Micro-light-emitting diodes (µLEDs) have recently gathered considerable interest for next-generation display applications [1,2]. The primary colors (red/green/blue) are required to be achieved separately to develop full-color displays. InGaN-based blue µLEDs are the best choice for blue color displays, because they have demonstrated remarkable performances, even when the device dimension decreases to 10 µm or less [3,4].

Two approaches could realize green and red colors. A simple method uses blue or violet µLEDs to excite color converters, such as phosphors or quantum dots [5,6]. Yet the color-conversion efficiency of hybrid devices must be improved [7]. Another approach chooses to fabricate red and green µLEDs using different materials, AlInGaP for red µLEDs [8] and InGaN for green µLEDs [9]. This material difference causes a mismatched angular distribution between the red and green µLEDs [10], leading to a noticeable color shift at different observed angles. Besides, AlGaInP red µLEDs typically suffer from great efficiency reduction because of the high surface recombination when the device dimensions shrink [11].

Therefore, expanding interest was spent on developing InGaN-based red µLEDs. The major challenge for achieving InGaN-based red µLEDs is the significant reduction of the external quantum efficiency (EQE) owing to high indium content in the active region. The large lattice mismatch between the InGaN active region and GaN cladding layers results in many defects in InGaN quantum wells. To reduce the lattice mismatch, partly relaxed InGaN pseudo-substrates were utilized in ${50} \times {50}\;{{\unicode{x00B5}{\rm m}}^2}$ sized red InGaN µLEDs, which were demonstrated with an on-wafer EQE of 0.09% at ${40}\;{{\rm A/cm}^2}$ with a peak emission wavelength of 616 nm [12]. A recent ${6}\times {6}\;{{\unicode{x00B5}{\rm m}}^2}$ InGaN red µLED was realized on porous GaN pseudo-substrates, and it achieved a record on-wafer EQE of 0.2% at ${10}\;{{\rm A/cm}^2}$ with a peak emission wavelength of 632 nm [13]. To the best of our knowledge, this on-wafer EQE is the highest reported value for InGaN red µLEDs. Recently, a high wall-plug efficiency (WPE) of 16.8% at 621 nm was obtained at ${0.8}\;{{\rm A/cm}^2}$ for ${1} \times {1}\;{{\rm mm}^2}$ InGaN-based LEDs [14]. This WPE value was the highest for standard InGaN red LEDs. Our group first developed a high-temperature growth technique for high-indium-content InGaN layers [15,16], and then employed strain engineering strategies from ${{\rm Ga}_2}{{\rm O}_3}$ substrates [17] and GaN templates [18] to active region structures [19]. We also improved the electrical and optical properties of InGaN LEDs in device design and fabrication [20,21]. Since InGaN-based green and red µLEDs do not perform as well as blue µLEDs, the investigation of green and red µLEDs is crucial to identify the direction for future improvements.

This Letter examined the performances of InGaN-based red/green µLEDs ranging from ${98}\times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ to ${17}\times \;{17}\;{{\unicode{x00B5}{\rm m}}^2}$. Current-voltage (I-V) curves were used to estimate the forward voltages and reverse leakage currents. We investigated the electroluminescence (EL) characteristics of red/green µLEDs at different current densities, including the EL peak wavelength, full-width at half-maximum (FWHM), on-wafer EQE, and color gamut. Finally, we also evaluated the temperature dependence of EL emission for red/green µLEDs.

InGaN-based red LED epitaxial wafers were grown on $c$-plane patterned sapphire substrates by metalorganic vapor-phase epitaxy [15]. The epitaxial structures of the InGaN-based red LEDs were reported in our previous work [18]. Indium tin oxide (ITO) films were deposited as a transparent conductive layer and subjected to two-step annealing to form ohmic contacts with p-GaN [20]. The µLED mesas ranging from ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ to ${17} \times {17}\;{{\unicode{x00B5}{\rm m}}^2}$ were patterned by standard photolithography and then etched through the ITO layer and InGaN active region by inductively coupled plasma (ICP) to expose the ${n}$-type layer. A ${{\rm SiO}_2}$ isolated layer was then deposited using plasma-enhanced chemical vapor deposition. Finally, we opened the ${{\rm SiO}_2}$ windows on the ITO and ${n}$-type layers by ICP etching and deposited Cr/Pt/Au as n-type and p-type contact pads. All the red µLEDs with different dimensions were fabricated on the same wafer. In the case of InGaN-based green µLEDs, we used commercial epitaxial wafers and performed the same fabrication process.

The electrical pumping of red and green µLEDs was carried out at a probe station using a semiconductor parameter analyzer. The EL characteristics were measured under direct current operation ranging from 295 (room temperature, RT) to 373 K. The temperatures were measured from the sample-holding stage of the probe station. We also calculated the on-wafer EQE by the output power collected from the top side of the µLEDs.

Figure 1(a) shows the absolute current density-voltage ($|J|-V$) curves for InGaN-based red µLEDs with the dimension from ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ to ${17} \times {17}\;{{\unicode{x00B5}{\rm m}}^2}$. The absolute current density was on a logarithmic scale. At the forward voltage, the $|J| - V$ curves were identical for all red µLEDs with different dimensions. The turn-on voltage, which was regarded as the transition point of the two linear parts in the semi-logarithmic scale in Fig. 1(a), was between 2.5 and 3 V. However, at the reverse voltage, the absolute current density increased with decreasing the dimension of red µLEDs. This behavior was opposite from our previous work [21] for red LEDs in different chip sizes larger than ${100} \times {100}\;{{\unicode{x00B5}{\rm m}}^2}$. We attributed this larger reverse leakage current for ${17} \times {17}\;{{\unicode{x00B5}{\rm m}}^2}$ µLEDs to the sidewall damages after ICP etching. Previous work [3] has shown that sidewall damage after ICP etching can be removed by suitable surface treatments.

 

Fig. 1. (a) Absolute current density of red µLEDs with the mesa areas of ${17} \times {17}\;{{\unicode{x00B5}{\rm m}}^2}$, ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$, and ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ at different bias voltages. (b) Averaged forward voltages of five red and green µLEDs at ${10}\;{{\rm A/cm}^2}$.

Download Full Size | PDF

We also compared the forward voltage of red and green µLEDs at ${10}\;{{\rm A/cm}^2}$. The average forward voltage was approximately 3.0 and 2.4 V for red and green µLEDs, respectively. Because we adopted AlN/AlGaN as the barrier layers to compensate for the strain in InGaN red quantum wells [22], a higher bias voltage was required for carriers to pass through the whole active region. Therefore, our red µLEDs exhibited a higher forward voltage compared to green µLEDs. Additionally, the average forward voltage for both the red and green µLEDs was independent of the µLED dimension, demonstrating good current spreading and low series resistance in our µLEDs [23].

We captured the emission patterns of the red and green µLEDs at ${4}\;{{\rm A/cm}^2}$, as shown in Fig. 2(a). Clearly, most device area of the red and green µLEDs from ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ to ${17} \times {17}\;{{\unicode{x00B5}{\rm m}}^2}$ exhibited uniform luminescence, although some bright spots were observed in the red µLEDs. The common red LEDs with a large mesa area in our previous work [21] displayed many dark points in the emission patterns. These dark points were usually corresponding to the defects in the InGaN red active region. However, these dark points were not found in the red µLEDs because of much smaller mesa areas, demonstrating high crystal quality for our red µLEDs. Figure 2(b) shows the typical EL spectra for ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ red and green µLEDs at ${4}\;{{\rm A/cm}^2}$. Two small additional peaks at around 550 and 690 nm were found at the tails of the spectrum of the red µLEDs. The additional peaks were regarded as the emission from localized states at low current densities and would disappear with increasing current density.

 

Fig. 2. (a) EL emission images of red and green µLEDs with the mesa areas of ${17} \times {17}\;{{\unicode{x00B5}{\rm m}}^2}$, ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$, and ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$. (b) Normalized EL spectra of red and green µLEDs with an area of ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$. The current density was ${4}\;{{\rm A/cm}^2}$.

Download Full Size | PDF

The current density dependences of the peak wavelength and FWHM were investigated from 2 to ${50}\;{{\rm A/cm}^2}$ at RT. Figures 3(a) and 3(b) show the average peak wavelength and FWHM for red and green µLEDs with the dimension of ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ and ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$. From Figs. 3(a) and 3(b), we discovered that the peak wavelength and FWHM behaviors at different current densities were independent of the dimensions of the red and green µLEDs. The similar behaviors reflected that the polarization field and the indium fluctuation in the InGaN red or green active region were identical despite the shrinking of the µLED sizes.

 

Fig. 3. (a) Average peak wavelength, (b) average FWHM, and (c) average on-wafer EQE of three red and green µLEDs with the areas of ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ and ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$, respectively, at different current densities. (d) CIE diagram of red and green µLEDs with the area of ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ at different current densities. The stars are the primary red and green colors defined in Rec. 2020.

Download Full Size | PDF

The red µLEDs exhibited a much larger blueshift (around 35 nm) compared to the green µLEDs (around 6 nm) at ${2 \!-\! 50}\;{{\rm A/cm}^2}$. The larger blueshift of the peak wavelength originated from the stronger quantum-confined Stark effect (QCSE) existing in the red InGaN active region. The average FWHM dropped to the minimum values of 50 nm for red µLEDs and 27 nm for green µLEDs at low current densities, respectively. This behavior was related to the saturation of the emission from deep localized states in the InGaN-based red and green active region [24]. After the minimum value, the average FWHM of the red µLEDs slightly increased with the current densities. Our previous work [21] found that the reason for the broad FWHMs was the heat generation in devices under high direct current injection. The heat was usually generated around the defect area because defects contributed as thermal resisters [25]. However, the green µLEDs did not show significant broadened FWHMs at high current densities. We presumed that the thermal effect could be ignored due to the low defect densities in the InGaN green active region.

We also calculated the average on-wafer EQEs for the red and green µLEDs at 2 to ${50}\;{{\rm A/cm}^2}$, as shown in Fig. 3(c). The on-wafer EQE first rose to the maximum values at the peak current density of 6 and ${8}\;{{\rm A/cm}^2}$ for ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ and ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ green µLEDs, respectively. Then it suffered from an efficiency droop and dropped quickly with the increasing current densities. The maximum on-wafer EQE was 3.7(1)% and 3.5(5)% for ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ and ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ green µLEDs, respectively. From these data, we concluded that the peak current density would increase, while the maximum on-wafer EQE would decrease for the green µLEDs when the device dimension shrank. The results were mostly caused by surface recombination at the active region edges, as explained in previous reports [26,27].

The red µLEDs behaved in a similar way at the low current densities in Fig. 3(c). The average on-wafer EQE rose to the maximum values of 0.62% and 0.56% at the peak current density of ${20}\;{{\rm A/cm}^2}$ for ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ and ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ red µLEDs, respectively. A slight reduction of the maximum on-wafer EQE for different device dimensions was also observed in the red µLEDs. However, unlike the green µLEDs, the red µLEDs did not show an obvious efficiency droop ($\lt\! 7 \%$) at the current density up to ${50}\;{{\rm A/cm}^2}$. We assumed that this phenomenon was due to higher dislocation densities in red µLEDs than in the green µLEDs [28].

Furthermore, we measured the absolute output power of ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ green µLEDs in the integrating sphere and determined the calibration factor for an on-wafer light output measurement vs. absolute output measurement in the integrating sphere. This calibration method was also used in other works [12,13]. The green µLEDs in this Letter were not encapsulated with resin when measured in the integrating sphere. From Figs. 3(a) and 3(c), the average on-wafer EQE of ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ red µLEDs was 0.36% at the peak wavelength of 626 nm at ${4}\;{{\rm A/cm}^2}$. Based on the calibration factor, an absolute EQE of approximately 0.87% could be expected from our ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ red µLEDs when measured in the integrating sphere. This estimated EQE value exceeds the reported one ($\gt$0.6%) calculated by the same method [13].

The CIE 1931 positions for the red and green µLEDs at ${2 \!-\! 50}\;{{\rm A/cm}^2}$ are shown in Fig. 3(d). According to the recommendation (Rec.) 2020 standard, the primary red and green colors were located at the coordinates of (0.708, 0.292) and (0.17, 0.797), respectively. They were marked as red and green stars as reference points in Fig. 3(d). Clearly, a small gap existed between the coordinates of the green µLEDs and the defined green color in Rec. 2020. This gap illustrated that the color saturation of the green µLEDs was insufficient, which was originated from the broad FWHM of the InGaN-based green µLEDs. Additionally, the distance between the coordinates of the green µLEDs and the reference point of green color in Rec. 2020 remained almost constant with the current density.

However, the red µLEDs exhibited different movements relative to the green µLEDs. At the low current density, the position was close to the reference point of red color in Rec. 2020. When the current density was increased, the coordinate position shifted farther away from the reference point. As a result, the color gamut area coverage would be reduced when the red µLEDs operated at higher current densities. Since this position shift of the red µLEDs is originated from a large blueshift of the peak wavelength, eliminating or reducing the QCSE in the InGaN red active region is necessary to decrease this shift. Although the FWHM of the red µLEDs was much larger than that of the green µLEDs, the values of 50–52 nm seemed to be acceptable, because all positions were located at the edge of the color space in the CIE 1931 diagram.

Finally, we investigated the temperature dependence of the EL integrated intensity and peak wavelength for the ${47}\;{{\unicode{x00B5}{\rm m}}^2} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ red and green µLEDs, as shown in Figs. 4(a) and 4(b), respectively. From the fitting dot lines in Fig. 4(a), the characteristic temperatures were obtained as 50 and 411 K at ${10}\;{{\rm A/cm}^2}$ for the red and green µLEDs, respectively. The smaller characteristic temperature for the red µLEDs indicated that the InGaN red active region has many Shockley–Read–Hall recombination centers [29], which are related to the defect density and become more effective with temperature.

 

Fig. 4. (a) Normalized intensity, (b) peak wavelength, and (c) CIE diagram of ${47} \times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ red and green µLEDs operated at different temperatures at ${10}\;{{\rm A/cm}^2}$.

Download Full Size | PDF

Figure 4(b) shows that the red and green µLEDs exhibit the redshift of the peak wavelength with increasing temperature. The redshift is caused by the shrinking of the InGaN bandgap. The redshift coefficients of the peak wavelength were 0.126 and ${0.038}\;{{\rm nmK}^{- 1}}$ for the red and green µLEDs, respectively. The redshift coefficient of InGaN red µLEDs was slightly smaller than that of AlInGaP red LEDs [30,31], indicating smaller temperature dependence of InGaN materials. However, InGaN red µLEDs show the larger coefficient compared to the green µLEDs. We attributed this larger coefficient to more SRH recombination centers in the InGaN red active region. The SRH recombination became effective and led to less carrier density for radiative recombination in the active region with temperature. As a result, the red active region suffered from more QCSE and had a larger redshift of the peak wavelength.

To estimate the influence of temperature in the color space, we plotted the temperature dependence of the color coordinates in the CIE 1931 diagram [Fig. 4(c)]. The movements of the red and green µLEDs are marked with arrows in Fig. 4(c). The green µLEDs moved away from the reference point of green color in Rec. 2020 [green star in Fig. 4(c)] with increasing temperature. Therefore, the gamut area coverage would be reduced due to this movement of the green µLEDs. In contrast, the red µLEDs moved towards the reference point of red color in Rec. 2020 [red star in Fig. 4(c)] with temperature. Although the movement of the red µLEDs led to the hue variation, the gamut area coverage could be increased.

In summary, we investigated the performance of the InGaN-based red/green µLEDs ranging from ${98} \times {98}\;{{\unicode{x00B5}{\rm m}}^2}$ to ${17} \times {17}\;{{\unicode{x00B5}{\rm m}}^2}$. A large blueshift of the peak wavelength ($\sim\! 35\,\,{\rm nm}$) and a broad FWHM ($\ge\! 50\,\,{\rm nm}$) at ${2\! -\! 50}\;{{\rm A/cm}^2}$ were observed in the red µLEDs. An on-wafer EQE of 0.36% (corresponding to an expected absolute EQE of 0.87%) at the peak wavelength of 626 nm at ${4}\;{{\rm A/cm}^2}$ was achieved from the ${47}\times {47}\;{{\unicode{x00B5}{\rm m}}^2}$ red µLEDs. Considering the positions in the CIE 1931 diagram, we recommend suppressing the peak wavelength blueshift for the red µLEDs and making the FWHM narrower for the green µLEDs to improve the color quality and stability. The InGaN red µLEDs suffer from a severe thermal droop compared to the green µLEDs, but the coordinates of the red µLEDs in the CIE 1931 diagram move towards the primary red color (Rec. 2020) with temperature. This opposite movement to the green µLEDs helps to increase the gamut area coverage in the color space.