1. Introduction

The presence of large densities of defects and dislocation [1–3], poor p-type conduction [4–7], and inefficient light extraction [8,9] have been identified as some of the major challenges for achieving high efficiency deep ultraviolet (UV) light emitting diodes (LEDs). Significant progress has been made in AlGaN quantum well LEDs, with the best reported external quantum efficiency ∼20% for devices operating at 275 nm [10]. It has remained extremely challenging, however, to achieve high wall-plug efficiency, particularly for devices operating at wavelengths ∼265 nm, or shorter, which are well suited for water purification and disinfection applications [11,12]. Light extraction efficiency of AlGaN deep UV LEDs can be fundamentally improved by incorporating photonic nanostructures. To date, the design of photonic crystal LEDs is often based on photonic bandgap engineering to prohibit light emission into guided modes and enable coupling into the ambient, thereby enhancing light extraction [13–17]. The device fabrication process, however, involves either etching through the active region, or bottom-up epitaxy of nanowire structures. The lack of deep UV transparent passivation materials [18,19] makes it difficult to demonstrate high efficiency UV devices. Such fabrication challenges can be addressed, to a certain extent, by employing photonic crystal structures as a diffraction grating layer to enhance light extraction efficiency for TM polarized emission [16]. Previously, the use of photonic crystal diffraction to increase LEE has been demonstrated for LEDs in the UV-A and visible spectra [16,17,20–23]. To date, however, there were very few reports on the use of photonic crystal diffraction properties to improve light extraction in the deep UV to our knowledge [24,25].

Another factor that fundamentally limits the wall-plug efficiency of deep UV LEDs is the poor p-type conduction of Al-rich AlGaN, due to the very large activation energy (up to 600 meV) for the p-type (Mg) dopant and the formation of extensive compensating defects under relatively high doping conditions. Recent studies suggest that, by tuning the Fermi level at the growth front away from the valence band edge during the epitaxy of Mg-doped AlGaN, the formation energy for compensating defects can be significantly enhanced, while the formation energy for Al (or Ga) substitutional Mg dopant incorporation can be reduced [26]. Therefore, large concentrations of Mg dopant can be incorporated without extensive defect formation. The resulting formation of an Mg impurity band enables hole hopping conduction and also effectively reduces the activation energy for a portion of Mg acceptors due to the dispersion of acceptor energy levels [27], thereby leading to significantly enhanced p-type conduction. Using this technique, relatively high hole concentration and small Mg-dopant activation energy has been measured for AlGaN with high Al content. The achievement of relatively efficient p-type conduction of AlGaN also provides the distinct opportunity to realize tunnel junction deep UV LEDs, which can further reduce the device contact resistance, enhance charge carrier (hole) injection, and reduce light absorption associated with conventional p-Ga(Al)N contact layer [28–31]. To date, however, there has been no demonstration of tunnel junction deep UV devices with the incorporation of photonic crystal structures.

Here we report on the first demonstration of tunnel junction deep UV photonic crystal LEDs operating at ∼267 nm. We have performed a detailed simulation of the light extraction efficiency (LEE) of AlGaN quantum well LEDs with the integration of photonic crystal. We show that the LEE can be enhanced by nearly a factor of two, reaching over 60% by integrating photonic crystal structures on a conventional AlGaN quantum well LED structure. Experimentally, we have investigated the MBE growth, characterization and fabrication of such deep UV LEDs. The device exhibits a peak wall-plug efficiency (WPE) ∼3.5% and external quantum efficiency (EQE) ∼5.4%, which are approximately 2.5 times higher compared to identical top-emitting LED structures but without the use of photonic crystal structures. This work demonstrates a viable path for achieving high efficiency deep UV LEDs through the integration of nanoscale structures with conventional planar AlGaN.

2. Device structure and photonic crystal simulation

Illustrated in Fig. 1(a) is the schematic of the AlGaN deep UV LED heterostructure, which was grown using a Veeco Gen 930 MBE system equipped with a radio frequency plasma-assisted nitrogen source on 1µm AlN-on-sapphire substrates from DOWA Holdings Co. Ltd. The device heterostructure consists of ∼480nm Si-doped Al_0.65Ga_0.35N and a graded Si-doped AlGaN layer with Al compositions varied from 65% to 85% within ∼25nm thickness. The active region consists of four Al_0.6Ga_0.4N quantum wells (QWs), showing photoluminescence emission peak at ∼267nm. A 25nm Mg-doped AlGaN layer with Al compositions varied from 85% to 65% was then grown, followed by a ∼100nm thick p-Al_0.65Ga_0.35N, 5nm GaN, and 360nm n⁺-Al_0.65Ga_0.35N. The incorporation of a thin GaN in the tunnel junction design can enhance the tunneling probability due to the large polarization induced band bending [30,32]. The samples were grown using the special technique of metal-semiconductor junction assisted epitaxy, which can significantly enhance Mg-dopant incorporation and enable relatively efficient p-type conduction of Al-rich AlGaN [26].

 

Fig. 1. Schematic of (a) tunnel junction deep UV LED heterostructure and (b) nanowire photonic crystal integrated deep UV LED. (c) Top view of the nanowire photonic crystal.

Download Full Size | PDF

Figure 1(b) shows the schematic of the device, wherein nanowire photonic crystal structures were realized by etching topmost n⁺-AlGaN layer of the planar deep UV LED structure. The nanowires have hexagonal shape and are arranged in a triangular lattice, illustrated in Fig. 1(c). The photonic crystal structure acts as a diffraction grating layer, and the light diffraction and coupling efficiency are directly related to the thickness and refractive index of each layer [16]. Important design parameters for achieving high LEE include nanowire diameter (d), lattice constant (a), nanowire height (L) and distance between active region and nanowire. In order to calculate the LEE, finite difference time domain (FDTD) simulation was performed using the software package Lumerical FDTD Solutions. To optimize top emitting light extraction, a perfect electrical conductor (PEC) substrate with nearly 100% reflectivity was included in the simulation, which can be experimentally realized by incorporating AlGaN/AlN distributed Bragg reflectors (DBRs) [33]. Perfect matched layers (PMLs) were used as boundary conditions, which inhibit outgoing electromagnetic waves reflecting back into simulation domain [34,35]. A minimum mesh step size of 0.25nm was used via adaptive meshing technique. Single polarized dipole source with 1:1 ratio of TE and TM polarized emission was placed in the active region to simulate the light emission [36]. The spectrum of dipole is Gaussian shape with 60nm full-width-at-half-maximum (FWHM) linewidth and a peak wavelength of 265nm. To increase the accuracy of simulation, the refractive indices for each layer are calibrated and further measured separately by Wollam M-2000 Ellipsometer, shown in Table 1, whereas average refractive indices were used for the active region and graded AlGaN layer. Absorption of emitted light in the p- and n-AlGaN layers was not considered, given their larger bandgap compared to the quantum well active region.

Table 1. List of refractive indices for tunnel junction deep UV LED structure

View Table

Figure 2(a) shows the calculated LEE, averaged in the wavelength range of 264nm to 271nm, as a function of nanowire diameter (d) and lattice constant (a) for nanowire height L = 240nm. The highest light extraction efficiency ∼60-64% was found at d = 180nm and a = 240nm, and at d = 160nm and a = 320nm, with the latter being used in the experimental studies in this work. Figure 2(b) shows the calculated LEE as a function of nanowire diameter at a = 320nm and L = 240nm. The LEE was also calculated as a function of nanowire height (not shown), based on which nanowire height L = 240nm was used in subsequent experimental studies. For comparison, the calculated LEE for an identical planar LED but without the incorporation of photonic crystal structures is ∼30%, which is approximately two times lower than the presented photonic crystal UV LED shown in Fig. 2(b).

 

Fig. 2. (a) Contour plot of average LEE as a function of nanowire diameter (d) and lattice constant (a) at nanowire height L = 240nm. (b) Variations of the average LEE vs. nanowire diameter (d) for a constant lattice constant (a) of 320nm and nanowire height (L) of 240nm.

Download Full Size | PDF

3. Device fabrication and measurement

The fabrication of AlGaN nanowire photonic crystal LED is described. AlGaN nanowire photonic crystal structures were defined by using e-beam lithography, BCl₃/Cl₂ plasma dry etching, and Tetramethylammonium hydroxide (TMAH) wet etching. Shown in Fig. 3 is the scanning electron microscopy (SEM) image of the fabricated nanowire photonic crystal structure, which has a diameter (d) ∼ 160nm, lattice constant (a) ∼ 320nm, and height (L) ∼ 240nm. Ti (40 nm)/Al (120 nm)/Ni (40 nm)/Au (50 nm) was deposited as the top as well as the bottom n-metal contact, schematically shown in Fig. 1(b). The samples were subsequently annealed at 700°C for 30 seconds in nitrogen ambient.

 

Fig. 3. 45°-tilted view SEM image of hexagonal nanowire photonic crystal arranged in triangular lattice with nanowire diameter (d) of 160nm and lattice constant (a) of 320nm. The inset shows a magnified image of photonic crystal.

Download Full Size | PDF

The fabricated deep UV LEDs were measured directly on wafer without any packaging. Shown in Fig. 4(a) are the current-voltage characteristics measured at room-temperature. The device has an areal size of 50µm × 50µm with metal grids deposited on the non-etched region to facilitate current spreading. A separate large contact pad was connected to the metal grids to facilitate electrical measurements. The same metal contact design was also used for LEDs without photonic crystal. The device turn-on voltage is ∼ 9V, and the current density reaches ∼20 A/cm² at ∼10V. The current-voltage characteristics can be further improved by optimizing the tunnel junction structure. It is also noticed that the current-voltage characteristics are nearly identical for the deep UV LEDs with and without the incorporation of AlGaN nanowire photonic crystals, shown in Fig. 4(a), due to the use of tunnel junction and efficient current spreading of the highly doped top n⁺-AlGaN layer having same metal contact design. Shown in Fig. 4(b) are the electroluminescence spectra of the LED with photonic crystal measured at room temperature, which were collected by a UV optical fiber, analyzed using high-resolution spectrometer and detected by charge coupled device. The peak emission wavelength is at ∼267nm, which stays nearly constant with increasing current injections. The measured emission spectra showed very small, or negligible variations compared to the LEDs without PhC. A Newport 818-ST2-UV silicon photodetector with Newport Model 1919-R power meter was used to measure the output optical power. Shown in Fig. 4(c), an output power density ∼3 W/cm² was measured at a current density of ∼23 A/cm², which is over two times higher than AlGaN deep UV LEDs without the incorporation of nanowire photonic crystal structures.

 

Fig. 4. (a) Current-voltage characteristics for both LEDs with and without incorporation of nanowire photonic crystal. (b) Electroluminescence spectra of the LED with photonic crystal measured with different current injections. (c) Top emitting optical power density measurement using 10kHz repetition rate and 1% duty cycle.

Download Full Size | PDF

The calculated EQE and WPE as a function of current density are shown in Figs. 5(a) and 5(b), respectively. Peak EQE ∼5.4% and WPE ∼3.5% were measured at low current densities 0.5∼1A/cm² for AlGaN photonic crystal UV LED, which is approximately ∼2.5 times higher compared to identical devices without the incorporation of photonic crystals. Given that the devices exhibit nearly identical current-voltage characteristics, the increase in WPE and EQE can be directly attributed to the enhanced LEE with the incorporation of nanowire photonic crystals. The measured EQE enhancement is slightly larger than that predicted by simulation, which can be explained by the assumption of a perfect electrical conductor (PEC) substrate in the simulation. A severe efficiency droop was measured with increasing current injection. The underlying causes for the efficiency droop may include electron overflow [37–39], Auger recombination [39,40], carrier delocalization [41], and/or heating effect [42]. Since the efficiency droop occurs at very low current density (∼1A/cm²), heating effect and Auger recombination are not likely the major causes. On the other hand, it is expected that electron overflow can become significant in AlGaN deep UV LEDs even at very low current densities, due to the highly asymmetric charge carrier transport properties between electrons and holes in Al-rich AlGaN. Similar efficiency droop was measured previously in AlGaN deep UV LEDs. Poor hole mobility, due to the hole hopping conduction in the Mg impurity band, was attributed to be the major cause [43]. It is expected that with further improvement of p-type conduction of AlGaN and optimization of the tunnel junction, the efficiency droop of AlGaN deep UV LEDs can be effectively reduced.

 

Fig. 5. (a) External quantum efficiency (EQE) and (b) wall-plug efficiency (WPE) as a function of injected current density.

Download Full Size | PDF

4. Summary

In summary, we have demonstrated AlGaN tunnel junction deep UV photonic crystal LEDs operating at ∼267nm. Significantly enhanced EQE and WPE were measured for AlGaN deep UV LEDs with the incorporation of nanowire photonic crystal structures. Although this work was primarily focused on devices operating at 267nm, such a unique design can be readily extended to AlGaN LEDs with higher Al compositions, i.e., shorter emission wavelengths, wherein TM polarized emission is more dominant. It is also observed that AlGaN deep UV LEDs suffer from severe efficiency droop, which is likely induced by electron overflow, due to the low hole mobility of Al-rich AlGaN. Enhanced efficiency and output power are expected by further optimizing the design, epitaxy and fabrication of AlGaN tunnel junction photonic crystal LED structures in the future.