Highly efficient blue InGaN nanoscale light-emitting diodes

Indium gallium nitride (InGaN)-based micro-LEDs (μLEDs) are suitable for meeting ever-increasing demands for high-performance displays owing to their high efficiency, brightness and stability1–5. However, μLEDs have a large problem in that the external quantum efficiency (EQE) decreases with the size reduction6–9. Here we demonstrate a blue InGaN/GaN multiple quantum well (MQW) nanorod-LED (nLED) with high EQE. To overcome the size-dependent EQE reduction problem8,9, we studied the interaction between the GaN surface and the sidewall passivation layer through various analyses. Minimizing the point defects created during the passivation process is crucial to manufacturing high-performance nLEDs. Notably, the sol–gel method is advantageous for the passivation because SiO2 nanoparticles are adsorbed on the GaN surface, thereby minimizing its atomic interactions. The fabricated nLEDs showed an EQE of 20.2 ± 0.6%, the highest EQE value ever reported for the LED in the nanoscale. This work opens the way for manufacturing self-emissive nLED displays that can become an enabling technology for next-generation displays. Using a sol–gel passivation method, the fabrication of blue InGaN nanorod-LEDs with the highest external quantum efficiency value ever reported for LEDs in the nanoscale is demonstrated.

Light-emitting diodes (LEDs) have been continuously seeing technological advances and are being used in various applications, such as solid-state lightings, displays, optical data communications and photonics [10][11][12][13] . In particular, the huge opportunity in the display applications for high brightness and contrast are leading the recent development of micro-LEDs (μLEDs) 3,12,13 . Moreover, rapid growth of the market for augmented reality and extended reality demanding high pixel density per inch (greater than 5,000) and high luminance (more than 10,000 cd m −2 ) 14-17 is also accelerating a shift from a conventional liquid crystal display to the self-emissive μLED displays.
For most display applications, especially high-resolution augmented displays, an LED chip size on the order of micrometres or submicrometres is required 15,16 . However, small-size μLEDs show the size-dependent EQE reduction problem, mainly owing to increased non-radiative recombination loss at the surface as the surface-to-volume ratio increases for smaller chip size [6][7][8][9] . Although the surface recombination velocity, a key parameter for surface recombination, of InGaN-based materials (about 3 × 10 2 -10 4 cm s −1 ) is several orders of magnitude lower than for other III-V semiconductors such as GaAs and InP (ref. 18 ), the Shockley-Read-Hall (SRH) recombination at surface defects becomes severe in the micrometre-scale InGaN LEDs [6][7][8][9][10] . In the manufacturing processes of μLEDs, defects, impurities and dangling bonds, which act as non-radiative recombination centres, are generated on the etched surface 7,9,15,16 . It was reported that the EQE of blue InGaN LEDs decreases from about 10% at the lateral dimension of 10 μm down to about 2-3% at the 1-μm size 9 . Therefore, it is essential to develop a technology that minimizes surface defects that occur when manufacturing μm or sub-μm-sized (nanoscale) LEDs and protects the surface with a robust passivation layer.
Development of highly efficient nanoscale LEDs can also make a breakthrough in reducing the manufacturing cost of large-sized display products. The current manufacturing process of μLED displays based on the 'pick-and-place' method has a high manufacturing cost because of slow tact time and low production yield 3 . To overcome this technical challenge, we are developing a new cost-effective method for fabricating nLED displays 19 . The nLEDs were dispersed on the substrate by means of inkjet printing and were horizontally aligned on prepatterned electrodes using dielectrophoretic forces [19][20][21] . Electric-field-assisted alignment of inkjet-printed nLEDs shortens the process time and lowers the unit cost 19 . This configuration not only increases the light extraction efficiency but also provides control over the luminance by adjusting the number of nLEDs in the pixel. Because the luminance is averaged out over many nLEDs connected in parallel in each pixel, the nLED display is less sensitive to luminance deviation owing to epitaxial defects compared with the μLED display having the pixel with a single μLED. Although the nLED display has a great potential of realizing high-performance displays, the development of highly efficient nLEDs by overcoming the size-dependent EQE reduction problem is a great challenge.
The sidewalls of the InGaN/GaN MQW nanorods contain various structural defects, such as surface dangling bonds, strain-relaxation-induced defects and residual damage from the dry-etch process 6,9,[22][23][24] . It was reported that surface passivation is an effective way to reduce defect states and thereby increase the EQE [22][23][24][25][26] . However, it is inevitable that the nLED is exposed to thermal energy, plasma and atomic reactions during the conventional passivation processes of chemical vapour deposition or atomic layer deposition (ALD) [22][23][24][25] . As a result, fragile sidewall surfaces are susceptible to structural damages, resulting in the formation of defects.
In this study, we will demonstrate that a low-temperature sol-gel process [27][28][29] can minimize the formation of defects during the surface passivation process, leading to blue InGaN nanoscale LEDs with a peak EQE of 20.2 ± 0.6%. Considering the trend of the decreasing EQE on size reduction (Extended Data Fig. 1), this is a remarkable achievement and offers the possibility of fabricating highly efficient nLED displays. Using extensive analysis methods such as photoluminescence (PL), electroluminescence (EL), cathodoluminescence (CL) and electron energy-loss spectroscopy (EELS), we identified the defects that cause the performance deterioration in the top-down-processed nLEDs. When the SiO 2 passivation layer is formed with an ALD process, gallium vacancy (V Ga ) complex defects are created on the sidewalls of InGaN/ GaN nanorods, resulting in the increased SRH recombination and the decreased EQE. This result was cross-validated using X-ray photoelectron spectroscopy (XPS) and deep-level transient spectroscopy (DLTS).
The fabrication process and surface treatment of the nLEDs consisting of an indium tin oxide (ITO)/p-GaN/MQW/n-GaN structure are shown in Fig. 1a-c. The eight pairs of InGaN/GaN MQWs were grown on a 4-inch c-plane sapphire substrate. A nanorod pattern with a diameter of about 600 nm was constructed through nanoimprint lithography and then it was dry etched by inductively coupled plasma reactive ion etching. Subsequently, potassium hydroxide (KOH) wet etching was conducted to remove the sidewall damages from the dry-etching process 25,30 . The nanorod shape is transformed from trapezoidal to a vertical cylinder because of the high etching barrier index of the m-plane surface 30 and the surfaces of the nanorods become smooth (Fig. 1b).
We applied two insulating layers on the nanorods because a single insulating layer is insufficient to simultaneously achieve the high efficiency, good reliability and processability required for display products. The inner SiO 2 (60 nm thickness) and outer Al 2 O 3 (20 nm thickness) layers were used for sidewall passivation and etch stopping during pixel processing, respectively, because of their high transparency, low leakage current and high bandgap energy 31,32 . Optimization of the inner insulating layer, which is directly deposited on the GaN surface, is of utmost importance for achieving highly efficient nLEDs. To minimize the surface damage during the passivation process, we developed a wet-chemical process for synthesizing the SiO 2 passivation layer by means of a sol-gel method, as shown in the inset of Fig. 1c. Performing the sol-gel process at room temperature could minimize the atomic reaction between the GaN surface and the SiO 2 layer and enhance the optical properties of the nanorods by passivating the dangling bonds.
The PL images and the fluorescence excitation-emission spectra (Fig. 1d) and the panchromatic (λ = 300-700 nm) CL images (Fig. 1e) are compared for the nanorods passivated by different methods; conventional plasma-enhanced ALD-deposited SiO 2 (left) and solgel-deposited SiO 2 (right). We chose the plasma-enhanced ALD SiO 2 as the reference for comparison with the sol-gel SiO 2 despite its EQE being slightly lower than the thermal ALD SiO 2 because of its superior thin film quality (Extended Data Fig. 2). Compared with the thermal ALD process, the plasma-enhanced ALD produces the SiO 2 layer with lower H 2 outgassing (Extended Data Fig. 2b), which deteriorates the device efficiency during the operation through defect creation, such as Mg-H complexes 33,34 . When the sol-gel process is applied for the SiO 2 passivation, the enhancement of the blue emission over the yellow emission is clearly observed in Fig. 1d. The intensity of the blue emission of the sol-gel SiO 2 -passivated nanorods is approximately 13 times higher than that of plasma-enhanced ALD SiO 2 -passivated nanorods (Fig. 1f).

Article
The slight increase of the yellow luminescence (YL) around 550 nm mainly originated from the n-GaN and u-GaN underneath the nanorods (Extended Data Fig. 3) owing to V Ga or substitutional carbon (C N ), and/or its complexes [35][36][37] . The YL intensity of nanorods dispersed on the wafer is higher than that of the glass specimen for both the plasma-enhanced ALD and sol-gel methods (Extended Data Fig. 3d,h). Furthermore, when we compared the PL spectra of the nanorods itself by normalizing the intensity of the glass specimens (Extended Data Fig. 3j), the YL intensity of the plasma-enhanced ALD SiO 2 -passivated nanorods is higher than that of the sol-gel SiO 2 -passivated nanorods, indicating that the sol-gel method produces fewer surface defects. The panchromatic CL images of the individual nanorod LEDs in Fig. 1e clearly visualizes the existence of the non-radiative recombination region depending on the passivation method 36,37 . The outer rim corresponds to the yellow emission from the insulator 38 and the inner circle corresponds to the blue emission of the MQW regions. In the case of plasma-enhanced ALD SiO 2 -passivated nanorods (Fig. 1e, left), the outside of the MQWs is darker owing to the non-radiative recombination centre (NRC). Furthermore, the nanorods show marked differences in brightness because of the different positions of the NRC. However, for the sol-gel SiO 2 -passivated nanorods (Fig. 1e, right), the overall brightness is uniform and the above-mentioned abnormal light emission is absent. Furthermore, the reduced carrier lifetime of the sol-gel SiO 2 -passivated nanorods verifies that the surface defects of GaN are reduced compared with that of the plasma-enhanced ALD-deposited SiO 2 (Fig. 1g). Figure 2a shows the EL and PL composite images of a single nLED within a pixel according to the passivation method. The pixel is composed of the nLEDs connected in parallel with transparent conductive metals. Because the EL of the horizontally aligned nLED is emitted along the radial and longitudinal directions owing to waveguide reflection, a bright blue and a relatively dark emission are observed at the top and bottom of the nanorods, respectively. The light extraction efficiency (LEE) is calculated for the pixel structure in Extended Data Fig. 4a using a finite-difference time-domain (Ansys Lumerical FDTD) method. The total LEE is approximately 25% and it comprises 71.0% radial emission, 16.8% p-side and 9.9% n-side waveguide emission, and the rest of the optical loss in the transparent electrodes (Extended Data Fig. 4b). The EL intensity profiles obtained from a region horizontally across the top of the nLEDs confirm that the EL intensity of the sol-gel SiO 2 -passivated nanorod is higher than that of the plasma-enhanced ALD SiO 2 -passivated nanorod (Fig. 2b). Figure 2c compares the EQE curves for the nLEDs for each surface passivation type. Each curve is obtained from a 60-pixel array (Extended Data Fig. 4c), and each pixel comprises an average of six and nine nLEDs for the plasma-enhanced ALD SiO 2 and sol-gel SiO 2 , respectively. The difference in the number of nLEDs per pixel originates from the dispersion in the aligned process. The average values of the peak EQE of the sol-gel and plasma-enhanced ALD SiO 2 -deposited nLEDs are 20.2 ± 0.6% and 8.9 ± 0.1%, respectively. Furthermore, the internal quantum efficiency (IQE), calculated by dividing the EQE by the LEE, was 81% and 36% for nLEDs with the sol-gel SiO 2 and plasma-enhanced ALD SiO 2 , respectively. The high EQE of the sol-gel SiO 2 nLED is remarkable because it is higher than the best EQE value of the μLED structure even at a larger size 9,23 . The primary reason for such a high EQE is the decreased GaN surface damage in the sol-gel SiO 2 nLED, as confirmed by the longer carrier lifetime in PL and decreased NRC region in CL in Fig. 1.
The nLEDs with the sol-gel SiO 2 show a lower leakage current than that of the plasma-enhanced ALD SiO 2 at below-threshold voltages (Fig. 2d,e) owing to the parallel resistance component attributed to the sidewall damage. This is consistent with the decrease in the ideality factor of nLEDs with the sol-gel SiO 2 compared with that of the plasma-enhanced ALD SiO 2 (Fig. 2f). Generally, the ideality factor of the LED changes greatly depending on its epi-structure 39 . However, we   have proved that the ideality factor can also be changed notably by the passivation method. The decrease in the ideality factor of sol-gel SiO 2 nanorods indicates a decrease in the effective SRH recombination 7,40 , thereby increasing the IQE to more than 80%. The marked change in the electrical and optical properties depending on the sidewall passivation method is closely related to the defects generated during the passivation process. Therefore, understanding the phenomenon occurring at the interface and controlling it are key to manufacturing nLEDs with outstanding performance. Figure 3 shows the analysis results of the interface between the sidewall of the nanorods and the insulator. The evolution of the morphology and atomic structure of the sidewall in the MQW region after each fabrication step is traced by means of high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) imaging (Fig. 3a). After the dry etching of the epi-layer, the nanorod developed a trapezoidal shape owing to the impact-energy difference on the surface. The white arrows indicate the amorphous region at the sidewall, which formed owing to successive ion bombardments. However, after KOH wet etching, the nanorod developed an inverted trapezoidal shape because the etch rate of p-GaN is slower than that of n-GaN (ref. 41 ). Furthermore, the semi-polar {1011} plane and m-plane appear alternately, resulting in a staircase-like sidewall morphology of the InGaN quantum wells. When the SiO 2 layer is deposited through plasma-enhanced ALD, the plasma creates point defects, such as nitrogen vacancies (V N ), nitrogen interstitials (N i ), nitrogen split interstitials (N-N) N , and V Ga and V Ga complexes 42,43 , resulting in the amorphization of the surface of the nanorods. This amorphous region is observed on the entire surface of the GaN nanorod. However, it is predominantly observed on the semi-polar facet of the InGaN quantum wells (yellow arrows), probably owing to the high density of dangling bonds. As the thickness of the SiO 2 layer increases from 2 nm to 60 nm (Extended Data Fig. 5e), the penetration depth of the plasma-induced amorphization increases to 4 nm at the InGaN quantum wells. However, when sol-gel SiO 2 is deposited on the nanorods, there is no amorphous region formed at the interface in addition to the deposited SiO 2 .  Fig. 6e), the latter shows oxygen deficiency because of oxygen atoms contributing to the Ga-O bond. Moreover, we obtained electron spin resonance (ESR) spectra to analyse the concentration of the (N-N) N defects (Fig. 3d,e), which are known as ambipolar defects, that is, with a deep acceptor in n-GaN and a deep donor in p-GaN (ref. 44 ). The (N-N) N 0 defects decreased after the wet etching of the nanorods and increased after plasma-enhanced ALD SiO 2 deposition. Because the defect concentrations are calculated by dividing the spins by the total weight of the nanorods and considering that most of the defects are concentrated at the nanorod surface, the actual differences in the defect concentrations of the nanorods after each fabrication step will be greater. Furthermore, the (N-N) N 0 defects in the sol-gel SiO 2 -coated nanorods were fewer than those in the wet-etched nanorod, demonstrating the passivation of surface dangling bonds without generating excess defects.
We also carried out DLTS analysis on a bulk LED chip and plasma-enhanced ALD SiO 2 -coated nLED-array chip (Extended Data Fig. 7). In the bulk LED chip, the main defects are electron traps owing to V N and N i with activation energies of 0.56 and 0.67 eV, respectively 45 . In the nLED-array chip, although the electron trap concentration is similar to that of the bulk LED chip, five types of hole traps (0.14, 0.39, 0.51, 0.56 and 0.93 eV) are observed and the defect concentration is greater than 10 15 cm −3 for the deepest defect level (E v + 0.93 eV). V Ga -related complexes (complexes of V Ga and oxygen that replace nitrogen sites, that is, V Ga -O N, V Ga -O N -2H) are reportedly responsible for these hole traps 46,47 . According to previous reports 41, 48 , these V Ga complexes increase the SRH coefficient, thereby increasing the non-radiative recombination rate and markedly decreasing the IQE.   Article Figure 4 shows a detailed comparison of two representative electron energy loss spectra obtained from the bulk and the surface region of the InGaN quantum wells. For the bulk InGaN, the N-K-edge electron energy-loss near-edge structure (ELNES) can be decomposed into four contributions, denoted by A, B, C and D (ref. 49 ). For the InGaN passivated by sol-gel SiO 2 , the observed spectral features of N-K ELNES are almost similar for the bulk and surface regions (top panel in Fig 4b).
Compared with those of InGaN passivated by the sol-gel method, pronounced changes are observed in the spectral features of the N-K ELNES (red line) of the surface region of InGaN passivated by plasma-enhanced ALD: the appearance of the small peak between the second and third peaks and the chemical shift of the third peak, labelled as C, to a lower energy (red shift).
To theoretically verify the observed spectral features of the N-K ELNES, density functional theory (DFT) calculations were performed by considering the V Ga -O N -2H complex in a 100-atom-based 3 × 4 × 2 GaN supercell, which corresponds to the most energetically favourable V Ga point defect complexed with O and H (ref. 47 ). The main changes caused by the V Ga -O N -2H complex are that the third peak (C) at approximately 405 eV (bottom-most panel of Fig. 4b) is shifted towards a lower energy by approximately 0.4 eV, and a small shoulder peak appears between B and C at approximately 403 eV, consistent with the experimental results. The red shift and the formation of the shoulder peak suggest that the V Ga -O N -2H complex locally breaks the wurtzite symmetry by changing the relative distance between the Ga and N atoms and then modifying the hybridization between the Ga and N 2p states, resulting in a chemical shift of the N-K edge and the formation of a shoulder peak. Therefore, we conclude that the V Ga -O N -2H complex is the dominant defect generated in the sidewalls of the nanorods during the plasma-enhanced ALD SiO 2 passivation, wherein a hydrogen-containing precursor is used.
Although the sol-gel SiO 2 passivation provides superior opto-electrical properties such as high peak EQE of 20.2 ± 0.6% (Fig. 2c), a low leakage current at below-threshold voltages (Fig. 2d,e) and lower ideality factor of 1.56 (Fig. 2f) compared with the plasma-enhanced ALD SiO 2 passivation, it has a few disadvantages, such as the presence of residual reaction by-products. To achieve good reliability, suitable post-treatments are required to complete the reaction and eliminate the reaction residue. We have found that this can be achieved while maintaining the high EQE by baking the sol-gel SiO 2 film (Extended Data Fig. 8). With these preliminary results, research on further improvement through process optimization is now under way.
In conclusion, we developed blue InGaN nLEDs with high peak EQE of 20.2 ± 0.6% by means of sol-gel SiO 2 passivation. The nanorods were aligned on prepatterned electrodes using the dielectrophoretic force and connected in parallel through transparent electrodes in each pixel. We demonstrated that passivation by means of conventional plasma-enhanced ALD induces amorphization at the InGaN quantum well surfaces and creates point defects on the sidewalls of GaN nanorods, which increases the SRH recombination and generates an NRC region at the InGaN sidewall of the nanorods. The sol-gel process provides effective passivation for the GaN surfaces because the SiO 2 nanoparticles are adsorbed on the GaN surface after the sol-gel reaction. Thus, the atomic interaction with the GaN surface is minimized and only dangling bonds of the surfaces are passivated, resulting in a low leakage current, decreased NRC regions and high EQE. We believe that our findings will accelerate the implementation of nLEDs in next-generation displays.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04933-5.

Fabrication of top-down-processed nLEDs
We used commercially grown 4-inch epitaxial wafers, produced by the Samsung Electronics LED Business Team. They had eight pairs of InGaN/GaN MQW LED structures on c-plane sapphire substrates. An ITO layer (110 nm thick) was deposited on top of the p-type GaN to obtain a p-type transparent conducting electrode. Nanoimprint lithography was performed to fabricate a rod-like pattern using a SiO 2 -based hard mask on the ITO layer, and this pattern was dry etched using inductively coupled plasma reactive ion etching. Subsequently, KOH wet etching was conducted to remove the surfaces of the mesa that were damaged during the dry-etching process. Then the nanorod surface was passivated by two insulating layers with the 60-nm-thick SiO 2 inner layer for sidewall passivation and the 20-nm-thick Al 2 O 3 outer layer for etch stopping during pixel processing. The average diameter and length of the fabricated nanorods were about 580 nm and 4 μm, respectively. The nanorods were separated from the wafer using a diamond cutter. The separated nanorods were then dispersed in solvents for depositing in the pixel area using an inkjet printing method. As shown in the schematic diagram in Extended Data Fig. 4a, the nanorods were placed between two bottom electrodes, which were used to align the nanorods horizontally by applying the dielectrophoretic force [19][20][21] .
The top transparent electrodes were connected with the nanorods for electrical biasing.

Synthesis of SiO 2 passivation layer by means of the sol-gel method
Tetraethyl orthosilicate (TEOS, Sigma-Aldrich, 98%) was used as a precursor for the synthesis of a SiO 2 passivation layer on the surface of nanorods. A schematic of the entire sol-gel process is shown in Fig. 1c. First, 210 ml of ethanol (EtOH, Sigma-Aldrich, anhydrous) and 290 ml of deionized water were mixed in a three-necked flask. The entire 4-inch wafer with nanorods was dipped in this solution. Then, 0.8 g of cetyltrimethylammonium bromide (CTAB, Sigma-Aldrich, ≥98%) was dissolved in the solution at room temperature to form micelles. After stirring the above mixture at 500 rpm for 5 min, 2.5 ml of an aqueous ammonia solution (Sigma-Aldrich, 28-30%) and 1.25 ml of TEOS were sequentially added dropwise into the reactants. The resulting mixture was then continuously stirred at 25 °C at 500 rpm for 2 h and the wafer was washed with deionized water and EtOH. The thickness of the SiO 2 layer increased linearly with an increase in the reaction time for 60 min and saturated at 23 nm owing to the limited quantity of sol particles (Extended Data Fig. 9g). The transmission electron microscopy image in the inset of Fig. 1c shows a well-defined and uniform SiO 2 layer on the nanorods from the top to bottom. By repeating the sol-gel reaction, the thickness of the SiO 2 layer increases. The PL intensity increased on doubling the thickness of the SiO 2 layer (Extended Data Fig. 9h) and then saturated. It is attributed to the reduction of defects and effective passivation of the dangling bonds (Extended Data Fig. 5c). To evaluate the passivation uniformity, we recorded the PL spectra from three different positions on the 4-inch wafer. We found that the variation in the PL intensity with respect to the position is minor.

PL analysis of nanorods and nLEDs in the pixel structure
Two-dimensional steady-state PL spectra were obtained using a FluoroMax-4 spectrofluorometer (Horiba) with a 150-W ozone-free xenon arc lamp as the excitation source. Steady-state PL/EL images and a two-photon spectrum of GaN nanorods were obtained using a modified confocal microscope (Leica TCS SP8). The excitation sources were an enhanced mercury arc lamp with a bandpass filter (λ = 340-380 nm) and a Ti:sapphire laser (Chameleon Vision) producing 140-fs pulses at a repetition rate of 80 MHz for the steady-state PL images and two-photon spectrum, respectively. An excitation wavelength of 740 nm was used for the two-photon excitation of InGaN MQWs. The lamp and laser were focused through a dry objective (NA 0.8). For the PL images in Fig. 1 and Extended Data Fig. 3, we applied a long-pass filter (λ > 425 nm) to acquire the blue and yellow overlapped image.
To avoid the PL saturation for the sol-gel SiO 2 nanorods, in Fig. 1d, we applied an attenuator to reduce the light intensity by a quarter. The PL decay traces and fluorescence lifetime images were obtained using another modified confocal microscope (Leica TCS SP8) with PicoHarp 300 (PicoQuant) by using the time-correlated single-photon counting technique. They were obtained by the one-photon excitation of a 405-nm diode laser at a repetition rate of 625 kHz. The objective lens was the same as that used in the steady-state PL measurements. All PL and EL measurements were performed at room temperature between 20 and 25 °C in a dark room. The PL and EL spectra for the plasma-enhanced ALD SiO 2 passivation and sol-gel SiO 2 passivation were compared in Extended Data Fig. 10. The red-shifted EL spectrum compared with the PL spectrum can be attributed to the influence of surface defects and different quantum confinement process between photoexcitation and electrical excitation. The EL spectrum broadening in the plasma SiO 2 -passivated device can be explained as the result of the increased dispersion owing to surface defects. The plasma SiO 2 passivation generates more surface defects compared with the sol-gel SiO 2 passivation or KOH wet etching ( Fig. 3 and Extended Data Fig. 11).

Quantification of defect concentrations using ESR
ESR measurements were performed using an X-band Bruker ELEXSYS-II E500 spectrometer. A helium gas flow cryostat was used at an operating temperature of 6 K. The GaN nanorod powder samples were placed in a quartz sample tube with a diameter of 4 mm and held in a microwave cavity. The experimental ESR settings were as follows: microwave power 0.002 mW; sweep width 1,000 G; modulation frequency 100 kHz; modulation amplitude 5 G; sweep time 200 s; number of points 2,048; number of scans 8. The spin density was calculated using the SpinCount/SpinFit software.

Surface-state analysis using XPS
The XPS data were obtained using an ESCALAB 250Xi electron spectrometer (Thermo Fisher Scientific) with a monochromatic Al-Kα X-ray (1,486.7 eV) source and a spot size of 400 μm. To obtain the Ga 3d and N 1s core-level spectra of the nanorod sidewall, the SiO 2 insulating layer was intentionally deposited with a thickness of only 2 nm using plasma-enhanced ALD and sol-gel synthesis. Then, the nanorods were separated from the wafer and dispersed on the Si substrate. Atomic concentrations were determined from the survey spectra (0-1,350 eV, 10 eV detector pass energy, 1 eV step size, 50 ms dwell time, two scans). Furthermore, high-resolution spectra were acquired (20 eV detector pass energy, 0.1 eV step size, 50 ms dwell time, ten scans) and the Thermo Fisher Scientific Avantage software was used for peak fitting; the Smart background option was used. The spectrometer was calibrated using the C 1s peak as a reference point (284.8 eV), which is in excellent agreement with the average values provided by the National Institute of Standards and Technology (NIST).

Trap-density analysis using DLTS
All the DLTS measurements in this study were conducted using a DLS-1100 (SEMILAB) system in the temperature range 80-600 K. The chips in which approximately 10,000 nanorods were p-n aligned in an area of 500 μm 2 were loaded in the DLTS chamber at a vacuum of 1 × 10 −2 torr. The measurement conditions were a temperature scan of 100 mK s −1 and a pulse voltage of 2 V to facilitate the formation of the depletion region in the nanorod device.

Measurement of electrical properties and EQE of nLEDs and LEE simulation
The current density-voltage ( J-V) curves for a single pixel in test element group (TEG) cells were recorded using a Keithley 2400 source measurement unit (Keithley Instruments). The ideality factor of the diode characteristics was calculated from the intermediate range of the J-V curves as shown in Extended Data Fig. 12. The EQE was measured for the TEG cell using a CAS 140CT Array Spectroradiometer (Konica Minolta) with an integrating sphere (ISP 40-101), which was calibrated with a standard light source (Extended Data Fig. 4c,d). The optical field distribution in Extended Data Fig. 4b was calculated for the pixel structure in Extended Data Fig. 4a with the simulation area of 30 × 30 × 5 μm 3 using an FDTD method (Ansys Lumerical). Isotropic dipoles from MQWs were used as a light source and the total amount of light emitted out of the pixel is defined as the LEE. The IQE was obtained by dividing the EQE by the LEE.

Observation of surface atomic structure using transmission electron microscopy
We used an aberration-corrected STEM system (Themis Z G3, Thermo Fisher Scientific) with an accelerated electron beam of 300 kV to observe the atomic structure of the sidewall region. The nanorods were directly dispersed on lacey carbon grids to prevent damage to Ga ions during focused ion beam (FIB) milling of the specimen. Furthermore, the insulating layer was intentionally deposited with a thickness of only 2 nm for electron transparency. To observe the interface between the nanorods and the SiO 2 layer, we performed STEM-EELS ( JEM-ARM200F, JEOL) measurement on an FIB specimen with 60-nm-thick SiO 2 -deposited nanorods. Further, we cleaned the specimen using an Ar-ion beam to remove the areas that sustained Ga-ion damage during FIB.
Electron energy-loss spectroscopic analysis EELS calculations were performed on the basis of the DFT implemented in VASP 50 . The projector augmented-wave method was used to mimic the ionic functional 51 , whereas the Perdew-Burke-Ernzerhof approximation was used for the electron exchange-correlation functional 52 . Plane waves were included up to an energy cut-off of 400 eV. We used a 9.0 × 11 × 11-Å supercell to consider defect complexes. The Brillouin zone for the supercell calculation was sampled using a 2 × 2 × 2 k-point grid. Ionic and electronic relaxations were performed by applying a convergence criterion of 5 × 10 −2 eV Å −1 per ion and 10 −5 eV per electronic step. The supercell core-hole method was used to obtain the electron energy loss spectra of GaN (ref. 53 ). The core electron of the oxygen atom was removed from the core and added to the valence/conduction bands, resembling the final state of the excitation process (also referred to as final-state approximation). We used a constant Lorentzian broadening of 0.5 eV to replicate the experimental broadening.

Data availability
All data generated or analysed during this study are included in the paper.

Code availability
We used the commercially available software of Ansys Lumerical for the FDTD simulation and VASP for the DFT calculations. The simulation settings are presented in Methods.