AlGaN-based ultraviolet light-emitting diodes (UV LEDs), with the advantages of compact structure, long lifetime, and controllable wavelength, are promising in numerous applications, such as air/water sterilization [1], surface disinfection [2], plant growth [3], and the detection of biochemical agents [1,4,5]. However, the development of AlGaN-based UV LEDs is still in its infancy compared to InGaN-based blue and green LEDs, owing to various scientific and technical challenges, such as inferior crystal quality [6], low doping efficiency of high Al composition p-type layers [7], poor light extraction efficiency due to dominant TM light polarization [8], and UV light absorption by p-GaN and metal contacts [9]. As a matter of fact, the widely reported external quantum efficiencies (EQEs) of AlGaN-based UV LEDs are mostly below 5% [10–13].

Internal quantum efficiency (IQE) is one of the most important factors contributing to the EQE of UV LEDs. Ban et al. have demonstrated that in order to increase IQE to a decent value such as 50%, threading dislocation density (TDD) needs to remain as low as 3×108cm−2 [14]. This poses a stringent requirement on AlN or AlGaN epitaxial growth, as the mobility of Al atoms is inherently low compared to Ga. Recently, researchers have demonstrated the defect-insensitive behaviors of the UV multiple quantum wells (MQWs) and UV LEDs with high IQE values and strong luminescence intensities. This was achieved by epitaxial growth of high-Al content AlGaN and heterostructures on a sapphire substrate with a high misorientation angle. In our previous study, IQE as high as 90% was obtained from the MQWs grown on a 4° misoriented sapphire substrate, even if the TDD was in the range of 109cm−2 [15,16]. It has been verified that composition fluctuation occurred along the step-bunching region of the epitaxial thin film. Higher Ga content is induced near the step edges during growth due to relatively higher surface diffusivity of Ga atoms [17–19]. This leads to enhanced carrier confinement within the potential minimum of the active region and greatly improved IQE [20].

Despite progress made in the fabrication of UV LEDs on misoriented substrates, the physics of carrier transportation and recombination in the active regions is still poorly understood. In our previous work, it was reported that radiative recombination center formed in the step edge area with the support of μ-cathodoluminescence (CL) characterization [15]. Kojima et al. further proposed a carrier localization model, together with a hypothesis of current paths in the uneven quantum wells (QWs) [18,21]. But this model is based upon optical pumping results such as photoluminescence (PL) characterization. Direct evidence on the carrier localization effect during electrical injection is still lacking. Under optical pumping, photogenerated carriers are uniformly produced in the exposed area. Nevertheless, under electrical injection, the transportation and recombination of carriers are dominated by the built-in electric field and heterojunction barrier. To make things even more complex, current will selectively flow through a potential minimum in a 3D configuration under external bias, yet the physics of vertical and lateral transport through MQW structures is still poorly understood [15]. In other words, whether the existence of inclined quantum wells (IQWs) will enhance or suppress the carrier localization effect under electrical injection needs to be further investigated, which is critically important during future design and fabrication of UV LEDs. Furthermore, since no information was given with regard to the relationship between threading dislocations and Ga-rich regions near the step-bunched regions, one would intuitively argue that step bunching would deteriorate the crystalline quality of AlGaN and lead to current leakage paths. This concern has yet to be elucidated.

In this work, the dependence of current distribution on the surface morphology in the sub-300 nm emission of UV-LEDs grown on misoriented sapphire substrate was investigated in detail. It is demonstrated that as forward bias increases, the current flows through the step edge first, and then expands to adjacent regions. The macro-step structure was also modeled and simulated according to transmission electron microscopy observations. Current paths are formed in the step edges owing to Ga enrichment. Further investigations found that the IQWs have better carrier confinement due to both smaller bandgap and spontaneous polarization compared to adjacent flat QWs. This comprehensive study clearly illustrates that a UV LED grown on misoriented substrate is only one of the many techniques to achieve phase separation and carrier localization, which provides a new perspective in the realization of high-efficiency UV emitters.

AlGaN-based UV LEDs were grown on 2-inch c-plane sapphire substrates with an approximate 1° misorientation angle toward the m plane using an AMEC Prismo HiT3 MOCVD reactor in H2 ambient. Trimethylaluminum, trimethylgallium, and ammonia (NH3) were used as precursors of Al, Ga, and N, respectively. Hydrogen (H2) and nitrogen (N2) were used as the carrier gases. Initially, a 20 nm thick low-temperature (LT) AlN nucleation layer (NL) was deposited on the sapphire substrate at 850°C. A 2.8 μm high-temperature (HT) AlN template was subsequently grown under 1250°C, followed by 50 pairs of AlN/AlGaN superlattices used as dislocation filters. Then, a 2 μm Si-doped n-Al0.6Ga0.4 contact layer was deposited. Afterwards, four pairs of Al0.4Ga0.6N/Al0.55Ga0.45N MQWs were grown under 1050°C. A 20 nm p-Al0.7Ga0.4N electron barrier layer (EBL), 50 nm p-Al0.65Ga0.45N layer, and 50 nm p+-GaN contact layer were subsequently deposited. Surface morphologies and spatially resolved current distributions of the UV LEDs were characterized by a Veeco Dimension 3100V conductive atomic force microscope (CAFM) in contact mode with a Pt-coated Si probe. Crystalline qualities were analyzed using a point-focused high-resolution X-ray (Cu Ka1) diffractometer (HRXRD, Bruker D8 DISCOVER) equipped with a four-bounce symmetric Ge (220) monochromator. PL studies were performed using a Coherent Ar-F (193 nm) excimer laser at pumping power of 50mW⋅cm−2, collected by a Horiba iHR550 spectrometer. Electroluminescence (EL) was analyzed by a Keithley 4200-SCS semiconductor characterization system. Electron channeling contrast imaging (ECCI) investigation was performed using a Verios G4 UC characterization system with voltage of 20 kV. Transmission electron microscopy (TEM) samples were prepared using an FEITM Helios dual-beam focused ion beam scanning electron microscope system with a Ga ion source. Thicknesses of MQWs and dislocation distributions were characterized by an FEI probe-corrected Titan high-angle annular dark-field imaging-scanning transmission electron microscopy (HAADF-STEM) system operated at an acceleration voltage of 300 kV. All measurements were performed at room temperature (RT).

The current-voltage (I-V) curves shown in Fig. 1(a) represent a local electrical behavior near the step edge within the voltage range of −10 to +10V. The I-V characterization was performed consecutively 4 times. When forward voltage bias (0–10 V) was applied to the surface, even though the turn-on voltage increases with the scan proceeding, the local current becomes saturated at higher bias, indicating a stable current pathway. However, when UV LED is under reverse bias (−10to0V), the reverse current reduces as the number of scans increases and finally becomes negligible. Prior CAFM studies show that dislocations may serve as channels for large and stable reverse-bias leakage current [24]. This is in contrast with the observation in this work, suggesting that the local current distributions in the UV LEDs are not associated with dislocations. From the surface morphology image shown in Fig. 1(b), large numbers of step bunching are formed on the surface. Spradlin et al. found a significant decrease in reverse bias current after several scans and attributed this phenomenon to “charge trapping effects” [25]. It is believed that the same phenomenon occurred in this work. When the reverse bias is applied to the surface, electrons tend to fill the potential minima. When the potential minima are saturated, the reverse-bias current becomes negligible. Figure 1(c) illustrates the current distribution under forward bias of +7V, in which case the LED is already turned on. Higher current is represented by the dark contrast in the map, which corresponds perfectly with the step edges, indicating that under forward bias, the current is mainly localized on the step edges due to carrier localization effect, and a strong dependence on the surface morphology is therefore obtained.

Figure 3(d) shows the zoom-in view of the MQWs near the step edge. The orientation and thickness of Ga-deficient and Ga-rich layers of the sample can be distinguished by the dark and bright contrast, respectively [30]. The MQW is 1° tilted relative to the (0001) crystallographic plane, as indicated by the white dashed line in Fig. 3(d), in accordance with the substrate misorientation angle of ∼1°. Thickness expansion or even twist of the QWs can be observed at the edge. The width of the AlGaN QWs grown on the flat surface is ∼3nm, while the width slightly expands to ∼3.5nm near the step edges. Even though the exact composition difference in the step edges is not known, there is no denying that thickness and composition fluctuation will result in lower bandgap. This would in turn lead to the different electrical and optical behaviors of the UV LEDs.

Figure 4(c) illustrates the EL spectrum of the UV LED under an injection current of 200 mA. There are also two emission peaks located at wavelengths of 287 and 298 nm. Due to strong UV absorption in the p-GaN contact layer, PL characterization was performed under high pumping power, leading to a strong band filling effect and thus blueshift of the peak position. More importantly, during optical pumping, both Ga-rich and Ga-deficient regions are uniformly excited. On the contrary, under electrical injection condition, photons mostly come from Ga-rich regions in UV LEDs due to the carrier localization effect. As a consequence, PL exhibits a shorter emission wavelength compared to EL. In addition, note that the intensity of the longer-wavelength emission peak is stronger than that of the shorter-wavelength one in Fig. 4(c). This is in contrast with the PL spectrum, suggesting that under current injection, carriers are more inclined to pass through the IQW region, where the carrier localization effect is enhanced. Note that from the CAFM result in Fig. 1, current is mainly localized in the step edges when the device is turned on, confirming that EL characterization is highly dependent on the surface morphology. To demonstrate this hypothesis, EL spectra under various injection currents from 50 to 300 mA are shown in Fig. 4(d). A shorter-wavelength peak can barely be seen under the low injection current, while as the current injection increases, the intensity of the shorter-wavelength emission peak increases significantly. This can be ascribed to the saturation of the current paths at the step edges, which will be further discussed later on.

Figure 6(f) shows the simulated EL spectra under the injection current densities of 3×103, 7×103, and 1×104A/cm2. Single peak emission at 290 nm is observed under 3×103A/cm2, while double peaks appear when the current increases. Note that the current is mainly concentrated in the step edge when a relatively low current density of 3×103A/cm2 at the bias of +6V is applied on the sample [Fig. 6(c)]. When the current density increases to 1×104A/cm2, current spreads to the periphery of the step edge region, as shown in the inset in Fig. 6(c). This is clearly demonstrated by the occurrence of a high-energy side peak from the simulation result. Therefore, under the condition of high current density injection, carrier delocalization occurs, and the current distribution becomes more uniform.

Interestingly, step-bunching-induced Ga phase separation not only appears in AlGaN grown on high misorientation sapphire. Similar phenomena have been observed in MQWs grown on low misorientation substrate or even on-axis sapphires, where nitride thin films exhibit step-flow growth morphology [6,17,32]. Carrier localization has always been a common observation in III-nitrides, especially InGaN. Currents prefer to reside in the potential minima. The investigations shown in this work are in fact valid for all ternary III-V semiconductor systems exhibiting phase separation, including III-arsenide. Understanding the carrier transport and recombination behaviors in such thin films can greatly help us to improve the performance of III-V semiconductor devices.

In summary, a sub-300 nm UVB LED with IQE of 63% and EQE of 1.23% was designed and fabricated on a 1° misoriented sapphire substrate. Direct evidence on carrier transportation and recombination on step-bunched surface of UV-LED was provided. The current distribution strongly follows the surface morphology of the thin film. Top view ECCI and cross-sectional HAADF-STEM demonstrated that these current paths are not associated with threading dislocations, but rather are related to the step edges with Ga-rich potential minima. This structure was further modeled and simulated by solving the Poisson equations and carrier transport equations. An efficient recombination center is formed in the IQWs owing to higher carrier concentration and larger radiative recombination rates. Finally, a schematic diagram of a 3D energy band structure of the active region of the UV LED across the step edges was proposed.

Acknowledgment. The authors appreciate the technical support from Nano Fabrication Facility, Platform for Characterization & Test in Ningbo Institute of Materials Technology and Engineering, CAS.