Green semipolar III-nitride light-emitting diodes grown by limited area epitaxy

InGaN-based light-emitting diodes (LEDs) grown on semipolar orientations of GaN are an attractive alternative to conventional c-plane devices for long-wavelength emission due to reduced polarization-induced internal electric fields,^1,2 increased optical polarization,^3–7 and improved surface morphology.^8–10 Cyan $(202¯1¯)$ LEDs have shown narrow spectral linewidths and small wavelength shifts with increasing current density,¹¹ and yellow-green $(112¯2)$ LEDs have demonstrated improved thermal performance compared with commercial devices.¹² However, semipolar LEDs have yet to displace conventional c-plane LEDs.¹³ This is largely due to the limited size and high cost of bulk semipolar substrates, as well as materials challenges such as the formation of misfit dislocations (MDs) and basal-plane stacking faults.^14–17 

An InGaN layer grown pseudomorphically on GaN is elastically strained due to the mismatched crystal interface. Once the layer reaches a certain critical thickness, compressive in-plane strain may be relieved by MD formation via glide on inclined slip planes. Slip occurs most easily on the basal c-plane. However, a nonzero resolved shear stress on the basal plane is required for slip to occur, and this condition is met only for semipolar growth orientations.¹⁸ In contrast, for c-plane orientations, there is no resolved shear stress on the basal plane, so relaxation by basal plane (BP) slip cannot occur.¹⁹ BP slip is observable in semipolar films as a one-dimensional array of MD lines parallel to the in-plane a-direction or m-direction of the growth surface.^14,20 Alternative non-basal plane (NBP) slip systems are also accessible such as prismatic slip along inclined nonpolar planes and pyramidal slip along semipolar planes.^21–24 These are typically observed as secondary slip mechanisms, occurring at thicknesses greater than the observed critical thickness for BP slip.^21,25 NBP slip has been associated with a severe degradation in performance for semipolar light-emitting devices.^21,26

Mitigating the detrimental effects of MD formation is important for continued development of semipolar III-nitride devices. In semipolar InGaN-based laser diodes (LDs) the growth of thick strained (Al,In,Ga)N cladding layers is typical, so strain management techniques are needed to suppress MD formation near the active region of the devices. These techniques include the growth of relaxed buffers to isolate MDs far below the emitting layers,²⁷ dislocation pinning by impurity hardening,²⁸ and the growth of strain-compensating layers.^26,29,30 The simplest MD prevention technique, however, is to keep device structures below the critical thickness for NBP dislocation formation. Consequently, state-of-the-art semipolar LEDs are typically composed of thin single quantum well (SQW) active regions.^31,32 Although Yamamoto et al. have demonstrated a green SQW semipolar LED with a peak external quantum efficiency (EQE) of 20%,³¹ this device design limits the volume of and increases the carrier density in the active region, which has been associated with efficiency droop at high current densities due to Auger recombination and carrier leakage out of the active region.^33,34

The thermodynamics of MD formation is given by the Matthews–Blakeslee critical thickness (h_c) criterion, which balances the energy of creating a dislocation with the elastic strain energy released in doing so.³⁵ The critical thickness is given by

where $b∥$ and $b⊥$ are the in-plane and out-of-plane components of the edge Burgers vector, respectively; $bs$ is the screw component of the Burgers vector; $εm$ is the component of misfit strain released by MDs; $ν$ is the Poisson ratio; and $r0$ is the dislocation core cutoff radius. This treatment assumes that a source of dislocations exists for MD formation and that the material is isotropic.^36,37 If the supply of dislocations is insufficient to relieve the misfit strain, it is theoretically possible to grow metastable layers many times the M-B critical thickness.³⁸ This was first explored experimentally by Fitzgerald et al. in systems where pre-existing substrate threading dislocations (TDs) served as the dislocation source.³⁹ It was shown that MD formation could be interrupted by introducing physical constraints to TD glide such as a mesa sidewall or dielectric layer. In doing so, the distribution of MDs, and therefore the degree of relaxation, could be controlled by restricting accessibility of TDs to certain regions of the substrate. This is a concept known as limited area epitaxy (LAE).

The maximum plastic strain relieved (⁠$εpmax)$ by MDs is given by

where $ρTD$ is the pre-existing TD density and $l$ is the average MD run length.⁴⁰ Using LAE, l can be reduced by patterning a mesa with sidewalls to act as barriers to glide. As the mesa dimension is reduced in the direction of glide, the linear density of MDs on the mesa decreases. This results in a localized reduction in the strain relieved by slip and allows for coherent layers beyond the critical thickness to be grown. Hardy et al. used LAE for the coherent growth of highly strained InGaN waveguides on $(202¯1)$⁠, leading to the demonstration of an InGaN-based LD with a lasing wavelength of 523 nm.⁴⁰ In the present work, we apply LAE to a $(202¯1)$ green LED to suppress relaxation in a highly strained multiple quantum well (MQW) active region. This MQW active region design has enabled the growth of $(202¯1)$ green LEDs with a larger active region volume than what has been previously achieved in SQW designs.^11,31

LEDs were grown by atmospheric pressure metalorganic chemical vapor deposition (MOCVD) on free-standing $(202¯1)$ GaN substrates provided by Mitsubishi Chemical Company (MCC) with a TD density of ∼5 × 10⁶ cm⁻². Figure 1 shows a cross-sectional schematic and perspective view of an LAE LED grown and fabricated on a $(202¯1)$ patterned substrate. Samples with and without substrate patterning were co-loaded during a single growth to study the effects of LAE on device performance. Substrates were patterned using a photoresist hardmask and a chlorine-based dry etch to form mesas with a height of 1 μm. The mesas were 300 μm long and 5 μm to 20 μm wide, with the long side of the mesa oriented parallel to the in-plane projection of the c-direction on the sample surface. This LAE geometry was previously determined to be the most effective at preventing MD formation.²⁵ Mesas of the same width were spaced 2 μm apart in a 300 μm wide linear array to define a single LED. LED epitaxial layers were grown on the tops of the mesas and were composed of a 1 μm n-type GaN:Si layer, an undoped In_0.25Ga_0.75N (4 nm)/GaN (4 nm) five-period MQW active region, an 8 nm p-type Al_0.15Ga_0.85N:Mg electron blocking layer, and a 40 nm p-type GaN:Mg layer. To confine current injection to the tops of the mesas, 100 nm of SiO₂ was sputtered in the trench between the mesas, on the mesa sidewalls, and 1 μm onto the top of each mesa. 300 × 300 μm² mesas with a height of 1 μm were patterned using a photoresist hardmask and a chlorine-based dry etch. A 100 nm thick indium tin oxide (ITO) layer was deposited by electron beam deposition to form a transparent p-contact. An n-contact of Ti/Al/Ni/Au (10/100/100/100 nm) was deposited onto the n-type GaN and a pad electrode of Ti/Au (20/300 nm) was deposited onto both the n-contact and the ITO-based p-contact. The presence of extended defects was evaluated by fluorescence microscopy (FLM) under a fixed excitation flux with an excitation band from 450–490 nm and a 520 nm longpass filter. To improve the extraction efficiency, the backside of each sample was thinned, polished, and roughened by forming truncated pyramids as described by Zhao et al.³² Samples were then diced and individual LEDs were mounted on a silver header and encapsulated with silicone. Electroluminescence (EL) measurements were conducted at room temperature under direct current bias in a calibrated integrating sphere.

Figure 2 shows the fluorescence micrographs for LAE mesas composed of 5 μm, 10 μm, and 20 μm wide mesas and for the co-loaded sample without patterning. Nonradiative dark line defects (DLDs) are visible on the mesas and increase in density with increasing mesa width. These DLDs are indicative of stress relaxation by the formation of MDs. One set of MDs has a line direction parallel to the in-plane a-direction $[12¯10]$ and corresponds to MD formation via BP slip. The other two sets are oriented at an inclination to the in-plane a-direction and are consistent with relaxation along a NBP slip system. Based on previous TEM and trace analysis on $(202¯1)$⁠, the NBP MDs observed here most closely correspond to relaxation along the prismatic, or m-plane, slip system.²² As shown by Fig. 2(a), 5 μm LAE mesas were able to significantly reduce the density of BP MDs and to completely eliminate the presence of NBP MDs. However, comparison of Figs. 2(b) and 2(c) with Fig. 2(d) indicates that LAE mesas 10 μm and wider were ineffective at reducing MD density. This is a result of increasing dislocation intersection and defect multiplication and will be further discussed below. The M-B critical thickness for In_0.12Ga_0.88N grown on $(202¯1)$ oriented GaN is 20 nm for BP slip and 10 nm for prismatic slip.¹⁸ The MD density on the 5 μm LAE mesas was too low to cause a measurable degree of relaxation, resulting in the nearly coherent growth of an active region with an average composition of In_0.12Ga_0.88N and a thickness of 40 nm that is at least twice the theoretically predicted limit.

To assess the degree to which MDs affected radiative efficiency, we observed their impact on luminescence during electrical injection. Figs. 3(a) and 3(b) show optical micrographs taken of a fully fabricated device during fluorescence and electrical injection, respectively. While BP MDs are confined to the glide interface at the bottom of the MQW, NBP MDs are associated with the generation of high densities of TDs that are not confined to the glide interface and can quench luminescence through all layers of the MQW.²¹ Under fluorescence conditions, carriers are excited uniformly throughout the active region and recombination occurs evenly throughout the MQWs, so both BP and NBP MDs appear to impact luminescence equally. When the QWs are electrically injected, however, only NBP MDs are visible. Since BP MDs are not observed under electrical injection, they must have a reduced impact on the electroluminescence of the device. The difference between fluorescence and EL can be explained by considering the mechanics of carrier injection in an MQW, which relies on carrier transport across quantum barriers for uniform carrier distribution to occur. It is possible that the MQW device structure, which was used in this study to increase the volume of and reduce the carrier density in the LED active region, caused non-uniform carrier injection throughout the QWs. As shown schematically in Ref. 41, in the case of a $(202¯1)$ LED, the built-in and polarization-induced electric fields in the QW are in the same direction, which results in a large thermionic barrier for hole transport. This large thermionic barrier, as well as a large hole effective mass, can cause an uneven carrier distribution with the majority of recombination occurring in the wells closest to the p-GaN, far from the glide interface where BP MDs would have their largest effect on radiative efficiency.^41,42

Figure 4 shows the dependence of EQE on drive current for the fully packaged LEDs. The peak EL wavelength at 35 A/cm² for these devices ranged between 532 nm and 536 nm. As expected from Figs. 2 and 3, the EQE increased for all measured drive currents with decreasing LAE mesa width. LEDs composed of 5 μm LAE mesas showed a 73% enhancement in peak EQE compared with LEDs composed of 20 μm LAE mesas. This improvement was caused by a reduction in the number of nonradiative MDs and TDs interacting with the device active region. As the mesa width was decreased, the MD run length decreased and the linear density of MDs along the length of the mesa was reduced. The decrease in MD density, along with the thinner mesa geometry, reduced the probability of dislocation intersection between BP and NBP MDs and between NBP MDs running in opposing directions. This intersection process is believed to generate additional TDs that may also bend at the glide plane, causing a runaway process of dislocation multiplication. It is likely that this accounted for the rapid rise in NBP MD density with increasing LAE mesa width. Additionally, a reduction in run length increased the proportion of TDs originating beneath the mesa that glided into the mesa sidewall, pinning the dislocation at that boundary and sweeping it away from the injected area of the device. This led to a local reduction in TD density for devices with narrower mesa widths, which may have further improved device performance for these devices compared to devices with wider mesa widths. A possible additional source of improved EQE in devices with narrower mesa widths was enhanced light extraction due to the increasing number of mesa sidewalls. Still, the peak EQE of the best devices in this study was approximately an order of magnitude lower than that of previous reported green semipolar InGaN-based LEDs.³¹ We attribute this to a reduction in radiative efficiency due to non-uniform carrier injection across the MQW active region, which likely caused an uneven distribution of electrons and holes near the bottommost and topmost QWs, respectively. Further optimization of this MQW active region design is needed to mitigate these effects and improve device efficiency.

In previous studies, where InGaN layers of lower indium composition were grown, a measurable amount of relaxation occurred by BP slip before the onset of NBP slip.^21,25 In the present work, however, where the average indium composition of the active region is higher, NBP slip was observed shortly after BP slip and quickly led to catastrophic defect propagation through dislocation intersection and subsequent multiplication. This can be explained by considering the competing forces that influence MD formation. The resolved shear stress on the m-plane (⁠$σm$⁠) and basal plane (⁠$σc$⁠) provide the mechanical driving force for dislocation glide.²⁴ For all compositions of InGaN grown on $(202¯1)$⁠, $σm$ is greater than $σc$⁠. However, the barrier to dislocation motion, known as the Peierls stress, is smallest on the close-packed basal plane. In the case of dilute InGaN layers, the difference in shear stresses is smaller than the difference in Peierls stresses on the two slip planes, leading to the onset of BP slip well before NBP slip. As the composition of the InGaN layer increases, the resolved shear stress on the m-plane increases faster than that on the basal plane until the difference in shear stresses is larger than the difference in Peierls stresses on the two slip planes. This results in the earlier onset of NBP slip for high indium content InGaN layers relative to dilute compositions.

In summary, we have used LAE to grow coherent MQW LEDs on a $(202¯1)$ GaN substrate with an active region thickness twice the theoretical limit. An improvement in device performance was observed as the mesa width was reduced from 20 μm to 5 μm due to a reduction in MD and TD intersection in the LED active region. LAE LEDs with 5 μm mesas exhibited a complete elimination of NBP MDs, yielding a 73% enhancement in peak external quantum efficiency relative to the devices with the widest mesas.

This work was supported by the Solid State Lighting & Energy Electronics Center (SSLEEC) and the KACST-KAUST-UCSB Solid State Lighting Program (SSLP). A portion of this work was done in the UCSB nanofabrication facility, part of the NSF funded National Nanotechnology Infrastructure Network (ECS-03357650). This work made use of the Materials Research Lab (MRL) Central Facilities at UCSB supported by the Materials Research Science and Engineering Center program of the NSF under Award No. DMR 1121053. C. D. Pynn was supported by the NSF Graduate Research Fellowship Program under Grant No. DGE 1144085. H. Gardner was supported by the CISEI program through the UCSB MRL and the International Center for Materials Research under NSF Award No. DMR 0843934.