1. Introduction

AlGaN-based deep-ultraviolet light-emitting diodes (DUV-LEDs) have become a potential DUV light source to take place of conventional mercury lamps due to their significant advantages such as small size, high flexibility, low power consumption, environmental protection, and so on [1–5]. They are expected to be widely used in the applications of medical sterilization, spectroscopic instruments, and non-line-of-sight communication [6–8]. However, the overall performance of AlGaN-based DUV-LEDs is still poor as their external quantum efficiency (EQE) usually cannot exceed 10% [9]. This value is far behind the EQE of 84.3% for mature InGaN-based blue LEDs [10]. Since the AlGaN multiple quantum wells (MQWs) could have a high crystal quality and realize an internal quantum efficiency of more than 80% [11,12], the light extraction is regarded as the main obstacle to hindering the efficiency of current DUV-LEDs [13,14].

Flip-chip (FC) packaging technologies have been commonly used for DUV-LEDs due to the efficient heat dissipation and the lack of a transparent conductor to spread current over the p-side of DUV-LEDs [15–19]. For FC DUV-LEDs, reflectors are generally preferred because they can enhance light extraction effectively [16–20]. Metal is the simplest reflector that can achieve high reflectance compared to reflective photonic crystals and distributed Bragg reflectors. Thus, fabricating a reflective p-electrode becomes the most popular method in the FC configuration of DUV-LEDs [21]. Among various types of metals, aluminum is undoubtedly an ideal candidate for the reflective p-electrode owing to its high reflectivity of more than 90% in the DUV region [22].

However, due to the low work function of Al, it is very difficult to form a good ohmic contact on p-GaN or high-Al-composition p-AlGaN [23]. Therefore, researchers usually inserted ultra-thin conductive layers in Al-based reflective p-electrodes to reduce contact resistivity with minimal light absorption. For example, Hirayama et al. used Ni/Al (1/150 nm) as p-type reflective electrode for 277 nm DUV-LEDs and reported a 70% higher EQE value than Ni/Au contact-based DUV-LEDs [21]. Jo et al. reported 279 nm DUV-LEDs with Ni/Al reflective p-electrode which showed an EQE of 7% [24]. Besides, Mo/Al and Pd/Al were also utilized to enhance the performance of DUV-LEDs [25,26]. However, all these high-reflectance p-electrodes increased the operation voltage of the device compared to those standard low-reflectance electrodes. Therefore, further investigation is still necessary to design a DUV-LED with a highly reflective p-electrode that allows for lower voltages.

In this work, we investigated the resistivity and reflectivity of various thin p-contacts and proposed a “Ni sacrifice” method to fabricate Al-based reflective p-electrodes for DUV-LEDs. The “Ni sacrifice” method consists of removing the deposited Ni after rapid annealing, which minimizes DUV absorption and results in a high-reflectance p-electrode in the DUV region. We also fabricated FC DUV-LEDs emitting at 275-277 nm with several kinds of reflective p-electrodes consisting of different p-contact conductors. The electrical and optical performance of these DUV-LEDs were carefully investigated and analyzed. Besides, we introduced a grid of vias in the device mesa and optimized the diameter of the vias to improve the output of DUV-LEDs.

2. Experimental details

The DUV-LED devices were fabricated from a commercial AlGaN-based LED wafer grown on a 2-inch double-sided polished sapphire substrate (0001). As shown in Fig. 1(a), the epitaxial structure included an AlN buffer layer (2 µm), an n-Al_0.6Ga_0.4N stress modulation layer (2 µm), an n-Al_0.5Ga_0.5N contact layer (600 nm), five pairs of Al_0.45Ga_0.55N quantum well layer/Al_0.6Ga_0.4N barrier layer (60 nm), a p-Al_0.7Ga_0.3N electron blocking layer (EBL) (50 nm), a graded p-Al_0.5→0.15GaN layer (30 nm) and a Mg-doped p-GaN thin layer (5 nm). The ultra-thin p-GaN layer not only effectively lowers the absorption of DUV light but also ensures good p-type ohmic contacts.

 

Fig. 1. (a) Schematic diagram of AlGaN-based DUV-LED grown on the sapphire substrate. (b) Device fabrication process flow charts. The n-electrode was first prepared before the p-electrode in step III. (c) top-view microscope image of the designed DUV-LED with 95-µm vias.

Download Full Size | PDF

Figure 1(b) shows the fabrication process flow charts of the designed back-emitting FC DUV-LEDs. The p-electrodes almost completely covered the p-GaN mesa, while the n-electrodes were formed in a grid of vias to connect the n-Al_0.5Ga_0.5N contact layer. This device model could improve the insufficient current spreading owing to the finite lateral conductivity in the high-Al-composition n-AlGaN layer. The p-electrode in this device model was composed of p-contact electrode covered with Al/Ni/Au (200/50/100 nm), which was designed as highly reflective contact to enhance the DUV light extraction. Considering that the p-electrodes were buried underneath the n-electrodes, an insulation layer (100 nm SiO₂) was inserted between the n- and p-electrodes to prevent electrical shorting of the two electrodes.

Figure 1(c) is a top-view microscope image of our fabricated DUV-LED with a grid of 95-µm vias. The device mesa was defined through standard photolithography and dry etch to the n-Al_0.5Ga_0.5N contact layer by an inductively coupled plasma (ICP) process using Cl₂/BCl₃ gas mixture. The sample was then immersed in 45°C HCl for about 15 minutes to eliminate dry-etching damages at the mesa sidewalls as well as to suppress the leakage current of DUV-LEDs. A V/Al/Ni/Au (30/150/50/100 nm) was evaporated onto n-Al_0.5Ga_0.5N contact layer in the grid of vias by photoresist mask and electron beam deposition, then subjected to rapid annealing at 850°C for 1 minute to achieve n-type ohmic contact and served as n-electrode.

We carried out transmission line method (TLM) measurements to examine the p-contact performance of different electrodes under different annealing conditions. Several kinds of electrodes, which yielded good p-contacts, were then chosen to be deposited on device mesa and annealed at optimized temperatures in an air atmosphere to form different p-type ohmic contacts. Note that we prepared the “Ni sacrifice” sample using 50°C HCl wet etching to wash off annealed Ni (10 nm) on the mesa. After forming p-contact, we further deposited Al/Ni/Au (200/50/100 nm) on it as a highly reflective layer. The sample was covered with 100-nm SiO₂ passivation layer by atomic layer deposition (ALD), which served as an insulation layer between the two electrodes as designed in Fig. 1(b). Then, we opened the SiO₂ windows to expose the n-electrodes in the grid of vias by reactive ion etching (RIE) and deposited a 200-nm Au layer to connect those V/Al/Ni/Au electrodes in the grid of vias as shown in Fig. 1(b). Finally, the device was directly connected to the external circuit by FC bonding using indium solder to measure the light output.

Electroluminescence (EL) spectra of these DUV-LEDs were taken at room temperature by an ultraviolet microscope collecting the light and then the light was transmitted through an optical fiber to the Horiba Jobin-Yvon iHR320 spectrometer with a Synapse CCD detector. An optical power meter (S130VC), which has been calibrated with an integrating sphere measurement system previously, was used to measure the light output power from the backside of DUV-LEDs that were on wafer instead of fully packaged. The DUV-LEDs were driven by Keithley 2426B source meter. All samples were tested under the same conditions, this would not affect the characteristic trends of optical power and EQE, and could ensure the reliability of comparison results.

We simulated the light field distribution of DUV-LEDs (275 nm) using the finite difference time domain (FDTD) method. A perfectly matched layer boundary condition is employed in all the x, y, and z directions in order to eliminate boundary effects. The simulation area is discretized using a 3D grid mesh. The light source is set as the dipole source with a wavelength of 275 nm. The scattered light is collected by a monitor right above the device to get the far-field light distribution.

3. Results and discussion

FC DUV-LEDs required highly reflective p-type ohmic contacts on p-GaN. Considering that the rapid annealing will decrease the metal reflectivity, we fabricated the reflective p-contacts in two steps. First, we choose materials with larger work functions to manufacture appropriate contacts on p-GaN, such as Ni, Pt, Pd, and ITO [27,28]. The annealing process was carefully optimized to reduce the contact resistance. Second, we deposited Al/Ni/Au thin films as the highly reflective mirror for DUV region to increase the light output. Notably, the ultrathin thickness of p-contact metals is more favorable because the contact metals will strongly absorb the DUV light emission and reduce the total reflectivity of the p-electrodes [29,30].

We first examined the electrical performance of different contact materials using TLM method. The Al/Ni/Au hybrid layers were deposited on the contact materials after rapid annealing process to form test pads. Figure 2(a) shows the calculated specific resistivity of different p-contacts based on the TLM measurements. The Ni/Au (1/1 nm), which serve as the most common ohmic contact materials on p-GaN, could achieve the lowest specific resistivity (7.03 × 10⁻⁴ Ω cm²). The good ohmic contact between Ni/Au and p-GaN was attributed to the NiO/Au alloy formed on p-GaN after annealing in an oxygen-containing environment [31–34]. Meanwhile, we noticed that the single Ni thin films without Au exhibited a very high specific resistivity exceeding 8 × 10⁻¹ Ω cm², as shown in Fig. 2(a). This result emphasized the critical role that the metal Au played during the forming of the p-contacts [32–35]. The specific resistance of 5 nm and 10 nm ITO was relatively close, with values of 1.8 × 10⁻³ and 1.25 × 10⁻³ Ω cm², respectively. In addition, we investigated the noble metals of Pd and Pt. With an optimal annealing process, the specific resistance of Pd could be reduced to 7.09 × 10⁻³ Ω cm², which was lower than that of Pt. This indicated that Pd had a superior contact performance with p-GaN than Pt.

 

Fig. 2. (a) Specific contact resistance tested by TLM between various electrodes and p-GaN. The insert shows the IV curve of p-electrode with “Ni sacrifice” annealed at 500°C in air for 5 minutes. (b) Reflectivity of contact electrodes covered with Al/Ni/Au (200/50/100 nm) to ultraviolet light. (c) SEM image of p-GaN with “Ni sacrifice”. The light field distribution simulation of DUV-LEDs (275 nm) with (d) “Ni sacrifice” and (e) Ni/Au (1/1 nm) p-contact covered with Al/Ni/Au (200/50/100 nm), respectively.

Download Full Size | PDF

To minimize absorption by the contact material, we completely removed the contact material Ni (10 nm) after annealing and prepared the “Ni sacrifice” sample. Interestingly, appropriate annealing conditions can help this “Ni sacrifice” sample reduce its specific resistance to 8.55 × 10⁻² Ω cm², which was lower than that of single Ni films. Notably, we have repeated the fabrication process of the “Ni sacrifice” p-electrode for multiple times and could obtain similar results. The mechanism of resistivity variation of the “Ni sacrifice” sample at different annealing temperatures can be explained as follows. At lower annealing temperatures, the hole concentration was lower and the contact resistance was higher because of the insufficient activation of Mg acceptors in p-GaN. As the temperature rose, the Ni significantly enhanced hydrogen depletion from the p-GaN and the degree of Mg activation increased, resulting in a decrease in contact resistance [36]. When the annealing temperature rose continuously, the donor nitrogen vacancy promoted by Ni film would compensate for the hole concentration and result in a decrease in contact resistance, which competed with the effect of hydrogen depletion, leading to changes in contact resistance. In Fig. 2(a), the I-V curve of p-electrode with “Ni sacrifice” method tends to be linear on the whole but slightly curved around the 0 V. It illustrated the Ohmic contact using “Ni sacrifice” method was not perfect but increased the voltage within 0.5 V. Therefore, the calculated specific resistance of “Ni sacrifice” electrode was not rigorous but implied the resistance level of “Ni sacrifice” p-electrodes. In addition, the p-GaN that underwent the “Ni sacrifice” process exhibited a slightly uneven surface in the scanning electron microscope (SEM) image (Fig. 2(c)). We guessed this uneven interface might be the reason for forming a good electrical contact between p-GaN and Al/Ni/Au in Fig. 2(a).

Furthermore, the reflectivity of these p-contacts covered with Al/Ni/Au (200/50/100 nm) was investigated carefully in the DUV region. All metal contacts were deposited on 500-µm-thick double polished sapphire substrates and the reflected light was collected from the sapphire side. The Al/Ni/Au was not annealed because an annealing process would degrade the reflectivity of Al/Ni/Au. The “Ni sacrifice” electrode could achieve the highest reflectivity (around 89.8% at the wavelength of 275 nm) among all these electrodes, as shown in Fig. 2(b). This ultrahigh reflectivity originated from the negligible absorption by the contact material, which had been completely removed for the “Ni sacrifice” electrode. Other contact materials exhibited different absorption in the DUV region. As a result, the reflectivity of these electrodes decreased significantly at 275 nm, especially when the thickness of the contact materials increases, such as Ni and ITO in Fig. 2(b). In addition, the Pd and Pt electrodes had a similar reflectivity of 47.8% and 49.0% at 275 nm, respectively.

Due to the specific contact resistance of Ni 2 nm and 5 nm film being different (Fig. 2(a)), we guess that the resistance of “Ni sacrifice” prepared from different thicknesses of Ni film may be different. And the previous paper has reported that the thickness of Ni thin films would influence the resistivity of “Ni sacrifice” [37]. Therefore, we believe the thickness of Ni films may need to be further optimized in future. In case of the reflectivity, the contact material Ni was completely removed by wet etching regardless of the thickness. So, we think the thickness of Ni perform a negligible effect on the reflectivity of “Ni sacrifice” p-electrode.

Based on the FDTD method, we compared the light field distribution of the DUV-LEDs with the “Ni sacrifice” or Ni/Au (1/1 nm) p-contact covered with Al/Ni/Au (200/50/100 nm), respectively. As shown in Fig. 2(d) and (e), the light field distribution angle was similar to both p-electrodes. However, the luminous intensity of “Ni sacrifice” DUV-LEDs is significantly stronger than that of Ni/Au (1/1 nm), which is well consistent with the reflectivity result in Fig. 2(b). Therefore, we believe that the “Ni sacrifice” p-electrode is able to obtain high reflectivity for DUV-LEDs.

Considering both the contact resistivity and reflectivity, we selected four typical contact materials (Ni/Au (1/1 nm), ITO (5 nm), Pd (5 nm) and “Ni sacrifice”) to prepare the p-contact electrodes of DUV-LEDs. Figure 3(a) reveals the current-voltage (I-V) curves of DUV-LEDs with different p-electrodes. The I-V characteristics behaved similarly for all the DUV-LEDs, which exhibited two linear regions in the semi-logarithmic scale. However, the voltage of the DUV-LED with “Ni sacrifice” p-contact is slightly higher than that of the other three at the same current. For example, the forward voltage of the DUV-LEDs with Ni/Au, ITO, Pd, and “Ni sacrifice” p-contacts at 20 mA are 6.5, 6.7, 6.85, and 7.1 V, respectively. This phenomenon is consistent with the trend of the contact resistance of these four p-electrodes shown in Fig. 2(a).

 

Fig. 3. (a) The comparison of I-V characteristics of DUV-LEDs (95-µm vias) with four different p-electrodes. (b) The normalized EL intensity spectra correspond to four DUV-LEDs at 150 mA.

Download Full Size | PDF

Figure 3(b) presents the normalized EL spectra of DUV-LEDs at 150 mA (∼20 A/cm²). The EL peak (λ_peak) of devices with the ITO, Pd, and “Ni sacrifice” electrodes are all located at 275 nm while that of DUV-LEDs with the Ni/Au electrode is 277 nm. Since all the devices were fabricated from the same 2-inch epi-wafer, the slight difference in the emission wavelength could be attributed to the non-uniformity of the Al composition in MQWs. And the thickness of the AlGaN QWs may be not uniform over the whole wafer. As a result, the quantum confined start effect (QCSE) was different and possibly resulted in a shift of the EL peak wavelength too. In addition, all these DUV-LEDs exhibited an identical full width at half maximum (FWHM) of ∼12.8 nm.

We further measured the light output power of these DUV-LEDs with different p-electrodes. The “Ni sacrifice” electrode-based DUV-LED exhibited a higher light output power compared to the other three devices operated at the same injection currents, as shown in Fig. 4(a). This increase in light output power is mainly attributed to the fact that the “Ni sacrifice” electrode has the lowest absorption in the DUV region and thus provides the best reflectivity of the device emission in Fig. 2(b). On the contrary, the ITO p-electrode will absorb the DUV emission significantly, resulting in a reduced output power of DUV-LEDs in Fig. 4(a).

 

Fig. 4. (a) Output light power and (b) EQE curves for different p-electrodes DUV-LEDs (95-µm vias). (c) Emission images (bottom-view) of “Ni sacrifice” p-electrode DUV-LED with 95-µm vias at 5-40 A/cm².

Download Full Size | PDF

Figure 4(b) displays the EQE of the DUV-LEDs with different p-electrodes at different currents. As the injection current increases, the EQE values behave with a trend of rising first and then falling for all the DUV-LEDs. The peak current is located at around 15 mA. These rising and falling regions are attributed to the Shockley–Read–Hall (SRH) [38] recombination and efficiency droop, respectively, which seem to be independent of p-electrodes. At low injection current, the EQE is low because the SRH non-radiative recombination is dominated. As the current increases, the SRH recombination becomes saturated, and radiative recombination dominates, leading to an increase in EQE. Then the EQE will reach the maximum value at the saturation of radiative recombination rate [39]. However, as the current continues to increase, the carriers will be possible to overflow the MQWs and cause current leakage for DUV-LEDs. The reduced current injection efficiency would result in a decrease in EQE. Besides, the auger non-radiative recombination was another reason for the decrease in EQE [39]. The peak EQE of the DUV-LEDs with Ni/Au, ITO, Pd, and “Ni sacrifice” p-electrodes are 1.64%, 1.43%, 1.51%, and 1.66%, respectively. The highest EQE for the “Ni sacrifice” p-electrode demonstrates that the reflectivity of the p-electrodes plays a more critical role in the light output of the DUV-LEDs than the contact resistance.

Figure 4(c) shows the emission images from the backside of DUV-LEDs with the “Ni sacrifice” p-electrode at the current density from 5 to 40 A/cm². The DUV-LED chip could obtain uniform emission across the whole chip area (around 1 mm × 1 mm). The uniform emission demonstrated identical p-type contacts for all the mesa areas although the contact materials were removed in the “Ni sacrifice” p-electrode. Besides, the n-AlGaN layer in our designed FC DUV-LEDs was connected by a grid of vias from the p-side surface with a gold-based n-electrode. The n-electrode grid led to an n-current spreading layer, which also contributed to the uniform emission of the DUV-LED chips.

Although the n-electrode grid improves the n-current spreading, the n-pad in the vias will absorb some of the emission when the light is scattered from the sidewalls of QWs in our designed DUV-LEDs. Thus, we further shrink the diameter of the vias to decrease the obstruction of n-pad to the emitted light, which will be beneficial for the light extraction efficiency (LEE) and EQE. Figure 5(a) shows the light output power and EQE of the DUV-LEDs (“Ni sacrifice” p-electrode) with the diameter of the vias ranging from 90 to 45 µm, respectively. We could clearly observe that the EQE of the DUV-LEDs increased as expected when the diameter of the vias decreased. The peak EQE of the DUV-LED with the grid of vias in a diameter of 45 µm is approximately 1.73%. In addition, shrinking the vias size could reduce the etching area and decrease the ratio of the sidewall damage area to the total emission area of DUV-LEDs, thereby contributing to the internal quantum efficiency (IQE) and improving the EQE.

 

Fig. 5. (a) Output light power and EQE curves of different apertures DUV-LED with “Ni sacrifice” p-electrode. (b) Peak EQE histogram of various p-electrodes DUV-LEDs at corresponding apertures. (c) I-WPE curves of various p-electrodes DUV-LEDs of 45-µm vias. (d) WPE comparison of DUV-LEDs of 45-µm vias at 10 mA and each point in the figure is an average WPE of five chips of DUV-LEDs.

Download Full Size | PDF

A similar grid of vias with different diameters was utilized in the DUV-LEDs with Ni/Au, ITO, and Pd p-electrodes. The peak EQE values of these DUV-LEDs were extracted and summarized in Fig. 5(b). It was observed that the “Ni sacrifice” p-electrode led to the highest peak EQE values among all the p-electrode types. Furthermore, we found that decreasing the diameter of the vias could effectively enhance the EQE values regardless of the p-electrodes, which suggests a feasible approach to improving the emission of DUV-LED FCs. The wall-plug efficiency (WPE) is another important index to evaluate the performance of DUV-LEDs. As shown in Fig. 5(c), the DUV-LEDs (45-µm vias) with a “Ni sacrifice” p-electrode could perform larger WPE at low currents below 25 mA. We measured five chips for every type of p-electrodes and presented the average WPE at 10 mA in Fig. 5(d), which demonstrated the enhanced performance of the DUV-LEDs with the “Ni sacrifice” electrode. However, as the current increased, the WPE of the DUV-LEDs with “Ni sacrifice” p-electrode decreased faster and would be lower than that of the DUV-LEDs with Ni/Au (1/1 nm) p-electrodes (see Fig. 5(c)). The reason was that the DUV-LEDs with “Ni sacrifice” p-electrodes exhibited a slightly higher voltage than that of the DUV-LEDs with Ni/Au (1/1 nm) p-electrodes as discussed above.

4. Conclusion

In summary, we systematically investigated various types of p-contact materials for highly reflective p-electrodes on AlGaN-based DUV-LEDs. The proposed “Ni sacrifice” p-electrode could achieve the highest reflectivity in the DUV region and provide identical p-contacts across the whole chip area for uniform emission. Although the “Ni sacrifice” p-electrode led to degraded electrical performance (higher resistivity and larger forward voltages at 20 mA), the light output power and EQE of the DUV-LEDs could be improved by utilizing the “Ni sacrifice” p-electrode. This indicated that the reflectivity of the p-electrodes played a dominant role in the emission of DUV-LEDs. Additionally, the reduced diameter of the vias in the device mesa could increase the emission area, which resulted in an enhanced peak EQE value up to 1.73%. Meanwhile, the WPE of DUV-LEDs (45-µm vias) with “Ni sacrifice” p-electrode at 10 mA performs higher than the other three.