Introduction

The combination of high-efficiency red, green, and blue (RGB) light emitting diodes (LEDs) can produce more efficient white LEDs by eliminating the Stoke’s losses related to the phosphor1,2. A peak external quantum efficiency (EQE) exceeding 83% has been reported for the GaN-based blue LEDs3. However, the EQE of the InGaN/GaN green LEDs is still relatively poor. The low efficiency of green LEDs, a phenomenon known as the ‘green gap’, is a critical factor hindering the realization of high-performance RGB white LEDs in solid-state lighting4,5,6,7. In general, there are two main reasons for the ‘green gap’ phenomenon. On one hand, InGaN/GaN MQWs with higher indium (In) composition are required for LEDs emitting at green spectra. Due to the 11% lattice mismatch between InN and GaN, the c-plane InGaN/GaN multiple quantum wells (MQWs) grown on sapphire substrate are subject to in-plane compressive stress and suffer from quantum confined Stark effect (QCSE) induced by the strain-induced piezoelectric field8,9,10,11,12,13,14,15,16. The QCSE limits the internal quantum efficiency (IQE) of green LEDs by reducing the overlap integral between the electron and hole wave functions, leading to a lower radiative recombination efficiency.On the other hand, low growth temperatures is necessary to incorporate higher In-concentration in the InGaN/GaN MQWs for achieving green emission, which introduces impurities and defects, resulting in an increase in Shockley-Read-Hall (SRH) non-radiative recombination17.

Several solutions have been proposed to enhance the efficiency of green LEDs including nanopatterned sapphire substrate18, staggered InGaN QW19,20, non- and semi-polar QWs21,22,23, lattice-matched AlGaInN barrier24, and AlGaN or AlInN interlayers25,26,27,28. An in-situ low-temperature GaN (LT-GaN) or AlN nucleation layer (NL) between the high-temperature GaN epilayers and substrate is generally introduced to reduce defect density in GaN film29,30. The ex-situ sputtered AlN NL is another choice for green LEDs with advantages over LT-GaN/AlN NL31,32. First, the sputtered AlN NL enables one-step growth of GaN film to reduce growth time and thermal cycles of epitaxy. Second, a low threading dislocation density (TDD) of GaN film can be obtained using sputtered AlN/sapphire template. However, Miyoshi et al. demonstrated that the InGaN/GaN MQWs grown on AlN/sapphire template has a larger in-plane compressive strain than those grown on LT-GaN/sapphire template33. The larger in-plane compressive strain in the InGaN QWs would induce a larger piezoelectric polarization electric field, resulting in a more severe QCSE for LED on sputtered AlN/sapphire template. Therefore, to realize high-efficiency green LEDs on sputtered AlN/sapphire template, the in-plane compressive strain in the green InGaN/GaN MQWs should be manipulated to alleviate the detrimental effect of the QCSE.

In this paper, we demonstrate that the in-plane compressive strain of green LEDs on sputtered AlN/sapphire templates can be modified by changing thickness of the sputtered AlN NL. Compared to the green LED on LT-GaN (19 nm)/sapphire template, the green LED on sputtered AlN (19 nm)/template has larger in-plane compressive strain and thus lower EQE. However, at an optimal thickness of sputtered AlN NL (33 nm), the in-plane compressive strain of green LED can be minimized. These findings reveal that in-plane compressive strain of green LEDs on LT-GaN and sputtered AlN/sapphire templates plays a dominant role in determining device efficiency, while the thickness manipulation of ex-situ sputtered AlN NL can be an effective method to relax in-plane compressive strain. As a result, the EQE of 526 nm green LED on sputtered AlN (33 nm)/sapphire template is 6.1% larger than that of the green LED on LT-GaN (19 nm)/sapphire template at 20 A/cm2. The higher efficiency of green LED on sputtered AlN (33 nm)/sapphire template accompanied with reduced growth time and thermal cycles of epitaxy suggest the ex-situ sputtered AlN NL may be a superior alternative solution relative to the in-situ LT-GaN NL.

Results and Discussion

Figure 1(a) shows schematic representation of the green LEDs epitaxial structure. Photoluminescence (PL) spectra of the green LED measured using a 405 nm laser diode are described in Fig. 1(b). As the sputtered AlN NL thickness increases, the PL peak intensity of green LEDs on sputtered AlN/sapphire templates increases and reaches a maximum value at a thickness of 33 nm, thereafter decreasing with a further increase of the thickness to 40 nm. While the PL peak intensity of green LEDs on sputtered AlN (19 nm)/sapphire template is weaker than that of green LEDs on LT-GaN (19 nm)/sapphire template, the PL peak intensity of green LEDs on sputtered AlN (33 nm)/sapphire template is approximately 1.02 times stronger than that of the green LED on LT-GaN (19 nm)/sapphire template. Such great improvement shows the advantage of using optimal AlN NL thickness for high-efficiency green LEDs. Figure 1(c,d) show cross-sectional TEM images of green LEDs. Thickness fluctuations are observed through a single QW, which could explain the broad PL spectra for these green LEDs.

Figure 1
figure 1

(a) Schematic representation of InGaN/GaN the green LED epitaxial structure. (b) PL spectra (T = 300 K) of the green LED on LT-GaN and sputtered AlN/sapphire templates. (c and d) Cross-sectional TEM images of the green LED epitaxial structure.

Full size image

We compared TDD of green LEDs on LT-GaN (19 nm) and sputtered AlN (19 nm)/sapphire templates. Screw (a-type), edge (c-type), and mixed-type (a+c-type) dislocations have a Burgers vector b = <0001>, b = 1/3 <11–20>, and b = 1/3 <11–23>, respectively. Cross section was cut out of the green LED to identify the screw, edge, and mixed type dislocations in the (10–10) zone axis using two perpendicular diffraction vectors g = (0002) and g = (11–20). Figure 2(a,d) demonstrate the cross-sectional bright-field (BF) TEM images of green LEDs on LT-GaN (19 nm) and sputtered AlN (19 nm)/sapphire templates, where the screw, edge, and mixed type dislocations are marked in S, E, and M, respectively. The cross-sectional BF TEM images taken under the condition of g = (0002) (Fig. 2b,e) reveal the screw and mixed dislocations and those taken under the condition of g = (11–20) (Fig. 2c and f) reveal the edge and mixed dislocations. The screw dislocation should be in contrast for g = (0002) and disappear for the perpendicular diffraction vector g = (11–20), while the edge dislocations should be in contrast for g = (11–20) and disappear for the perpendicular diffraction vector g = (0002). By comparing same areas of the sample with two perpendicular diffraction vectors, most of TDs can be identified as mixed dislocation since they should be in contrast for both these vectors. We can clearly see that the green LED on sputtered AlN/sapphire template has a lower TDD than the green LED on LT-GaN/sapphire template.

Figure 2
figure 2

Cross-sectional TEM images of green LEDs cut in the (10–10) zone axis. (a,d) Cross-sectional BF TEM images showing screw (S), edge (E), and mixed (M) dislocations. (b,e) Cross-sectional BF TEM images with g = (0002) showing screw and mixed type dislocations. (c,f) Cross-sectional BF TEM images with g = (11–20) showing edge and mixed type dislocations.

Full size image

To investigate the effects of the NLs on crystal quality of green LEDs, we performed high-resolution X-ray diffraction (XRD) rocking curves measurement. Figure 3 compares the full widths at half-maximum (FWHMs) of symmetric (002) and asymmetric (102) reflections for green LEDs on LT-GaN and sputtered AlN/sapphire templates. The (002) plane X-ray rocking curve FWHMs are 249.1 and 190.6 arcsec for the green LED on LT-GaN (19 nm)/sapphire template and the green LED on sputtered AlN (19 nm)/sapphire template, respectively. The (102) plane X-ray rocking curve FWHMs are 247.6 and 233.4 arcsec for the green LED on LT-GaN (19 nm)/sapphire template and the green LED on sputtered AlN (19 nm)/sapphire template, respectively. Compared to the green LED on LT-GaN (19 nm)/sapphire template, we observed a decrease of the (002) FWHM and (102) FWHM for the green LED on sputtered AlN (19 nm)/sapphire template. It is well known that the FWHMs of symmetric (002) and asymmetric (102) reflections are affected by screw and edge dislocation densities, respectively34,35. According to the measured (002) and (102) FWHMs, the screw dislocation densities of green LEDs on LT-GaN (19 nm) and sputtered AlN (19 nm)/sapphire templates are determined to be 1.25 × 108 and 7.29 × 107 cm−2, respectively; the edge dislocation densities of green LEDs on LT-GaN (19 nm) and sputtered AlN (19 nm)/sapphire templates are determined to be 3.25 × 108 and 2.89 × 108 cm−2, respectively. These results indicated that the screw dislocation density of the green LED on sputtered AlN (19 nm)/sapphire template was much lower than that of the green LED on sputtered AlN (19 nm)/sapphire template.

Figure 3
figure 3

(a) Symmetric (002) and (b) asymmetric (102) reflections for green LEDs on LT-GaN and sputtered AlN/sapphire templates.

Full size image

We also analyzed the effect of sputtered AlN NL thickness on crystal quality of green LEDs. When the thickness of sputtered AlN NL was varied from 19 nm to 40 nm (19, 26, 33 and 40 nm), the (002) FWHMs of green LEDs on sputtered AlN/sapphire templates were 190.6, 128.3, 121.9, and 122.1 arcsec; the (102) FWHMs of green LEDs on sputtered AlN/sapphire templates were 233.4, 202.3, 197.4, and 192.6 arcsec. The corresponding screw dislocation densities of green LEDs on sputtered AlN/sapphire templates were 7.29 × 107, 3.31 × 107, 2.98 × 107, and 2.99 × 107 cm−2; the corresponding edge dislocation densities of green LEDs on sputtered AlN/sapphire templates were 2.89 × 108, 2.17 × 108, 2.07 × 108, and 1.97 × 108 cm−2. This revealed that the TDD of green LEDs on sputtered AlN/sapphire template would be significantly increased when using a thin sputtered AlN NL (19 nm).

High-resolution XRD was exploited to extract out-of-plane lattice parameter c and strain state of green LEDs on LT-GaN and sputtered AlN/sapphire templates. Figure 4(a) shows ω-2θ scans through (0002) reflection of GaN for green LEDs. The lattice constants c for the green LEDs with LT-GaN NL (19 nm) and with various thicknesses of sputtered AlN NLs (19, 26, 33, and 40 nm) on sapphire substrates are estimated to be 5.189500 Å, 5.190320 Å, 5.189862 Å, 5.189186 Å, and 5.189316 Å, respectively. The lattice contact c for the stress-free GaN bulk substrate is estimated to be 5.184980 Å. We find that the lattice constant c for the green LED on LT-GaN (19 nm)/sapphire template is smaller than the value for the green LED on sputtered AlN (19 nm)/sapphire template. Furthermore, the lattice constants c for the green LEDs on sputtered AlN/sapphire template are found to vary with the thickness of the sputtered AlN NL. According to the Poisson effect, the expansion of out-of-plane lattice indicates the shrinkage of in-plane lattices at the same time. Hence, the larger lattice constant c reveals the presence of larger in-plane compressive strain as described in the following.

Figure 4
figure 4

(a) XRD ω-2θ scans and (b) Raman spectra for green LEDs on LT-GaN and sputtered AlN/sapphire templates.

Full size image

Figure 4(b) shows Raman spectra around the high-energy E2(high) mode and the whole Raman spectra of the samples were shown in Supplementary Figure S1. The E2(high) mode is sensitive to strain and has been extensively used to quantify the stress in GaN epilayersCharacterization and measurement

Cross-sectional TEM samples were prepared by focus ion beam milling using Ga ions at 30 kV, and then micrographs were taken with an FEI Talos F200X system at 200 kV (FEI, Hillsboro, OR, USA). For the time-integrated PL measurements, we used the 405 nm line of a 5 mW semiconductor laser as an excitation source. The detection part of the PL system consisted of a charge-coupled device detector (Princeton Instruments PIX IS256, Trenton, NJ, USA) connected to a spectrometer (Princeton Instruments SP2500i, Trenton, NJ, USA). PL spectra were measured using a reflex confocal microscope to focus the laser beam on the sample to a spot size of 500 μm in diameter. The PL signal was dispersed in a 50 cm monochromator with an 1800 grooves/mm grating and detected by charge coupled device detector. High-resolution XRD was performed with Cu Kα1 radiation of wavelength 1.54056 Å using a PAnalytical X’Pert Pro MRD diffractometer with a mirror, a four-bounce asymmetric Ge (220) monochromator and a triple-bounce analyser. Micro-Raman measurements were conducted by a Renishaw InVia spectrometer with the 633 nm frequency laser as the excitation source. The sub-bandgap laser illumination and low laser power guarantee that there is a negligible laser absorption and no interference with the device operating during the measurement. A 100× objective (N.A. = 0.85) microscope objective was used for focusing the laser and collecting the Raman scattered photons. Before each measurement, the wavelength and intensity were calibrated by silicon standard using the calibration system. All the Raman spectra were acquired with a diffraction grating of 1800 line/mm and slit width of 65 μm in back scattering geometry. The light output power versus current and the current versus voltage characteristics of the green LEDs were measured by using a home-built probe station system40.