1. Introduction

Owing to the issues of energy shortages and environment, white LEDs have been recognized as a promising next-generation general lighting source [1]. The most popular method for generating white LEDs is to bring GaN-based blue LEDs and yellow phosphor together. Therefore, blue LED is one of the key components for developing high-efficiency white LEDs. Recently, the vertical thin-GaN LED have been identified as the most promising structure for improving the efficiency of high-power LEDs over other traditional LED structures with sapphire substrates [2–4]. Unlike conventional LED structures, conventional thin-GaN LEDs are designed with a conductive substrate and a metal reflector to improve light-extraction efficiency (LEE) and reduce thermal resistance [2–4]. For such LEDs, the well-known laser lift-off (LLO) and wafer bonding techniques are used to remove the epitaxial substrates (i.e., sapphires) and to support the thin-GaN films through an electrical and thermal conductive carrier substrate (such as a Si substrate), respectively. To separate GaN thin films from the sapphire substrate, laser pulses are directed through the transparent sapphire substrate and are absorbed throughout the sapphire/GaN interface. This results in the decomposition of the GaN layer, and hence in the peeling of epitaxial layers from the sapphire substrate [2]. However, the transfer of GaN films to receptor substrates results in a considerable release in stresses, which originate from the large lattice and thermal mismatch between the GaN-based layers and sapphire substrate. The released stresses may result in structure defects and even cracks in the epitaxial layers that degrade device performance [3]. In addition to the inherent mismatch, laser-induced damage also results in residual stresses after the LLO process, thereby degrading the electrical and optical properties of LEDs [5]. On the other hand, an additional wet etching process is used to texture the exposed GaN surface layer for further enhancing the light-extraction efficiency [6]. To avoid the aforementioned problem, a vertical-type GaN-based LED (VLED), without the removal of the sapphire substrate, is demonstrated in the present study. That is, thinning and etching were performed on the sapphire substrate instead of the LLO process, to expose the n-GaN layer for the fabrication of VLEDs. In addition, patterned sapphire substrates (PSS) were used in this study to feature the VLEDs with in situ textured GaN/sapphire interface for enhancing the LEE. The detailed fabrication procedures and characterizations of experimental LEDs are presented in the following sections.

2. Experiments

The GaN-based LEDs used in the present study were grown on PSS using metal-organic vapor-phase epitaxy [7]. The layer structure of the LEDs was similar to our previous reports [8]. Multiple metal layers of Ni/Ag/Ni/Au (1/200/50/800 nm) were deposited onto the p-GaN top layer of GaN-based LED wafers to serve as the reflector/ohmic contact layer. After the formation of the reflector/ohmic contact layer, a 1.5 μm thick indium layer was used as the bonding layer. In addition, a bilayer Ti/Au (20 nm/1500 nm) metal was deposited onto the surfaces of Si substrates to serve as ohmic contacts, and these wafers served as receptors for the wafer bonding process. The GaN-based LED wafers were then bonded onto the Si substrates at 350 °C for 60 min. After the bonding process, the sapphire substrates were lapped and polished to 7 μm. To expose the n⁺-GaN layer, the remaining sapphire layer and the undoped GaN(u-GaN) layer were selectively etched away using the dry-etching technique. BCl₃ of 30 sccm was used as the etching gas, associated with RF (radio frequency) power of 300 W and ICP (inductively coupled plasma) power of 100 W. The process pressure was 5 × 10⁻³ torr. Ti/Al/Ti/Au (2/20/100/2000 nm) metal layers were then deposited onto the exposed n⁺-GaN layer to form the n-type ohmic contacts (cathode electrodes) on the LED wafers. Finally, the Si substrates were thinned to 150 μm and coated with Ti/Au (20 nm/200 nm) metals to serve as the backside ohmic contact layer. LEDs fabricated using the aforementioned process were of the vertical type, similar to the conventional vertical LEDs fabricated using the LLO and wafer bonding processes. These LEDs were labeled as LED-I. Figure 1(a) shows a schematic structure of LED-I. For comparison, lateral-type LEDs grown on PSS were prepared and labeled as LED-II. Figure 1(b) shows the schematic structure of LED-II, with interdigitated electrodes on the same surface as the sapphire substrate. In addition, conventionnal vertical LEDs(CV-LEDs) were also prepared for comparison.

 

Fig. 1 (a) Schematic structure of LED-I with a bonded Si substrate (b) schematic structure of LED-II with sapphire substrate.

Download Full Size | PDF

After the LLO of sapphire substrate, the CV-LEDs with planar and textured top n-GaN surface were labeled as CV-LED-I and CV-LED-II, respectively. To texture the n-GaN surface of the CV-LEDs, 6 M KOH solution at an elevated temperature of 60 °C for 180 seconds. All the experimental LEDs used in this study were in bare-chip form on TO can without encapulant and had a chip area of 1 mm × 1 mm. Current-voltage (I-V) characteristics of experimental LEDs were measured using the HP-4156C semiconductor parameter analyzer. Light output power-current (L-I) of the LEDs was measured using a calibrated integrating sphere combined with a source meter of Keithly 2400.

3. Results and discussions

Figure 2(a) shows the typical cross-sectional view scanning electron microscopy (SEM) image of LED-I, which was taken to inspect the etching profile of sapphire and the bonding interface between the Si substrate and GaN layer. The typical SEM image indicates that the bonding interface was void-free. Vertical thin-GaN LEDs with a smooth bonding interface facilitates heat dissipation, mechanical support, and current conduction. The presence of voids around the bonding interface reduces the reliability and performance of the LEDs [9]. On the other hand, periodic convex patterns were observed on the bottom surface of the trench, as shown in Fig. 2(b). To clearly examine the feature, an enlarged SEM image taken from a trench of the etched sapphire substrate on the emitting surface of LED-I is shown in Fig. 2(c). The formation of convex patterns is due to that fact that the conformal etching of PSS took place during the formation of via trenches through the remaining sapphire layer. Figure 2(d) shows a top-view image taken from LED-I by optical microscope.

 

Fig. 2 (a) A typical cross-sectional view SEM image of LED-I (b) a focus at the trench (c) enlarged SEM image taken from the bottom surface of trench (d) a top-view image taken from the LED-I by optical microscope.

Download Full Size | PDF

Figure 3 shows the typical L-I characteristics of experimental LEDs. With an injection current of 350 mA, the light output powers were 360, 310, 250 and 309 mW for LED-I, LED-II, CV-LED-I and CV-LED-II, respectively. This corresponded to have external quantum efficiencies of 37.2%, 32.1%, 25.8% and 32.0% for LED-I, LED-II, CV-LED-I and CV-LED-II, respectively. The significant improvement in the output power of LED-I was attributable to the increase in its LEE, compared with those of LED-II and conventionnal vertical LEDs. Considering the LED-II with p-side-up emitting surface, the indium tin oxide (ITO) layer possesses a relatively lower critical angle for light escaping into the air, because the refractive index of ITO is larger than that of sapphire. In addition, the remaining top sapphire layer, with a thickness of 7 μm, was significantly thicker than that of the ITO layer. Therefore, the top sapphire layer served as a window layer to facilitate light extraction. Although the LED-I and LED-II both possess textured GaN/sapphire substrate to reflect and/or redirect light for facilitating light extraction, the nature of thin-GaN LEDs for LED-I would be further beneficial to light extraction [2–4]. That is, downward-propagating light is reflected up by a metal reflector. On the other hand, the light-shadowing effect by opaque electrodes in LED-I is relatively insignificant to that in LED-II, as shown in the insets of Figs. 1(a) and 1(b). On the other hand, CV-LED-II with textured n-GaN emitting surface exhibited a marked enhancement in light output power compared with CV-LED-I. Although the light output power of CV-LED-II could be improved by further texturing the surface of LED, the texture process may cause the increase of reverse leakage current, as shown in Fig. 4(a) . Considering the CV-LED-II and LED-I, the difference of light output powers could be attributed the fact that they had different emitting surface layers. In addition, the substantial roughness of light scattering interface was also different between the LEDs.

 

Fig. 3 Typical L-I and I-V characteristics of the experimental LEDs.

Download Full Size | PDF

 

Fig. 4 (a) Reverse I-V characteristics of the experimental LEDs (b) emission peak wavelength against on injection current for the experimental LEDs.

Download Full Size | PDF

Figure 3 also shows the typical forward I-V characteristics of experimental LEDs. The forward voltages at a driving current of 350 mA (V_f) were 3.28, 3.48, 3.28 and 3.27 V for LED-I, LED-II, CV-LED-I and CV-LED-II, respectively. The slightly higher V_f observed in LED-II can be attributed to its relatively higher series resistance (Rs), compared with LED-I and CV-LEDs. In fact, the Rs extracted from the I-V characteristics were approximately 1.1 and 1.6 Ω for LED-I(or CV-LED) and LED-II, respectively. As shown in Fig. 1(b), the current path between the anode and cathode electrodes in the vertical LEDs(LED-I and CV-LEDs) could be expected to be shorter than those in LED-II. Although the vertical LEDs consist of a greater number of metal/semiconductor interfaces and an additional Si substrate, these factors in the vertical LEDs with adequate bonding and ohmic contact layers contribute limited magnitude in Rs. Therefore, the higher Rs of LED-II were attributed to the relatively larger distance between anode and cathode electrodes. In other words, the low Rs in the vertical LEDs can be attributed to a relatively better current-spreading effect, compared with that of the lateral LEDs. In addition to the superior forward I-V characteristics, typical reverse leakage current (Ir) of the LED-I was comparable with the LED-II, but markedly lower than the CV-LEDs(CV-LED-I and CV-LED-II), as shown in Fig. 4(a).

The relative higher Ir could be attributed to the fact that strain release and laser-induced damage on the GaN-based epitaxial layers led to the degradation of electrical properties of the conventional vertical LEDs. Another benefit of the design scheme of LED-I is that the heat generated in the LED flows directly from the p-n junction to the Si substrate. However, heat extraction through the thermally resistive sapphire substrate is inevitable for conventional GaN/sapphire-based LEDs, i.e., LED-II. The degradation of light output power of LED-II, attributable to junction heating, is expected when the LED-II is operated under a high driving current. This is because of the low thermal conductivity of sapphire. As shown in Fig. 3, LED-I and CV-LEDs can operate up to 1.0 A without significant power degradation. However, LED-II exhibits substantial power degradation when the injection current exceeds 850 mA. Furthermore, the heating effect on the redshift of the emission wavelength was also observed. Figure 4(b) shows the typical curves of emission peak wavelengths against the injection current for the experimental LEDs. At a low injection current, the quantum-confined Stark effect (QCSE), attributable to the polarization field in the QWs, reflects on the blueshift of the emission wavelengths [10]. The redshift of the emission wavelength, resulting from band-gap shrinkage at a high injection current, is attributed to the junction heating effect. LED-I and CV-LEDs exhibited low Rs and excellent heat conduction, resulting in low junction heating. As shown in Fig. 4(b), the redshift of the emission wavelength was not observed in LED-I and CV-LEDs, even when the injection current reached 1000 mA. In contrast, LED-II exhibited a significant redshift as the injection current exceeded 200 mA. The redshift occurring in the relatively lower current levels revealed that junction temperature increased rapidly with increasing injection current, and that the heat extraction rate of LED-II through sapphire was lower than the heat generation rate. Thus, the band-gap shrinkage rate, resulting from junction heating versus injection current, was higher than that of the band-gap widening rate (i.e., band-filling effect).

3. Conclusions

GaN/Si-based vertical LEDs emitted from a sapphire surface were demonstrated to have higher light output power and lower forward voltage. Owing to the retained sapphire layer of the vertical LEDs, the use of a patterned sapphire substrate was shown to be able to mitigate the light-guided effect, and thereby increase the light-extraction efficiency. In addition, the proposed vertical LEDs behaved in a manner similar to flip-chip GaN/sapphire LEDs with excellent heat conduction, and exhibited low power degradation and minor red shifts in emission wavelength at high current driving. The proposed vertical GaN LEDs can potentially be applied to high-power LEDs used for future solid-state lighting.