Design and demonstration of efficient transparent 30% Al-content AlGaN interband tunnel junctions

In recent years, III-nitride ultraviolet (UV) light emitting diodes (LEDs) and lasers have attracted great research interest due to a wide range of applications in air/water purification, disinfection, sterilization, and sensing.^1–3 LEDs and lasers are advantageous over conventional gas-based lamps due to compact size, low power consumption, and safety.⁴ Considerable efforts have been made in increasing the radiative efficiency by improving substrate and active region quality.^5–8 However, conventional UV LEDs still have significantly low external quantum and wall plug efficiency compared to their visible counterparts.

One of the major challenges in conventional LEDs is the difficulty to make ohmic contact to the p-AlGaN layer due to its doping limitations. Hence, a p-GaN^9–13 layer or an AlGaN/AlGaN(GaN)^14,15 superlattice is adopted to make ohmic contact which leads to absorption and electrical losses, respectively. In addition to this, the acceptor activation energy in AlGaN is high and increases with the increase in the Al content in AlGaN.¹⁶ This results in low hole concentration in p-layers. An efficient solution to overcome the above-mentioned challenges associated with the poor conductivity and p-type doping in AlGaN in a conventional LED is the use of tunnel junctions (TJs).^17,18 Although GaN based transparent tunnel junctions with low resistances have extensively been studied,^19–22 reports on AlGaN based transparent TJs are relatively few. Previous work on AlGaN TJs with InGaN^23–28 and GaN^29,30,33 as interlayers have shown very low voltage drop, but these interlayers have their own disadvantages as they absorb UV light. In addition, InGaN can only be grown at lower temperatures leading to growth interruptions which may lead to deterioration in the crystal quality.³¹ These result in a lower external quantum efficiency (EQE). On the other hand, previously reported transparent homojunction tunnel junctions (HJ TJs)^32–34,38 showed significantly higher voltage losses leading to a lower wall plug efficiency. The higher tunnel junction resistance is because of the increase in tunneling barrier height with the increase in the material bandgap. The poor tunneling probability is also due to the lower hole concentration due to higher activation energy with the increase in the AlGaN composition. Therefore, in addition to degenerate doping, polarization induced charges could also be used in the tunnel junction region by compositional grading. In this work, we show that a combination of compositional grading and high doping can enable a fully transparent Al_0.3Ga_0.7N (Al-content ≥ 30%) transparent tunnel junction with a voltage drop as low as 1.86 V across the tunnel junction at 20 A/cm².

To analyze the effects of grading, two PN + TJ (NPN) diodes (displayed in Fig. 1) were first grown on metal–organic chemical vapor deposition (MOCVD)-grown n-type Al_0.3Ga_0.7N templates (5 × 10¹⁸ cm⁻³ Si doping) with a threading dislocation density (TDD) of 2 × 10⁹ cm⁻². The samples were grown using Veeco Gen 930 N₂ plasma assisted molecular beam epitaxy (PAMBE) using standard effusion cells for Ga, Mg, Al, and Si at a plasma power of 300 W and a N₂ flow rate of 2.25  sccm corresponding to a growth rate of 246 nm/h. The PN part of the device consists of the following epilayers: a 400 nm n Al_0.3Ga_0.7N buffer layer (1 × 10¹⁹ cm⁻³ Si doping), a 225 nm n-Al_0.3Ga_0.7N (1 × 10¹⁷ cm⁻³ Si doping), and a 200 nm p Al_0.3Ga_0.7N (1 × 10¹⁹ cm⁻³Mg doping). The TJ epistack varies for the two samples. The graded tunnel junction has $p++−Al0.5Ga0.5N→p++−Al0.3Ga0.7N$ (Mg = 1 × 10²⁰ cm⁻³) and $n++−Al0.3Ga0.7N→n++−Al0.5Ga0.5N$ (Si = 3 × 10²⁰ cm⁻³) to take advantage of induced 3D polarization charges. The ungraded TJ has $p++−Al0.3Ga0.7N$ (Mg = 1 × 10²⁰ cm⁻³) and $n++−Al0.3Ga0.7N$ (Si = 3 × 10²⁰ cm⁻³). Both the growths were terminated by a 170 nm n $Al0.3Ga0.7N$ and a 25 nm $n+−Al0.3Ga0.7N$ (Si = 1 × 10²⁰ cm⁻³).

The device structures were then fabricated starting with defining square mesas by inductively coupled plasma and reactive ion etching (ICP-RIE) using BCl₃/Cl₂/Ar etch chemistry. This was followed by metal deposition for the top and bottom contacts. The bottom metal stack consists of Ti (20 nm)/Al (120 nm)/Ni (30 nm)/Au (50 nm) annealed in N₂ atmosphere at 850 °C, and the non-alloyed top contact consists of Al (30 nm)/Ni (30 nm)/Au (50 nm). Current density–voltage (J–V) characteristics were measured using a Keysight B1500A semiconductor device analyzer.

Transfer length measurements were performed to extract the contact resistances of the top and bottom layers. The bottom contact resistance was $9×10−6 Ω cm2$ for both the device structures and the top contact resistances were $1.61×10−4$ and $2.2×10−6 Ω cm2$ for the ungraded and graded tunnel junction structures, respectively. The difference in the contact resistances can be due to the variations in growth. The total voltage drop at 20 A/cm² for is 5.26 and 5.77 for the graded and ungraded TJs, respectively (Fig. 2a). The differential resistance for the two devices was calculated by subtracting the contact resistances and neglecting the diode resistance. This sets an upper threshold limit of $8.9×10−5 Ω cm2$ for the graded TJ structure and $6.9×10−3 Ω cm2$ for the ungraded TJ structure (Fig. 2b). The graded TJ structure has the lowest reported differential resistance for any AlGaN based TJ (Fig. 2e).^{23,25,26,37,38} Simulations of the TJ structures were carried out using Silvaco Atlas TCAD. The non-local band to band tunneling model was used to calculate the tunneling rates in both the heavily doped n and p regions. When the n and p layers are compositionally graded, the polarization charges induce large concentrations of free carriers in the tunnel junction region,^39–41 reducing depletion related tunneling barriers. It also improves the tunneling probability at reduced bias. The modeling predicts that linearly grading from 50% Al to 30% Al on the p side increases the hole concentration eightfold, and tunneling rate fourfold compared with a non-graded junction (Fig. 2c,d).

Following the NPN diode structures, the graded TJ was then grown on top of LED devices The epitaxial structure of the tunnel junction-based UV LED is shown in Fig. 3(a) and the corresponding equilibrium energy band diagram is shown in Fig. 1(b). The growth was initiated with a 100 nm thick n-Al_0.3Ga_0.7N buffer layer (1 × 10¹⁹ cm⁻³ Si doping). The active region consists of three pairs of 2.5 nm Al_0.15Ga_0.85N quantum wells and 7 nm Al_0.3Ga_0.7N quantum barriers, which is followed by a 4 nm Al_0.7Ga_0.3 N electron blocking layer (EBL). The transparent TJ was grown at 700 °C immediately on top of the active region. The grading gives rise to a high concentration of holes very close to the active region. The growth was completed with a 170 nm n-type Al_0.3Ga_0.7N (2 × 10¹⁹ cm⁻³ Si doping) layer and a 25 nm n⁺-type Al_0.3Ga_0.7N (1 × 10²⁰ cm⁻³ Si doping) acting as a contact layer.

The surface morphology of the as-grown epitaxial structure was analyzed using Bruker Icon Dimension atomic force microscopy (AFM). An rms roughness of 1.33 nm was extracted for a scan area of $5×5$ μm². The absence of step-flow features is attributed to the high Si doping, since Si can act as an anti-surfactant.⁴² The $2θ−ω$ scan of the LED was measured using a Bruker D8 Discover x-ray diffraction (XRD) system. Simulated thickness/composition values [Fig. 3(a)] and experimental peaks/fringes were found to match fairly well, as shown in Fig. 3(d).

The LED device structures were fabricated using the same process as that of the PN + TJ didoes. Apart from J–V measurements, capacitance–voltage (C–V) measurements were carried out by reverse biasing the top n-contact with an excitation frequency of 5 MHz and an amplitude of 30 mV. Electroluminescence (EL) peaks were obtained from on-wafer measurement at room temperature using a calibrated Ocean Optics USB 2000 spectrometer coupled with a fiber optic cable. The external quantum efficiency (EQE) of the device was measured from the output power collected using a Thorlabs PM100D optical power meter fitted with a S120VC photodiode power sensor.

The TJ and the LEDs were separately simulated using doping density values similar to the MBE grown LED. The polarization charges in the graded layers of the tunnel junction and at the quantum well/barrier interface in the LED was based on previously calculated values.⁴³ Since the tunnel junction is at a significant distance from the active region, we expect that simulating these two components separately and adding the voltage can predict the voltage drop of the full TJ-LED structure.

The on-wafer room-temperature electrical characteristics are shown in Fig. 4. Measured electrical characteristics of the fabricated LED (⁠$100×100$ μm²) are shown in Fig. 4(a), together with the simulated characteristics for the LED. In the case of the simulation, the voltage drop for TJ, LED, and the sum of the simulated TJ and LED voltage drops are shown. At 20 A/cm², the experimental device exhibits a forward voltage of 5.55 V. This compares well with the simulated voltage drop (∼5.7 V), which is the combination of the tunnel junction loss (1.4 V) and the voltage drop across the active region (4.3 V) at 20 A/cm². Top-down CV measurements were done on $40 μm$ diameter circular pads, and a zero-bias depletion width of 41 nm [Fig. 4(b)] was extracted, approximately matching the expected depletion width from the equilibrium energy band diagram [Fig. 3(b)]. The C–V profile is relatively flat with respect to bias due to the heavy doping in the p- and n-regions. The effective charge density with respect to the depletion width is shown in Fig. 4(c). The peak near zero bias depletion width may be attributed to electron accumulation in the bottom-most quantum well. At larger reverse bias, depletion width expands, and the C–V profile suggests an apparent carrier density ∼5 × 10¹⁸ cm⁻³, which is similar to the predicted doping density. Capacitance at higher reverse bias could not be measured due to leakage and increase in the loss tangent of the test structures. Transfer length measurement patterns were characterized on the top and bottom layers. The ohmic behavior was observed for both contacts, with a sheet resistance of $786$ and $92Ω/□$ and a contact resistance of $1.24×10−4$ and $1.29×10−5 Ω cm2$ for the bottom and top contact layers, respectively.

Electroluminescence measurements carried out on a $100×100$ μm²LED at different current levels ranging from 100 to 350 A/cm² indicate a single peak at 339 nm at 350 A/cm² which blue-shifted from 343 nm at 100 A/cm² due to both quantum stark confined effect and band filling effect [shown in Fig. 5(a)]. No secondary peaks were observed in the measurement. On-wafer measurements show a peak EQE of 1.02% at a current density of 80 A/cm². The emission power spectrum with respect to the current density is shown in Fig. 5(b)—the values reported here correspond to direct measurements from the calibrated detector with no corrections for light extraction were made. An optical output power density of 10.7 W/cm² is recorded for a current density of 358 A/cm². Since these measurements were made on-wafer without an integrating sphere, the EQE and power density values may be underestimated. Figure 5(c) shows the voltage drop across the tunnel junction at 20 A/cm² for previously reported tunnel junction-based UV LEDs as a function of the Al composition (%) across the tunnel junction layers. The tunnel junction voltage drop was estimated as the difference between the total voltage drop of the LED measured at 20 A/cm² and the bandgap of the quantum wells (assumed to be the same as the emission spectrum peak photon energy) except for Ref. 32. In Ref. 32, the tunnel junction drop is calculated by taking the total voltage drop at 20 A/cm² and subtracting the voltage drop at 20 A/cm² from the PN junction mentioned in Ref. 32. Previously reported transparent tunnel junctions (red spheres) showed relatively high operating voltages, while the use of InGaN and GaN interlayers within TJs has typically led to better performance. Our results show that a combination of high doping and compositional grading can enable low forward voltage drop, at least at Al-content in the range discussed here.

In summary, we have demonstrated low voltage drop transparent tunnel junctions with Al content ≥ 30% in the tunnel junction grown by molecular beam epitaxy for UV-B LEDs. With a 30% Al composition, emission wavelengths in the UVA and UVB region (up to 300 nm wavelength) can be obtained for various applications. The device exhibited a peak EQE of 1.02% and an output power density of 10.76 W/cm², which indicates excellent hole injection through the tunnel junction. The efficiency of these LEDs could be further improved by optimizing the active region design. Such low voltage drop transparent tunnel junction at high Al-content will be beneficial for higher current density applications like lasers and cascading LEDs.

This research was funded by the National Science Foundation (NSF) under Grant No. 2034140.