2.3.1. 1984-1994: XPS and thermal stability

Techniques that have been used for the evaluation of GaN polarity are listed in chronological order in Table I (a). The relationship between the growth conditions (and also the growth method) and the resulting polarity is presented. The first characterization results came out in 1988, two years after better quality GaN films were deposited by the development of the LT-buffer layer. Sasaki et al. deposited GaN films on both the Si and C faces of 6H-SiC substrates by MOCVD, and their polarities were determined by the dependence of the Ga signal intensity detected with angle-resolved x-ray photoelectron spectroscopy (XPS) Reference Sasaki and Matsuoka[12]. The surface morphologies of the samples on Si- and C-face SiC substrates were either featureless or hexagonal facetted. GaN films deposited on Si- and C-face 6H-SiC substrates were expected to show +c and −c polarity, respectively, according to the relationship of the electron negativity of Si and C Reference Ren and Dow[13] Reference Ren and Dow[14]. However, they determined the polarity of the former and the latter as −c and +c polarity, respectively. This misjudgment might possibly have been caused by an immature understanding of oxygen adsorption on the polar surface of GaN (refer to Sec. 5.1).

7 years later, Sun et al. found that a GaN film deposited on the Si-face of a 6H-SiC substrate was unstable in an H₂ ambient at 600°C Reference Sun, Kung, Saxler, Ohsato, Bigan, Razeghi and Gaskill[15]. This GaN sample was determined to have +c polarity due to the electron negativity of Si and C, in contrast with Sasaki's conclusion. We also confirmed the instability of +c GaN under these conditions Reference Fuke, Teshigawara, Kuwahara, Takano, Ito, Yanagihara and Ohtsuka[16]. However, we felt that the window of the conditions used for annealing temperature, time and gas ambient seemed to be very narrow. Therefore, this technique was difficult to use generally for determining the polarity. In 2001, Koukitsu et al. measured the decomposition rate of +c and −c GaN samples in N₂ or H₂ gas ambients using the microgravity method Reference Koukitu, Mayumi and Kumagai[17]. GaN with +c polarity decomposed faster at lower temperatures (800-850°C) than GaN with −c polarity (900-950°C), which was consistent with Sun's report and Hellman's standard framework.

2.3.2. 1996: CBED

The investigation of polarity began intensively in '96, when the first GaN blue laser diode was developed using MOCVD and when better quality GaN was also being attempted to be grown by MBE. A convergence beam of electron diffraction (CBED) technique based on TEM Reference Ponce, Bour, Young, Saunders and Steeds[18] Reference Liliental-Weber, Kisielowski, Ruvimov, Chen, Washburn, Grzegory, Bockowski, Jun and Porowski[19] was mainly used to determine the polarity by studying asymmetric diffraction spots of (0002) and (000

). The quality of the GaN material must be sufficiently high in order to achieve clear asymmetric diffraction spots. Important aspects concerning polarity were identified through CBED observations of several kinds of GaN films in the following work;

1. 1. MOCVD- and MBE-GaN films were likely to have +c and mixed polarity-containing inversion domains (IDs), respectively Reference Romano, Northrup and O’Keefe[20].

2. 2. Nitridation of the sapphire substrate in MOCVD might result in a −c GaN film with a rough surface Reference Rouviere, Arlery, Niebuhr, Bachem and Briot[21]

3. 3. The dependence of various properties on the polarity was revealed Reference Keller, Keller, Wu, Heying, Kapolnek, Speck, Mishra and DenBaars[22] Reference Ponce, Bour, Gotz and Wright[23].

Claims 1 and 2 indicate the importance of the growth process, as discussed in Sec. 4, and claim 3 is related to the dependence of impurity adsorption on the polar surface, as discussed in Sec. 5

In order to extend the possibilities of techniques based on the TEM, efforts have been made continuously to observe atomic alignment with high-resolution image matching Reference Stirman, Ponce, Pavloska, Tsong and Smith[24], to introduce micro-channeling effects during EDS analysis Reference Jiang, Eustis, Cai, Ponce, Spence and Silox[25], to detect the N K-edge and Ga L-edge during EELS analysis Reference Kong, Hu, Duan, Lu and Liu[26], or to count diffracted electrons quantitatively Reference Zandbergen, Janzen, Zauner and Weyher[27]. The polarity can also be evaluated by these new approaches.

2.3.3. 1997-1998: Surface reconstruction, chemical stability and CAICISS

There were two interesting reports in '97. One of these, which highlighted the temperature dependence of the surface reconstruction on the polarity, was carried out by Smith et al. Reference Smith, Feenstra, Greve, Neugebauer and Northrup[28]. Held et al. confirmed this surface reconstruction by RHEED observations of bulk GaN samples for which the polarity had been rigorously identified Reference Held, Nowak, Ishaug, Seutter, Parkhomovsky, Dabiran, Cohen, Grzegory and Porowski[29]. RHEED patterns that were characteristic of (1×1) and (2×2) reconstructions for −c and +c polar surfaces respectively were observed against the [11

0] azimuth under a constant supply of NH₃ gas. The quality of the GaN samples needs to be sufficiently good to achieve these surface reconstructions. When the surface reconstructions were analyzed theoretically by taking into account the polarity and surface-termination atoms, the most favorable element to terminate the surface was evaluated as being Ga, in spite of the polarity Reference Smith, Feenstra, Greve, Shin, Skowronski, Neugebauer and Northrup[30]. The relationship between the termination and the reconstruction are described in detail in an excellent review Reference Feenstra, Northrup and Neugebauer[31].

Another technique involves the chemical stability of the −c GaN surface in alkali solution Reference Weyher, Müller, Grzegory and Porowski[32]. Apparently, −c GaN films are etched in KOH or NaOH, while +c GaN is inert to these solutions. The polarity that was determined by investigating the chemical stability was consistent with that determined by using hemispherically-scanned x-ray photoelectron diffraction (HSXPD) Reference Seelmann-Eggebert, Weyher, Obloh, Zimmermann, Rar and Porowski[6]. We reported the mechanism for the selective etching, in that OH^- in solution would promote etching, attacking one back-bond of the Ga that was bonded to the nitrogen on the −c polar surface Reference Li, Sumiya, Fuke, Yang, Que, Suzuki and Fukuda[33]. The etching effects that originate from side facets or dislocations are still unknown, especially for GaN with rough surface morphology Reference Losurdo, Giangregorio, Capezzuto, Bruno, Namkoong, Doolittle and Brown[34]. This chemical stability, however, is the easiest way to determine the polarity of GaN

The methods mentioned above seem not to be suitable for determining the polarity of thin and poor quality GaN such as LT-buffer layers. In '98, we used CAICISS to determine their polarity Reference Sumiya, Ohnishi, Teshigawara, Tanaka, Ohkubo, Kawasaki, Yoshimoto, Ohtsuka, Koinuma and Fuke[35]. The polarity of LT-GaN buffer layers was successfully evaluated for the first time by CAICISS Reference Sumiya, Tanaka, Ohtsuka, Fuke, Ohnishi, Ohkubo, Yoshimoto, Koinuma and Kawasaki[36]. This technique has been used for analyzing the polarity of MBE-GaN by Shimizu et al. Reference Shimizu, Suzuki, Nishihara, Hayashi and Shinohara[37]. The various features and advantages of CAICISS will be discussed in Sec. 2.4.

2.3.4. Recent movements

The other methods are briefly commented on according to the data in Table I (b). The polarity can be determined by observation of both the angular dependence of the Ga K-edge Reference Kazimirov, Faleev, Temkin, Bedzyk, Dmitriev and Melnik[38] and the crystal truncation rod Reference Tabuchi, Matsumoto, Takeda, Takeuchi, Amano and Akasaki[39] generated by the standing wave of the x-rays from synchrotron radiation. Recently, a GaN film with both polarities was deposited by MBE on a single sapphire substrate Reference Dimitrov, Tilak, Murphy, Schaff, Eastman, Lima, Miskys, Ambacher and Stutzmann[40]. The properties in the two regions were measured, and the relative differences in the polarity were observed. A1 (TO) mode appeared in Raman spectroscopy in the −c GaN region Reference Cros, Joshi, Smith, Cantarero, Martines-Criado, Ambacher and Stutzmann[41] due to a higher level of impurities and defects (also discussed in Sec. 5.1). The surface potential detected by KPFM was +25±19, −30±10meV Reference Jones, Visconti, Yun, Baski and Morkoc[42], and the surface charge (σ/e) evaluated by PFM was −1.78×10¹³, 1.83×10¹³cm⁻³ Reference Rodriguesz, Gruverman, Kingon, Nemanich and Ambacher[43] for the +c and the −c GaN regions, respectively. These polarity dependences are important for device design with respect to the potential at the interface Reference Jang, Lee and Lee[44].

2.4.1. Principle and advantages

Figure 3 shows a schematic illustration of the CAICSS equipment that we used in this research. The He⁺ beam is pulsed by chopping the aperture (150ns duration at 100kHz repetition rate) and is accelerated at low energy (2keV in this article). The He ion beam impinges on the sample surface with a 2mmϕ spot-size, focused by an Einzel lens. The diameter of the shadow cone formed by the low energy ion beam is calculated to be several Å, which is much larger than the figure of 0.1Å that is used for high energy ions Reference Daudin, Rouviere and Arlery[45], such as in Rutherford back-scattering. The larger shadow cone enhances the cross-section of the scattering, and detection becomes very sensitive to atoms in the surface region. By changing the incident angle, both shadowing and focusing effects take place according to the atomic arrangement on surface, as illustrated in Figure 4. CAICISS is different from time-of-scattering and recoiling spectrometry (TOF-SARS Reference Sung, Ahn, Bykov, Rabalais, Koleske and Wickenden[46]), using the principle of elastic recoil detection.

Figure 3. Schematic illustration of CAICISS apparatus with a chamber and block diagram. The incident and azimuth angles of the ion beam can be altered by moving the sample holder, which is equipped with a heater.

Figure 4. Schematic illustration explaining the focusing and shadowing effect. According to the definition given in Ref. Reference Katayama, Nomura, Kanekawa, Soejima and Aono[5], which was made by Katayama et al., the incident angle α was changed from 90° (normal to the sample surface) towards the lower angle. The angular dependence of the CAICISS signal can be obtained because both effects correspond to the atomic arrangement on surface, as observed in the middle. Variations in the TOF spectra are induced by changing the incident angle, as shown on the right. The dependence of the integrated peak (colored area) of the TOF spectra on the angle corresponds to the CAICISS result shown on the bottom of the left-hand side, which can be used to determine the polarity and the surface structure.

According to the procedure in Ref. Reference Katayama, Nomura, Kanekawa, Soejima and Aono[5] as carried out by Katayama et al., we changed the incident angle α from 90° (normal to sample) towards a lower angle. In CAICISS, He⁺ and He⁰ (He particles) are back-scattered along an angle of 180°, due to impact collisions with atoms on the surface, and are detected by a multi channel plate. The intensity of the back-scattered He ions depends largely on the incident angle. Consequently, an angular dependence against an identical atom is obtained, as shown in Figure 4. The better the quality, the deeper the dip ((1) in Figure 4) and the narrower the width of the peak ((2)).

A specific feature of CAICISS is that it is a simple way of quantitatively analyzing the atomic arrangement on the surface, such as the distance and the angle made with neighboring atoms, because the analysis of the scattering orbitals can be extremely simplified by the focusing and shadowing effects and by taking only ions that have impact-collided with atoms into account. Using these features, the atomic structure of the surface (several nm deep) can be non-destructively analyzed in real space with CAICISS. The potential of CAICISS for determining the surface atomic arrangement in real space has been demonstrated for Si surfaces, compound semiconductors Reference Katayama, Williams, Kato, Nomura and Aono[47] [48] and oxide thin films Reference Ohnishi[49].

The polarity can be determined by the positions of the parabolic shadowing dip and the focusing peak of the He⁺ beam in CAICISS analysis. When the incident angle of a He⁺ beam that is irradiated from the [11

0] azimuth is changed, the angular dependence of the cation signal intensity is represented by one or two peaks around 70° for +c and −c polarity GaN, respectively. These angular dependences can be calculated by computer simulation Reference Sonoda, Shimizu, Suzuki, Balakrishnan, Shirakashi, Okumura, Nishihara and Shonohara[50]. However, we have experimentally determined the polarity of a GaN film from the angular dependence measured by CAICISS, by comparing it to those shown in Figure 5 for bulk ZnO with Zn- (+c) and O- (−c) face polarity Reference Ohnishi, Ohtomo, Kawasaki, Takahashi, Yoshimoto and Koinuma[51]. This was achieved without using the results of simulations. This is possible because ZnO has the same crystalline structure and very similar lattice constants to those of GaN, since they are neighboring elements in the periodic table.

Figure 5. Incident angular dependence of the Zn signal intensity when the specimen was tilted along the <110> azimuth. (a) single crystal (0001) Zn face (+c), (b) simulated curve of (a), (c) single crystal (0001)O-face (−c), and (d) simulated curve of (c). Simulation was based on a three-dimensional two-atom model for a virtual surface cut from an ideal bulk structure without any reconstruction. (Ohnishi Dr. Thesis p.70 Ref. Reference Ohnishi[49]) The polarity of the GaN was determined from the angular dependence revealed by CAICISS by comparing these results with these results of ZnO, because ZnO has the same crystal structure and lattice constants that are very close to those of GaN.

We have to comment here that it is not the termination atoms but the polarity, which can be analyzed by the CAICISS method that we use in this section. The competence of CAICISS for determining the polarity of III-nitride can be best demonstrated when it is applied to very thin films, such as LT-buffer layers and quantum well structures. When the CAICISS technique is to be used for thicker III-nitride samples, cross-checking should be implemented using either CBED or the chemical stability in alkali solution.

2.4.2. Application of CAICISS to an InGaN SQW

The types of atoms that the He⁺ beam collides with can be distinguished by determining the TOF (calculated from the distance that the He has traveled), the acceleration voltage and from the weight of the atoms. Figure 6 shows a TOF spectrum for a 3-nm In_0.5Ga_0.5N single quantum well (SQW) on a GaN film at normal incidence to a He ion beam Reference Sumiya, Nakamura, Chichibu, Mizuno, Furusawa and Yoshimoto[52]. Since the length of travel and the energy of the He ions are 836mm and 2keV, respectively, the TOFs of the He that is backscattered due to collisions with the In and the Ga can be calculated theoretically to be 6417 ns and 6550 ns, respectively. The position of the two sharp peaks in Figure 6 can thus be assigned to the In and Ga signals. TOF spectra with higher mass-resolution for cases where both atoms have atomic weights that are close to each other can be obtained by using Ne ions instead of He ions. The broader TOF spectrum after the Ga signal is attributed to the multiple scattering of He ions. Since the cross-section of light atoms such as nitrogen is smaller, their signals are detected after a longer time-of-flight, and are likely to be in the noise level in Figure 6. An In signal, which is barely detectable, for the 6-nm GaN capped SQW (as indicated in the inset) is probably derived from the interface between the SQW/GaN, due to the channeling effect at normal incidence. This In signal was hardly detectable at other incidence angles. This indicates that CAICISS analysis can detect information from the surface region of III-nitrides, down to a depth of several nm.

Figure 6. TOF spectrum of the backscattered He⁺ ions used in the CAICISS analysis of an In_0.5Ga_0.5 SQW when the ion beam was irradiated at normal incidence to the sample. The inset depicts the TOF spectrum for In_0.2Ga_0.8N capped with 6-nm-thick GaN as a reference [after Reference Ohnishi, Ohtomo, Kawasaki, Takahashi, Yoshimoto and KoinumaRef. 51]. The In and Ga signals can be detected separately for each time-of-flight. The inset indicates how CAICISS analysis detects a region several nm deep below the surface.

The integrated signal for the III-group element (shadow area in Figure 4) from each TOF spectrum with changing angle of incidence was obtained to evaluate their polarity. Surface contaminants such as C and O, which have small cross-sections, would not have any influence on determining the polarity. Figure 7 shows the angular dependence of the intensities of Ga and In in CAICISS-TOF spectra for an In_0.5Ga_0.5N SQW. It was confirmed from comparison with the results in Figure 5 that the alloyed In has +c polarity, while the variation in the Ga signal also indicates +c polarity. The indium atoms incorporated into the In_xGa_1-xN SQW were found to occupy substitutional sites for Ga, and they exhibited +c polarity.

Figure 7. Incident angle dependence of Ga and In scattered intensity at the [11

0] azimuth for In_0.5Ga_0.5N SQW. Variation of both the Ga and In signals indicates +c polarity judging from the CAICISS results on ZnO bulk given in Fig. 5. Indium atoms incorporated into an In_xGa_1−xN SQW were found to occupy substitutional Ga sites. Reference Sumiya, Nakamura, Chichibu, Mizuno, Furusawa and Yoshimoto[52]]

3.1.1. Conditions used for treatment of the sapphire and for the buffer layer

3.1.2. Mg accumulation layer

Ramachandran et al. have investigated the phenomenon of Mg adsorption on the surface of a +c GaN layer grown on Si-face 6H-SiC by PAMBE. When the film was exposed to 1.2±0.4ML or more of Mg during growth under Ga-poor conditions, they found that the polarity for the subsequent GaN film switched to −c polarity Reference Ramachandran, Feenstra, Sarney, Salamanca-Riba, Northrup, Romano and Greve[74]. When the inversion takes place, the incorporated Mg concentration in the layer was confirmed to be approximately 8×10¹⁹cm⁻³. At the point where the polarity changed, the interface was rough, with a zigzag inversion domain boundary with mainly {1113} facets and a few (0001) segments Reference Romano, Northrup, Ptak and Myers[75]. Recently, Grandjean et al. have controlled the polarity conversion, not only from +c to −c, but also from −c to +c polarity in NH₃-MBE Reference Grandjean, Dussaigne, Pezzagna and Vennegues[76]. The interface in the latter conversion was confirmed as being flat.

A monolayer of Mg deposits on the Ga-terminated surface of the +c GaN film. Since the local structure of the Mg₃N₂ is more favorable than that of the bulk GaN, the Mg is likely to bond with the N atoms. Consequently, the displacement of the Ga and the N atoms in the outermost layer should occur, forming the configuration Mg-N-Ga/Ga-N from Mg-Ga-N-Ga-N on the +c polar surface. The sign Ga/Ga indicates IDB, consisting of a plane of Ga-Ga bonds. The N atoms are six-fold coordinated with the outermost Mg and the underlying Ga. This surface structure has been theoretically calculated to be the most stable Reference Feenstra, Northrup and Neugebauer[31] Reference Ramachandran, Feenstra, Sarney, Salamanca-Riba, Northrup, Romano and Greve[74]. Recently, a model of pyramidal (zigzag) inversion domains originating from the Mg on the (0001) segment of the boundary was theoretically performed by first-principles pseudopotential density functional calculations Reference Northrup[77]. The most favorable structure of the Mg boundary inserted into the GaN was evaluated as abcab stacking across the (0001) segment, corresponding to the atomic sequence GaNMgNGa, where the side of the boundary lies along the {1113} direction, corresponding to the zigzag inversion domain boundary. In this structure, the concentration of Mg in the boundary layer was calculated to be 3/4 monolayer, occupying H3 sites.

3.1.3. Deposition of GaN film: V/III ratio and growth rate

Tarsa et al. deposited GaN films homo-epitaxially on +c GaN templates by plasma-assisted MBE Reference Tarsa, Heying, Wu, Fini, DenBaars and Speck[78]. GaN films deposited with a low V/III ratio (Ga-stable) had a quality comparable with the underlying GaN template, while GaN films with a high V/III ratio (N-stable) showed a faceted surface and poor crystalline quality. Held et al. Reference Held, Crawford, Johnston, Dabiran and Cohen[79] and Myers et al. Reference Myers, Hirsch, Romano and Richards-Babb[80] deposited on −c GaN templates by MBE. The former obtained GaN films with 3-D growth under excess NH₃ (Ga-limited growth) and films with a step-flow growth under excess Ga (NH₃-limited growth). The latter obtained GaN samples with pyramidal hillocks or a flat surface morphology under N- and Ga-stable conditions, respectively. The Ga-rich condition is likely to result in GaN films with a smooth surface, regardless of the polarity of the template.

The dependence of the polarity on growth rate is also observed for GaN and AlN buffer layers, shown in Table II (a), as well as in ZnO films with the same wurtzite crystal structure Reference Ohkubo, Ohtomo, Ohnishi, Matsumoto, Koinuma and Kawasaki[81]. Takahashi et al. deposited GaN films on GaAs (111) A and B-face substrates by MOMBE. The deposition was carried out at 700°C by changing the Ga flux (beam equivalent pressure (BEP); 2~8×10⁻⁸Torr) under a constant supply of DMHy as the N source. The growth rate of the GaN on both substrates increased up to 5×10⁻⁸Torr of Ga supply, and then saturated at 400nm/h above that level. This indicates that the growth was promoted with a supply-limit (N-rich) for the lower BEP, while it occurred with a surface kinetic limit (Ga-rich) for the higher BEP. GaN films on A-face GaAs (111) were found to have +c polarity, and were independent of the growth rate. On the other hand, the polarity on the GaAs (111) B-face was −c for the supply-limited condition, and it was +c when it was kinetic-limited. GaN grown under III-rich conditions (kinetic limited, Ga- stable or limited growth) is considered to predominantly display +c polarity Reference Takahashi, Nakayama, Souda and Hasegawa[82].

3.1.4. Factors for controlling the polarity in MBE-GaN on sapphire substrate

The following are considered to be the key points that are decisive in determining the polarity of MBE-GaN;

1. 1. Use of an AlN buffer layer deposited on a sapphire substrate at higher temperature with a higher growth rate (III-rich conditions)

2. 2. Insertion of an Mg, Al and Ga metal layer at the interface

3. 3. Use of a thicker LT-GaN buffer layer.

The vapor pressure of GaN is much higher than that of AlN, as shown in Figure 8 Reference Edgar, Strite, Akasaki, Amano and Wetzel[83]. AlN buffer layers, with their lower vapor pressure, should be suitable to act as nucleation layers for MBE-GaN grown under high vacuum conditions. Felice et al. calculated theoretically the atomic structure of films consisting of approximately 1 bilayer of AlN on c-plane sapphire substrates Reference Di Felice and Northrup[58]. Under equilibrium conditions, the Al layer in the H3 sites lying between the last O plane (blue region) and the first N plane (yellow region) maintain the stoichiometry of bulk sapphire for 2/3 monolayers, as shown in Figure 9 (a) and Figure 9(b). This favorable structure could be changed by the amount of Al in the initial growth, from Al-rich in Figure 9 (a) to Al-poor in Figure 9 (b). Both of the geometries of the AlN on the outermost layer (between the green brackets) corresponded to +c polarities (from previous calculations) Reference Northrup, Di Felice and Neugebauer[84]. In contrast, the alignment of the AlN (between the red brackets) in Figure 9(c) and Figure 9(d) corresponded to −c polarity. This structural difference is observed in the complete interface of Al adatoms lying in the T4 sites between the last O plane and the N plane. The calculations predicted that the polarity of these very thin films on sapphire substrates would be attributed to that of the III-V nitrides. Moreover, since the two structures in Figure 9 (b) and Figure 9 (c) have very similar formation energies, independent of Al abundance, it is assumed that a slight fluctuation in conditions in the initial stages of deposition can alternately switch the structure. The situation that is shown in Figure 9 (c) must occur under non-equilibrium conditions, such as nitridation of the sapphire under Al-deficient conditions. Consequently, the nitridation of sapphire is regarded as resulting in −c GaN, though moderate nitridation as carried out by the AIST group Reference Sonoda, Shimizu, Shen, Hara and Okumura[56] might form the interface structure shown in Figure 9 (b). Increasing the supply of group III implies that the growth of nitride materials would approach the equilibrium state, forming the structures with +c polarity shown in Figure 9 (a) and Figure 9 (b).

Figure 8. Equilibrium N₂ pressure over III-V nitrides (solid) + III-metal (liquid). Lines for each nitride material are plotted together from Ref. Reference Ohkubo, Ohtomo, Ohnishi, Matsumoto, Koinuma and Kawasaki81

Figure 9. Models for the AlN thin films on c-plane sapphire substrates given in Ref. Reference Di Felice and Northrup[58]. The color regions are added to explain the theoretical predictions. The structures in the brackets correspond to the polar structures of AlN described in Ref. Reference Northrup, Di Felice and Neugebauer[84]. (a) and (b) correspond to a +c polar surface. (c) and (d) to −c polarity.

Figure 10. Nitridation-temperature dependence of the atomic concentration on a sapphire surface determined by XPS analysis. When HT-GaN films were deposited on sapphire substrates nitrided at temperatures of more than 700°C, the films represented −c polarity and hexagonal facetted surfaces.

3.3.1. Until 1992: Direct growth, GaCl treatment, and ZnO insertion

The first GaN film was deposited on a sapphire substrate by HVPE in '69 Reference Maruska and Tietjen[93]. Judging from its hexagonal-facetted surface morphology, this GaN film had −c polarity. Intensive efforts have been made to grow smooth large area HVPE-GaN films directly on sapphire substrates by Monemar et al. Reference Monemar, Lagerstedt and Gislason[94].

In Table III, reports of GaN growth on sapphire substrates by HVPE are listed with respect to the relationship between initial growth conditions and surface morphology. The first smooth HVPE-GaN was achieved by Naniwae et al. in 1990 Reference Naniwae, Itoh, Amano, Itoh, Hiramatsu and Akasaki[95]. The key technology was the treatment of the sapphire substrate by Ga+HCl at 1030°C. In order to obtain smooth GaN, Ga+HCl treatment should be carried out for more than 20min, as shown in the Table. Two years later, Detchprochm et al. found that the insertion of a ZnO layer on the sapphire substrate made it possible to grow HVPE-GaN that was transparent, with a smooth surface Reference Detchprochm, Hiramatsu, Amano and Akasaki[96]. A 10-300nm thick ZnO buffer layer was deposited on the sapphire substrate by sputtering at room temperature, and then the sapphire substrate that was covered with the ZnO layer was introduced into the HVPE system.

The vapor pressure of ZnO is so high at 1000°C in an HVPE reactor that there was no evidence of ZnO at the GaN/sapphire interface, as reported by Molnar et al. Reference Molnar, Götz, Romano and Johnson[97]. Ga atoms left on the sapphire by the GaCl treatment should be desorbed as well. However, it is supposed that a very small fraction of the Zn or the Ga could form compositions that would play a role in converting the polarity at the interface of the sapphire.

3.3.2. 1997-1999: AlN and GaN buffer layers

AlN and GaN buffer layers began to be used in 1997. The effectiveness of AlN buffer layers was confirmed by Lee et al. for the first time Reference Lee, Yuri, Ueda, Harris and Sin[98]. Paskova et al. made a comparison of the effects of sapphire nitridation, GaCl treatment and AlN buffer layers for HVPE-GaN growth Reference Paskova, Birch, Tungasmita, Becard, Heuken, Svedberg, Runesson, Goldys and Monemar[99]. (The growth of AlN buffer layers was carried out by sputtering in a separate apparatus due to the AlCl₃ corroding the quartz tube of the HVPE reactor). It was concluded that the smoothest GaN was achieved by the use of an AlN buffer layer.

Wagner et al. reported that HVPE-GaN grown on a GaN buffer layer was smoother than that on an AlN buffer layer deposited by MOCVD Reference Wagner, Parillaud, Buhlmann and Ilegems[100]. Here, it is worth noting that HVPE-GaN on an LT-GaN buffer layer less than 10nm thick had a hexagonal-facetted surface (discussed in Sec. 4.3 Recipe 1-(2)). The GaN buffer layer could be desorbed during the ramping of the substrate temperature under an NH₃ ambient, and then the unintentional direct-growth of HVPE-GaN on the sapphire substrate was supposed to take place. In contrast, smooth HVPE-GaN was deposited on AlN buffer layers. This is why AlN buffer layers, which have a lower desorption rate (as shown in Figure 8) were expected to work as nucleation layers even at the high temperatures used for HVPE-GaN growth.

The importance of variations in the LT-buffer layer caused by annealing can be found in the report by Tavernier et al., who applied a similar buffer layer technology to that used in MOCVD to HVPE in a single chamber Reference Tavernier, Etzkorn, Wang and Clarke[101]. Their studies of buffer layers revealed that layer thickness and annealing conditions are crucial to obtaining HVPE-GaN of high quality.

3.3.3. 2000-2003: Kinetic effect and a new approach

Small islands that were typically several nm in height were formed on sapphire substrates by the nitridation treatments used in HVPE Reference Gu, Zhang, Shi, Zheng, Zhang, Dwikusuma and Kuech[102] and in MOCVD. Nitridation of the substrate is likely to result in −c GaN growth with hexagonal-facetted surfaces, as discussed in the section on MBE-GaN. However, such a trend is not observed in HVPE-GaN. Namerikawa et al. reported that +c polarity was observed for HVPE-GaN samples deposited on both A- and B- face GaAs (111) substrates Reference Namerikawa, Sato, Takahashi, Suemasu and Hasegawa[103]. This is probably due to the use of growth rates as high as a few μm/min. (Murakami et al. found that when an LT-GaN buffer layer on a B-face substrate was annealed in an NH₃ ambient, the substrate was deteriorated due to etching of the buffer layer, suggesting that GaAs (111) B-face substrates are not suitable for GaN deposition Reference Murakami, Kumagai, Seki and Koukitsu[104].)

In HVPE-GaN, the coalescence of the GaN islands occurs rapidly (within 10 sec of their growth) corresponding to 0.4-0.5μm of film thickness Reference Golan, Wu, Speck, Vaudo and Phanse[105]. The quality of the HVPE-GaN can be divided into two regimes, thinner than 0.4-0.5μm and thicker than that, which correspond to destructive and better quality, respectively. Gu et al. obtained better quality HVPE-GaN by changing the optimum conditions for the two regions Reference Gu, Zhang, Shi, Zheng, Zhang, Dwikusuma and Kuech[102]. In addition, in a similar way to the case of MEE Reference Kikuchi, Yamada, Kusakabe, Sugihara, Nakamura and Kishino[69] given in Sec. 3.1, flow modulated growth (FMG) with a periodically interrupted HCl flow under a constant flow of NH₃ can also improve HVPE-GaN, as reported by Zhang et al. Reference Zhang, Riemann, Alves, Heuken, Meister, Kriegseis, Hofmann, Christen, Krost and Meyer[106].

3.3.4. Features for controlling polarity in HVPE-GaN

Nitridation of the sapphire substrates in HVPE does not have a great influence on the surface morphology (polarity) of the GaN, probably due to the high growth rate induced by the kinetic effect. However, the interface compositions between the substrate and the initial very thin film are crucial to +c polarity, as provided in the following:

1. 1. GaCl treatment of the sapphire substrate

2. 2. Deposition of ZnO and AlN layers in an isolated system

3. 3. Use of a thicker LT-GaN buffer layer.

These features are supposed to be essentially equal to those used in MBE. In order to understand the general context of polarity control beyond these growth techniques, it will be necessary to be able to analyze a sapphire surface treated with GaCl and also these buffer layers.

4.3.1. Article 1: Surface of the treated sapphire substrate

It is most important to recognize that the surface of the substrate is the first hetero-interface in GaN film growth. The surface of the sapphire substrate was investigated using XPS when the substrate was treated by cleaning under H₂ or by nitridation under flowing NH₃.

Although the surface of the sapphire substrate was sometimes covered with undesirable contamination (Ga, N, Al, Si etc.) due to hysteresis in the MOCVD apparatus Reference Golan, Fini, DenBaars and Speck[116], oxygen was conventionally removed from the sapphire surface during H₂ cleaning. Consequently, the surface was slightly rougher, and an Al-rich surface was formed, which was confirmed to be Al:O = 50:50% by XPS. When it was subsequently nitrided under flowing NH₃ at various temperatures between 600°C and 1080°C, the surface compositions of the Al, O and N were changed, as shown in Fig. 10 . Nitrogen was detected for sapphire nitrided at even 600°C, and the nitrogen composition increased with higher temperature, while the oxygen decreased.

Thus, an Al-rich surface was formed on the sapphire substrate by the removal of oxygen during H₂ cleaning. In contrast, AlO_xN_1−x was induced by nitridation, depending on the temperature used. These chemical states at the surface of the substrate play a decisive role in the polar structure of the buffer layer and also the evaporation behavior. (Refer to Recipes 1 and 2).

4.3.2. Article 2: Structure of the LT-GaN buffer layer on the nitrided sapphire substrate

To understand the correlation between the MOCVD-process and the polarity, it is necessary to evaluate the polarity of the LT-buffer layer itself. Figure 11 shows CAICISS results for LT-GaN buffer layers on H₂-cleaned and nitrided sapphire substrates Reference Sumiya, Tanaka, Ohtsuka, Fuke, Ohnishi, Ohkubo, Yoshimoto, Koinuma and Kawasaki[36] Reference Sumiya, Ohnishi, Tanaka, Ohtomo, Kawasaki, Yoshimoto, Koinuma, Ohtsuka and Fuke[117]. The dependence of the intensity on the incident angle shows less variation due to the poor crystalline quality of the buffer layer (refer to Sec. 2.4). However, the dominant polarity of buffer layers as thin as 20nm can be judged from these results by comparison with the results in Figure 5. There are two interesting features observed from this result. Firstly, both as-deposited buffer layers show +c polarity, independent of the substrate treatment. The other is that the resulting polarity for the layer on the nitrided sapphire was converted to −c polarity after the annealing, while the polarity was +c for the film on the H₂ cleaned substrate.

Figure 11. Angular dependence of Ga signal intensity for the buffer layers in CAICISS analysis: (a) as-deposited 20-nm buffer layer on non-nitrided sapphire, (b) as-deposited 20-nm buffer layer on nitirided sapphire and (C) annealed buffer layer of (b). The polarity of a buffer layers on nitrided sapphire substrates changed from +c to −c polarity, while there was no change for buffer layers on non-nitrided sapphire substrates. [after Reference Sumiya, Ohnishi, Teshigawara, Tanaka, Ohkubo, Kawasaki, Yoshimoto, Ohtsuka, Koinuma and FukeRef. 35]. This is the first determination of the polarity for LT-GaN buffer layers, which has led to an understanding of the correlation between the MOCVD process and the polarity.

Our concern was focused on buffer layers on nitrided sapphire. The non-stoichiometric AlO_xN_1−x layer formed by the nitridation is likely to have −c polarity, as indicated by theoretical calculations Reference Di Felice and Northrup[58]. Both +c and −c polarity nucleate simultaneously on the nitrided sapphire, probably due to either inhomogeneous nitridation Reference Sumiya, Ohnishi, Tanaka, Ohtomo, Kawasaki, Yoshimoto, Koinuma, Ohtsuka and Fuke[117] or favorable formation energy. The surface of the as-deposited buffer layer would be covered with a +c layer with higher growth rate Reference Yamane, Shimada, Endo and DiSalvo[118]. Subsequent annealing of the film would make the −c domains rise to the surface due to the sublimation of the film. Based on these considerations, buffer layers on nitrided sapphire substrate are assumed to be covered with a +c layer grown laterally over the −c domains.

To confirm this assumption, thicker buffer layers (210nm) were prepared on nitrided sapphire and annealed for various times at 1040°C. An LT-GaN buffer layer on a nitrided substrate was found to evaporate with a layer-by-layer mode due to the AlO_xN_1−x on its surface Reference Sumiya, Ogusu, Yotsuda, Itoh, Fuke, Nakamura, Mochizuki, Sano, Kamiyama, Amano and Akasaki[119]. Figure 12 shows the angular dependence of the Ga signal intensity in CAICISS for the thicker buffer layer annealed for 0, 10, 20 and 30min Reference Sumiya, Yoshimura, Ogusu, Fuke, Mizuno, Yoshimoto and Romano[120]. The CAICISS result for the as-deposited sample shows predominantly +c polarity [Figure 12 (a)]. The sample continues to have +c polarity after annealing for 10min. Sharpening of the peaks is also observed in the result, suggesting an improvement in the crystal quality near the surface [Figure 12 (b)]. The peak at 72° splits into two peaks and the peak at 35° intensifies with further annealing, as shown in Figs. 12 (c) and (d). This indicates that the film surface is transforming from +c polarity (triangles) to −c polarity (squares). The lines in Figs. 12 (c) and (d) present the weight ratio of +c: −c polarity at 5:5 and 2:8, respectively, with the assumption that they share the same crystal quality (i.e., the same intensity of the CAICISSS signal for +c and −c domains).

Figure 12. Incident angular dependence of Ga signal intensity at [11

0] azimuth of He⁺ beam in CAICISS analysis for a 210nm GaN buffer layer on nitrided sapphire: (a) as-deposited, (b)10min, (c) 20min, and (d) 30min annealing time. The lines in (c) and (d) are calculated according to the weight ratios of the +c:−c polarity material as being 5:5 and 2:8, respectively, assuming that they are of the same crystal quality i.e., the same intensity of CAICISS signal for +c and −c domains. The sharpening peak in (b) indicates crystallization, and the peak splitting at 72° in (c) suggests the existence of −c domains that are becoming exposed due to the evaporation of the buffer layer. [after Reference Yamane, Shimada, Endo and DiSalvoRef. 118]

Since CAICISS analysis detects the atomic arrangement of only the surface region (as discussed in Sec. 2.4), the films were further investigated by TEM Reference Sumiya, Yoshimura, Ogusu, Fuke, Mizuno, Yoshimoto and Romano[120]. Figure 13 shows X-TEM images for the same samples that are shown in Figs. 12 (c) and 12 (d). Columnar IDs were found to extend to the surface, and dome-shaped domains were found near the interface, similar to the observations reported by Wu et al. Reference Wu, Brown, Kapolnek, Keller, Keller, DenBaars and Speck[121]. With increasing annealing time (sublimation of the film), the dome-shaped domains are exposed to the surface. The TEM images are consistent with the CAICISS spectra. Since the −c signal component of the CAICISS result increased after annealing, the inverted domains (dome shaped) can be considered to have −c polarity.

Figure 13. Cross-sectional TEM images for the buffer layers annealed for (a) 20min and (b) 30min. The samples correspond to (c) and (d) in Figure 12, respectively. [after Reference Yamane, Shimada, Endo and DiSalvoRef. 118]

4.3.3. Article 3: Interface between the GaN buffer and the treated sapphire substrate

The interface between the GaN buffer layer and the substrate was investigated by XPS. Figure 14 shows the annealing-time dependence of the peak position of N1s for buffer layers on both H₂ cleaned and nitrided sapphire substrates. The peak positions for both as-deposited layers are observed at around 397.4 eV, corresponding to GaN. The sample on nitrided substrate mostly evaporated after annealing for 20min, because the position of N 1s at 396.5eV, corresponding to the AlN and Ga signals, was in the noise level. In contrast, Al_xGa_1−xN was formed at the interface between the GaN buffer layer and the H₂ cleaned sapphire substrate. The peak position for the annealed buffer layer on the H₂ cleaned sapphire stayed at 396.8eV, regardless of the time, and the Ga 3d peak was shifted to slightly lower binding energy from GaN (19.8eV) Reference Sumiya, Ogusu, Yotsuda, Itoh, Fuke, Nakamura, Mochizuki, Sano, Kamiyama, Amano and Akasaki[119].

Figure 14. Dependence of (a) N1s and (b) Ga 3d peak positions for a 20-nm-thickness GaN buffer layer on H₂ cleaned (open circles) and nitrided (closed circles) sapphire substrates on the annealing time under the N₂ and H₂ mixed ambient. The positions were detected by XPS analysis. [after Reference Sumiya, Ohnishi, Tanaka, Ohtomo, Kawasaki, Yoshimoto, Koinuma, Ohtsuka and FukeRef. 117] The sample on the nitrided substrate evaporated completely, and AlGaN was formed at the interface between the sample and the non-nitrided substrate.

Recently, we have deposited a GaN film by MOCVD on an H₂ cleaned sapphire substrate that had once been exposed to air. Although the surface should have been terminated with oxygen, a +c GaN film with sufficient quality was obtained on the sapphire by two-step MOCVD without the need for a second H₂ cleaning process [122]. The sapphire surface becomes rougher by H₂ cleaning, represented by the weak RHEED pattern. In addition, a comparable GaN film with high quality was even grown on a sapphire substrate cleaned in an N₂ ambient at more than 1000°C Reference Sato, Ogawa, Ohtsuka, Kuwahara, Sumiya, Takano and Fuke[123]. It is supposed from these facts that the thermal roughening of sapphire at higher temperature might be important as a nucleation site for the growth of the LT-buffer layer.

4.3.4. Recipe 1: Thickness and annealing of the GaN buffer layer on nitrided substrates

Taking into account the IDs structure of the GaN buffer layer on the nitrided sapphire reported in Article 2, we demonstrate how to control the polarity of the HT-GaN by changing the thickness of the buffer layers and the conditions used for the annealing. The mixed polarity of the buffer layers grown by MBE on nitrided sapphire has also been observed by CAICISS analysis Reference Sonoda, Shimizu, Suzuki, Balakrishnan, Sirakashi and Okumura[59]. Since GaN film growth on nitrided sapphire substrates is supposed to be similar to growth by MBE, our recipes in MOCVD will be compared with the features found in MBE, which are listed in Table II (a) and Table II (b).

4.3.5. Recipe 2: Growth on H2 cleaned substrates

H₂ cleaning of sapphire substrates is indispensable for the growth of GaN with a smooth surface by two-step MOCVD. GaN films and GaN buffer layers on this type of substrate have +c polarity. The basis of the recipe for preventing growth with −c polarity on H₂ cleaned sapphire is in preventing the substrate from undergoing unintentional nitridation.

It was confirmed that nitridation using NH₃ gas at temperatures higher than 700°C resulted in −c GaN films in our deposition system. In addition, the introduction of a flow of NH₃ into the reactor, even for several seconds at 1080°C, resulted in the growth of −c GaN films. Therefore, unintentional nitridation probably takes place in the following circumstances; 1) when NH₃ is introduced into the reactor during the decrease in substrate temperature for the deposition of the LT-buffer layer after H₂ cleaning of the sapphire, and 2) when a longer time is used and annealing is carried out in an H₂ ambient for a thin LT buffer layer (exposure of the sapphire surface). These factors indicate the importance of correct timing of the switching of the source gases when shifting to a subsequent part of the growth process.

Seelmann-Eggebert et al. have already pointed out that the occurrence of inversion domains in the films could mainly be attributed to poor process control during substrate cleaning and in the very initial stages of the nucleation process, preceding buffer growth Reference Seelmann-Eggebert, Weyher, Obloh, Zimmermann, Rar and Porowski[6]. It is supposed that the GaN with the hexagonal facets on a thin buffer layer that was reported in Ref. Reference Nakamura[3] might originate from the unintentional nitridation of the sapphire substrate.

4.5.1. Article 5: Correlation between the polarity and the growth process

The polarity of HT-GaN was found to be attributable to that of the annealed buffer layer from the observations in the previous sections. During re-growth on a GaN template, the polarity could also attributed to that of the template, as recognized by Weyher et al. Reference Weyher, Brown, Zauner, Muller, Boothroyd, Foord, Hageman, Humphreys, Larsen, Grezegory and Porowski[128]. In our deposition system, the deposition rate of MOCVD-GaN films is limited by the arrival rate of the TMG source gas. Films are grown with changing deposition rates, ranging from 0.9 to 5.0 μm/h. The polarities of these MOCVD-GaN films were identical with those of the annealed buffer layer on nitrided sapphire substrates Reference Sumiya, Yoshimura, Ito, Ohtsuka, Fuke, Mizuno, Yoshimoto, Koinuma, Ohtomo and Kawasaki[129]. This is inconsistent with MBE-GaN in terms of the kinetic effect highlighted in Sec. 3.1.3. Although the range of conditions used for the HT-GaN might be insufficient, we insist here that the polar structure at the interface of one of the underlying layers (the annealed LT-buffer layer or the template) would be the most crucial factor in determining the polarity of GaN films in MOCVD.

We have investigated the implications in the exact growth sequence and conditions in MOCVD. When summarizing them in Articles 1-5 and Recipes 1-3with respect to the polarity, the relationship between the growth conditions and the polarity can be represented as a branching road map on a timing chart of the MOCVD process, as shown in Figure 21 Reference Sumiya and Fuke[130]. The route is divided by the nitridation of the sapphire substrate. In order to achieve +c GaN in MOCVD, the sapphire substrate must not be nitrided after the H₂ cleaning. In the case where nitridation occurs, however, the polarity can still be controlled by modifying the preparation of the LT-buffer layer.

Figure 21. Summary of the key points for controlling the polarity of GaN films on sapphire substrates (as detailed in the Articles and the Recipes) in the time chart of the two-step MOCVD process. Our typical growth conditions are represented. The polarity at the end of each process is remarked with blue and red indicating +c and −c polarity, respectively, which can be used as the road map to control the polarity of GaN in MOCVD. [after Reference Weyher, Brown, Zauner, Muller, Boothroyd, Foord, Hageman, Humphreys, Larsen, Grezegory and PorowskiRef. 128]

4.5.2. Recipe 4: Application of the road map

The road map can be examined in several ways, not just for controlling the polarity, but also for depositing GaN films on various substrates. When GaN films are deposited on Si-face 6H-SiC substrates in MOCVD, it is well known that LT-AlN buffer layers should be used. When an LT-GaN buffer layer was used, part of the SiC surface was exposed during the time that the temperature was ramping because of the absence of a non-volatile intermediate layer (such as AlGaN) at the interface of the sapphire substrates (Article 3). To overcome this unsuitable feature of using an LT-GaN buffer layer on a SiC substrate, it was annealed in an N₂ ambient to suppress evaporation during the ramping of the temperature (Recipe 1(3)). As expected, a GaN film with a smooth surface and of better quality was obtained on the SiC substrate using the LT-GaN buffer layer Reference Kurumasa, Ito, Ohtsuka, Kuwahara, Sumiya, Takano and Fuke[131].

By modifying the timing of the introduction of the source gases according to Recipe 2, growth of GaN was achieved on a Si (111) substrate using only an AlN buffer layer. The key point of this was that the TMA and the NH₃ gas should arrive simultaneously at the Si substrate in order to prevent the Al and N from alloying with the Si Reference Matsuura, Ochi, Sumiya and Fuke[132].

GaN films were deposited on (La_0.29, Sr_0.71)(Al_0.65, Ta_0.35) O₃ (LSAT) (111) substrates, which have a lattice constant that corresponds to the 3×3 structure of GaN (0001) and a thermal expansion coefficient close to that of GaN. Since the LSAT substrate was deteriorated by NH₃ and TMG gases at high temperature, an AlN layer was used as a blocking layer to protect the surface. The GaN film on the LSAT had +c polarity and its GaN [1

00] // LSAT [10] orientation was rotated in-plane by 30° against the expected orientation (GaN [2 0] // LSAT [10]) Reference Sumiya, Chikyow, Sasahara, Yoshimura, Ohta, Fujioka, Tagaya, Ikeya, Koinuma and Fuke[133]. This was probably caused by the bond configuration of the surface of the LSAT substrate. In addition, GaN films on metal-face and O-face LiGaO₂ (001) substrates had +c and −c polarity, respectively Reference Matsuoka and Ishi[134].