Solid-state lighting (SSL) holds the promise of a more energy-efficient, longer-lasting, more compact, and lower maintenance substitute for today’s incandescent and fluorescent light sources. Since lighting currently represents about 22% of all electricity consumption, the adoption of SSL could significantly reduce greenhouse gas emissions. Light-emitting diodes (LEDs) based on In_xGa_(1−x)N alloys are currently the most promising candidates for realizing efficient SSL. InGaN is a direct wide bandgap semiconductor with an emission which can span the entire visible spectrum via compositional tuning. However, InGaN LED performance is highly wavelength dependent. Indeed, ultra-bright and efficient blue InGaN-based LEDs are readily available but the efficiency of InGaN based green LEDs is still far from adequate for use in SSL.

The lack of economical lattice-matched substrates for the growth of III-nitrides necessitates the usage of GaNmismatched silicon carbide (SiC) or sapphire (Al₂O₃) substrates, which leads to dislocation densities on the order of 10⁸ cm⁻². The high performance of blue LEDs in spite of these dislocations is attributed to indium segregation in the InGaN layers that produces nanometer-wide indium-rich regions that behave like quantum dots. These quantum dots (QDs) localize the carriers, and prevent them from recombining non-radiatively at the dislocation sites. Difficulty in realizing high-power green LEDs has three major parts: (1) the limited solubility of indium in InGaN imposes a restricted growth window for the green-emitting InGaN active layer, (2) InGaN with high indium content becomes unstable at elevated growth temperatures required for other layers in the device leading to indium migrating out of the active layers, which reduces the LED spectral quality, and (3) InGaN with high indium content generates dislocations leading to lower performance.

A five step high quality growth scheme is developed that elimininates the dislocations on the laterally overgrown area. Figure 1 displays the AFM of the fully coalesced LEO GaN templates. The surface above the opening region (Fig. 1a) is similar to that of conventional GaN, exhibiting chaotic atomic steps, the surface termination of which identify screw/mixed type threading dislocations. Contrarily, the wing region (Fig. 1b), where lateral growth occurs, possesses well-ordered parallel atomic steps with no atomic step terminations. The entire surface, including the coalescence region, where the neighboring atomic steps interfere, is observed in Fig. 1c. In order to study the dislocations, a hot (170°C) phosphoric acid (85% H₃PO₄) treatment for 15 min was used. This etch-pit-density study reveals no discernable dislocations in the wing areas, whereas in the LEO GaN coalescence area and opening region dislocation densities of (2 ± 1) × 10⁸ cm⁻² and (9 ± 2) × 10⁸ cm⁻² are observed, respectively (Fig. 1d). For comparison, conventional GaN was observed to have a dislocation density of (9 ± 1) × 10⁸ cm⁻². This phosphoric acid treatment is capable of distinguishing between edge and screw or mixed type dislocations. The bigger pits correspond to dislocations with a screw component whereas smaller ones correspond to edge-type dislocations. It is known that edge-type dislocations may exist in the wing region as their bending is very sensitive to growth conditions. The non-existence of any discernable dislocations in our wing regions shows the quality of the five-step LEO GaN developed, and establishes a baseline from which we can study the effect of dislocations on blue and green InGaN-based LEDs. For LEO GaN, no GaN peak separation is observed, which shows that there is no significant plane tilt , supporting our high-quality growth scheme described.

Figure 1: AFM images of coalesced LEO GaN. (a) Opening (2 μm × 2 μm) and (b) wing (2 μm × 2 μm) have root mean square (RMS) roughness of 1.9 Å and 1.5 Å, respectively. (c) Larger area (12 μm × 23 μm) AFM scan. (d) Dislocations, revealed by hot phosphoric acid treatment, which are seen as dark spots. ‘W’, ‘O’, and ‘C’ correspond to wing, opening, and coalescence regions, respectively

Electroluminescence (EL) spectra were acquired for the LEDs under pulsed current injection (duty cycle of 10% and frequency of 5 kHz) in order to help reduce heating effects under higher current injection (Fig. 2). The minimal role of heating was confirmed by measuring the peak intensity versus power and noting that, for the currents used in this study, no thermal roll-over was observed to occur. The EL spectra of blue LEDs are shown in Fig. 2(left). The inset (left) shows that both devices demonstrate a blue shift (from 465 to 446 nm) with increasing current. The decrease of wavelength with injection current is attributed to band-gap renormalization (due to free-carrier screening of the piezoelectric (PE) field). At all but the lowest currents, B_LEO has a slightly longer wavelength emission than B_Conv(Fig. 2 inset). This observation is in agreement with the photoluminescence studies. Device B_LEO has a narrower EL spectra than B_Conv (Fig. 2). This suggests a more uniform indium distribution throughout the active layer. Indeed, our atomic force microscopy studies directly illustrates the more uniform surface of blue MQWs on LEO GaN, supporting the advantage of LEO templates.

In Fig. 2 (left) inset the EL FWHMs of the blue LEDs are plotted as a function of the injection current; the FWHM of the device on conventional GaN can be seen to decrease with current, while the FWHM of the device on LEO GaN can be seen to increase. The EL FWHM broadening in B_LEO could be related to dislocation alignment in the LEO openings resulting in an electrical field build up. The EL spectra of the green LEDs are shown in Fig. 9. Device G_LEO is observed to have longer peak wavelength than device G_Conv. (Fig. 2 (right) inset). With increasing current, the EL FWHM increased for both devices; however, the EL peak shift of sample G_LEO is larger than that of G_LEO, possibly due to the effects of piezoelectric fields. These piezoelectric fields may also be responsible for the broader EL FWHM observed in green LEDs.

Comparing blue and green LEDs, a more pronounced EL peak shift is observed for green LEDs. This suggests stronger piezoelectric effects in the active layer. Piezoelectric effects are expected to be more pronounced in green MQWs due to the higher indium content of the layers. The broader EL FWHMs of green LEDs than those of blue ones indicates a bigger indium fluctuation through the green active layer. This is in agreement with the AFM measurements we have reported.

Figure 2: Electroluminescence spectra of blue (left) and green (right) LED on conventional ( _Conv. ) and LEO GaN ( _LEO ).
Insets display the peak wavelength at different current injections.

A five-step LEO GaN growth technique for high-quality LEO GaN growth is introduced. Blue- and green-emitting active layers and LEDs on conventional GaN and five-step grown LEO GaN templates are realized. AFM, XRD, and PL are used to study the structural and optical properties of the active layers, and the effects of dislocations on blue and green active layers are identified. Significant differences in blue and green active layer surfaces are analyzed. The high-quality LEO templates are observed to be important for smoother active layer surface morphologies. Blue and green LEDs on conventional GaN and high-quality LEO GaN are studied. Green LEDs are observed to be leakier than blue ones, and no significant differences between green LEDs on conventional GaN and LEO GaN in terms of I–V behavior and peak power are observed. Green MQW quality is determined to be the bottleneck for high-performance green emitters, not template dislocation density.