Device-Quality InGaN Templates With In Content Greater Than 10%: A Potential Solution for Red LED Problems

InGaN-based red emitters for light-emitting diode (LED) applications continue to be a topic of interest for full-spectrum RGB (red-green-blue) emitters. Current RGB LED technology utilizes InGaN for blue and green emitters, and AlInGaP for red emitters. As LED sizes decrease with the current motivation to achieve micro-LED emission as required by increased-resolution display technologies, traditional transfer printing becomes a challenge due to equipment precision, yield loss, and high production times.^1,2 In addition, AlInGaP has its own drawbacks with miniaturization; as chip size decreases, so does the efficiency of AlInGaP, as well as observed temperature sensitivity.³ Thus, the development of higher-efficiency red emitters based on InGaN is important for achieving a monolithic LED system.

As the target wavelength of InGaN-based emitters is pushed into red emission with increased indium content, the quantum efficiency decreases, due to the quality of the heteroepitaxy growth, as well as a polarization-induced quantum-confined Stark effect (QCSE).⁴ Large lattice mismatch between GaN and the In-rich active regions of an LED result in device performance degradation and reduced efficiency.⁵ Compressive strain also causes the growing InGaN crystal lattice to reject In atoms during the deposition process, resulting in a decrease in indium incorporation as well as indium segregation at the surface of the growing film, referred to as the compositional pulling effect.⁶ In addition, the lower temperature required to achieve high-In-content InGaN grown on GaN leads to reduced material quality, and phase separation when the temperature reaches the miscibility gap threshold. Alternative templates that reduce strain present in the active region and support higher-growth-temperature InGaN growth are of interest to support high-quality, high-In-content InGaN-based devices.

Researchers have attempted to reduce the lattice mismatch between InGaN active regions and the growth template via several unique approaches: short-period superlattices,⁷ graded InGaN buffers,⁸ decomposition layers,⁹ InGaN pseudo-substrates,^10,11 tensile strained GaN buffers,¹² and porous GaN.^13‐15 The addition of SiN treatment on sapphire prior to the GaN buffer growth has also been shown to enhance the luminescence of LEDs.¹⁶

Foreign substrates such as ScAlMgO₄ and ZnO have shown promise in InGaN optoelectronics.^17‐21 The use of ScAlMgO₄ substrates has shown that In_xGa_1−xN/In_yGa_1−yN multiple quantum wells (MQWs) can achieve red emission with In content of x ~ 0.26–0.42 in the QW.¹⁹ However, foreign substrates add complexity to the fabrication process, can be difficult to scale for industry applications, and frequently suffer from impurity diffusion from the underlying substrate, although recent work has shown that AlInN can serve as an effective diffusion blocking layer.¹⁸

Approaches where strain relaxation is addressed have shown promise in supporting higher wavelength LED emission.^22‐24 Dussiagne et al. demonstrated that utilizing InGaN-on-sapphire (InGaNOS) templates with 4–8% In content achieved peak LED emission of ~625 nm, with external quantum efficiency (EQE) of 0.14%, but it was still limited to In content of less than 10%.²³ Pasayat et al. utilized a porous GaN underlayer, where a high degree of InGaN relaxation could be achieved via reduced stiffness of the underlying GaN.¹³ Wong et al. demonstrated red-emitting micro-LED growth on an 85% strain-relaxed template consisting of a high-In-content InGaN decomposition layer.⁹ However, none of these approaches were able to achieve InGaN templates with close to 10% In content, and they require complex processing.

InGaN semi-bulk growth slowly and periodically relaxes the film, taking advantage of the steady relaxation of the crystal lattice to accommodate higher-In-content InGaN and the lattice mismatch with the underlying GaN, as described elsewhere.²⁵ The relaxation process in these templates is accompanied by the formation of V-pits. GaN interlayers (ILs) are expected to back-fill and reduce the size of these formed V-pits, as was reported by earlier transmission electron microscopy (TEM) studies.²⁶

As discussed previously by our group, relaxation can be difficult to quantify via traditional x-ray diffraction (XRD) methods such as reciprocal space mapping (RSM) due to the periodic nature of the templates; as the film thickens, the satellite peaks smear with the zeroth-order peaks seen in the RSM, making quantification of relaxation challenging.²⁵ Using high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM), we previously demonstrated that the a-lattice parameter increases as the growth progresses, demonstrating highly relaxed templates; a 20-period template displaying gradual strain relaxation of 30% to ~85% at the topmost layer, accompanied by an increase in In content, was achieved.²⁶ We also demonstrated a redshift of Δλ = 100 nm in emission wavelength at injection current < 10 mA when an MQW grown on GaN was grown on InGaN templates.²⁷

All samples in each study were grown in a custom-built metal–organic chemical vapor deposition (MOCVD) reactor. The reactor comprises a double-walled quartz tube containing a bi-channel inlet tube housed in the upper region that supplies the precursors to a showerhead at the lower region, just above the rotating TaC-coated graphite susceptor. The heat is supplied via inductive radio-frequency (RF) heating and monitored via a thermocouple housed in the base of the susceptor. As described previously, N₂ and H₂ were used as carrier gases, with trimethylgallium (TMGa), trimethylindium (TMIn), and ammonia (NH₃) as the group III and group V precursors, respectively.²⁷ All samples were grown on commercially supplied GaN on c-plane sapphire with dislocation density of ~10⁸ cm⁻². Approximately 1 μm of GaN was grown at 1045°C, followed by the InGaN semi-bulk template. A period of semi-bulk (inset of Fig. 1a) consists of an InGaN layer (~20 nm), followed by a GaN IL (~1.5–4.5 nm), in which the thickness gradually increases, as detailed elsewhere.²⁷ Higher In content was targeted using three different temperature profiles. The first (SB1, Fig. 1a) involved growth with a susceptor temperature targeting ~20°C lower (at 730°C) than our optimized semi-bulk growth for all 30 periods, as reported previously.²⁷ In SB2 (Fig. 1b), a sudden temperature drop was employed, where the first 24 periods were grown at 750°C, and then the next 16 periods were grown 17°C colder. In SB3 and SB4 (Fig. 1c and d), the first 30 periods of the template are grown 750°C. In SB3, the growth temperature in region 2 is decreased by 11°C at a rate of 1°C per period. In SB4, the growth temperature in region 2 is also decreased at a rate of 1°C per period, for a total temperature reduction of 17°C. Region 3 of both SB3 and SB4 contains four additional periods, where the growth temperature matches the final temperature of the graded region.

Full size image

Photoluminescence (PL) spectra were acquired with a 325 nm He-Cd laser. High-resolution XRD (HRXRD) of the (00.4) reflection was taken on a Philips X’Pert diffractometer for all samples, as shown in Fig. 2. For SB4, indium content was characterized by time-of-flight secondary ion mass spectrometry (SIMS), as described elsewhere.²⁷ A sputter rate of ~2 nm/s was used during data acquisition, corresponding to approximately 10 data points per period. PL (Fig. 3a) or SIMS (Fig. 3b) was then coupled with (00.4) HRXRD 2θ scans to deduce the strain relaxation state, when applicable.²⁸ The surface of each semi-bulk template was compared using atomic force microscopy (AFM).

Full size image

Full size image

The HRXRD data for the (00.4) reflection for SB1–SB4 can be seen in Fig. 2a. Satellite peaks are indicated by numerals, and the zeroth-order InGaN peak by the ★ symbol. Compared to SB1, SB2–SB4 display a peak closer to the GaN peak, consistent with the initial higher InGaN growth temperature. In SB2, an additional peak is seen, as indicated by the ☆ symbol. It was previously reported that the addition of an abrupt temperature step of 20°C in a semi-bulk template resulted in an additional InGaN peak, which was verified by SIMS In content correlation.²⁸ The first region grown at 750°C corresponds to the peak marked by the ★ symbol. In SB3–SB4, the zeroth-order peak is not as well defined, which can be attributed to the compositional grading occurring during the growth.

The PL for SB1 and SB2 and for an analogous structure for SB3 with a different doping profile can be seen in Fig. 3a, and the SIMS for SB4 can be seen in Fig. 3b. SB1 displayed photoluminescence emission at 439 nm with FWHM of 25.2 nm (Fig 3a). SB2 displayed PL emission of 425 nm with FWHM of 16.4 nm, indicating that SB1 achieved a higher In content or relaxation state than that of SB2. The structure similar to SB3 exhibited emission of 442 nm, with FWHM of 22.6 nm. Since SB4 displays low RMS roughness, it was a good candidate for SIMS measurements. The strong and clear oscillations seen in Fig. 3b for SB4 indicate smooth interfaces in the topmost layers of the template. The indium content for the topmost periods is determined via the local maximum of each oscillation. This compositional grading can be clearly seen in the SIMS data for SB4. SIMS shows ~6.1% In content at the bottommost layer and 13.4% at the topmost layer, a 120% increase in In content, assisted by a gradual temperature grading. The first 30 periods display a steady increase in indium content, but as the temperature grading is employed, the slope of the In content increase rate increases.²⁹ Using SIMS for SB4 to estimate In content in SB3 is justified since the two have very similar structures (as shown in Fig. 1). Thus, we estimate the In content of the topmost period for SB3 to be near 12%, based on the SIMS for SB4 (Fig. 3b). The high In content is a promising result, as other techniques such as InGaN pseudo-substrates have been limited to In content of 4–8%.¹⁰

To normalize the discussion of indium content by accounting for relaxation, we also present an effective In content, x_eff, which is the indium equivalency for a fully relaxed film. For example, an 85% relaxed film with In content of 10% in the topmost layer will have an equivalent lattice constant of a fully relaxed film with 8.5% In content, resulting in x_eff of 8.5%.

The relaxation can be estimated by assuming a simplified model for XRD; the measured InGaN peak can be thought of as a convolution of different strain states and In content throughout the film. Thus, using the SIMS value from the topmost layer in SB4, the strain relaxation state of the topmost layer can be estimated. Alternatively, we can estimate the relaxation state based on the PL wavelength emission, since the PL emission is from the topmost layers of the template. However, this method should be treated as an approximation of the relaxation, as the In content and relaxation state cannot be decoupled without quantifying the In content. A bowing parameter of ~2.5 eV is used for estimating relaxation, to reflect the literature range of 1.3–2.8 eV.^30,31

For SB1, the range of possible In content calculated from XRD is from 8.6% fully strained to 14.2% fully relaxed. Calculating the relaxation based on the 2θ peak and PL emission, we expect SB1 to be fully relaxed. As compared to a previously reported template that emits at 425 nm with ~11% In content and relaxation of ~85%, the template emits at ~15 nm higher at 439 nm; thus we estimate that the lower growth temperature led to nearly full relaxation.³² At 439 nm emission (b = 2.5 eV), ~12% In content corresponds to ~95% relaxation. In addition, as further discussed below, the large pit coalescence observed in this sample also supports nearly full relaxation.

For SB2, the first few periods have lower In content than expected, at a range of ~5.5% In content for a fully strained structure and ~10% for fully relaxed. For the second InGaN peak, marked with the ☆ symbol, we see a range of 8.4% (fully strained) to 13.8% for fully relaxed. At PL emission of 425 nm with FWHM of 16.4 nm, the PL emission is lower than SB1; thus we expect lower In content or relaxation state, or both. At 425 nm, we estimate relaxation of the topmost layer of ~85%, correlating to In content of 11%, resulting in an x_eff value of 9.4%.

A zoomed-in view of the XRD peaks for SB3 and SB4 can be seen in Fig. 2b and c, respectively. Because of the temperature grading, the zeroth InGaN peak is not a clearly defined peak, but can be fit using multiple peaks. The peaks were fit using OriginPro fitting software, with a PsdVoigt1 model, as described elsewhere.⁶ For the (00.4) reflection, the zeroth-order peak for SB4 was fit with three peaks, at 72.10(±0.03)°, 71.92(±0.05)°, and 71.7(±0.6)°, from right to left (Fig. 2b). The resulting fit has an R-squared value of 0.99327. Looking at the rightmost peak, for fully strained InGaN, the indium composition corresponds to 6.1%, consistent with the measured In content in the bottommost layer as measured by SIMS for SB4. Using SIMS at the topmost layer of SB4 to inform the calculation for the leftmost peak, the c-parameter corresponds to an InGaN film with 13.4% In content at relaxation of 92%, resulting in an x_eff of 12.3%.

Applying the same analysis to SB3 (Fig. 2c), the (00.4) zeroth-order peak was fit with two peaks, at 72.03(±0.02)° and 71.78(±0.03)°, with an R-squared value of 0.98537. The right peak, corresponding to the strained InGaN state, results in In content of 6.1%, which agrees with the SIMS data measured for SB4. However, SB3 is five periods thinner in the graded region than SB4, so we would expect less relaxation and less In content in the topmost period using this indium value, where the left peak corresponds to a relaxation state of ~90%, resulting in an x_eff of 10.8%. Considering the PL emission at 442 nm, at a bowing parameter of 2.5 eV, an 80% relaxation rate corresponding to an indium content range of 13% is in good agreement with the x-ray analysis. However, note that the PL emission of a structure similar to SB3 has comparable emission to SB2, so the true In content and relaxation state of the topmost layer may be higher, and using the SIMS for SB4 results in a more conservative analysis.

To support the XRD analysis, multiple superlattice models were constructed to show the potential indium content and relaxation of SB4 (Fig. 4). Three superlattices containing 210 nm of In_xGa_1−xN and 3.5 nm GaN are shown with varying indium and relaxation states. The GaN layer was set to be fully strained to the InGaN layer. The superlattice containing 7.2% In content in the InGaN layer was set to be fully strained for a total of 30 periods, where the zeroth peak is seen around −1300 arcsec. Next, a superlattice with four total periods was set to In content of 14.3% and a relaxation state of 100%. The zeroth-order peak is observed at around −1700 rel. arcsec, supporting that the graded semi-bulk may show multiple zeroth-order peaks within proximity of each other. An intermediate superlattice was constructed at In content of 10.7% and a relaxation state of 50%, with 17 periods. This superlattice appears to align quite well with the strained superlattice. Thus, we can conclude that XRD data may display blended zeroth-order peaks when the In content range within one semi-bulk template due to grading is larger than an ungraded semi-bulk template. Although these simulated superlattices do not fully reflect SB4, they do provide a reasonable visualization of the changing In content and relaxation state.

Full size image

The AFM of 5 × 5 µm² areas of SB1–SB4 can be seen in Fig. 5a–d. Isolated V-pits are usually observed in the InGaN templates; however, due to the increased In content and thickness of the templates, the pits coalesce into elongated valleys. Because of the depth and coalescence of the V-pits in the samples, the pit density was quantified by looking at 1 × 1µm² areas and determining the maximum pit depth, where pit depth is defined as the minimum height measured, subtracted by the median z height of the area surveyed. The resulting histograms can be seen in Fig. 4e–h, where the mean of the normal distribution is represented by µ and the standard deviation by σ. Total area of 50–75 µm² was measured for each sample.

Full size image

Reducing the growth temperature by ~20°C from the optimized growth temperature drastically roughened the surface in SB1 to 9.2 nm, and a large coalescence of pits was observed. The average pit depth measured was 22.1 nm (σ = 7.97 nm). The deep V-pit formation and coalescence present in SB1 indicates that the material quality degrades early on in the growth; as seen in TEM previous studies, V-pit formation in semi-bulk occurs when the stored strain energy in the template reaches a threshold in which relaxation is more energetically favorable to the system; in other words, when the strain energy in the beginning of the semi-bulk is increased, the threshold at which V-pit formation occurs can be reached sooner. In the case of SB1, the growth temperature is lower, resulting in higher In content in the beginning of the SB growth; this results in higher strain energy between the template and underlying GaN. Thus, V-pit formation occurs earlier in the growth in SB1. These V-pits then worsen as the growth progresses, as seen by the depth, width, and coalescence of the pits shown in Fig. 5a. These deep, coalesced V-pits can be challenging in device applications if present within the depletion region of the junction, since they can act as shortage pathways for active regions grown on top and reduce the usable epitaxial growth for device processing, and the indium variation be challenging for device uniformity.^33,34 However, recent studies have shown that controlled growth of nanometer-sized V-pits can enhance hole injection and screen dislocations, where V-pits act as a 3D p–n junction.^34‐36

As seen in SB2, the first 24 periods grown at 750°C, followed by a sudden temperature decrease across one period (ΔT = 17°C), does improve the surface morphology as compared to growing the whole structure at a lower temperature, as shown for SB1. RMS roughness of SB2 is improved to 5.1 nm, and the average pit depth decreases to 15.2 nm (σ = 9.62 nm). However, a large coalescence of pits is still present, with depths of ~40 nm, but at a lower frequency than that of SB1. Comparing the histograms of SB1 (Fig. 5e) and SB2 (Fig. 5f), the mean pit depth is decreased by ~30%.

Utilizing a step gradient of 1°C/period in SB3–SB4 improves the surface morphology. The RMS roughness is measured as 2.2 nm and 3.4 nm across a 5 × 5µm² area for SB3 and SB4, respectively. The average pit depth is also drastically improved as compared to the abrupt temperature change of SB2: SB3 and SB4 display an average pit depth of 5.54 nm (σ = 2.07 nm) and 9.02 nm (σ = 9.31 nm), respectively. For SB4, this is an average pit depth reduction of ~59% compared to a traditional approach of growth at a lower temperature. The pit depth is lower in SB3 than in SB4, as the indium content and relaxation states are lower. Further optimization of grading at lower temperatures may be needed to reduce pit formation as temperature decreases. A summary of relaxation state, indium content, x_eff, RMS roughness, and pit depth can be seen in Table I. An RMS roughness of 3.4 nm with In content of 13.4% shows promise for future device fabrication, as compared to the RMS roughness of 8.2 nm for InGaN films grown on ScAlMgO₄.²¹

We previously reported that with an InGaN template grown with a constant susceptor temperature containing x_eff ~10.8%, a redshift of 0.33 eV was observed.³² In this study, a linear relationship of redshift versus x_eff was found. Thus, with an x_eff ~12.3%, LEDs fabricated using a graded temperature approach may demonstrate a higher redshift; however, further work is needed.

Efforts towards increasing the indium content above 10% in InGaN templates while maintaining device-quality surface morphology have been presented. Several approaches were presented for achieving higher-In-content templates: fixed low-growth temperature, sudden temperature reduction in the middle of the growth, and a gradual temperature reduction as the growth progresses. A semi-bulk template that utilizes a temperature decrease in the middle of the growth is shown to improve surface morphology to 4.6 nm and average pit depth of 15.2 nm when compared to conventional lower-temperature semi-bulk growth with RMS roughness of 9.2 nm and average pit depth of 22.1 nm. An InGaN semi-bulk template with 13.4% In content and relaxation of ~92% is achieved with RMS roughness of 3.4 nm and average pit depth of ~9 nm via a temperature gradient approach.