1. Introduction

Group III-nitride-based semiconductors have been recently used as high-brightness light-emitting diodes (LEDs) in large full color outdoor displays, signal lights and high performance back light units in liquid crystal displays [1, 2 ]. Even though high brightness GaN-based LEDs are commercially available, it is still difficult to manufacture highly efficient LEDs. The main limitation on the light output power (LOP) is due to the internal quantum efficiency (IQE) and light extraction efficiency (LEE). To improve the LEDs performance, several methods have been proposed, such as epitaxial lateral overgrowth (ELOG) [3], surface roughening [4, 5 ], metal mirror reflect layer [6], photonic crystal structures [7–9 ] and patterned sapphire substrate (PSS). PSS technology has been widely used because of it can effectively reduce the threading dislocation (TD) density by the ELOG and increased LEE by scattering [10–12 ]. Especially, the LEDs grown on nanopatterned sapphire substrate (NPSS) have shown better epitaxial film quality and higher LEE as compared to the LEDs grown on micropatterned sapphire substrate [13, 14 ]. However, only a few approaches is applied to produce NPSS [15, 16 ], such as nanoimprint [17], e-beam lithography [18], and holographic lithography [19]. But due to high cost and low throughput of these approaches, nanoscale-PSS is rarely used in commercial applications.

Carbon nanotubes (CNTs) are promising nanomaterials due to their high aspect ratio, superlative mechanical, thermal and electronic properties [20]. These properties make nanotubes ideal, not only for a wide sphere of applications but also as a laboratory bench for fundamental science. Previously, we have proposed CNTs patterned sapphire substrate (CNPSS) [21], which has the many advantages of a low cost, mass-producible, scalable, etc. In our previous researches, different lateral strain distributions and stress reductions are analyzed in the GaN film on CNPSS. It will be of great value for the CNPSS technology to further optimize the growth procedure, investigate the nucleation characteristics on CNPSS, and analyze the influence of light absorption and redirection induced by CNTs layer.

In this paper, we mainly analyze the impact of CNTs on GaN deposition during the nucleation procedure and present that growth procedure without rough layer is better to improve the crystal quality of the GaN films by reducing TDs. The optical and electrical performance of these LEDs on CNPSS and conventional sapphire substrate (CSS) are also discussed in detail. Moreover, combined with the PL measurement and finite difference time-domain (FDTD) simulation, the influence of CNTs on external quantum efficiency of LEDs is mainly attributed to the increase of IQE, with an eye to the light absorption of CNTs weakening LEE enhancement. Annealing time are as changed conditions in comparison to reduce the light absorption of CNTs. High temperature annealing can effectively remove part of CNTs to increase the LOP, while overlong annealing time has caused degradation of the quantum well resulting in the attenuation of optical power.

2. Experiment

Super aligned multi-walled carbon nanotube arrays were grown on 4-inch silicon wafers by low-pressure chemical vapor deposition using a 5 nm iron film as a catalyst and acetylene as a precursor [22]. The diameter of the CNTs in the super-aligned arrays was about 15 nm. Aligned CNTs films were continuously dry spun from the CNTs array and then coated on 2-inch c-plane sapphire wafers with CNTs alignment parallel to both the GaN [11–20 ] and [1-100] directions, respectively using a focused laser beam. The wafers were dipped into electronic grade ethanol solution and dried in air at room temperature to ensure that the CNTs film was well-coated on the sapphire substrates. Neighboring CNTs shrunk into bundles and then resulted in micrometer-scaled intervals by the ethanol treatment, as shown in Fig. 1(a) . The width of each bundle was around 2-4 μm. In each bundle, CNTs were tightly bonded together with 200 to 500 nm gaps. The CNPSS was then ready for the growth of GaN by metal-organic chemical vapor deposition (MOCVD) [21].

 

Fig. 1 The surface morphologies of the substrates and the GaN films: (a) SEM images of CNPSS; (b) and (c) SEM images of GaN nucleation layers on CSS and CNPSS, respectively; (d) and (e) SEM images of GaN rough layers on CSS and CNPSS, respectively; (f) cross sectional SEM image of CNPSS-LED.

Download Full Size | PDF

In order to confirm the validity of the CNPSS, we prepared multi-layer of CNPSS, as shown in Fig. 1 (a). In this experiments, the multi-layer of CNPSS mean that carbon nanotubes were double vertical cross. For comparison, CSS was also prepared. At first, we used CNPSS and CSS to grow in two different growth processes to optimize the growth process. The differences between two growth processes were that whether it contained rough layer which was grown after nucleation layer. The complex growth process would be introduced in the following. The Trimethyl gallium (TMGa) and ammonia (NH₃) were used as precursors and pure hydrogen was used as the carrier gas. At first, all the samples were annealed at 1100°C in H₂ atmosphere, followed by the growth of a 30-nm thick GaN nucleation layer at 525°C. Then rough layer was grown under a higher temperature about 1035°C for 6 min. Then 3.5-μm-thick undoped GaN and 2.3-μm-thick Si-doped n-GaN layers were sequentially grown at a temperature of 1055°C. An InGaN/GaN multi-quantum well (MQWs) active region, which consisted of nine pairs of 3-nm-thick undoped InGaN wells and 10-nm-thick Si-doped GaN barriers, was then grown on the n-GaN layer. A p-type cladding layer, which consisted in Mg-doped strained layer AlGaN/GaN superlattice, was grown directly on top of the active region, followed by the growth of a 80 nm p-type GaN contact layer.

To make sure the absorption effect of CNTs, we removed the CNTs first and then processed by conventional chip technology, the optimized chip technology would be introduced in the following. At first, all the epitaxial wafers were cleaned to remove all kinds of contaminants, oil and other hydrocarbons. Then 3000 Å thick SiO₂ film were deposited on wafer surface by using plasma-enhanced chemical vapor deposition to avoid lithography and etching damage p-GaN. Followed by deep etching process, the wafer is then etched into sapphire, so that oxygen could diffuse into the chip along GaN / sapphire interface, then parts of CNTs are removed efficiently. The pattern size of reticle is 45mil × 45mil, and the spacing between the patterns of the reticle is 40μm. After spin coating, exposure and the development process, the pattern is transferred to the wafer. In the masking effect of the photoresist, BOE solution (a mixed solution of HF and NH₄F) was first used to corrode cleanly the exposed SiO₂ film of the spacing. Using inductively coupled plasma etching process, the GaN epitaxial material of the spacing is etched to the sapphire substrate. Then the GaN epitaxial wafer was corroded in 6 mol/L KOH solution at 70 °C. Because of the quantum well growth temperature was 800 °C, in order to avoid the damage to quantum well active region, followed by annealing at 700 °C oxygen environment. All samples were corroded 3 h in KOH solution and annealed in 0 h, 2 h and 10 h, respectively. Annealing time was as changed conditions in the following comparison. Subsequently, conventional chip technology was processed to get chips with a square mesa of 45 mil × 45 mil in size.

3. Results and discussion

Figure 1 (b)-(e) clearly show the SEM images of surface morphology of GaN on CSS and CNPSS at different growth stages. Figure 1(b) shows the morphologies of the low temperature nucleation layer on CSS. The GaN islands are deposited uniformly on the CSS after the nucleation procedure. After the high temperature rough procedure, platelet-shaped islands are spatially random deposited on CSS as shown in Fig. 1(d). Figure 1(c) shows the morphologies of the low temperature nucleation layer on CNPSS. The GaN islands were distributed on both sides of the lowermost layer of CNTs. And in the vicinity of the CNTs, the number of GaN islands distributed less. The closer to the CNTs is, the fewer of the number of GaN islands is. In Fig. 1(e), the morphologies of the rough layer on CNPSS are shown. There is no significant change near the CNTs on the CNPSS, but in the area away from the CNTs flat film distributes sparsely. After the procedure, the GaN layers grown on CNPSS and CSS become a uniform and low roughness film surface. Figure 1(f) shows the cross sectional SEM image of CNPSS-LED. Lots of voids are formed upon the interface between the epilayer and substrate. The lateral coalescence of GaN on the CNTs results in these voids, thus improving the crystalline quality and LEE.

To confirm crystalline quality of GaN epitaxial layer, the omega-scan X-ray diffraction (XRD) rocking curves and the TD densities of all the samples are shown in Table 1 . It has been reported that the density of screw dislocations including screw component of mixed dislocation and edge dislocation including edge component of mixed dislocation correspond to the DCXRD full widths at half maximum (FWHM) of (002) and (102) planes, respectively [23]. The densities of screw-type dislocations D_s and edge-type dislocations D_e can be estimated from the XRD FWHM values using the following formulas [24]:

Table 1. FWHM values and dislocation density of samples

View Table

Where |b_s| and |b_e| are the Burgers vector magnitudes of the screw-type dislocations (|b_s| = 0.5185 nm) and edge-type dislocations (|b_e| = 0.3189 nm), respectively, and β₀₀₂ and β₁₀₂ are XRD FWHM values for (002) and (102) planes, respectively. α is the angle between the reciprocal lattice vector (K_hkl) and the (001) surface normal. The calculated TD densities are listed in Table 1. Compare to CSS-LEDs, the (102) rocking curve of CNPSS-LEDs is lower obviously. The low (102) FWHM values of CNPSS-LED could achieve with a low density of edge and mixed TDs. Overall, compared to the CSS-LEDs, the TD density of CNPSS-LEDs is reduced. The growth process without rough layer is better for CNPSS-LED. So we could use less time to reduce the dislocation density and thus improve the crystal quality of the GaN epitaxial layers by using the CNPSS template.

To explain the results of XRD, the mechanism of nucleation is discussed. For CSS, as the nucleation layer recrystallizes under a high temperature, the GaN materials are redistributed. The smaller GaN islands are buried by overgrowth of the larger islands, or that material transport occurs from the smaller to the larger islands, causing the smaller islands to disappear [25]. Subsequent growth produces rapid coalescence and smoothening in two-dimension (2D) growth mode because of the higher growth temperature and the lower growth pressure. The interaction of TDs with the side facets generated at coalescence procedure causes the TDs to bend over. This lateral growth greatly decreases the TD density in the grown film, similarly to the ELOG mode. So, growth procedure with rough layer is better for CSS to reduce the TDs. Schematic diagrams of CNPSS growing in different growth procedure are showed in Fig. 2 . For CNPSS, the CNTs impact the distribution of the GaN islands because of the absence of adhesion of reactant gas on CNTs. It is hard for reactant gas to flow between adjacent CNTs. Figure 2 (a) shows the schematic diagrams of growth procedure with rough layer on CNPSS. After the rough procedure, there is no significant change near CNTs on CNPSS, but in the area away from the CNTs flat film distributes sparsely. This is due to the GaN islands are too sparse to coalesce a platelet-shaped islands near the CNTs. As the influence of CNTs, reactant gas are mainly deposited and coalesced to arise overgrown islands between large intervals of CNTs. Overgrown islands influence the TD density because coalescence procedure generates a large amount of TDs arising at the coalescence boundaries of misoriented islands [26]. The height of these overgrown islands and CNTs has no significant difference. As a result, there is no obvious lateral growth mode during subsequent growth procedure, so many dislocations cannot bend over. Therefore, for CNPSS, growth procedure with rough layer will produce more TDs. In Fig. 2 (b), the GaN islands are coalesced directly in lateral growth mode after nucleation layer as CNTs act as a native mask during the GaN films growth process. So TDs arising at the coalescence boundaries have enough time to bend over. So CNPSS can improve crystal quality by reducing TD density through lateral growth mode without the rough layer.

 

Fig. 2 Schematic diagrams of GaN grown on CNPSS (a) with rough layer and (b) without rough layer.

Download Full Size | PDF

Figure 3(a) shows the I-V curves of samples CNPSS-LEDs and CSS-LEDs in different growth process. In the logarithmic I–V curves, the leakage currents at the reverse bias voltage of −10 V were 0.35 µA, and 0.672 µA, 2.74 µA and 1.37 µA for CNPSS-LED without rough layer, CNPSS-LED with rough layer, CSS-LED without rough layer and CSS-LED with rough layer, respectively. Obviously, the leakage currents of samples CNPSS-LEDs are decreased compare to those of CSS-LEDs. The leakage current of CNPSS-LED without rough layer is the smallest in all the samples. Moreover, it has been reported that leakage current in the reverse voltage region increases with TD density. The smaller leakage current again indicates that CNPSS-LED without rough layer has better crystalline quality. At the forward injection current of 350 mA, the forward voltage for CNPSS-LEDs and CSS-LEDs do not show any significant differences, which are about 3.62 V. By comparing the leakage current of each sample at the reverse and forward regions, we note that introducing the CNTs is very important in decreasing both the TD density of the GaN template and the leakage current. The LOP has been measured to investigate the optical properties of the LEDs. Figure 3(b) shows the LOP of samples CNPSS-LEDs and CSS-LEDs in different growth process. In order to ensure the validity of the data, we selected more than 100 LEDs for testing for each set of samples and selected representative LEDs to plot. The samples have relatively good uniformity and the standard deviation of each set of samples is less than 5%. At 350 mA, LOP of the samples are 266.3 mW, 252.4 mW, 194.2 mW, and 221.8 mW for CNPSS-LED without rough layer, CNPSS-LED with rough layer, CSS-LED without rough layer and CSS-LED with rough layer, respectively. As compared with sample CSS-LED without rough layer, the LOP of sample CNPSS-LED without rough layer and sample CNPSS with rough layer-LED exhibit 37.1% and 30.0% enhancement, respectively. Clearly, it is better for CNPSS to grow in the growth process without rough layer. In contrast, the growth process with rough layer is better for CSS. Compared with CSS, we produce devices on CNPSS with a higher LOP in a shorter growth time. This significant enhancement is mainly due to the stress relaxation that causes the reduction of TD density and improvement of crystalline quality. Therefore, photons could experience more opportunities to be extracted outside, as mentioned above. To clearly confirm that the improvement of LOP is due to an increase in the IQE through embedding CNTs, the temperature-dependent PL is measured at temperatures of 10 and 300 K. The IQE of the MQWs can be estimated by comparing the integrated PL intensities measured at these two different temperatures and assuming that the IQE is 100% at a low temperature of 10 K [27, 28 ]. The IQEs of CNPSS-LED without rough layer and CSS-LED without rough layer are calculated to be 22.84% and 17.98%, respectively, indicating that the IQE of CNPSS-LED without rough layer is increased by 27% compared with that of CSS-LED without rough layer due to the embedded CNTs. According to the enhancement of LOP about 37.1% and IQE about 27% for samples CNPSS-LED without rough layer to CSS-LED without rough layer, the increase in light extraction efficiency is estimated to be 8% for the improvement of light scattering effected by voids formed besides CNTs. For the conventional PSS, the LEE enhancement is about 40%, which is the main reason of the increase of LOP [29, 30 ]. Here, CNTs may absorb a part of light to weaken LEE. Thus, it is concluded that the improvement of the LOP is mainly attributed to the reduction of TD density and improvement of crystalline quality as mentioned above. To further study the influence of the CNTs structure on the devices, the light output radiation patterns have been measured at a driving current of 350 mA, as shown in Fig. 3. The divergent angle of LEDs is identified as the angle of half-maximum emission intensity. The divergent angles for samples CNPSS-LED and CSS-LED are 142°, and 158°, respectively. The decrease in the divergent angle for samples CNPSS-LED means that rays are extracted toward to the front side to a greater extent for samples embedded CNTs than for samples CSS-LED without embedded CNTs. 2D FDTD simulations are carried out on CSS and CNPSS to numerically predict the far-field emission patterns in Fig. 3(d). The CNPSS for FDTD calculations are simplified by periodic CNTs which interval between the bundles is 200nm. The refractive index of the CNTs measured by ellipsometer is 1.29. The duty ratio is 40% and the height of each layer of the CNTs is 100nm. Point sources are placed near the sample surface where the QWs are located. The emission wavelength of the source is set as 450 nm referring to the electroluminescence (EL) measurements. For the CSS-LED, photons emitting outside the critical angle are seen to be totally reflected at the GaN/Air flat interface. Most of the light emitted from the active layer is trapped within the GaN layer. In contrast, the CNPSS-LED are seen to suppress lateral guiding modes and redirect the trapped photons into radiated modes. In the area of GaN, CNTs for total internal reflection of light has obvious absorption. It is also noted that CNPSS exhibits more distinct converging effect than CSS, which is far from the MQWs. Therefore, the enhancement of LOP of CNPSS-LED is mainly attributed to the improvement of IQE_. The CNTs absorb a portion of light, and at the same time affect the output direction of light and LEE. Compared with CSS-LED, the improvement of the LOP for CNPSS-LED is mainly due to the impact of the IQE. The reason for LEE has less effect on LOP is that the CNTs absorb light, weakening the LEE enhancement.

 

Fig. 3 (a) Forward and reverse I-V curves of CNPSS- LEDs and CSS-LEDs; (b) Light output power for LEDs on CNPSS and CSS in different growth process; (c)Far-field radiation patterns of CSS-LED and CNPSS-LEDs with different number of layers of CNTs at an injection current of 350 mA; (d)Simulated emission patterns of (a) CSS-LED and (b) CNPSS-LED.

Download Full Size | PDF

To further study the absorption of CNTs, the transmittance and LOP of annealed samples are measured, as showed in Fig. 4 . In Fig. 4(a), the transmittance of CNPSS is measured based on the baseline of transmittance of CSS. Compared with the CNPSS without annealing, the transmission of annealed wafer has increased significantly, especially in 450 nm. The inset in Fig. 4(a) shows the cross sectional SEM image of CNPSS-LED after annealing process. In the inset, the CNTs are partly removed and some air gaps are formed, compared to that without annealing in Fig. 1(f). Therefore, the annealing process by carbon nanotubes reacting with oxygen at a high temperature to remove the CNTs has significant effect. However, there is no obvious difference in transmittance of samples CNPSS annealing 2 h and CNPSS annealing 10 h. That indicates that 2 h annealing time is enough to remove the CNTs nearby spacing. Thus, to further increase the annealing time will not remove more CNTs. LOP of LEDs is also measured to investigate optical properties of the annealed LEDs. Figure 4(b) shows the LOP of samples CNPSS without annealing-LED, CNPSS annealing 2h-LED and CNPSS annealing 10h-LED as a function of injection current. The LOP of sample CNPSS annealing 2h-LED exhibits 11% enhancement than that of sample CNPSS without annealing-LED, and The LOP of sample CNPSS annealing 10h-LED exhibits 6% enhancement than that of sample CNPSS without annealing-LED. The results of transmittance and LOP show that samples CNPSS annealing 2h improved the optical power due to weakening the absorbance of carbon nanotubes by removing part of CNTs in the annealing process. For CNPSS annealing 10 h, the overlong annealing time does not eliminate more carbon nanotubes, and long annealing time has caused degradation of the quantum well resulting in the attenuation of optical power. High temperature annealing can effectively remove part of CNTs to increase the LOP.

 

Fig. 4 (a) The transmittance of CNPSS with different annealing time based on the baseline of transmittance of CSS. Inset shows the cross sectional SEM image of CNPSS-LED after 2h-ennealing. (b) Light output power for LEDs on CNPSS with different annealing time.

Download Full Size | PDF

4. Conclusion

We use the CNPSS method to fabricate InGaN/GaN MQWs CNPSS-LED by MOCVD. The influence of CNTs during the procedure of nucleation is studied in to explain the improvement of crystal quality introduced by the CNPSS. The LEDs on CNPSS exhibit smaller reverse-bias current and divergent angle, and large enhancement of the LOP compared to the conventional LEDs. The improvement of the LOP for CNPSS-LED is mainly due to the impact of the IQE according to the low temperatures PL measurements and 2D FDTD simulation results. All results demonstrate that the CNTs improve crystalline quality and affect the light output direction and intensity. We carry out different annealing time under oxygen environment to remove part of carbon nanotubes. Under 350 mA current injections, the LOP of CNPSS-LED annealed 2 h and 10 h can be improved by magnitudes of approximately 11%, and 6%, compared to that of the CNPSS-LED without annealing, respectively. The results of transmittance and LOP illustrate that overlong annealing time does not eliminate more carbon nanotubes and has caused degradation of the quantum well resulting in the attenuation of optical power.