Wafer-scale emission uniformity of InGaN-based red light-emitting diodes on an in situ InGaN decomposition template

There has been significant interest in red, green, and blue (RGB) full-color micro-light-emitting diode (μLED) display technology, which has advantages, such as lower power consumption, longer lifetime, higher contrast, and higher resolution than liquid crystal displays (LCDs) and organic light-emitting diodes (OLEDs).^1–4 Blue and green LED devices are commonly realized from indium gallium nitride (InGaN), whereas red LED devices are realized from aluminum gallium indium phosphide (AlGaInP).⁵ Although the standard InGaN-based blue/green and AlGaInP-based red LEDs in large chip sizes have high efficiency, RGB μLEDs with a dimension less than 100 μm commonly suffer from a reduced external quantum efficiency (EQE), especially for AlGaInP-based red LEDs.^6,7 The significant decrease in EQE of AlGaInP-based red LEDs is mainly due to the high surface recombination and long carrier diffusion length.^8,9 Furthermore, AlGaInP-based red LEDs are not compatible with InGaN-based blue/green LEDs due to mismatches in light angular distribution and large inconsistencies in the device operating voltage range.^10,11 In contrast, InGaN-based red LEDs have a smaller size effect and are compatible with InGaN-based blue/green LEDs. Thus, great attention has been paid to high-efficiency InGaN-based red LEDs, which potentially replace AlGaInP-based red μLEDs to achieve full-color displays.

To achieve efficient InGaN-based red LEDs, the In content of InGaN quantum wells (QWs) should exceed 0.3, while the InGaN-based red LEDs must maintain good crystal quality in the active region.¹² However, when the In content of the QWs increases, the lattice mismatch between InGaN QWs and the underlying GaN gradually increases,^13,14 which causes the composition pulling effect and forms many defects in the active region. These issues hinder the realization of efficient red LEDs.^15,16 Therefore, SRTs are commonly utilized below the active region to reduce the lattice mismatch of the InGaN QWs. SRT can effectively increase the indium incorporation of QWs effectively and reduce the quantum confined Stark effect (QCSE) in the active region.¹⁷ 

Several methods to implement SRTs have been reported. Pasayat et al. used electrochemical etching to prepare porous GaN pseudo-substrates and modulated the relaxation of the InGaN template on the GaN pseudo-substrates by various porosities.^18–21 Soitec developed an InGaN pseudo-substrate (InGaNOS) based on Smart Cut^TM and the epitaxial layer transfer technology to control the lattice constant of the InGaN seed layer on InGaNOS substrates.^22–25 Although these two SRTs could provide good relaxation and enable the red emission of InGaN-based LEDs, they required complicated fabrication processes.

Recently, Chan et al. reported a simple and efficient SRT that consisted of a 2–3 nm low-temperature InGaN decomposition layer (DL) and a high-temperature decomposition stop layer (DSL) above the DL. The high-temperature growth of the DSL decomposed the InGaN DL and created voids, which acted as a strain relaxation layer.^26–29 The introduction of the DL-DSL SRT achieved a large redshift in the peak wavelength of LED devices, but the template exhibited significant negative impacts on the crystal quality and surface morphology, which eventually dramatically degraded the LED performance.^26,27

In this work, we have redesigned the DL-DSL SRT using three different temperatures of GaN protecting layers grown as DSL to recover the crystal quality and surface morphology of the epitaxial layer from its dramatic degradation caused by the DL during high-temperature thermal decomposition. We evaluated the surface morphology and emission wavelength uniformity across the entire 4-in. wafers with the optimized DL-DSL SRT by atomic force microscopy (AFM) and photoluminescent (PL) mapping measurements. Then, the effect of a DL-DSL SRT on the electroluminescence (EL) performance metrics, including the peak wavelength, full widths at half-maximum (FWHM), EQE, and wall-plug efficiency (WPE), was investigated by fabricating packaged LED die chips. Finally, we inserted a DL-DSL SRT into the LED epitaxial structure to produce an InGaN red LED that could achieve efficient pure red emission at higher current densities.

All samples were grown on 4-in. c-plane patterned sapphire substrates (PSS) using a commercial vertical metalorganic vapor phase epitaxy (MOVPE) system. Trimethylgallium (TMGa)/triethylgallium (TEGa), trimethylaluminum (TMAl), trimethylindium (TMIn), and ammonia (NH₃) were used as the sources of gallium, aluminum, indium, and nitrogen, respectively. Silane (SiH₄) and dicyclopentadienylmagnesium (CP₂Mg) were used as n-type and p-type dopant sources, respectively. Nitrogen, hydrogen, and their mixture were used as carrier gases during the growth process.

We first grew samples A [Fig. 1(a)] and B [Fig. 1(b)] to verify the effect of the DL-DSL SRT. The unintentionally doped GaN layer (uid-GaN) in both samples A and B was 4-μm thick. The InGaN multiple quantum wells (MQWs) consisted of three pairs of 3-nm In_0.2Ga_0.8N/9-nm GaN (In_0.2Ga_0.8N QWs grown at 765 °C). The mentioned layers of both sample A and sample B were grown under identical conditions for comparison.

The DL-DSL SRT for sample A consists of a 2.2-nm thick InGaN DL and GaN DSL (grown at three different temperatures). The In_0.3Ga_0.7N DL was grown at a V/III ratio of 1088, a temperature of 720 °C, and a growth pressure of 200 Torr. The first part of GaN DSL was of 3-nm thick and grown at the same temperature as the InGaN DL at a V/III ratio of 2733. This cap layer could prevent the InGaN DL from prematurely decomposing during the subsequent growth of a GaN medium-temperature layer (MTL) and a high-temperature layer (HTL). The GaN MTL (grown at 865 °C and a V/III ratio of 1141) was 125 nm thick, and it was carefully optimized to repair the poor crystal quality and surface morphology caused by the low-temperature InGaN DL and GaN cap layer. The following 25-nm GaN HTL was grown at a V/III ratio of 506 and a higher temperature (1000 °C), much higher than the growth temperature of the InGaN DL. As a result, the InGaN DL would decompose and create voids inside at this high temperature, which contributed to the strain relaxation of the GaN MTL and HTL.

Then, InGaN red LED epi-structures with/without a DL-DSL SRT were prepared and marked as samples C and D, as shown in Figs. 1(c) and 1(d), respectively. The DL-DSL SRT of sample C was grown under identical conditions to sample A. We used InGaN hybrid QWs as the active region, which consisted of a single low-In-content QW (3-nm In_0.2Ga_0.8N/9-nm GaN) and a double high-In-content QWs (3-nm In_0.3Ga_0.7N/6-nm Al_0.1Ga_0.9N).³⁰ All growth conditions of other epi-layers in samples C and D were identical for comparison, in which the In_0.2Ga_0.8N and In_0.3Ga_0.7N QWs were grown at 765 and 720 °C, respectively.

We used scanning transmission electron microscopy (STEM) and energy dispersive x-ray spectroscopy (EDS) to investigate the InGaN DL structure of sample A, as shown in Fig. 2. Voids in the InGaN DL are visible in Fig. 2(a), which indicates that the inserting InGaN layer thermally decomposed as expected.³¹ The decomposition process can be explained by the previous work²⁷ and form an In-rich region in this InGaN DL, as shown in Figs. 2(b) and 2(c). Although the indium phase separation occurs in the InGaN DL, the InGaN QWs in Figs. 2(b) and 2(c) exhibited a uniform indium distribution. This guarantees the uniform emission wavelength from InGaN QWs, which is quite important for mass production.

To further investigate the effect of the DL-DSL SRT on the strain in the upper epitaxial layers, we tested the Raman spectra of samples A and B, as shown in Figs. 3(a) and 3(b), respectively. The strain of GaN was obtained by analyzing the E₂ (high) phonon peak. The blue dashed line in the figures is at a wave number of 566.2 cm⁻¹ and is considered the E₂ (high) phonon peak for strain-free GaN.³² The E₂ (high) phonon peak is observed at 571.1 cm⁻¹ for sample A and 572.8 cm⁻¹ for sample B, both of which are located at higher wave numbers than the strain-free GaN phonon peak. Thus, the GaN epi-layers in both samples A and B are subject to compressive strain.

Then, the surface morphologies of samples A and B were analyzed using an AFM. Figures 4(a) and 4(b) show the AFM images of samples A and B with a scanned area of 5 × 5 μm², respectively. The root mean square (RMS) surface roughness values of sample A and sample B were 0.794 and 0.455 nm, respectively. Both samples A and B exhibited smooth surfaces with a roughness value below 1 nm. We attribute the good surface morphology of sample A to the introduction of the GaN cap layer and GaN MTL. The GaN cap layer avoided surface deterioration caused by the unexpected decomposition of the DL layer during the growth of the GaN MTL. Moreover, the GaN MTL could effectively reduce the defects caused by the low-temperature DL and GaN cap layers and greatly improved the surface morphology. Notably, the dark spots in the AFM images in Figs. 4(a) and 4(b) are V-pits that formed during the epitaxial growth of QWs. The densities of V-pits in samples A and B were 7.44 × 10⁶ and 1.78 × 10⁶ cm⁻², respectively. The higher V-pits density of sample A suggests that the DL-DSL SRT reduces the crystal quality and increases the density of dislocations. However, this change was maintained at an acceptable level by the DSL.

Figure 4(c) shows the PL spectra at room temperature (RT) using a 325-nm He–Cd laser. The measured position was selected to be the similar region at the 4-in. wafer for both samples A and B. The peak wavelength of samples A and B was 484.0 and 451.9 nm, respectively. A 32.1-nm peak wavelength redshift was obtained for sample A in the presence of DL-DSL SRT. This large wavelength shift demonstrates the higher indium incorporation in the InGaN QWs. We believed that the higher indium incorporation originated from the strain relaxation of the underlying DL-DSL SRT, which has previously been explained.

AFM and Raman spectroscopy of sample A with a DL-DSL SRT and sample B without a DL-DSL SRT confirmed that DL-DSL SRT hardly affects the epitaxial layer crystal quality, and surface morphology and can effectively release stress. PL spectra further verify that the DL-DSL SRT enhances the indium incorporation. We introduced it into our LED epi-structures (named sample C) to achieve long-wavelength emission exceeding 600 nm. Figures 5(a) and 5(b) show the PL mapping results of the emission wavelength over the 4-in. wafer for samples C and D. Both samples C and D exhibited uniform emission wavelength over the entire 4-in. wafer excluding the edge regions (around several millimeters wide). The emission wavelengths of sample C are mostly located in the orange-red region, which is 15–20 nm longer than the emission wavelength of sample D in the yellow region.

The standard deviation (STD) of the emission wavelengths for samples C and D were calculated as 3.4 and 1.9 nm, respectively. The higher STD of sample C reflects the slight degradation in indium distribution uniformity in QWs, which was presumably caused by inserting the DL-DSL SRT. As discussed above, the DL-DSL SRT degraded the surface morphology and introduced more dislocations as shown in Figs. 4(a) and 4(b). The degraded surface morphology and more dislocations would cause the indium fluctuation at these regions, reduce the uniformity of the emission wavelength, and eventually increase the STD value. Notably, the STD value (3.4 nm) of the emission wavelength over the 4-in. wafer for samples C remained low for long-wavelength LED wafers, which reveals that our DL-DSL SRT method is greatly promising for future commercial production.

We fabricated LED devices on 4-in. wafers and diced them into single chips for samples C and D. The EL properties of the LED chips were tested in an integrating sphere. Figure 6(a) shows the EL spectra of the LED chips for sample C at 0.8–10 A/cm². Figure 6(a) shows the single-peak emission with a clear blueshift at high current densities. The inset of Fig. 6(a) shows the red emission from the packaged LED die chip (sample C) at 1 A/cm². Figure 6(b) shows the characteristic curves of peak wavelength vs current density for sample C and sample D. The peak wavelength blueshift values for sample C and sample D were 42.6 and 43.8 nm, respectively, when the current density was 0.8–100 A/cm². The large blueshift was due to the energy band-filling effect and QCSE.³⁶ The peak wavelengths of sample C and sample D were 612 and 594 nm at 1 A/cm², respectively. Sample C also exhibited an average 15.1-nm redshift of the peak wavelength at different current densities compared to sample D, which is consistent with the PL mapping results in Figs. 5(a) and 5(b).

Figure 6(c) shows the FWHM of samples C and D as a function of the current density. At current densities below 10 A/cm², the FWHM slightly decreased. Then, as the current density increases, the FWHM increases due to the increased heat generation of the LED. Sample C had a wider FWHM than sample D, because the degraded surface morphology and more dislocations for sample C (as discussed in Fig. 2) caused a higher indium fluctuation. Notably, the FWHM value of 55 nm at 100 A/cm² for sample C with the DL-DSL SRT is narrower than the FWHMs (>60 nm) of the InGaN red LEDs using the DL SRT of other groups.^26–29 

Figure 7(a) shows the current density–voltage (J–V) curves of samples C and D. The forward voltages of sample C and sample D at 20 A/cm² were 3.15 and 3.42 V, respectively. Sample C had a lower forward voltage due to its narrow bandgap of QWs and more V-pits on the surface to improve the hole injection. Figures 7(b) and 7(c) show the EQE and WPE for samples C and D, respectively. At 1 A/cm², sample C had an EQE of 8.2% and a WPE of 7.2%, whereas sample D had an EQE of 9.6% and a WPE of 8%. The lower EQE and WPE of sample C are reasonable because there was more indium incorporation in QWs but the crystal quality degraded after the DL-DSL SRT had been inserted for sample C. Nevertheless, sample C with the DL-DSL SRT had a significantly improved efficiency compared to other works.

Finally, we reduced the growth temperature of the high-In-content QWs from sample C to produce sample E and achieved pure red emission. (The high-In-content QW growth temperature of sample E was 713 °C.) Figure 8 shows the EL spectrum of sample E, which exhibited a peak wavelength of 634 nm with an FWHM of 59 nm at 10 A/cm². The EQE was obtained as 1.3% at 10 A/cm², which demonstrates significantly higher device performance compared with the previous work.^26–29 The inset of Fig. 8 illustrates the pure red emission from the packaged die chip (sample E) at 10 A/cm².

In summary, we have demonstrated a DL-DSL SRT that can be used to achieve high-efficiency red InGaN-based LEDs. The epi-wafer containing this structure has a uniform emission wavelength across the entire 4-in. wafer with an STD of ∼3.4 nm, excluding a few millimeters of edge area. Compared to the control sample, the peak wavelength could be extended by 15–20 nm on average when inserting the DL-DSL SRT. Based on this DL-DSL SRT, we achieved InGaN-based red LEDs with a peak wavelength of 634 nm and an EQE of 1.3% at 10 A/cm². These results suggest that the proposed DL-DSL SRT structure is very effective in producing long wavelength and highly efficient InGaN LEDs and holds promise for future mass production.

The authors acknowledge the financial support from Pangna Micro Semiconductor Technology Co. Ltd, the Fundamental Research Funds for the Central Universities, the National Nature Science Foundation of China (Nos. 62204073, 62274083, 62074077, T2221003, 61921005, and 61974062), and the National Nature Science Foundation of Anhui Province (No. 2208085QF210).