Introduction

Although there is a large lattice mismatch of about 16% between GaN and sapphire substrate, device-quality GaN has been achieved with use of LT-buffer layer [Reference Amano, Sawaki, Akasaki and Toyoda1]. It was soon followed by the success of control of n-type conductivity by Si-doping [Reference Amano and Akasaki:2], and achievement of p-type nitride films using Mg as a dopant [Reference Amano, Kitoh, Hiramatsu and Akasaki3] and the following special treatment [Reference Amano, Kito, Hiramatsu and Akasaki4]. These successes have led to the commercialization of nitride-based near-UV, blue and green light-emitting diodes and violet laser diodes.

There are another new field waiting for us. Optical devices in near-UV to vacuum-UV region are one of the most attractive target. The applications of which are chemical sensing [Reference Han, Crawf, Shul, Hearne, Chason, Figiel and Banas5], flame detection [Reference Pernot, Hirano, Amano and Akasaki6,Reference Pernot, Hirano, Iwaya, Detchprohm, Amano and Akasaki7], ozone-hole sensing, remote sensing, high-density optical storage, excitation source for phosphors, fine lithography, etc. In order to realize such device applications, thick Al _x Ga_1−x N films with high-AlN molar fraction x and high-crystalline quality are essential.

There are several reports concerning the crystalline quality of Al _x Ga _1−x N films. Koide et al. reported that the crystalline quality of Al _x Ga _1−x N on a sapphire substrate covered with an LT-AlN buffer layer [Reference Koide, Itoh, Itoh, Sawaki and Akasaki8] is much improved in comparison with that of Al _x Ga _1−x N directly grown on sapphire. It is also reported that its crystalline quality progressively worsens with increasing x reported by Itoh [Reference Itoh9]. He also reported that the crystalline quality of Al _x Ga _1−x N could be significantly improved by growing it on a GaN. At the same time, however, crack network generated with high-density if Al _x Ga _1−x N exceeded its critical thickness. Therefore, it had been quite difficult to achieve crack-free, high-quality and thick Al _x Ga _1−x N layers with a high AlN molar fraction.

Very recently, we reported that insertion of LT-deposited layer between HT-GaN reduces threading dislocation density [Reference Iwaya, Takeuchi, Yamaguchi, Wetzel, Amano and Akasaki:10,Reference Amano, Iwaya, Kashima, Katsuragawa, Akasaki, Han, Hearne, Floro, Chason and Figiel11]. LT-deposited layer inserted between HT-GaN is called the ”LT-interlayer” are GaN film with the dislocation density as low as 5×10⁶ cm⁻² has been achieved. This LT-interlayer technique was applied to grow thick Al _x Ga _1−x N. By using this technique, high quality in terms of narrow X-ray diffraction profile and crack-free Al _x Ga _1−x N with a whole compositional range was achieved [Reference Amano, Iwaya, Hayashi, Kashima, Katsuragawa, Takeuchi, Wetzel and Akasaki12-Reference Kashima, Nakamura, Iwaya, Kato, Yamaguchi, Amano and Akasaki14]. The purpose of this work is to understand the relationship between the microscopic structure and the performance of the photodetectors based on these Al _x Ga _1−x N layers. For the application of flame sensing, the device should be blind to wavelength longer than 280 nm and have sensitivity on the order of 1 nW/cm² for shorter wavelength in order to detect only the flame luminescence. Furthermore, the photocurrent to dark current (PC/DC) ratio should be as large as possible in order to prevent misdetection when the dark current level varies due, for instance, to temperature fluctuation. A response speed on the order of a few ms is sufficient for safety application. Although several groups [Reference Muñoz, Garrido, Izpura, Sánchez, Sánchez-Garcia, Calleja, Beaumont and Gibart15,Reference Binet, Duboz, Rosencher, Scholtz and Härlle16] have reported on the evolution of photocurrent with optical power, to our knowledge, there had been no discussion on Al _x Ga _1−x N based UV detector operated under low illumination intensity (< 1 μW/cm²) with high PC/DC-ratio. The metal-semiconductor-metal (MSM) structure, with fabrication simplicity and need for only a single active layer, is useful tool for characterizing the quality of the detection layer. The results obtained by using LT-interlayered-Al _x Ga _1−x N are very promising for the development of UV high sensitivity detectors.

Experiment

Unintentionally doped Al _x Ga _1−x N films with were grown by organometallic vapor-phase epitaxy (OMVPE) at pressures around 200 hPa. Trimethylaluminum, trimethylgallium and ammonia were used as source gases. (0001) c-plane sapphire was used as the substrate. Figures 1(a) - 1(e) schematically show the structure of each sample. In all the growth, we used LT-buffer layer between nitrides and sapphire. Thickness of Al _x Ga _1−x N was fixed around 1 μm. Samples A was single Al_0.43Ga_0.57N layer, which is the same as that reported by Koide et al. Sample B was Al_0.43Ga_0.57N grown on HT-GaN, which is the same as that reported by Itoh et al. Sample C and D were Al_0.43Ga_0.57N and GaN grown on HT-GaN, but the “LT-AlN interlayer” was inserted between them. Sample E was single HT-GaN. This reference sample is the underlying layer in samples B, C and D. The alloy composition x were precisely determined high-resolution X-ray diffraction. 2 θ/ω scans of the (0002) and (20-24) diffractions were used to determine the lattice constants c and a at room temperature. For details, see Reference Amano, Sota, Takeuchi, Kobayashi, Akasaki, Burm, Schaff, Floro and Eastmanref. 17.

Fig. 1 Structures of the samples.

Plan-view and cross-sectional transmission electron microscopy (TEM) observations were carried out using a HITACHI H-9000NAR TEM system at an acceleration voltage of 300 kV. Ar⁺-ion milling and focused Ga⁺-ion beam milling were used for plan-view and cross-sectional TEM samples preparation.

The photodetector structure consists of interdigitated electrodes occupying an area of 1 mm². The fingers are 10 μm wide with 10 μm spacing. Using a conventional lift-off process, Ti/Au contacts were deposited by electron-beam and thermal evaporation, respectively.

Results and Discussion

Figures 2(a) through 2(e) show surface SEM micrographs of five samples, respectively. In these five samples, cracks were formed only in sample B. It is identical to the results of Itoh [Reference Itoh9]. Although cracks were not generated in sample A, atomic force microscopy showed that the surface was quite rough with a root mean square (RMS) roughness of 12.4 nm. In the image of sample C, which was Al_0.43Ga_0.57N grown on “LT-AlN interlayer”, no cracks are observed. The surface is much smoother than sample A, with the RMS roughness of about 0.4 nm. Both samples D and E are crack-free and have smooth surface.

Fig. 2 SEM micrographs of various samples

The mechanism of crack formation was studied by stress observation during growth using multi-beam stress sensor system technique [Reference Chason, Lee, Twesten, Hwang and Freund18]. In case of high-quality crystal, it is found that steep relief of the tensile stress, or in other words crack generation, occurred if thickness*tensile stress product exceeds 0.8 GPa*μm. The tensile stress during growth of Al _0.43 Ga _0.57 N on “LT-AlN interlayer” shown as sample C is 0.05 GPa. Therefore, critical thickness is much thicker than 1 μm. In case of Al _0.43 Ga _0.57 N grown on GaN shown as sample B, we could not observe the clear strain relief during growth because the critical thickness is too thin. Details was discussed elsewhere [Reference Iwaya, Terao, Hayashi, Kashima, Amano and Akasaki13].

The crystalline quality of two types of crack-free Al_xGa_1−xN, that is, the same structure of sample A and sample C, is compared. In case Al_xGa_1−xN was grown on sapphire using LT-buffer, when the same structure of sample A, the FWHM becomes wider with increasing x, which means the tilting component of the mosaicity increases rapidly with increasing AlN molar fraction. The same tendency has already been reported by Itoh [Reference Itoh9]. In constant, when Al_xGa_1−xN was grown on “LT-AlN interlyaer”, the FWHM of XRC remained unchanged over the entire compositional range. This clearly indicates that the crystalline quality of Al _x Ga _1−x N film grown on GaN covered with “LT-AlN interlayer” is superior to that grown on sapphire covered with the LT-buffer layer.

Further characterization of the crystalline quality was carried out using TEM. Figures 3(a) and 3(b) show the plan-view TEM images of sample A and C, respectively. Grain boundaries of about 50-250 nm in size were observed in Fig. 3(a). In comparison, such grain boundaries could not be observed in Fig. 3(b). The structure of Al_0.43Ga_0.57N shown in Fig. 3(b) is quite similar to that of the underlying GaN except for the density of threading dislocations. When the plan-view image shown in Fig. 3(b) was taken, the sample was slightly tilted from the [0001] axis in order to identify the type of threading dislocations. Using this observation technique, pure-edge and mixed dislocations exhibit strong contrast and can be easily distinguished. For details, see ref. 15. The magnified image in Fig. 3(b) shows that almost all the threading dislocations are pure-edge type ones. The densities of screw and mixed type dislocations are quite low. These results are further confirmed by cross-sectional dark-field image observations. The majority of threading dislocations is in contrast with g =[Reference Amano, Sawaki, Akasaki and Toyoda1-100] and out of contrast with g =[000-Reference Amano and Akasaki:2]. From this result, it is concluded that the majority of threading dislocations in the top Al _0.43 Ga _0.57 N layer has Burger’s vector b =1/3[Reference Amano, Iwaya, Kashima, Katsuragawa, Akasaki, Han, Hearne, Floro, Chason and Figiel11-20]. Three types of threading dislocations propagate to the c-direction in hexagonal GaN were reported, that is, screw-type with Burger’s vector b =[0001], edge-type with b =1/3[Reference Amano, Iwaya, Kashima, Katsuragawa, Akasaki, Han, Hearne, Floro, Chason and Figiel11-20] and mixed-type with b =1/3[Reference Amano, Iwaya, Kashima, Katsuragawa, Akasaki, Han, Hearne, Floro, Chason and Figiel11-23]. The results confirmed that the majority of threading dislocations in the uppermost Al _0.43 Ga _0.57 N layer is pure-edge-type dislocations. Fig. 4 shows the compositional dependence of the density of threading dislocations in the same structure of sample A and sample C [Reference Kashima, Nakamura, Iwaya, Kato, Yamaguchi, Amano and Akasaki14].

Fig. 3 Bright-field plan-view TEM image of the top surface

Fig. 4 Compositional dependence of the density of threading dislocations. Solid and dotted lines provide guides for the eye.

The effects of each type of threading dislocations on the electrical and optical properties have not been clarified yet. Table I summarizes the density of each threading dislocations in these samples. Compared to the underlying GaN layer, Al _0.43 Ga _0.57 N grown on “LT-AlN interlyaer” contains higher density of edge type dislocations, although pure screw type and mixed type dislocation is much reduced. In this study, we fabricated and characterized photodetectors based on these four crack-free samples (samples A, C, D and E). Measurements were carried out at room temperature under a bias voltage of 10 V with and without illumination from mercury lamp (λ = 254 nm). The dark current for each sample is listed in Table I. Due to high dark current level, PC/DC of sample A was less than one even for 100 μW/cm². The other samples showed good uniformity from one detector another, and act as high sensitivity UV detectors with PC/DC greater than one even at very low weak illumination of 40 nW/cm². Using the mercury lamp at power density of 10 μW/cm² and mechanical shutter, we measured the photoconductive build-up and decay time under 10 V bias at room temperature. Before each experiment, the samples were kept in the dark for at least 12 hours. After the dark current reached a quasi-steady-state value, the excitation light was turned on for 2 hours. When starting measurement, we observed a transient response time as the time needed for the photoresponse signal to drop to 1 % of its maximum value (Fig. 5), compared to the 83 s needed for sample E of HT-GaN grown on LT-AlN buffer layer, samples grown on “LT-AlN interlayer” present much faster decay with response time of 1.6±0.05 s and 2.0±0.1 s, for sample D and C respectively. For sample A, the response time exceeds the measurement time. [Reference Pernot, Hirano, Iwaya, Detchprohm, Amano and Akasaki7].

Fig. 5. The normalized photocurrent as a function of time shows a great improvement of the response speed for LT-AlN interlayer.

From these results, it is expected that pure-screw-type and mixed-type threading dislocations are electrically active and relates to deep traps, while pure-edge type dislocation is not so much.

Conclusions

The relationship between microscopic structure of Al _x Ga _1−x N and the UV photoconductive properties have been clarified. The “LT-AlN interlayer” is found to act as the dislocation filter, especially for dislocations of which Burger’s vector contains [0001] components. The fabricated MSM detectors show a very low dark-current level, below 50 fA at 10 V. These results show that “LT-interlayered-Al _x Ga _1−x N” are very promising for the development of optical devices in the near-UV and vacuum UV region.