Strategy toward the fabrication of ultrahigh-resolution micro-LED displays by bonding-interface-engineered vertical stacking and surface passivation

Dae-Myeong Geum a, Seong Kwang Kim b, Chang-Mo Kang c, Seung-Hyun Moon c, Jihoon Kyhm d, JaeHoon Han e, Dong-Seon Lee c and SangHyeon Kim *ab
aInformation and Electronics Research Institute, Korea Advanced Institute of Science and Technology (KAIST), Daehak-ro 291, Yuseong-gu, Daejeon 34141, Republic of Korea. E-mail:
bSchool of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daehak-ro 291, Yuseong-gu, Daejeon 34141, Republic of Korea
cSchool of Electrical Engineering and Computer Science, Gwangju Institute of Science & Technology, chumdan gwagiro 123, buk-gu, oryongdong, Gwangju 61005, Republic of Korea
dQuantum Functional Semiconductor Research Center, Dongguk University, Joong-gu, Seoul 04620, Republic of Korea
eCenter for Opto-Electronics Materials and Devices, Korea Institute of Science and Technology (KIST), Hwarangro 14-gil 5, Seongbuk-gu, Seoul 02792, Republic of Korea

Received 27th May 2019 , Accepted 5th September 2019

First published on 10th September 2019

In this study, we proposed a strategy to fabricate vertically stacked subpixel (VSS) micro-light-emitting diodes (μ-LEDs) for future ultrahigh-resolution microdisplays. At first, to vertically stack the LED with different colors, we successfully adopted a bonding-interface-engineered monolithic integration method using SiO2/SiNx distributed Bragg reflectors (DBRs). It was found that an intermediate DBR structure could be used as the bonding layer and color filter, which could reflect and transmit desired wavelengths through the bonding interface. Furthermore, the optically pumped μ-LED array with a pitch of 0.4 μm corresponding to the ultrahigh-resolution of 63[thin space (1/6-em)]500 PPI could be successfully fabricated using a typical semiconductor process, including electron-beam lithography. Compared with the pick-and-place strategy (limited by machine alignment accuracy), the proposed strategy leads to the fabrication of significantly improved high-density μ-LEDs. Finally, we systematically investigated the effects of surface traps using time-resolved photoluminescence (TRPL) and two-dimensional simulations. The obtained results clearly demonstrated that performance improvements could be possible by employing optimal passivation techniques by diminishing the pixel size for fabricating low-power and highly efficient microdisplays.


In the recent years, with the increasing demand for next-generation visual systems such as ultrahigh-resolution augmented reality/virtual reality (AR/VR) and holographic displays, inorganic light-emitting diodes (LEDs) have been extensively used as a self-emitting light source in these applications due to their high quantum efficiency and robust durability.1–3 In the past few decades, significant efforts and developments, involving inorganic LEDs, have led to the expansion of solid-state lighting systems and outdoor digital signage.1 However, the requirement of dimensional scaling for the fabrication process of conventional LEDs has been stagnant at around a few hundred micrometers, mainly targeting lighting applications using phosphor-coated blue LEDs. Different than lighting applications, present and future displays need high-resolution full-color LEDs for red (R), green (G), and blue (B) on the same backplane to represent bright and broad color combinations.4,5 Therefore, miniaturized and low-power-consuming micro-LEDs (μ-LEDs) will be a crucial technology for fabricating future high-resolution displays, which require above 2000 PPI.

Among several approaches for fabricating full-color μ-LEDs, many research groups have investigated the epitaxy of InGaN/GaN-based materials on a single substrate with various structures, such as multifaceted nanorods and pyramids, because the emission wavelengths of InGaN can be theoretically tuned from 1790 nm to 360 nm by varying the indium compositions.6–9 However, the emission wavelengths of these structures are normally impacted using an external bias, resulting in a limited range of color tunability.10 Moreover, these types of devices showed a rapid decrease in the quantum efficiency beyond the green spectral range.11 Because of these reasons, the use of epitaxy-based methods would be difficult for fabricating highly efficient full-color displays. Another approach involves the use of local strain engineering of the InGaN/GaN quantum well in nanostructures.12 It showed different colors in the same quantum well on the basis of the nanopillar diameters. Although it could alleviate the epitaxy problem, broadening the wavelength coverage was still challenging.

Therefore, these problems have motivated researchers to investigate assembly-like pixel arrangement using a massive number of individual μ-LEDs, called “pick-and-place” technology. For this technology, InGaN/GaN-based blue and green LEDs and AlGaInP-based red LEDs are separately grown on proper substrates (sapphire or GaAs). Then, each RGB subpixel is separated from the mother substrates (picked) and arranged (placed) within a size of few tens of micrometers on the target substrate; therefore, the phrase “pick-and-place” is used.13,14 Generally, these subpixels are placed in the form of laterally arranged subpixels (LASs). However, several critical issues persist, such as transfer yield and difficulty of mechanical alignment due to the requirement of handling small-sized pixels. Fundamentally, its resolution would be critically limited by the alignment accuracy of the pick-and-place machine, which typically ranges within a few tens of micrometers.14–16 To satisfy the requirements of high-resolution displays (such as AR/VR), further improvement in machine accuracy and transfer yield are imperative.

To circumvent these issues of the pick-and-place method, wafer-bonding-based monolithic integration has become a fairly promising strategy.17–21 This approach offers the vertical integration of different thin films on a single substrate, which can yield vertically stacked subpixel (VSS) arrays of RGB LEDs with very high pixel density as compared to that obtained with LAS arrays. Naturally, it eliminates accuracy problems inherent in conventional semiconductor top-down fabrication strategies, yielding nanometer-scale precise alignment between the pixels using lithographic vertical alignment. The concept of lithographic-alignment-based monolithic 3D (M3D) technology would be very similar to conventional mechanical-alignment-based 3D packaging technology in the complementary metal–oxide–semiconductor (CMOS) society.22 For a brief comparison of each technology, we calculated the theoretical resolutions provided by a single pixel area determined with each subpixel (size: 5 μm), as shown in the inset of Fig. 1. It shows the theoretical resolution in PPI as a function of alignment accuracy, which, in this case, is the subpixel pitch. For the pick-and-place strategy, when the alignment accuracy is reduced to 10 μm, 500 PPI can be obtained, which limits applications only to smartphones, smartwatches, and TVs. To be applicable for ultrahigh-resolution displays (which necessitate more than 2000 PPI), the alignment accuracy has to be reduced to 1 μm. However, this alignment accuracy is fairly challenging to achieve with the pick-and-place approach as mechanical alignment involves multiple transfers. In contrast, the higher resolution of 2000 PPI is already achievable with our approach with alignment accuracy of 10 μm due to the vertically stacked pixel structure; this is suitable for the implementation of consumer VR displays. In single-color μ-LED integration approach, 5000 PPI μ-LED devices were demonstrated with subpixel pitch of 5 μm and color-converting materials.20 Therefore, this calculation based on VSS μ-LEDs seems to be fairly reasonable. Furthermore, lithographic alignment in our wafer-bonding-based approach allows more precise alignment. It can realize much higher resolution (in excess of 4000 PPI) by reducing the alignment accuracy down to approximately 1 μm despite the pixel dimensions of 5 μm, which is relatively easy in conventional semiconductor process technologies. It not only facilitates the formation of ultrahigh-resolution lighting sources but also yields closer integration with Si CMOS driving circuits using conventional semiconductor process technologies.21

image file: c9nr04423j-f1.tif
Fig. 1 Comparison of expected PPIs for pick-and-place and wafer-bonding-based approaches.

Even though there are many advantages in wafer-bonding-based high-resolution displays, only a few studies have been reported.10,18,23 This can be attributed to the interference (color modulation) between LEDs due to the vertically stacked nature. Earlier, we have reported adhesive-bonded multicolor LEDs, where we observed photoluminescent responses in red LEDs due to electroluminescence in green LEDs.18 When high-energy photons are emitted from blue or green light, they can excite red AlGaInP materials (which have a lower energy bandgap) through the transparent bonding medium in the visible region. This results in unintentional color modulation, which can be considered as a problem of VSS LEDs. Therefore, to utilize the full potential of VSS full-color LEDs, color modulation should be eliminated by proper design of the optical device. Moreover, since the surface-to-volume ratio involves scaling at the subpixel size, surface properties are more critical than that in conventionally large LEDs.24,25 Nonradiative recombination centers, such as dangling bonds, at the surface are one of the reasons for the reduction in the quantum efficiency of μ-LEDs.26 Generally, red LEDs as well as blue and green LEDs exhibit efficiency drops as the LED size decreases. Therefore, understanding and passivating the surface effects of μ-LEDs are essential in order to maximize the performance of μ-LEDs.

In this paper, we propose the device structure and fabrication of ultrahigh-density VSS μ-LEDs using distributed Bragg reflectors (DBRs) as the bonding medium for the removal of color modulation and employing the top-down semiconductor process. First, we designed DBR structures consisting of SiO2/SiNx and systematically analyzed the optical properties of DBR-engineered μ-LEDs via both simulation and experiment with different numbers of pairs. The measurement results revealed the desired stopband properties, i.e., reflecting blue light and transmitting red light at the bonding interface. After optimizing the DBR property, vertical stacked LED structures were successfully fabricated via the thin-film transfer of red LEDs on blue substrates with a designed DBR structure as the bonding medium. Transmission electron spectroscopy (TEM) images revealed that there is not only a good-quality bonding interface, but also possess reliability and no unintentional excitation of red LEDs from blue light in the photoluminescence (PL) measurements. Furthermore, we showed the promising potential of this process for fabricating high-resolution full-color LEDs, yielding record-high resolution of 63[thin space (1/6-em)]500 PPI by only using a typical semiconductor process. Finally, using the optical characterization of the time-resolved photoluminescence (TRPL) method, we have systematically investigated the impact of pixel size on the performance by reducing the number of LED pixels, together with 2-dimensional technology computer-aided design (TCAD) simulations. These results strongly indicate that we pioneered feasible methods for manufacturing ultrahigh-resolution full-color displays in future applications such as AR, VR, hologram, etc.

Results and discussion

Simulations and experiments for designing optimal DBR structures

Fig. 2 shows the schematic diagram of monolithically integrated blue and red LEDs by using wafer bonding. In order to avoid the undesired optical excitation of red LEDs by blue LEDs, the bonding medium should be designed by carefully considering the optical transparency and reflectivity in red and blue lights, respectively. Here, to meet this requirement, we selected DBR as the bonding medium. The DBR should be optimized depending on the bonding interfaces, such as blue/green, blue/red, and green/red regions. Among them, we investigated the blue/red LEDs for understanding the feasibility of engineering the bonding interface. Therefore, the purpose of DBR is not only the selective transmission of red light, but also the reflection of blue light, as shown in Fig. 2(a).
image file: c9nr04423j-f2.tif
Fig. 2 (a) Schematic of the VSS LEDs and its color mixing problem in the experiment. (b) Simulated structure with tabulated SiO2/SiNx DBR pair. (c) Reflectance spectra with increasing DBR pairs. (d) Specific reflectance and transmittance values at 450 and 630 nm in terms of number of DBR pairs and specific reflectance depending on the incident angles (inset). (e) Measured spectra of the fabricated DBRs on sapphire substrates as a function of different DBR pairs with light incident from the DBR side and (f) substrate side.

The simulation of the DBR structure was carried out for satisfying this requirement. As shown in Fig. 2(b), we assumed the structure of the silicon dioxide layer (SiO2)/silicon nitride layer (SiNx) on the sapphire substrate for performing optical simulations. Here, the thickness (t) of SiO2 and SiNx should be determined by the following equation:

t = λ/(4·nmaterial)(1)
where λ and nmaterial are denote the target wavelength in the stopband and refractive index of the dielectric material, respectively. When considering the target wavelength of 450 nm and dielectric constant of the material, the thicknesses are determined to be 77.7 and 51.34 nm for SiO2 and SiNx, respectively. Optical simulations were conducted for a spectral reflectance calculation by using the transfer matrix method, where the incident light was perpendicular to the sapphire surface. Fig. 2(c) shows the simulation results by varying the number of DBRs ranging from 3 to 9 pairs. As the number of DBR pairs increases, the stopband obviously becomes narrower. The full width at half maximum (FWHM) values of the stopband were 200, 140, 120, and 110 nm for 3, 5, 7, and 9 pairs, respectively. The FWHM values were relatively narrow and similar for DBR pairs higher than 7. The required number of pairs can be reduced by using a combination of materials with sufficiently different refractive indices. In addition, the calculated reflectance and transmittance values of 450 and 630 nm, respectively, are described as a function of the number of DBR pairs, as shown in Fig. 2(d). The reflectance was rapidly increased from 8.9% to 99% as the number of DBR pairs increased at 450 nm. At the wavelength of 630 nm, no significant difference was observed in the transmittance depending on the number of DBR pairs; however, the highest transmittance value was as low as 97% when using 7 DBR pairs. This indicates that there is 95% reflection of blue emission and only 3% loss of red emission through the DBR structure between the actual blue LED and red LED. However, there is angle-dependent reflectance and transmittance of DBR at viewing angles, which could induce an angular color shift due to the combination of angular dependency, optical cavity effect, and sidewall extraction of DBRs in the μ-LEDs.27–29 In order to investigate the angle-dependent reflectance of the designed DBR structure, we additionally performed the simulation by varying the incident angles ranging from 0° to 75°, as shown in the inset of Fig. 3(d). While the incident light perpendicular to the surface could be reflected and transmitted at 450 and 630 nm, respectively, as designed, the reflectance at 450 nm decreased beyond 45°. As the angle increased, the reflectance at 630 nm increased, reaching 60% at 75°, suggesting the extraction of red emission could be fairly difficult at higher angles. On the other hand, it should be noted that this DBR could perform the designated role of the reflection of blue emissions and transmission of red emissions below 45°, although with slight performance degradation. Further, it is very important to note that ultrahigh-resolution displays are used in near-eye applications, such as AR and VR, which do not require very broad viewing angles. Therefore, these results suggest that the proposed VSS LED structures could be satisfactory for near-eye applications (e.g., AR and VR), which require relatively low viewing angles.

image file: c9nr04423j-f3.tif
Fig. 3 (a) Schematic diagram of the fabrication process for integrating red LED on blue LED with an engineered bonding medium (DBR). (b) Cross-sectional TEM image with low magnitude for a red LED on a blue substrate. Magnified TEM image at the (c) GaP/DBR interface and (d) bonding interface.

Nevertheless, in order to reduce the color shift, a new configuration can be added to the final μ-LED structure, such as sidewall metal reflectors as well as optimized taper angle and thickness.30 Such structures can minimize the sidewall emission of μ-LEDs, as well as maintain or enhance the top emissions, thereby reducing power consumption. Obviously, a novel device design is imperative by considering the physical parameters (RGB subpixel thickness, taper angle, and DBR). Therefore, further investigation of the proposed structures needs to be performed to alleviate the angular dependency of μ-LEDs.

To experimentally verify the light propagation performances of DBRs, dielectric materials (e.g., SiO2 and SiNx) were alternately deposited on the sapphire substrate by using plasma-enhanced chemical vapor deposition (PECVD) on the basis of the simulation results. The measurements of the transmittance spectra were carried out with 3, 5, and 7 pairs of DBRs ranging from 300 to 800 nm, as shown in Fig. 2(e) and (f). When the beam was incident on the DBR surface (indicating blue emission at the top), low transmittance values of 37%, 15%, and 5.5% at 450 nm were observed as the number of DBR pairs increased, as shown in Fig. 2(e). At 630 nm, all the samples showed high transmittance values (higher than 90%).

In addition, the sample with 7 DBR pairs exhibited a stopband width of approximately 110 nm. The measured optical properties were effectively matched with the results predicted by the simulations. In order to compare the situation of blue emission at the bottom, a beam incident from the substrate side was also measured, as shown in Fig. 2(f). Almost the same transmittance spectra were obtained in the entire wavelength range, demonstrating that the fabricated DBR could perform the desired functionality of blue light reflection and red light transmission. This eliminated the problem of color modulation. Consequently, we successfully fabricated a DBR structure for engineering the bonding interface.

Ultrahigh-density VSS LED structures with DBR bonding interface

Fig. 3(a) shows the fabrication process for the monolithic integration of red and blue LEDs by using the engineering of the bonding interface with a designed DBR. By using VSS LEDs, we analyzed the material and its optical properties. Fig. 3(b) shows a cross-sectional TEM image of the fabricated sample. It has been confirmed that the red LED structure was bonded to the blue substrate with an interfacial DBR and bonding materials, exhibiting excellent uniformity and a clear bonding interface, as shown in Fig. 3(b). Further, voids and visible defects were not observed in the bonding materials, suggesting a highly stable integration process induced by the low-temperature wafer bonding. Fig. 3(c) and (d) show the magnification images of GaP/DBR and Al2O3 bonding interfaces, respectively. At the GaP/DBR interface, a slightly roughened interface was observed due to the lattice mismatch between GaP and GaAs. The DBR was deposited subsequent to the generation of this roughness, as shown in Fig. 3(c). Since it could deteriorate the bonding quality of the final structure, we planarized the surface roughness by the chemical–mechanical polishing (CMP) process until achieving a low root mean square surface roughness (Rrms) value of less than 1 nm. Owing to this process, a nearly flat surface of DBR and Al2O3 bonding layers could be obtained across the entire substrate, as shown in Fig. 3(d). These results reveal the fact that we successfully fabricated fully VSS LED structures using an engineered bonding interface. When compared with the earlier results regarding adhesive bonding (>4 μm), a thin and functional intermediate bonding layer was obtained, which could enhance the device performances by potentially improved heat dissipation and light extraction.10,18,23

To evaluate the actual DBR characteristics in the VSS LED structures, micro-photoluminescence (μ-PL) measurements were carried out for different numbers of DBR pairs. A laser diode with an emission wavelength of 405 nm corresponding to a blue excitation source was vertically incident from the substrate side, as shown in the inset of Fig. 4(a). To carefully analyze the performance, optical pumping powers of 51, 230, 524, 720, and 1260 μW were calibrated with a Si photodiode (Thorlabs). The PL response of the red LED clearly decreased and/or disappeared with an increase in the number of DBR pairs from 3 to 7, whereas sharp PL responses of red LED were observed without DBR, as shown Fig. 4(b)–(d). When the photons with higher energy than that of the bandgap of the red LED were illuminated on the sample without DBR, the photons could generate electron–hole pairs, resulting in red PL responses in MQWs. A similar result can be seen in the sample with 3 pairs of DBRs with an increase in the laser power because the DBR cannot completely block the blue light, as shown in Fig. 2. On the other hand, PL responses were obviously decreased in the sample with a higher number of DBR pairs even for relatively high power of 1260 μW. These results strongly suggest that the optical engineering of the interface using DBR can be used to fabricate full-color VSS LEDs without undesirable color interference for ultrahigh-resolution microdisplays by taking full advantage of inorganic-based μ-LEDs.

image file: c9nr04423j-f4.tif
Fig. 4 Micro-photoluminescence (μ-PL) measurement results depending on the various laser powers in the fabricated VSS LED structure: (a) without DBR, (b) 3 pairs of DBRs, (c) 5 pairs of DBRs, and (d) 7 pairs of DBRs.

For investigating the feasibility of the fabrication of ultrahigh-resolution displays, post-integration processes need to be performed on the DBR-inserted samples by using electron beam (e-beam) lithography. Obviously, it can be replaced by other mass-production lithography technologies in the semiconductor process to define tiny patterns. Fig. 5(b) and (c) show the tilted top view of the pixel array with a pitch of 1.2 μm and bird's-eye view of the pixel array with a pixel pitch of 710 nm, respectively, which showed excellent uniformity and process reliability for larger dimensions. When an ultraviolet laser (365 nm) was illuminated for samples with widths of 1.0 and 0.6 μm, the apparent red emission dependent on optical pumping is shown in the inset of Fig. 5(c). Uniform red emission from μ-LEDs with deeply scaled dimensions was obtained, indicating high potential for use in ultrahigh-resolution displays. Fig. 5(d)–(g) show the successful pixel fabrication with aggressive scaling of the pixel pitch from 1.2 to 0.4 μm. Even with a minimal pixel pitch of 0.4 μm in our approach, the singulation of the pixel and uniformity were successfully demonstrated, whereas the samples with reduced pixel pitch showed a conical shape near the surface. This shape could be attributed to mask erosion under our present etching conditions, which induces the undesired transformation of pattern shapes.31,32 However, since mask erosion would be mainly affected by the process conditions (such as rf power, time, and gas flow), the pattern shape could be easily improved by using further optimization of the etching conditions.31

image file: c9nr04423j-f5.tif
Fig. 5 (a) Schematic diagram of the fabrication process of ultrahigh-density VSS-type μ-LEDs. (b) Tilted top view of the SEM image for fabricating μ-LEDs with a pitch of 1.2 μm. (c) Bird's-eye view of the μ-LED with a pitch of 0.71 μm. (d–g) Cross-sectional SEM images of μ-LEDs with pitch values of 1.2, 0.71, 0.5, and 0.4 μm, respectively.

Moreover, it is evident that the resolution would be a critical parameter in this work. Therefore, we estimated the resolutions of the fabricated pixels by assuming a single rod corresponding to a single pixel of the VSS RGB. Consequently, we obtained surprisingly high resolutions ranging from a pitch of 1.2 μm (21[thin space (1/6-em)]200 PPI) to a pitch of 0.4 μm (63[thin space (1/6-em)]500 PPI)—the smallest pixel pitch ever reported among any material system.1,33 It is a significant first step in the fabrication of high-resolution displays because producing such high resolutions and high yields would be very difficult to obtain by using other conventional methodologies. We believe that this can only be fulfilled by using sequential integration processes after the transfer of the entire layer, which is in contrast to the pick-and-place method, fundamentally limited by the accuracy of mechanical alignment. Therefore, we have proposed and demonstrated a viable method—for the first time—for fabricating future high-resolution microdisplays.

Finally, in order to fabricate complete μ-LED displays, the integration with driving circuits is necessary, including electrical contacts in the μ-LEDs themselves. However, conventional flip-chip integration based on indium bump cannot properly correspond to the reduction in the pixel size due to limited alignment accuracy.20 With regard to monolithically integrated VSS μ-LEDs, we have a plan to address each pixel to the driving circuit by utilizing CMOS technology that can enable several fabrication processes at the scale of several tens of nanometers. Therefore, we propose a new approach to fabricate μ-LED displays using sequential 3D integration (monolithic 3D integration) processes. In the sequential 3D integration (monolithic 3D integration) process, the device layer is initially formed on top of the prefabricated CMOS device and the device layer is sequentially processed. By doing this, we could achieve very precise alignment accuracy, resulting in high-resolution displays. Subsequently, we finally landed the interconnection metals through the CMOS process after forming the device layers (LED thin films for R, G, and B in this work). Initially, four lines and three pads for each R/G/B pixel were formed on the driving circuits before the transfer of the LED thin films to address each other, as shown in Fig. 6(a). When the metal pad and lines were formed, a proper insulation process was employed, such as polymer coating, PECVD, or ALD process. Then, thin-film transfer of R/G/B LEDs was carried out by using the proposed DBR bonding process, as shown in Fig. 6(b). After that, the device fabrication process could start with the formation of electrodes and device isolation. Fig. 6(c) shows the schematic of the fabricated devices on driving circuits, which comprise the R/G/B electrodes. Each electrode used different cathodes and anodes for various LEDs. Lastly, the interconnection process was carried out by the landing metals after the opening of the pad region for driving the circuit and devices. As shown in Fig. 6(d), interconnecting metals (black pattern) would be landed in place, e.g., G1 electrode of the device-G1 pad of the bus line. In this approach, the pixel pitch would mainly rely on the fabrication capability and arrangement designs, which lead to reducing the LED sizes beyond the limitation of the flip-chip integration method. For example, the number of metal lines would be decreased when the common cathode pads for R/G/B LEDs are fabricated in this approach.

image file: c9nr04423j-f6.tif
Fig. 6 Conceptual fabrication process of full-color VSS μ-LEDs on the driving circuit. (a) Formation of bus lines and pads on driving circuits for landing interconnection metals. (b) Transfers of R/G/B thin films by using interface-engineered bonding technology. (c) Fabrication of each μ-LED for electrical operation. (d) Interconnection between electrodes on the devices and each pad.

Analysis of surface properties by optical measurements and simulations

In order to evaluate the effect of nonradiative recombinations, TRPL was measured for the top red LEDs for different surface-to-volume ratios at room temperature by employing a time-correlated single-photon system (Halcyone, Ultrafast systems). This is because nitride (N)-based materials have fundamentally lower diffusion coefficients and higher bandgaps than those of arsenic (As)-based materials, which makes the surface passivation for blue/green emitters less important than that for red emitters. While the samples were optically pumped with an excitation source at 405 nm, the detection wavelength was centered at 630 nm. Fig. 7(a) shows the measurement results for each LED pixel size of 1.0, 0.6, 0.4, and 0.3 μm. It should be noted that the decay time (τ) gradually decreased with the dimensions of red LEDs. This decreased the decay time originating from the increased ratio of nonradiative recombinations at the surface since surface-related PL lifetimes are typically short.26,34 For an accurate evaluation, the effective decay times (τeff) of each dimension were estimated by using a linear sum of the weighted multiple exponentials as follows:
image file: c9nr04423j-t1.tif(2)
where ai and τi denote the weighting coefficients and fitted decay characteristic times, respectively.35 As shown in the inset of Fig. 6(a), we have the best fitting result for 1 μm LED patterns by utilizing i = 3 for the triple exponential decay. After proceeding to fit all the TRPL signals, effective decay times were found to be 2.34, 2.11, 1.33, and 1.31 ns in the order from 1.0 to 0.3 μm, respectively, as shown in the inset table of Fig. 7(b). These values seem to be reasonable because they are fairly similar to the decay times of 1.39 and 1.98 ns in the earlier report involving unpassivated III–V nanowires with similar dimensions of 300 and 700 nm, respectively.34 Therefore, elongated effective decay times strongly reveal that the contribution of nonradiative recombination processes would be small due to the reduced surface-to-volume ratio.34,36

image file: c9nr04423j-f7.tif
Fig. 7 (a) Experimental TRPL decay curves measured at room temperature for different μ-LED sizes and fitting results of the decay curve for 1 μm-sized LED, as shown in the inset. (b) Inverse effective decay time as a function of inverse LED widths for calculating the SRV value and fitting parameters for calculating the effective decay times as inserted in the inset table. (c) Comparison of the TRPL decay curves of bulk-layer red LED layer and unpassivated and passivated 1.0 μm-width LEDs with sulfur passivation and enhancement ratio shown in the inset.

For additional quantitative analyses, we extracted the surface recombination velocity (SRV) of the sample by using the measured decay times, which are dependent on the LED size. The effective decay time is simply expressed as the contributions of bulk and surface recombinations as the following relation:

image file: c9nr04423j-t2.tif(3)
where τbulk and d refer to the carrier lifetime in bulk material and width of the pillar with a square cross-section, respectively.36,37Fig. 7(b) shows the effective decay times as a function of inverse LED widths. Therefore, the linear fitting curve enables us to extract the SRV value of 4100 cm s−1, which is a moderate value when compared with the other results involving unpassivated GaAs- and InP-based nanowires in the range from 103 to 105.36–38 Finally, we calculated the interface trap density (Dit) from the SRV value for quantitatively analyzing the performances of the LEDs. The relationship between SRV and Dit is defined as = σvthDit, where σ and vth refer to the capturing cross section and thermal velocity, respectively.39 We found that Dit at the surface of our sample was approximately 7.4 × 1012 eV−1 cm−2, with σ = 2.4 × 10−15 cm2 for AlGaInP, indicating relatively poor surface quality.40 This deduction was fairly meaningful because it enabled the estimation of the interfacial properties by simple optical measurements, although the power-dependent TRPL measurements and deep-level transient spectroscopy should be carried out for obtaining more accurate SRV and Dit values.40–42 Further, it should be noted that since the electron diffusion length in GaAs-based systems would reach several tens of micrometers, the reduction in Dit and proper surface passivation should be very important to fully utilize the emission properties, particularly for red LEDs. Ultimately, this approach could lead to strengthening the rapid evaluation for scaled μ-LEDs by combining the actual electrical characterization.

Therefore, to reduce the surface trap density, we carried out sulfur passivation for red LEDs with 1 μm size, which is a fairly common surface treatment for III–V-based electronic and optoelectronic devices.43,44 After the native oxide was removed in NH4OH[thin space (1/6-em)]:[thin space (1/6-em)]DI (1[thin space (1/6-em)]:[thin space (1/6-em)]1) for 1 min, the sample was dipped in (NH4)2S[thin space (1/6-em)]:[thin space (1/6-em)]DI (1[thin space (1/6-em)]:[thin space (1/6-em)]10) for 10 min. After this passivation, TRPL measurements were immediately performed. Fig. 7(c) shows the TRPL results for three samples: the bulk reference layer without any patterning, unpassivated LEDs, and passivated LEDs. Here, as a reference, we measured the TRPL response of the bulk layer, which could act as the final target of the passivated sample. It was observed that there was a considerable discrepancy in the decay curves between the unpassivated sample and bulk layer, indicating a substantial contribution of nonradiative recombinations. In contrast, the decay behavior of the passivated sample came very close to the decay curve of the bulk layer with simple sulfur passivation. Therefore, it is evident that sulfur treatment leads to an increase in the effective decay time due to effective surface passivation.45 As shown in the inset of Fig. 7(c), a dramatic enhancement in the effective decay time was found to be 83% when compared with the unpassivated initial state, yielding results similar to those obtained from the bulk reference layer. Although this enhancement in optical measurement does not linearly produce an improvement in the LED performances, it surely confirms that there is an obvious improvement in the electrical performances in earlier reports.46,47

Lastly, in order to investigate the effects of Dit at the surface on the device performances in electrically pumped LEDs, we carried out two-dimensional TCAD simulations. The structure was assumed to be a p-GaAs/p-cladding layer/MQW/n-cladding layer/n-GaAs layer from the top to bottom, as shown in Fig. 8. Here, to investigate the impact of LED dimensions, we varied the LED widths from 1 to 20 μm. Fig. 8 shows the relative efficiency as a function of the current density for different LED widths as a parameter of Dit. Radiative efficiency was determined from the portion of radiative recombinations over the total recombination including nonradiative recombinations. Evidently, there is approximately 20% efficiency drop between the LED without Dit and LED with Dit (1013 eV−1 cm−2) even for relatively large LED widths (20 μm). The impact of Dit would be more critical with decreasing LED sizes, as predicted. With a reduction in the LED width from 20 to 1 μm, the efficiency drops increased and radiative efficiency was less than 10% for LED widths of 1 μm due to the increased surface-to-volume ratio. This strongly implies that Dit reduction must be considered for μ-LED fabrication and the performance of the fabricated red LEDs in this work would be further improved through optimal passivation. This would be the key to realize low-power/high-resolution microdisplays.

image file: c9nr04423j-f8.tif
Fig. 8 Schematic structure of the simulation and relative efficiency vs. current density results for different surface trap densities depending on the μ-LED sizes.


Formation of VSS μ-LED structures

The AlGaInP-based red LED structure comprises p+-GaP contact layer/p-GaP window layer/p-AlGaInP tensile strain barrier and reducing layer (TSBR)/p-AlInP cladding layer/AlGaInP MQWs/n-AlInP cladding layer/n-AlGaInP layer/n-GaInP layer/n+-GaAs contact layer/InGaP etch stop layer, from the top to bottom; they are grown on a GaAs substrate by using metalorganic vapor-phase epitaxy (MOCVD) method, as shown in Fig. 3. The integration process started from standard wafer cleaning with acetone, methanol, and isopropyl alcohol (IPA). After the cleaning, dielectric materials of SiO2 and SiNx were deposited on the top surface of the red LED. Subsequently, an additional SiO2 layer was deposited on the DBR top surface and a blue substrate for forming the surface protection layer during the planarization process. Then, the CMP process was performed on both the samples for achieving a smooth surface, followed by 5 nm-thick Al2O3 deposition for wafer bonding because Al2O3 provides strong bonding strength. Then, oxygen plasma treatment was conducted on both the Al2O3 surfaces for surface activation, subsequently performing wafer bonding at a pressure of 2.2 MPa at room temperature in a wafer bonder. Here, this low-temperature bonding process allowed the introduction of DBR engineering, which cannot be achievable with high-temperature bonding causing the material's intermixing at the bonding and material interfaces. Finally, after removing the GaAs substrate, the monolithic integrated LED structure with a DBR bonding interface was completed, as shown in Fig. 3.

Fabrication of VSS μ-LEDs with ultrahigh density

First, PECVD-SiO2 was deposited on the surface as a dry etching mask, as shown in Fig. 5(a). Then, every single pixel was patterned using an e-beam resist of PMMA; subsequently, 100 nm-thick nickel was deposited and lifted off on a 500 nm-thick SiO2 etch mask. After defining the patterns with various LED sizes of 200 nm, 300 nm, 500 nm, and 1 μm, the SiO2 mask was patterned by the selective etching of a SiO2 layer through CF4-gas-based reactive-ion etching (RIE). Finally, the pixel singulation process was carried out by using inductively coupled plasma reactive-ion etching (ICP-RIE) with Cl2-based chemistry, resulting in isolated pixels at different pixel pitches, as shown in the scanning electron microscope (SEM) images.


We successfully demonstrated the monolithic integration of AlGaInP red μ-LEDs on a blue substrate with SiO2/SiNx DBR layer for eliminating color modulation. The resulting VSS μ-LED structure showed negligible color modulation under high emissions by the optimized DBR to selectively reflect blue emission at 450 nm and transmit red emission at 630 nm at the bonding interface. Moreover, we fabricated optically pumped ultrahigh-density VSS μ-LED arrays yielding 63[thin space (1/6-em)]500 PPI, which cannot be achieved in LSS μ-LED arrays utilizing the standard semiconductor process technology. Finally, we investigated the surface contribution of the recombination in dense-red LED arrays. The fabricated red LEDs showed relatively high SRV values of 4100 cm s−1, corresponding to a Dit value of approximately 7.4 × 10−12 eV−1 cm−2, addressing the importance of surface passivation for high-resolution LEDs. Further, by conducting a systematic investigation of the impact of Dit on radiative recombinations using TCAD simulations, we found that proper passivation must be taken into account in μ-LED fabrication and it can provide surprising enhancement in quantum efficiency. From these results, we firmly believe that a bonding-interface- and surface-engineered monolithic integration method, as suggested in this study, can be a powerful strategy for fabricating future energy-efficient/ultrahigh-resolution microdisplays.

Conflicts of interest

There are no conflicts to declare.


This work was partly supported by KAIST startup funding (G04180061), BK21plus, and the National Research Foundation of Korea (NRF) grant (Grant No. 2017M1A2A2048879, 2019R1F1A1061927).

Notes and references

  1. K. Zhang, D. Peng, K. M. Lau and Z. Liu, SID Int. Symp. Dig. Tech., 2017, 48, 358–361 Search PubMed .
  2. T. Lin, H. C. Kuo, X. D. Jiang and Z. C. Feng, Nanoscale Res. Lett., 2017, 12, 2–7 CrossRef PubMed .
  3. J. R. Bonar, G. J. Valentine, Z. Gong, J. Small and S. Gorton, Proc. SPIE, 2016, 9768, 97680Y CrossRef .
  4. Z. J. Liu, W. C. Chong, K. M. Wong, K. H. Tam and K. M. Lau, IEEE Photonics Technol. Lett., 2013, 25, 2267–2270 CAS .
  5. S. Zhang, Z. Gong, J. J. D. McKendry, S. Watson, A. Cogman, E. Xie, P. Tian, E. Gu, Z. Chen, G. Zhang, A. E. Kelly, R. K. Henderson and M. D. Dawson, IEEE Photonics J., 2012, 4, 1639–1646 Search PubMed .
  6. M. Funato, K. Hayashi, M. Ueda, Y. Kawakami, Y. Narukawa and T. Mukai, Appl. Phys. Lett., 2008, 93, 021126 CrossRef .
  7. M.-L. Lee, Y.-H. Yeh, S.-J. Tu, P. C. Chen, W.-C. Lai and J.-K. Sheu, Opt. Express, 2015, 23, A401 CrossRef CAS PubMed .
  8. I. K. Park, J. Y. Kim, M. K. Kwon, C. Y. Cho, J. H. Lim and S. J. Park, Appl. Phys. Lett., 2008, 92, 90–93 Search PubMed .
  9. H. S. El-Ghoroury, M. Yeh, J. C. Chen, X. Li and C. L. Chuang, AIP Adv., 2016, 6, 075316 CrossRef .
  10. J. Chun, K. J. Lee, Y. C. Leem, W. M. Kang, T. Jeong, J. H. Baek, H. J. Lee, B. J. Kim and S. J. Park, ACS Appl. Mater. Interfaces, 2014, 6, 19482–19487 CrossRef CAS PubMed .
  11. M. Auf Der Maur, A. Pecchia, G. Penazzi, W. Rodrigues and A. Di Carlo, Phys. Rev. Lett., 2016, 116, 027401 CrossRef PubMed .
  12. K. Chung, J. Sui, B. Demory, C. H. Teng and P. C. Ku, Appl. Phys. Lett., 2017, 110, 111103 CrossRef .
  13. M. Choi, B. Jang, W. Lee, S. Lee, T. W. Kim, H. J. Lee, J. H. Kim and J. H. Ahn, Adv. Funct. Mater., 2017, 27, 1606005 CrossRef .
  14. M. A. Meitl, Z. T. Zhu, V. Kumar, K. J. Lee, X. Feng, Y. Y. Huang, I. Adesida, R. G. Nuzzo and J. A. Rogers, Nat. Mater., 2006, 5, 33–38 CrossRef CAS .
  15. B. H. Kim, S. Nam, N. Oh, S. Y. Cho, K. J. Yu, C. H. Lee, J. Zhang, K. Deshpande, P. Trefonas, J. H. Kim, J. Lee, J. H. Shin, Y. Yu, J. Bin Lim, S. M. Won, Y. K. Cho, N. H. Kim, K. J. Seo, H. Lee, T. Il Kim, M. Shim and J. A. Rogers, ACS Nano, 2016, 10, 4920–4925 CrossRef CAS PubMed .
  16. J. Xiao, R. R. Chaudhuri and S. W. Seo, IEEE Trans. Compon., Packag., Manuf. Technol., 2016, 6, 1283–1289 CAS .
  17. S. K. Kim, J. P. Shim, D. M. Geum, J. Kim, C. Z. Kim, H. S. Kim, J. D. Song, S. J. Choi, D. H. Kim, W. J. Choi, H. J. Kim, D. M. Kim and S. Kim, IEEE Trans. Electron Devices, 2018, 65, 1862–1868 CAS .
  18. C. M. Kang, J. Y. Lee, D. J. Kong, J. P. Shim, S. Kim, S. H. Mun, S. Y. Choi, M. Do Park, J. Kim and D. S. Lee, ACS Photonics, 2018, 5, 4413–4422 CrossRef CAS .
  19. D. M. Geum, M. S. Park, J. Y. Lim, H. D. Yang, J. D. Song, C. Z. Kim, E. Yoon, S. H. Kim and W. J. Choi, Sci. Rep., 2016, 6, 20610 CrossRef CAS PubMed .
  20. L. Zhang, F. Ou, W. C. Chong, Y. Chen and Q. Li, J. Soc. Inf. Disp., 2018, 26, 137–145 CrossRef CAS .
  21. S. H. Kim, S. K. Kim, J. P. Shim, D. M. Geum, G. Ju, H. S. Kim, H. J. Lim, H. R. Lim, J. H. Han, S. Lee, H. S. Kim, P. Bidenko, C. M. Kang, D. S. Lee, J. D. Song, W. J. Choi and H. J. Kim, IEEE J. Electron Devices Soc., 2018, 6, 579–587 CAS .
  22. S. Panth, S. Samal, Y. S. Yu and S. K. Lim, 2014 SOI-3D-Subthreshold Microelectron. Technol. Unified Conf. S3S 2014, 2014, pp. 1–2 .
  23. C. M. Kang, S. J. Kang, S. H. Mun, S. Y. Choi, J. H. Min, S. Kim, J. P. Shim and D. S. Lee, Sci. Rep., 2017, 7, 1–9 CrossRef PubMed .
  24. A. Daami, F. Olivier, L. Dupré, F. Henry and F. Templier, SID Int. Symp. Dig. Tech., 2018, 49, 790–793 CrossRef CAS .
  25. Y. Lee, C. Lee and C. Chen, IEEE J. Quantum Electron., 2011, 47, 636–641 CAS .
  26. S. L. Chen, W. M. Chen, F. Ishikawa and I. A. Buyanova, Sci. Rep., 2015, 5, 11653 CrossRef PubMed .
  27. F. Gou, E.-L. Hsiang, G. Tan, P.-T. Chou, Y.-L. Li, Y.-F. Lan and S.-T. Wu, Opt. Express, 2019, 27, A746 CrossRef PubMed .
  28. R. H. Horng, W. K. Wang, S. Y. Huang and D. S. Wuu, IEEE Photonics Technol. Lett., 2006, 18, 457–459 CAS .
  29. T. Fujii, A. David, C. Schwach, P. M. Pattison, R. Sharma, K. Fujito, T. Margalith, S. P. Denbaars, C. Weisbuch and S. Nakamura, Jpn. J. Appl. Phys., Part 2, 2004, 43, L411 CrossRef CAS .
  30. J. W. Lee, J. H. Park, D. Y. Kim, E. F. Schubert, J. Kim, J. Lee, Y. L. Kim, Y. Park and J. K. Kim, ACS Photonics, 2016, 3, 2030–2034 CrossRef CAS .
  31. S. Guilet, S. Bouchoule, C. Jany, C. S. Corr and P. Chabert, J. Vac. Sci. Technol., B: Microelectron. Nanometer Struct.–Process., Meas., Phenom., 2006, 24, 2381 CrossRef CAS .
  32. K. Kennedy, K. M. Groom and R. A. Hogg, Semicond. Sci. Technol., 2006, 21, L1–L5 CrossRef CAS .
  33. T. Fujii, C. Kon, Y. Motoyama, K. Shimizu, T. Shimayama, T. Yamazaki, T. Kato, S. Sakai, K. Hashikaki, K. Tanaka and Y. Nakano, J. Soc. Inf. Disp., 2018, 26, 178–186 CrossRef .
  34. D. Tedeschi, M. De Luca, H. A. Fonseka, Q. Gao, F. Mura, H. H. Tan, S. Rubini, F. Martelli, C. Jagadish, M. Capizzi and A. Polimeni, Nano Lett., 2016, 16, 3085–3093 CrossRef CAS PubMed .
  35. S. Y. Jeong, J. Kyhm, S. K. Cha, D. K. Hwang, B. K. Ju, J. S. Park, S. J. Kang and I. K. Han, Small, 2019, 1900008, 1–9 Search PubMed .
  36. A. Higuera-Rodriguez, B. Romeira, S. Birindelli, L. E. Black, E. Smalbrugge, P. J. Van Veldhoven, W. M. M. Kessels, M. K. Smit and A. Fiore, Nano Lett., 2017, 17, 2627–2633 CrossRef CAS PubMed .
  37. S. Arab, C. Y. Chi, T. Shi, Y. Wang, D. P. Dapkus, H. E. Jackson, L. M. Smith and S. B. Cronin, ACS Nano, 2015, 9, 1336–1340 CrossRef CAS PubMed .
  38. D. Ren, Z. Rong, B. Liang and D. L. Huffaker, Proc. SPIE, 2018, 10543, 1054306 Search PubMed .
  39. D. K. Schroder, Material and device semiconductor material and device, 3rd edn, 2006, vol. 44 Search PubMed .
  40. A. Tukiainen, J. Dekker, T. Leinonen and M. Pessa, Mater. Sci. Eng., B, 2002, 91–92, 389–392 CrossRef .
  41. L. E. Black, A. Cavalli, M. A. Verheijen, J. E. M. Haverkort, E. P. A. M. Bakkers and W. M. M. Kessels, Nano Lett., 2017, 17, 6287–6294 CrossRef CAS PubMed .
  42. N. Khelifati, D. Bouhafsa, A. Mebarek-Azzem, S. E. H. Abaidia, B. Palahouane and Y. Kouhlane, Acta Phys. Pol., A, 2016, 130, 188–190 CrossRef CAS .
  43. Y. K. Su, H. C. Wang, C. L. Lin, W. B. Chen and S. M. Chen, IEEE Photonics Technol. Lett., 2003, 15, 1345–1347 Search PubMed .
  44. H. K. Kang, Y. S. Kang, M. Baik, K. S. Jeong, D. K. Kim, J. D. Song and M. H. Cho, J. Phys. Chem. C, 2018, 122, 7226–7235 CrossRef CAS .
  45. M. Latzel, P. Büttner, G. Sarau, K. Höflich, M. Heilmann, W. Chen, X. Wen, G. Conibeer and S. H. Christiansen, Nanotechnology, 2017, 28, 055201 CrossRef CAS PubMed .
  46. X. Li, S. Okur, F. Zhang, V. Avrutin, Ü. Özgür, H. Morko, S. M. Hong, S. H. Yen, T. S. Hsu and A. Matulionis, J. Appl. Phys., 2012, 111, 063112 CrossRef .
  47. Y. Narukawa, S. Saijou, Y. Kawakami, S. Fujita, T. Mukai and S. Nakamura, Appl. Phys. Lett., 1999, 74, 558–560 CrossRef CAS .

This journal is © The Royal Society of Chemistry 2019