1. Introduction

With steady decrease of the component cost, the future of solid-state lighting is no longer solely focusing on the efficiency [1]. One emerging trend for future solid-state lighting is color- or spectrally tunable lighting to address the needs of well-being (e.g. circadian rhythm), indoor agriculture, and special lighting for retails and entertainment. Color-tunable lighting also increases versatility of solid-state lighting, enabling the integration of lighting with building facades (e.g. windows and interior walls), information displays, and wireless communication network (e.g. LiFi) [2]. To meet these needs, an LED luminaire needs to consist of multiple color channels, individually addressable, as well as color-mixing optics in order to dynamically tune the output color or spectrum. To date, organic LEDs (OLEDs) have been much more advanced in achieving this goal compared to gallium nitride (GaN) based LEDs. Yet, OLEDs are limited in optical intensities, requiring a large device which increases the cost. In this paper, we study a novel GaN nanopillar LED array comprising of individually addressable color channels, focusing on its spectrum tunability for color-tunable lighting, transparency for window lights, and fast dynamic response for integration with data display and communication. We will demonstrate the feasibility of integrating red, green, and blue color channels on the chip while still maintaining good transparency. We will also discuss technology challenges unique to nanostructured LEDs that must be overcome in order to meet the demanding performance for the next-generation solid-state lighting.

The focus of this work is on the monolithic integration of multiple individually addressable color LEDs. The technology is also highly relevant for emerging augmented reality applications. Various approaches have been proposed to assemble multiple micro-LED devices and materials on the same substrate including massive transfer [3,4], stacked quantum wells [5], heterogeneous integration of quantum dots [6], selective-area epitaxy [7,8], wafer-level thin-film transfer combined with epitaxial overgrowth [9], and local strain engineering [10–14].

In this work, our building block is a nanopillar LED consisting of a high-density array of dot-in-wire III-nitride heterostructures. Vertically, the dot-in-wire structure is identical to a conventional thin-film LED while horizontally, the dimension of individual structures is on the order of 100nm to 1µm. The color is tuned via controlling the strain in the InGaN/GaN nanopillars by changing the diameters of individual nanopillars. Previously, tuning from 470–650nm has been shown from optically pumped In_0.32Ga_0.68N/GaN nanopillars. More recently, monolithic integration of individually addressable red-, green-, and blue-emitting nanopillar LED “pixels” has also been demonstrated [14]. Good performance of color-mixing has been shown, even without any external optics [13]. However, a detailed study of the electronic and dynamic properties as well as interconnectivity of multiple large-area color channels have not been shown. This includes interleaving multiple color channels for uniform color mixing output, potential crosstalk between different color channels when individually modulated, analysis of electrical characteristics and power efficiencies, and strategies for improvements. These properties are important to apply the nanopillar LEDs toward spectral-tunable lighting and beyond.

2. Methodology

Figure 1(a) shows the schematic of the device which consists of several individually addressable alphabets. The purpose of the alphabets is to allow us to easily detect any interference between the color channels. The dimensions of each color channel are 12µm×20µm. For simplicity, we opted for a passive matrix driving scheme. A top-down fabrication method was employed. The emission color of individual nanopillar LEDs was tuned by controlling the nanopillar diameter. The red channel is simply a continuous thin film; green and blue consist of 200- and 45-nm diameter nanopillars, respectively. The sample was grown on a c-plane sapphire substrate using metal-organic chemical vapor deposition (MOCVD) by a commercial foundry (NovaGaN). The epitaxial structure consists of Si-doped n-GaN, 5-periods of InGaN/GaN multiple quantum wells (MQWs), Mg-doped AlGaN electron-blocking layer, and Mg-doped p-GaN. For patterning, an 80nm-thick Ni film was used as an etch mask. The red and blue/green channels were patterned using optical and electron-beam lithography, respectively. The nanopillar etching was performed by inductively-coupled-plasma-reactive-ion-etching (ICP-RIE). The sample was then etched in a diluted KOH solution to create a vertical sidewall for the nanopillars. Figures 1(b)–1(d) show the scanning electron microscope (SEM) images of the red, green, and blue channels, respectively.

 

Fig. 1. (a) Monolithic integration of red, green, and blue color channels using nanopillars of different diameters. (b)–(d) SEM images of the red (b), green (c), and blue (d) channels. The green and blue channels consist of nanopillars of 200nm and 45nm diameters, respectively.

Download Full Size | PDF

To electrically isolate adjacent color channels, a 500nm-thick SiO₂ layer was coated over the entire sample followed by planarization and etch-back using 800nm-thick polymethyl methacrylate (PMMA). The PMMA has a similar etching rate as SiO₂ using ICP-RIE. For p-type contact, we used 250nm indium tin oxide (ITO) on top of thermally annealed Ni/Au (10/10nm). Additionally, 3µm-wide Ni/Au (35/200nm) lines were deposited on the ITO layer for electrical interconnects. All color channels shared a common n-contact of Ti/Au (35/200nm).

3. Results and discussions

Figures 2(a) and 2(b) compare optical images of the sample before and after the nanopillar etching. Due to the high indium composition in the MQWs, the sample exhibits a yellowish tint before etching. But after the formation of the nanopillars, the yellowish tint disappeared. The transmittance across the visible spectrum is around 50–55%, as shown in Fig. 2(c). The measurement was performed using a white light source through a sample area where roughly 70% of the area was covered with ITO and Ni/Au layers. The uniform transmittance from 400–750nm wavelength range suggests the transmittance degradations were mainly from metal electrodes instead of light absorption from the MQWs. No antireflection coating or structure was added to the substrate in this work which accounted for about 15% of the optical loss due to reflection. Optical transparency is expected to be further improved with optimizations of the metal electrodes [15,16]. The excellent transparency can be attractive for window lights and see-through displays.

 

Fig. 2. (a), (b) Optical images of the LED sample (a) before and (b) after the top-down fabrication process. (c) Transmittance of the LED sample at the metal contact region measured across the visible spectrum. The inset shows an optical image of the measurement area where the illuminated area by the incident white light is denoted by a 300µm-radius dotted circle.

Download Full Size | PDF

Figure 3(a) shows the optical images of the color channels being individually turned on. All channels showed light emission with fairly uniform brightness. No crosstalk was observed, that is color channels biased at 0V had no observable light emission. This suggests good electrical isolation can be achieved between adjacent color channels even with a passive matrix. Figure 3(b) shows the electroluminescence (EL) measured with the sample uncooled with no heat sinks in the indoor environment. The dominant EL peaks were observed at 602nm for red, 525nm for green, and 493nm for blue channels. Single peaks were observed in all channels, suggesting a good color purity. Mixing of these color channels allows the generation of white light for illumination. For example, 2850K white light can be generated by mixing the blue and red channels at a 28/72 ratio. Meanwhile, the red channel showed a relatively broad spectral linewidth of 80nm presumably due to indium compositional and quantum well width fluctuations in the MQWs. However, the linewidth is decreased to 42nm for the green and 38nm for the blue channels due to the strain relaxation which reduces the piezoelectric field in the quantum wells. The linewidth broadening due to well-width fluctuations is reduced by ${\Delta }{F_{piezo}}{\Delta }{t_{QW}}$ where ${\Delta }{F_{piezo}}$ is the reduction of the piezoelectric field due to strain relaxation and ${\Delta }{t_{QW}}$ is the well-width fluctuations. A small linewidth is desirable for better luminous efficacy but with a trade-off to color rending quality. The Purcell effect may be leveraged to further reduce the linewidth by arranging the nanopillars into a photonic crystal pattern [17,18].

 

Fig. 3. Electroluminescence characteristics of the red, green, and blue color channels. (a) Light emission images. (b) Normalized EL spectra of the red, green, and blue channels measured without any active cooling or heatsink. (c) Modulation response of the red, green, and blue channels as a function of modulation frequency.

Download Full Size | PDF

We also measured the modulation response of the red, green, and blue channels. Each channel is turned on and off (0V) with a 50% duty cycle. The light output is detected by a photomultiplier tube. Figure 3(c) shows the amplitude of the modulation transfer function as a function of modulation frequency. The data is the overall response including electrical wires and probes. The 3dB bandwidths for all device well exceeded 1kHz and therefore are suitable for both lighting and display applications. The red channel has the slowest response, with a 3dB bandwidth of 700kHz which is much slower than a typical blue InGaN LED [19]. This is due to the strong quantum-confined Stark effect (QCSE) in the red channel which results in a long carrier lifetime in the MQW region [20]. The green and blue channels exhibit a much larger modulation bandwidth due to suppressed QCSE.

Figure 4(a) shows the external quantum efficiency (EQE) of the red, green, and blue channels at various current densities. Note that we did not optimize the light extraction efficiency in the current sample. The EQE of the red channel peaked around 1.2% at 10mA/cm² corresponding to a power efficiency of 1%, which was similar to the previously reported values of red InGaN LEDs [21]. As the current density increases, the EQE drastically decreases which is known as the efficiency droop. Meanwhile, the peak EQE for the green and blue channels were observed at a much higher current density around 13‒14A/cm². Nevertheless, the peak EQE of the blue channel was lower than that of the red and green channels. It is likely the reduced droop for the green channel is due to the suppression of QCSE and the lower EQE for the blue channel is due to enhanced non-radiative recombinations which will be discussed below.

 

Fig. 4. (a) External quantum efficiency (EQE) and (b) Current density-voltage (J-V) curves of the red, green, and blue LED channels. The inset in (b) shows the equivalent circuit model used to fit the J-V curves.

Download Full Size | PDF

To understand the lower EQE observed for the blue channels, we fitted the current-voltage (J-V) curves shown in Fig. 4(b) using an equivalent circuit model shown in the inset of Fig. 4(b) where R_S represents the series resistance that includes both the contact resistance and the bulk resistance through the nanopillar. R_P represents the shunt resistance due to surface leakages. When fitting R_S, we used the same resistivity ρ_S of the RGB channels because they were fabricated on the same epitaxial wafer. We obtained R_S to be 230Ω (red), 1.9×10⁴Ω (green), and 1.3×10⁵Ω (blue); and R_P to be 10⁹Ω (red), 4×10⁹Ω (green) and 5×10⁶Ω (blue). The smaller values of R_S compared to R_P suggest dominant carrier transport occurred through the p-MQW-n junction inside the nanopillar. Near the perimeter of the InGaN nanodisk region, there is a potential barrier formed in the radial direction between the center of the nanopillar and the sidewall. This potential barrier, a result of nonuniform strain relaxation along the radial direction [20], helps direct the electrons (holes) toward the center of nanopillar when they are injected into the InGaN region.

Further observing the R_S and R_P values for the three color channels, the difference between R_S and R_P is much smaller for the blue channel. In particular, R_P for the blue channel is three orders of magnitude smaller than that for the red and green channels, suggesting a non-negligible surface leakage for the blue channel. We attribute this enhanced surface leakage for the blue nanopillars to the plasma damage during the ICP-RIE rather than intrinsic to the epitaxial materials since all red, green, and blue LEDs share identical device structure and fabrication processes. It has been reported that the penetration depth of the plasma damage due to the Ar and/or Cl₂ etch gas is approximately 50–100nm [22–24], which is larger than the diameter of the blue-emitting nanopillars. Although the top region of the p-GaN was protected by the Ni etch mask, the plasma damage can still affect the sidewall of the InGaN active region, leading to large leakage current through surfaces and tunneling current through MQWs even if electrons and holes are transported through the inside of the nanopillar. This is further confirmed by the larger ideality factor η associated with the blue channel. A large ideality factor for the InGaN LED is often attributed to defect-assisted tunneling through the MQW region. From the J-V curves, the ideality factor η was estimated to be 6.5, 9, and 28 for the red, green, and blue channels, respectively. Meanwhile, the green-emitting nanopillar has a diameter larger than the plasma damage penetration depth. Therefore, the green channel exhibits better optical and electrical characteristics compared to the blue channel. To reduce the impact of plasma damages, one can increase the initial dimensions of the nanopillars after ICP-RIE and remove the outer 50–100nm of materials using wet etch [22,24]. Surface passivation has also been shown to reduce the surface recombination [25,26].

4. Summary

In summary, a color-tunable LED chip comprising of multiple color channels, each of which connecting a series of nanopillar LED pixels was fabricated and characterized. The chip design utilized local strain engineering to achieve co-planer monolithic integration of red, green, and blue color channels which enables simpler fabrication and electrical interconnect layout, and better optical transparency. The integration of red, green, and blue color channels at the microscale also reduces the complexity of color mixing, even at a large viewing angle. Our results show that each color channel exhibits good color purity and uniform brightness, negligible crosstalk even with a passive matrix, and uniform transparency (>50%) across the visible spectrum. Although our demonstration has focused on red, green, and blue integration but the design can be easily scaled to more colors, simply by adding nanopillar arrays with different diameters. At the present stage, the power efficiency of individual color channels is still low. The MQW materials quality and plasma damages during ICP-RIE etching were identified as the major cause. Extended wet etching step can be a possible solution to the plasma damages. In addition, current efforts for fabricating monolithically integrated full-color pixels for display applications are expected to improve the efficiency of red-emitting InGaN MQWs in the future. With these improvements, the nanopillar LEDs may be suitable for future color-tunable lighting and integration of lighting with windows, data displays and data communication.