has strong potential for the low-cost fabrication of large-diameter films on +c-GaN substrates.