Impact of Sidewall Conditions on Internal Quantum Efficiency and Light Extraction Efficiency of Micro‐LEDs

The sidewall condition is a key factor determining the performance of micro‐light emitting diodes (µLEDs). In this study, equilateral triangular III‐nitride blue µLEDs are prepared with exclusively m‐plane sidewall surfaces to confirm the impact of sidewall conditions. It is found that inductively coupled plasma‐reactive ion etching (ICP‐RIE) causes surface damages to the sidewall and results in rough surface morphology. As confirmed by time‐resolved photoluminescence (TRPL) and X‐ray photoemission spectroscopy (XPS), tetramethylammonium hydroxide (TMAH) eliminates the etching damage and flattens the sidewall surface. After ICP‐RIE, 100 µm2‐µLEDs yield higher external quantum efficiency (EQE) than 400 µm2‐µLEDs. However, after TMAH treatment, the peak EQE of 400 µm2‐µLEDs increases by ≈10% in the low current regime, whereas that of 100 µm2‐µLEDs slightly decreases by ≈3%. The EQE of the 100 µm2‐µLEDs decreases after TMAH treatment although the internal quantum efficiency (IQE) increases. Further, the IQE of the 100 µm2‐µLEDs before and after TMAH treatment is insignificant at temperatures below 150 K, above which it becomes considerable. Based on PL, XPS, scanning transmission electron microscopy, and scanning electron microscopy results, mechanisms for the size dependence of the EQE of µLEDs are explained in terms of non‐radiative recombination rate and light extraction.

not significant. This behavior changes when the LED size gets smaller. For example, a sample with 2.5 µm ITO and mesa (i.e., no extra spacing) can have ≈5000 PPI with a 2.5 µm pixel distance. However, with 2.5 µm ITO and 8.5 µm mesa (i.e., 3 µm spacer) would only have ≈2300 PPI with a 2.5 µm pixel distance. Therefore, for ultra-high PPI displays, both p-contact and mesa should have an identical size. Thus, the performance of µLEDs having identical size of p-contact and mesa need to be studied.
It is well known that an ICP-RIE causes µLEDs to suffer from the sidewall damage, which reduces internal quantum efficiency (IQE) with shrinking chip size because a surface recombination velocity is proportional to the ratio of peripheral length to area. [14][15] Since external quantum efficiency (EQE) is defined as the product of IQE and light extraction efficiency (η e ), it is important to enhance both IQE and η e to achieve high EQE. [16] More importantly, since η e is an independent parameter of sidewall damage and increases with decreasing size, [17][18] it is necessary to accurately analyze which of IQE or η e has a more decisive effect on EQE. In this study, we analyzed the EQE of blue µLEDs having the same p-contact and mesa size by comparing blue µLEDs with areas of 100 µm 2 and 400 µm 2 . To clearly understand the effect of TMAH etching, we designed equilateral triangle µLEDs structures with m-planes and confirmed that TMAH not only removed the lattice distortion of the sidewall, but also changed the sidewall morphology, resulting in changes in IQE and η e simultaneously.

Design of µLEDs
To remove the sidewall defects induced by ICP-RIE, the sidewalls were treated with tetramethylammonium hydroxide (TMAH), which etches non-polar GaN. [19][20] Thus, the sidewall morphology could change after treatment and the performance of TMAH-treated µLEDs, such as η e , may be affected by the crystal plane orientations. To accurately demonstrate the effect of TMAH, while excluding the effect of the morphology of different orientations, we designed an equilateral triangle structure that has only m-plane orientation sidewall as illustrated in Figure 2a. This is possible, because of the hexagonal structure or (0001) oriented GaN. Figure 2b displays designed µLEDs with areas of 100 and 400 µm 2 (referred here to 100 µm 2 µLEDs and 400 µm 2 µLEDs, respectively), peripheral lengths of 45.6 µm and 91.2 µm, and identically sized ITO electrodes and mesas. A crosssectional high angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image showing the structure is displayed in Figure 2c. As shown in Figure 2d,e, although ICP-RIE slightly etched the ITO layer near the sidewalls, the ITO p-electrode clearly extended to the width of the mesa, as confirmed by the In and Sn Energy-dispersive X-ray spectroscopy (EDS) mapping. The peak wavelength was ≈476 and 470 nm at 5 and 25 A cm −2 , respectively indicating the blue µLEDs.

Influence of the TMAH on IQE and Non-Radiative Recombination
Since µLEDs are fully covered by an Al layer as shown in Figure 2b, most of the light is emitted toward the bottom where the light output (L) was measured by a silicon photodiode (PD) at room temperature. The measured L by PD gives the relative EQE as a function of injected current (I).
(1)    www.advopticalmat.de as a function of current density. For the 400 µm 2 µLEDs, TMAH etching increases the peak EQE by ≈10% in the low current regime, similar to our previous report, [21] whereas for the 100 µm 2 µLEDs, TMAH etching slightly decreases EQE by ≈3%. EQE was reported to decrease with shrinking µLEDs size and attributed to increased Shockley-Read-Hall (SRH) non-radiative recombination rate at the sidewall. [14,22] However, Figure 3a shows that before TMAH etching, the 100 µm 2 µLEDs yield higher EQE than the 400 µm 2 µLEDs, which is apparently different from the previous results. [22] To explain this discrepancy, we first discuss the change in IQE of 100 µm 2 µLEDs. Temperature-dependent photoluminescence (TDPL) was measured from 20 to 300 K with resonant optical excitation (405 nm continuous-wave laser) at 5 mW of excitation power and the integrate intensity was normalized to 1 at 20 K. For the sample structure for PL analysis, only ICP-RIE was carried out without a passivation layer, p-pad, and n-pad. Figure 3b shows the normalized integrated PL intensity (i.e., relative IQE [23][24][25] ) for the 100 µm 2 µLEDs. The difference between before and after TMAH etching is not significant at low temperatures, meaning that the sidewall surface recombination is not active because the carriers are localized too much to move large distances. However, the effect of TMAH becomes meaningful above 150 K.
In order to quantify the non-radiative recombination, we measured time-resolved photoluminescence (TRPL). Figure 3c shows the decay curves of the PL signal of 100 µm 2 mesa structures before and after TMAH treatment. The TRPL can be fitted by two exponentials with the decay parameters τ fast and τ slow , a common technique for AlGaN QW [26] and InGaN QWs. [27] where A fast and A slow are fitting constant. The non-radiative lifetime (τ nr ) correlated to SRH non-radiative recombination rate is calculated following Equation (3) [26][27] nr 1 fast τ nr was estimated to be 4.62 and 6.73 ns before and after TMAH, respectively. The longer lifetime demonstrates that the TMAH treatment reduces the non-radiative recombination. Since the total lifetime at room temperature is mainly dominated by the non-radiative lifetime [28][29] and the non-radiative recombination rate is the inverse of the lifetime, one could try to quantify the sidewall recombination [15,21] : where A is the non-radiative recombination rate, A 0 is the nonradiative recombination rate in the QW, A s is the surface recombination rate, λ is the carrier diffusion length, l is the peripheral length of device, and S is the area of the device. A s depends on the sidewall surface and should be higher without TMAH treatment. However, above 150 K, IQEs are different before and after TMAH treatment. Thus, λ becomes sufficiently long to catch bigger numbers of carriers (especially holes) only above 150 K. From normalization at 20 K, the IQEs is estimated to be 0.36 and 0.49 before and after TMAH at 5 mW, respectively (0.32 and 0.37 before and after TMAH at 1 mW, respectively, Figure S1, Supporting Information). This implies that TMAH considerably reduces sidewall recombination by eliminating the sidewall damage, resulting in a larger IQE after TMAH. Indeed, Finot et al. [30] systemically investigating carrier lifetime characteristics cathodoluminescence, showed that the carrier lifetime near the sidewall could be enhanced with better sidewall conditions. This result is consistent with our TRPL and consequently the fact that better sidewall conditions improve the IQE of µLEDs. In the ABC model, the IQE is calculated from three different recombination parameters and carrier density given by: [31][32] Bn An Bn Cn IQE n is carrier density, B is radiative recombination rate, and C is Auger recombination rate. As mentioned above, reduced sidewall recombination related to both A s and λ will increase IQE. However, it should be noted that other recombination parameters, such as B, might change depending on sidewall conditions. For example, B is different for damaged (i.e., sidewall damage) and undamaged areas. Chen et al. [33] showed that the TRPL lifetime at 15 K, where the radiative recombination is mainly dominant, decreased with better sidewall conditions. Therefore, if the sidewall condition is improved through the TMAH treatment (namely, by removing the damaged area), the proportion of the undamaged area relatively increases, and thereby B may increase. Finally, as illustrated in Figure 3d, the ICP-RIE induced surface shows a broad yellow luminescence (YL) band near 550 nm that is mostly related to Ga vacancy, impurity, and/or vacancy-impurity complexes. [34][35] The YL band disappears after TMAH treatment, i.e., the ICP-RIE-induced surface damage is effectively etched by the TMAH treatment. Similar result of suppressing YL band by sol-gel SiO 2 was reported elsewhere. [16]

Effect of TMAH on m-Plane GaN Surface States
The most important thing in the performance of µLEDs is the sidewall condition. [36][37][38][39][40][41][42] Except for special cases, such as semipolar and nonpolar GaN LEDs, most GaN-based LED wafers are grown on c-plane sapphire, and hence the sidewall should have a nonpolar surface such as m-plane orientation ( Figure 2). X-ray photoelectron spectroscopy (XPS) analysis is suitable for confirming the surface states. However, since for our samples, the m-plane sidewall surface is perpendicular to c-plane GaN, it is difficult to analyze vertical sidewall surfaces with XPS. Therefore, we deliberately etched the planar m-plane GaN wafer using ICP-RIE to study the worst-case effect (i.e., vertical etching damage) in detail using XPS analysis. We prepared four samples: reference (pristine), TMAH-treated, after ICP-RIE, and after ICP-RIE/TMAH treatment. The surface roughness of all samples was measured by atomic force microscopy (AFM), as shown in Figure 4a. Scratch lines from mechanical polishing of m-GaN substrates of bulk GaN crystals [43] were www.advopticalmat.de visible in AFM images (as marked by the arrows in Figure 4a and Figure S1, Supporting Information). All samples reveal slightly rough surfaces with RMS roughness in the range of 3.3-6.6 nm. It is noted that regardless of ICP-RIE, TMAH similarly etches m-plane GaN surfaces, resulting in similar surface features. (For all samples, information including depth, root mean square, and surface smoothness is available in Figure S1, Supporting Information). Figure 4b shows that XPS spectra of Ga 3d for each sample, which shows the contribution from the N 2s core level and Ga-Ga, Ga-N, and Ga-O bonds. The Ga-Ga intensity is weak, implying no metallic droplets. The N 2s core level is broad and remains unchanged as it mainly comes from the area way from the surface. However, only the ICP-RIE surface shows lower Ga-N and higher Ga-O intensities than the other three samples. This behavior can be easily seen in Figure 4c. For example, the ICP-RIE surface exhibits Ga-N ratio of 51.45% and Ga-O of 44.67%. This indicates that the ICP-RIE increases the oxidized surface area and the formation of vacancies-related defects. [16,44] Furthermore, as shown in Figure 4d, the O 1s core level intensity is almost doubled after ICP-RIE but recovers after TMAH. Finally, only the ICP-RIE surface shows a strong signal of Cl 2p (Figure 4e), which is caused by Cl implantation in the Cl 2 etching gas. The Cl 2p signal disappears after TMAH treatment. Consequently, this shows that TMAH completely removes the ICP-RIE-damaged surface of m-polar GaN. Bartos et al., [44] investigating the surface band bending behaviors of polar, semi-polar, and nonpolar GaN after several surface treatments by means of XPS, reported an upward surface band bending behavior when the surface states increased due to disorder. Hence, the XPS and AFM results suggest that the surfaces after ICP-RIE could show an upward band bending due to a high density of surface states, resulting in preferential accumulation of holes.

Effect of Sidewall Microstructure on IQE and η e
We have presented evidence that the sidewall non-radiative recombination is due to carriers that diffuse to the sidewall and then recombine there. As further evidence, we observed the sidewall morphology using scanning electron microscopy (SEM) and STEM.

www.advopticalmat.de
ICP-RIE-treated m-plane sidewalls before and after TMAH etching. Before the TMAH treatment, the sidewall morphology is very rough because of the ICP-RIE process after deposition of ITO on the epi wafer. (The effect of ITO on the sidewall morphology during ICP-RIE process is discussed in Figure S2, Supporting Information). It is noted that the rough sidewall is flattened by the TMAH treatment (Figure 5b). Similar results were recently reported by Liu et al. [45] Figure 5c,d shows HAADF-STEM image. After ICP-RIE, there is a blur toward the sidewall (yellow arrow), i.e., the lattice distortion. High resolution bright field-STEM image evidently shows that the distortions of the atomic arrangement (Figure 5e) (as marked by the white dots) is more severe closer to the sidewalls and extends around ten lattice constants from the surface. However, after TMAH, the arrangement of atoms near the sidewall is less distorted (Figure 5f), indicating that TMAH etches the distorted regions induced by ICP-RIE. (Focused ion beamed (FIB) regions for STEM are shown in the Figure S3, Supporting Information). The HAADF-STEM images observed at >300 nm away from the sidewall are not significantly different before and after TMAH (Figure 5g,h, respectively). Thus, the ICP-RIE etching only damages the sidewall surface, confirming that the surface recombination is predominantly determined by surface damages (Figure 5c,e). The mechanism of surface recombination via carrier traps is sketched in Figure 5i. Olivier et al. [14] reported a sharp increase of A as µLEDs size decreased. Jiang et al. [46] Adv. Optical Mater. 2023, 11, 2203128 www.advopticalmat.de found from simulations that the surface band bending caused holes to diffuse several µm to non-radiatively recombine at the sidewall surface, depending on the density of surface states and the current density. Moreover, Yamada et al. [47] reported that the lattice distortions induced by ICP-RIE caused an introduction of gap states, degrading the performance of III-nitride devices. Thus, as shown in Figure 5e, the observed lattice distortion caused by ICP-RIE increases the density of surface states, inducing band bending. Their effect on the recombination can be described as follows. First, it increases the recombination rate (A s ). Then, the increased number of surface states also can accumulate more carriers and these carriers recombine nonradiatively. Both lattice distortion and surface states cause more holes to migrate to the trap sites and recombine non-radiatively. The TMAH treatment removes the lattice distorted regions and sharpens the sidewall (Figures 4 and 5d), thus alleviating surface states (reducing A s ) and reducing surface band bending as illustrated in Figure 5i. All these effects increase IQE via Equations (4) and (5) that is consistent with the temperature dependent PL data ( Figure. 3b).
However, the effect of Fermi level pinning could not be negligible. Lymperakis et al. [48] showed numerically that even an ideal (1010) GaN surface had a surface Fermi level intrinsically at E c = −(0.6 ± 0.2) eV (upward band bending). Hence, even if all oxide bonds are removed, some surface states could exist. Thus, the surface recombination occurs non-radiatively at the sidewalls even when the sidewall morphology is fully recovered. Consequently, as the ratio of peripheral length to area increases, the IQE would decrease due to an intrinsic surface state. However, if the number of surface states is small, only small number of carriers would reach the surface. Thus, surface recombination is unavoidable, but a sufficiently low density of surface states means that surface recombination (and thus sidewall damage) is insignificant at typically operating carrier densities. This also means that passivation techniques should be chosen according to the distribution of surface states.
It was shown that the TMAH treatment improved the IQE of 100 µm 2 -µLEDs (Figure 3b), but rather reduced the EQE (Figure 3a). This can be understood if TMAH affects light extraction η e , because EQE is proportional to both η e and IQE. [31][32] e EQE IQE η = × As stated above, surface recombination is not significant at high carrier densities. Recalling the denominator of Equation (5), it can be deduced that nonradiative recombination current is negligibly small when the carrier density is too high (i.e., high current density). [49] This makes us take into account that nonradiative sidewall surface recombination insignificantly affects EQE in the high current regime, resulting in EQE that is dominantly proportional to η e at the same n. The inset image of Figure 3a exhibits that the EQE at high current density is in the droop regime. The effect is most pronounced for the 400 µm 2 -µLEDs. The EQE behavior before and after TMAH is different at low and high current densities. In other words, at low current densities, the peak EQE after TMAH treatment is higher than before TMAH treatment, but it is the other way around at the high current densities. The increase at low current results from the increased IQE (as confirmed by PL, Figure 3b), whereas the increase at high current before TMAH treatment results from higher η e . Thus, the change of the sidewall morphology not only reduces non-radiative recombination but also decreases η e . As shown earlier, TMAH removes sidewall surface lattice distortions induced by ICP-RIE. However, the smoothening of the sidewall morphology by TMAH (Figure 5b) rather reduces η e , and results in a lower EQE at the high current regime.
According to the Snell's law, the light escape cone (θ) can be defined by Equation (7): where n in is the refractive index of the material from where light is generated and n out is the refractive index of the one from which light is extracted. A rougher surface has a wide distribution of surface angles, and thus Equation (7) is almost always fulfilled, and strongly increases η e , as displayed in Figure 5j. Fujii et al. [50] reported that the roughened surface of LEDs could improve η e , which is in good agreement with our result. This implies that the sidewall roughening causes better light extraction via sidewall emission. Furthermore, Gou et al. [51] suggested that the contribution of sidewall emission to the overall emissions became significant as µLEDs size decreased. Ley et al. [52] also demonstrated that the EQE increased with shrinking size beyond a critical value due to the increase of η e . These findings are consistent with our EQE data. Thus, the emission of the 100 µm 2 -µLEDs is strongly influenced by light extraction than that of the 400 µm 2 -µLEDs. [51][52][53][54] Taken together, the EQE of the 100 µm 2 -µLEDs decreases after TMAH treatment even though the IQE increases because the sidewall smoothening reduces the light extraction.

Conclusion
We systemically investigated the influence of sidewall conditions on the EQE of specially designed equilateral triangular µLEDs. The TMAH treatment increased the IQE of µLEDs by eliminating lattice distorted regions of the sidewalls. Non-radiative recombination at the sidewall became active only above around 150 K, pointing to long distance carrier diffusion. The XPS results showed a large number of Ga-O bonds and even Cl after ICP-RIE, which was removed by TMAH. Further, the STEM results revealed lattice distorted regions of a few nm at the sidewalls. No damage was found at the QWs distance away from the sidewall surface. Thus, it is suggested that the size dependent increased non-radiative recombination rate of µLEDs is caused by a large number of surface defects induced by ICP-RIE. While TMAH etching removed most surface defects, the resultant TMAH etching reduced η e and even reduced EQE. Therefore, it is important to strategically choose which of IQE and η e will effectively enhance EQE depending on the chip size.

Experimental Section
Device Fabrication Process: To define the structure of µLEDs, first a 100 nm-thick ITO layer as a p-electrode was deposited on the epi wafer www.advopticalmat.de by sputtering. The ITO was then annealed at 600 °C in N 2 for 5 min. The mesa was etched with the ITO on top of the epi-structure by ICP-RIE (RIE-200iPN-2, SAMCO) at ICP power of 150 W, bias power of 15 W, Cl 2 gas (30 sccm), and total pressure of 2.0 Pa. For TMAH treatment, the µLEDs were etched by 25 wt.% TMAH at 80 °C for 10 min. For n-electrode metal, Cr/Au (30/160 nm) was deposited on the n-GaN layer. To form the p-electrode, the samples were covered by SiO 2 layer with plasma enhanced chemical vapor deposition. Then a hole was opened by a CF 4 -based RIE system. Finally, Al/Ni/Au (400/20/160 nm) was deposited. The Al layer served as a reflection layer to enhance downward light extraction. The sample fabrication processes are similar to those presented in the previous work. [21] To investigate the effect of ICP-RIE on the surface states of the (10-10) GaN surface, m-plane GaN substrate (purchased from Mitsubishi chemical) was etched by ICP-RIE for 5 min under conditions similar to those described above.
Measurement and Characterization: EQEs were measured with a large area silicon photodiode (PD, Hamamatsu S1337-1010BQ) placed under the sample. The current was stepped from high to low in DC mode at room temperature. The distance between samples and PD was same, which allows for relative comparison. The sidewall morphology was observed by SEM (Hitachi SU-9000). The specimens for STEM analysis were thinned to ≈50 nm or less by a focus ion beam technique (FEI Helios 660). STEM images were captured with a Hitachi HD2700 with an accelerating voltage of 200 kV. XPS data were measured by ESCALAB250 with Al Kα (1486.6 eV) X-ray excitation. Ga 3d spectra were obtained at a pass energy of 20 eV with a step size of 0.025 eV, integrating 10 scans with a dwell time of 50 ms. The step size for the Cl 2p and O 1s was core levels was 0.1 eV.
All PL measurements were performed in a closed-cycle cryostat (Advanced Research System Co., DMX-1AL) to maintain and control the operating temperature from 20 to 300 K. A thermal grease was used to attach the samples to cryostat. Excitation was performed by a 405 nm continuous-wave laser (Coherent Inc., CUBE 405-100C) where the spot size was ≈35 µm of radius. For time resolved PL measurement, the sample was excited by using the Hamamatsu PLP-10 pulse laser at 405 nm at room temperature, which has an operation pulse width of less than 100 ps and a repetition rate up to 100 MHz. The signal was measured by using a Hamamatsu C4780 Picosecond fluorescence lifetime measurement system with a temporal resolution of 5 ps.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.