1. Introduction

Highly efficient III-nitride based light-emitting diodes (LEDs) have been employed extensively in solid-state lighting [1]. Micro-sized LEDs (µLEDs) have gained expanding research interest as the technologies become more advanced and complex [2–6]. µLEDs have advantages in long operating lifetime, high luminous efficiency, and chemical robustness. Compared to organic light-emitting diodes and liquid crystal display, µLEDs are a promising candidate for display applications, such as near-eye head-mounted display and large-area self-emitting display [2,7]. Monochromatic display using µLEDs have been demonstrated for high resolution, efficiency, and contrast ratio [3,8,9]. In addition to display applications, µLEDs have been developed for high-speed transmitters in visible-light communication (VLC) application, offering GHz modulation bandwidth [5,6].

Although µLEDs yield outstanding performances in display and VLC applications, previous studies have reported that the peak external quantum efficiency (EQE) decreases as the µLED dimension decreases [2,7,10]. This decrease is due to surface recombination and the sidewall damage of the mesa from the plasma-assisted dry etching, which creates sidewall defects for non-radiative recombination. Therefore, surface recombination and sidewall damage contribute to Shockley-Read-Hall non-radiative recombination and the effect is more significant at small dimensions [7,11,12]. Sidewall passivation using dielectric materials is a common approach to minimize the effect of plasma damage in LEDs with large emission area. Sidewall passivation is performed conventionally using plasma-enhanced chemical vapor deposition (PECVD), which uses hydrogen-containing precursors, for providing rapid deposition rates [13-16]. In this work, we report the important enhancements of µLEDs with sidewall passivation utilizing atomic layer deposition (ALD). ALD employs metalorganic precursors that gives excellent dielectric quality and yields precise thickness control in atomic-level. The two different passivation techniques were employed to determine their influence on leakage current due to surface recombination and sidewall damage. The lowest leakage current with uniform light emission was achieved by using ALD followed by buffered hydrofluoric acid (HF) wet etch. For 20 × 20 µm² µLEDs, the EQE was 33% with ALD passivation compared to 24% without sidewall passivation, which the EQE improvement was due to a combination of enhancement in light extraction and reduction in leakage current caused by surface recombination and sidewall damage.

2. Experimental

The µLED structures were grown on a (0001) c-plane patterned sapphire substrate by metal-organic chemical vapor deposition (MOCVD). The structural details of the epitaxial wafer and device structure are shown in Fig. 1(a). and listed in Table 1. All µLEDs were processed together from the same epitaxial wafer to minimize sample variations. µLEDs were fabricated in squares with the edge length ranging from 10 µm, 20 µm, 40 µm, 60 µm, 80 µm, and 100 µm. The geometrical details of the devices have been reported elsewhere [7]. After the MOCVD growth, 110 nm indium-tin oxide (ITO) was deposited via electron-beam evaporation as a transparent and ohmic p-contact. The µLED mesa structures were defined using reactive-ion etching (RIE) to etch ITO and etch down to the n-GaN layer. An omnidirectional reflector (ODR) with 95.5% reflectance in the blue wavelength range (430-450 nm) was deposited using ion beam deposition as a metal isolation dielectric layer. The goals of employing the ODR were to serve as an isolation layer between n- and p- contacts, and to further reflect light from the metal layers. The ODR composed of silicon dioxide (SiO₂) and tantalum pentoxide (Ta₂O₅), followed by aluminum oxide (Al₂O₃) as the capping layer. 50 nm of SiO₂ was blanket deposited using ALD or PECVD at 300°C to cover the mesa. The silicon precursors were tris(dimethylamino)silane for ALD and silane for PECVD. After SiO₂ deposition, HF or inductively coupled plasma (ICP) etching was employed to remove SiO₂ to open a window for metal contacts. The metal consisted of 700/100/700 nm of Al/Ni/Au for common p- and n- metal contacts by electron-beam evaporation. The fabricated µLED structures with different sizes are shown in Fig. 1(b). After the µLED fabrication, the current-voltage characteristics and electroluminescence (EL) images were obtained by on-wafer testing. The µLEDs were singulated into 750 × 750 µm² dies, mounted onto silver headers, wire-bonded, and encapsulated using Dow Corning OE-6650 resin with a refractive index of 1.54 to perform EQE measurement in a calibrated integrating sphere.

 

Fig. 1 (a) Cross section schematics of the µLED design and (b) SEM images of the six µLEDs

Download Full Size | PDF

Table 1. Structural details of µLED

View Table

3. Results and discussion

To distinguish different sidewall passivation and etching techniques, LED-1 is used to designate the µLEDs without sidewall passivation, LED-2 is used for µLEDs with ALD sidewall passivation and ICP etching, LED-3 is labelled for µLEDs with PECVD sidewall passivation and HF etching, and LED-4 is designated for µLEDs with ALD sidewall passivation and HF etching. The µLEDs with combination of PECVD and ICP etch resulted in extremely low light emission that was unable to capture by camera at low current density and yielded worse electrical performances. Therefore, this work focused on the effects of PECVD and ICP etch separately.

The difference between employing ALD and PECVD as the sidewall passivation method can be distinguished from the EL images of µLEDs as shown in Fig. 2(a). Each row of images represents one size of the devices and each column of images represents the different sidewall passivation/etch method. Figure 2(b) demonstrates the light output power characteristics of ALD and PECVD passivation methods at different current density for 20 × 20 µm² µLEDs. For LED-1, the performance of the devices were in good agreement with previously reported data [7]. It was observed that the injected current was crowded around the edges of mesa with a dimmed center for 100 × 100 µm², 80 × 80 µm², and 60 × 60 µm² µLEDs [2,7]. However, for devices smaller than 60 × 60 µm², the light-emitting area was uniform due to effective current spreading across smaller mesa areas, even though the light emission of smaller devices was dimmer than that of the larger devices. It was expected that smaller devices suffer with greater influence from surface recombination and non-radiative recombination caused by sidewall damages, hence the light emission was dimmer in smaller devices [2,11,12,17]. For LEDs-2 and −4, uniform light-emitting area was observed for all sizes. Since the nature of isotropic wet etching of HF could be problematic as the µLED dimensions shrink, ICP was considered to compare with wet etching. The light emission remained uniform across the mesa for the 40 × 40 µm², 20 × 20 µm², and 10 × 10 µm² sizes, unlike LED-1. This uniformity indicated the non-radiative recombination caused by sidewall damage and surface recombination was reduced after ALD sidewall passivation. The effects of ALD versus PECVD were shown when looking at the smaller sizes of the µLEDs. The light emission of the smaller LED-3 was much dimmer compared with the larger ones, whereas the light emission of LEDs-2 and −4 remained high and uniform among all device sizes.

 

Fig. 2 (a) Electroluminescence images of the µLEDs with different sidewall passivation and etch methods at 1 A/cm² and (b) light output power characteristics of ALD and PECVD passivation methods at different current density for 20 × 20 µm² µLEDs.

Download Full Size | PDF

Enhancing light extraction is another advantage of using dielectric sidewall passivation [13,14]. By depositing dielectric layer with lower refractive index than III-nitrides, the light extraction is increased due to a larger critical angle is allowed. According to Snell’s law, θ_crit = sin⁻¹(n₁/n₂) where θ_crit is the critical angle, n₁ is the refractive index of air and n₂ is the refractive index of III-nitrides. The critical angle of total internal reflection depends on the ratio of the refractive indices, which most dielectric materials have higher refractive indices than air and yield greater critical angle. The improvement of light extraction is demonstrated from the light-output power characteristics of 20 × 20 µm² µLEDs shown in Fig. 2(b). Both LEDs-2 and −4 provided more light output power than that of LED-1, which increased 40% light output power at 20 A/cm² and 20% at 95 A/cm². However, PECVD passivation did not give the expected trend and resulted in the least light output power. This decrease was caused by the reduction of ITO transparency after PECVD. During the PECVD process, hydrogen radicals, generated from the decomposition of hydrogen-containing silicon precursor, reacted with ITO to form metallic indium and tin oxides at the interface and lowered the ITO transparency [18-20]. The reaction between hydrogen radical and ITO was not observed in ALD passivation, because metalorganic precursors were used in ALD. With metalorganic precursors, organic molecules were created as a byproduct and no generation of hydrogen radicals during the deposition.

In addition to the improvement in light extraction of µLEDs, dielectric sidewall passivation also served as a surface-insulating layer to reduce leakage current caused by surface recombination and sidewall damage. Figure 3(a) shows the current density-voltage characteristics of 20 × 20 µm² µLEDs with different sidewall passivation techniques from −4 V to + 4 V. The current density is normalized by the light emission area. Between 0 V and −2 V, the leakage current densities of LEDs-2 and −3 were an order of magnitude higher than that of LEDs-1and −4. The high leakage current density in LEDs-2 and −3 indicated PECVD and ICP etch created more leakage paths in the µLEDs. The leakage paths may have caused by the damaging in the ITO layer or in the p-GaN layer. Beyond −2 V, LED-1 started to breakdown sooner than µLEDs with sidewall passivation. For LED-4, the device did not show any breakdown characteristics at −4 V, which demonstrated the significance in reliability by employing the combination of ALD sidewall passivation and wet etching.

 

Fig. 3 (a) Current density-voltage characteristics from −4 V to + 4 V of 20 × 20 µm² µLEDs with different sidewall passivation techniques and (b) the dependence of leakage current at −4 V on the dimensions of µLEDs with different sidewall passivation methods.

Download Full Size | PDF

The effectiveness of the two deposition methods and the relationship between leakage current and the µLED sizes are demonstrated in Fig. 3(b). LED-1 suffered from the highest amount of leakage current at −4V and the leakage current density increased as the size of µLED decreases. Leakage current was suppressed when applying sidewall passivation; however, the capacity of reducing leakage depended significantly on the sidewall passivation technique. PECVD sidewall passivation reduced leakage effectively in the 100 × 100 µm², 80 × 80 µm², and 60 × 60 µm² µLEDs and yielded the same order of magnitude of leakage current as LED-1 for sizes smaller than 40 × 40 µm². The increase in leakage current in smaller µLEDs may be due to the dielectric quality of PECVD. This increase revealed PECVD sidewall passivation was insufficient to reduce leakage current for µLEDs with large perimeter/area ratio. For LEDs-2 and −4, the performance of reducing leakage current behaved disparately by employing different etch methods. LED-4 resulted in the least amount of leakage among all sizes, which surface recombination sites, such as dangling bonds, were removed by ALD sidewall passivation. However, LED-2 yielded similar behavior as the LED-4. The dissimilarity between wet and dry etching can be explained by the damage of the ITO layer. Figure 4 shows a cross-sectional scanning electron microscopy (SEM) image, prepared by focused ion beam (FIB), of the ITO interface after Al/Ni/Au metal deposition. The Pt layer is the platinum deposited on the sample during FIB for surface protection purpose, which is not part of the device structure. The ITO layer exposed to the ICP etch was thinner and rougher than the ITO covered by SiO₂, which indicated the ITO layer was etched unintentionally by ICP. As a result, the ITO interface was damaged by the plasma from ICP, and the ITO exposed to ICP was anticipated to induce more leakage. On the other hand, the ITO was undamaged by using HF, because the etch rate selectivity between SiO₂ and ITO was higher in HF etching than in ICP etching. Therefore, LEDs-2 and −4 were passivated by ALD, yet different etching techniques introduced leakage to the device performance. From Fig. 3(b), although ALD sidewall passivation yielded least leakage among all sizes, leakage current increased as the µLED dimensions decreased. The increase in leakage current indicated that sidewall passivation did not completely erase the effects of sidewall damage.

 

Fig. 4 Cross sectional scanning electron microscopy (SEM) image of ITO layer after exposing to ICP etch. The left side was exposed to ICP etch to remove SiO₂ and the right side was covered by photoresist during the etch. The layer labelled “Pt” is the platinum layer deposited during focused ion beam (FIB) for cross-sectional SEM imagining.

Download Full Size | PDF

The EQEs of 100 × 100 µm² and 20 × 20 µm² devices are presented in Fig. 5(a) and (b), respectively, to demonstrate the effects of different sidewall passivation techniques. For the 100 × 100 µm² µLEDs, Fig. 5(a), all the passivated µLEDs reached similar peak EQE (LED-1: 40%, LED-2: 36%, LED-3: 38%, LED-4: 41%). As the perimeter/area ratio was small, the influence of sidewall damage to the device performance was insignificant. Therefore, the effect of sidewall passivation was not critical for large µLEDs. Moreover, the EQE is less sensitive to sidewall damage and did not vary with sidewall passivation techniques for large devices. Although the maximum EQE remained unchanged regardless of sidewall passivation methods in the 100 × 100 µm² samples, the EQE droop was different. The differences in the EQE curves may be related to changes in the carrier concentration in the active region by employing sidewall passivation. Further investigations will be conducted on the droop mechanism using different passivation techniques.

 

Fig. 5 Dependence of EQE on current injection for (a) 100 x 100 µm² and (b) 20 x 20 µm² devices with different sidewall passivation methods

Download Full Size | PDF

Furthermore, the EQE performance for the 20 × 20 µm² devices, shown in Fig. 5(b), behaved distinctly from the 100 × 100 µm² samples. Besides the µLED with PECVD passivation, all the µLEDs experienced less droop at high current density, which was observed in previous reports [2,7]. The PECVD sample yielded dissimilar performance than the other samples because of the less transparent ITO layer. From Fig. 2, the light emitted from the device was almost unnoticeable at 1 A/cm², since light was blocked by the ITO layer. As the emission area shrinks, light emitted from the small devices is less and more light output is required from small samples than from large devices to overcome the less transparent ITO barrier. For LED-3, light was obstructed by the reduced ITO at low current density. Thus, the EQE curve increased moderately and did not exhibit a peak EQE at low current density like the other devices. For LED-1, the peak EQE dropped 40% by decreasing the size from 100 × 100 µm² to 20 × 20 µm². The decrease in EQE as the µLED dimensions becomes smaller has been observed by other researchers and the reduction in peak EQE deviates from 10% to 50% [2,7,12]. For LEDs-2 and −4, regardless of the SiO₂ etch method, the peak EQE were higher than the peak EQE without sidewall passivation, and the maximum EQE difference was 32% between the devices with ALD sidewall passivation and without sidewall passivation. The increase in EQE was a consequence of increasing light extraction and reducing the leakage current. However, the EQE in the 20 × 20 µm² devices with ALD sidewall passivation was lower than the EQE in the 100 × 100 µm² devices. Since the influence of the sidewall defects and surface recombination was suppressed by ALD sidewall passivation, as demonstrated in Fig. 3, but the plasma damage was still presented after sidewall passivation, Shockley-Read-Hall nonradiative recombination through sidewall damage still contributed to the efficiency loss.

4. Conclusion

We demonstrated the efficiency of µLEDs can be recovered by ALD sidewall passivation. The EQE was 24% without sidewall passivation and 33% with ALD sidewall passivation for 20 × 20 µm² µLEDs. By employing ALD sidewall passivation, the µLEDs resulted in homogenous light emission at low current density and the least amount of leakage current. The EQE in the 20 × 20 µm² µLEDs was not as high as that in 100 × 100 µm² µLEDs, since ALD sidewall passivation only suppressed the leakage current generated from surface recombination and sidewall defects. On the other hand, PECVD sidewall passivation was beneficial to reduce leakage current for large devices but failed to reduce leakage current for µLEDs smaller than 60 × 60 µm², because the perimeter/area ratio was small and the sidewall damage was trivial for large devices.