1. Introduction
Recently, deep ultraviolet light-emitting diodes (deep UV-LEDs) have attracted considerable attention for being widely used in many commercial products, such as water purification, sterilization, communication, sensing, UV-curing, medical and lithography[1]. High-quality AlN underlying layer with low-density threading dislocations is indispensable for high-efficiency deep UV-LEDs[2]. Conventionally, for lack of bulk AlN substrate[3], AlN layers are grown on sapphire by metalorganic vapor phase epitaxy (MOVPE), but the large lattice mismatch and thermal expansion mismatch[4] between the AlN layer and the sapphire substrate induce high-density threading dislocations during AlN layer growth. High density of threading dislocations degrades the internal quantum efficiency of the active layer of the multiple quantum wells (MQW). This degradation limits the light output efficiency of deep UV-LEDs. In order to reduce the density of threading dislocations, many research studies focus on growth temperature[5], V/III ratio[6] and chamber pressure in reactors[7]. Additionally, some techniques have been developed, such as nucleation layers (NL)[8], pulsed ammonia[9] and epitaxial lateral overgrowth (ELO)[10]. Nevertheless, these techniques are not enough to put high-performance deep UV-LEDs into commercial production of high throughput at low cost. At present, only the low-yield MOVPE system has the ability to produce high-performance commercial deep UV-LEDs. If the high-yield industrial MOVPE was able to produce commercial deep UV-LEDs, the production cost would be greatly reduced. Hence it is necessary to find a process to produce commercial deep UV-LEDs by high-yield industrial MOVPE.
Generally, reactors of high-yield industrial MOVPE devices are larger than that of low-yield MOVPE devices, and it is difficult to control the uniformity of temperature and steady the gas flow fields in huge reactors. However, the growth rate of AlN is very sensitive to temperature especially in the growth process of nucleation layer. The difference of temperature would result in the difference of thickness of AlN layer. In addition, the different mobility of Al atom at different temperatures would result in the difference of nucleation island size[11]. Furthermore, the pre-reaction of Al atoms is serious in the MOVPE growth process, especially in huge reactors. The harmful pre-reaction not only consumes Al atoms, but also produces AlN particles which maybe fall on the surface of wafer and form crystal nucleus[12–14]. Hence, it is difficult to grow high-quality nucleation layer (NL) by high-throughput industrial MOVPE devices. However, the AlN epitaxial layer is sensitive to the quality of NL[8]. Therefore, it is difficult to grow high-quality AlN epitaxial layers by high-yield MOVPE. However, high-quality AlN layers with low-density threading dislocations are indispensable for high-efficiency deep ultraviolet LEDs. It has been reported that ex-situ reactive plasma deposited (RPD) AlN nucleation layer[15] and ex-situ sputtered AlN nucleation layer[16–18] can improve the material quality of GaN epitaxial layers. In our study, we demonstrate that the sputtered nucleation layer (SNL), which is deposited on sapphire substrate by the reactive magnetron sputtering system, can greatly improve the material quality of AlN epitaxial layer that was grown on SNL by the industrial MOVPE. Utilizing a sputtered AlN nucleation layer on the sapphire substrate to enhance the quality of the epitaxial layer, commercial deep UV-LEDs at 282 nm were successfully grown on sapphire substrate by the high-yield industrial MOVPE system.
2. Experiment
In this work, two systems were used for growing AlN layers. One was an iTops A230 AlN sputtering system, which was a 19 × 2" reactive magnetron sputtering reactor and designed by Beijing NMC Co., Ltd. Another system was a homemade 48 × 2" MOVPE reactor. Both systems were high-yield industrial equipment. Trimethylaluminum (TMAl) and ammonia (NH 3) were used as aluminum and nitrogen precursors, respectively, while hydrogen (H2) was used as carrier gas. All samples were grown on 2 in c-plane sapphire substrates (0.2° ± 0.01° miscut towardsm-plane).
We prepared two samples to study the effect of SNL on material quality of AlN epitaxial layer. The first sample was grown by MOVPE under a general two-step process and optimized conditions. The first step, after thermal cleaning of the sapphire substrate at 1120 °C in hydrogen flow for 5 min, a 25-nm-thick low temperature AlN nucleation layer (in-situ NL) was directly deposited on sapphire at 630 °C with relatively high V/III ratio (1000) and reactor pressure keeping at 50 Torr. The second step, a high-temperature AlN epitaxial layer of 900 nm thickness was grown by MOVPE at 1250 °C under optimized condition. For the second sample, firstly, a 25-nm-thick AlN layer was directly deposited on sapphire by reactive magnetron sputtering, which was acting as a nucleation layer (simply calling SNL). Then, a 900-nm-thick high temperature AlN epitaxial layer was grown by MOVPE at 1250 °C under the same condition as the high-temperature AlN epitaxial layer of the first sample. In addition, another two samples were added for investigating the effect of different thickness of SNL on the crystal quality of AlN layer. The thickness of SNL of these two samples was 50 and 100 nm, respectively. All other conditions were kept the same with the second sample. Atom force microscopy (AFM) and scanning electron microscopy (SEM) were carried out for getting morphological information of the surfaces of in-situ NL and SNL. High-resolution X-ray diffraction (HRXRD) was used to determine the quality of high-temperature AlN epitaxial layers of all samples. Finally, a deep UV-LED was grown on the SNL of 50 nm thickness. Except for the nucleation layer (or buffer layer) replaced by SNL, the remaining growth process of the deep UV-LED was similar to Ref. [ 19]. The LED light output power (LOP) was measured with a calibrated integrating sphere at room temperature.
3. Results and discussion
SEM and AFM were carried out to determine the morphologies of the surface of AlN nucleation layers. Fig. 1 shows the SEM and AFM images of surface topography of nucleation layers, while (a, c) were grown by MOVPE and (b, d) were grown by reactive magnetron sputtering system. Measured by AFM, the root mean square (RMS) roughness of in-situ NL and SNL is 1.5 and 0.47 nm, respectively. It is found that the surface of SNL is smoother than the in-situ AlN NL. In general the surface morphology of AlN NL influences the growth of AlN epitaxial layer[8, 11].
Fig. 2 shows the normalized XRD rocking curves of (002) and (102) peaks from the 925 nm AlN (900 nm AlN epitaxial layer with 25 nm SNL or in-situ NL). AlN epitaxial layer with in-situ NL shows 1081 and 1735 arcsec of FWHM of the (002) and (102) peaks, respectively, and the corresponding screw and edge dislocation densities are calculated to be 2.54 × 10 9 cm−2 and 2.88 × 10 10 cm−2[20]. The FWHM and TDD values suggest that high-yield industrial MOVPE is difficult to grow high-quality AlN epitaxial layers under general two-step process. Compared with the in-situ NL, the AlN epitaxial layer with SNL shows narrower FWHM values of 67 and 1101 arcsec from (002) and (102) peaks, respectively, and the corresponding screw and edge dislocation densities are calculated to be 9.77 × 10 6 and 1.46 × 10 10 cm−2[20]. Generally, the declining FWHM of the (002) peak implies the reduction of the screw and mixed-type dislocation density of the AlN epitaxial layer[21]. Furthermore, the decreased FWHM value of (102) peak implies the reduction of the edge threading dislocation density of the AlN epitaxial layer[22]. Utilizing an SNL, the results clearly show that relative high-quality AlN material can be grown on sapphire through the high-yield industrial MOVPE system.
The growth mode of high-temperature AlN epitaxial layer on nucleation layer possibly includes three-dimensional growth, island merging, and a quasi-two-dimensional growth[11, 23]. During the process of island merging, many dislocations will steer and disappear. Thus the island merging process can reduce the dislocations and promotes the quality of high-temperature AlN epitaxial layer. However, if the quality of the nucleation layer is too poor, such as the nucleation density is too small or the size of crystal islands is very large, it may cause the epitaxial layer to remain three-dimensional growth[11, 23]. In this case, the crystal quality will be poor. The in-situ NL has very large grain-size, fairly small nucleation density and extremely inhomogeneous surface. Hence the merging of sparse isolated seeds may be difficult to be completed during the growth of the 900 nm high temperature AlN. However, the SNL has a homogeneous surface and higher nucleation density than in-situ NL. Thus the growth process of high-temperature AlN on SNL[24] is easier to complete the transition to the quasi-two-dimensional growth than in-situ NL.
The thickness of SNL is one of the most critical growth parameters for high-temperature AlN. As the thickness of the SNL increases, significant changes will take place on the crystal quality and surface morphology. In our work, the thickness of SNL was varied from 25 to 100 nm. All growth conditions remained the same. AFM was carried out to observe the surface morphology of SNL. We investigated the RMS roughness of the SNL surfaces by AFM (shown in Fig. 3). Comparing these three samples, when the thickness of SNL is increasing from 25 to 100 nm, the surface of SNL will be rougher.
In order to investigate the quality of AlN grown by MOVPE in high temperature with SNL of different thickness, XRD ω-scan was carried out to determine the FWHM values of (002) and (102) peaks, which are shown in Fig. 3. While the thickness of SNL is increased from 25 to 100 nm, the FWHM value of (002) is increased from 67 to 226 arcsec, and the FWHM value of (102) is reduced from 1101 to 762 arcsec. For the AlN templates with 25, 50 and 100 nm thick SNL, the calculated screw dislocation densities[20] values are 9.77 × 10 6, 1.65 × 10 7 and 1.11 × 10 8 cm−2, respectively, and the calculated edge dislocation densities[20] values are 1.46 × 10 10, 8.91 × 10 9 and 6.70 × 10 9 cm−2, respectively. There are two situations broadening FWHM value[11]. One situation is that the grain growth direction tilts (Tilt), deviating from the c axis of sapphire substrate. Tilt will introduce screw dislocation paralleling to the direction of growth, which can broaden the FWHM value of (002) peak. The other situation of producing the non-conforming orientation of crystal columns is the crystal column in the c axis torsion (Twist). Twist will introduce edge dislocation paralleling to the growth direction[25] and broaden the FWHM value of (102) peak. When the thickness of SNL is increased from 25 to 100 nm, it seems that the tilt of grains increases and the twist of grains reduces.
Trade-off between the FWHM values of these AlN templates and the RMS roughness of these SNL surfaces, we selected the 50-nm-thick sputtering AlN as nucleation layer. Finally, a commercial deep UV-LED was grown on the 50-nm-thick SNL. Fig. 4(a) shows the electroluminescence (EL) spectra of the deep UV-LED, measured under an injection current of 20 mA. The emission peak wavelength for the UV-LED is 282 nm. Then the V–I–LOP characteristics of the finished devices were measured. Fig. 4(b) presents the output power as a function of injection current. The light-output power of the deep UV-LED reaches 1.65 mW at 20 mA with external quantum efficiency[26] of 1.87%. In addition, the saturation LOP of the deep UV-LED is 4.31 mW at an injection current of 60 mA. Due to utilizing a 50-nm-thick SNL on the sapphire substrate to enhance the quality of the epitaxial layer, the 282 nm commercial deep UV-LED was realized by using high-throughput industrial MOVPE.
4. Conclusion
Compared with low-yield industrial MOVPE, it is difficult for high-throughput industrial MOVPE to grow high-quality AlN due to hardly controlling the uniformity of temperature and harmful pre-reaction in huge reactors, especially during AlN growth. But an SNL avoids these difficulties. When we used an SNL to replace a general in-situ NL, the FWHM values of the XRD rocking curves of the (002) and (102) peaks were reduced greatly. Utilizing an SNL, relatively high-quality AlN material can be grown on sapphire through the high-yield industrial MOVPE. When we varied the thickness of SNL, we got the smallest FWHM values of the (002) and (102) peak from different samples, which were 67 and 762 arcsec, respectively. Finally, we have demonstrated the 282 nm commercial deep UV-LEDs first fabricated on a 50 nm-thick SNL by high-yield industrial MOVPE. The deep UV-LED has reached LOP of 1.65 mW and EQE of 1.87% at I = 20 mA. In summary, our studies supply a possible process to grow high-quality AlN and commercial deep UV-LEDs in high throughput industrial MOVPE, which can increase yield, all at lower cost.