Impact of etching conditions on the sidewall quality of InGaN/GaN micro-LEDs investigated by cathodoluminescence imaging Open Access

Blue light-emitting diodes (LEDs) based on the InGaN/GaN material system are known for their extremely high efficiency, achieving values of over 90% for internal quantum efficiency.^1–3 Research is currently being carried out to miniaturize them for applications in high-resolution displays, but many other applications,⁴ such as optical neuromorphic computing,⁵ are also of interest. However, there are difficulties in maintaining the efficiency when the LEDs are miniaturized into micro-LEDs, as demonstrated in various publications.^6–13 The main reason for this is that the surface ultimately limits the device performance, first because the surface acts as a strong perturbation to the ideal crystal, resulting in naturally occurring surface states that can mediate non-radiative recombination, and second because the sidewall is typically defined by inductively coupled plasma reactive ion etching (ICP-RIE) of planar LEDs, creating a damaged surface layer due to high-energy ion bombardment. As a consequence, several techniques have been developed to reduce the non-radiative recombination activity at the sidewall. This includes surface passivation with a dielectric such as Al₂O₃^14,15 or SiO₂^7,12 to saturate the dangling bonds at the sidewall and thereby reduce the density of the surface states, or removal of the damaged surface by wet chemical etching, e.g., with a KOH solution.^7,16 However, in some cases, it is not preferable to use one of these surface recovery techniques as they introduce another step in the processing chain and some techniques also change the surface morphology and dimensions, e.g., wet chemical etching. Hence, it is advisable to optimize the ICP etching process so that no further surface processing is necessary, which requires a profound understanding of the impact of etching parameters on the surfaced quality in terms of quantum efficiency. In this publication, this challenge is addressed by comparing the cathodoluminescence (CL) behavior of micro-LEDs from the same wafer, but subjected to stronger physical etching conditions or stronger chemical etching conditions, keeping all other process parameters as comparable as possible. This was achieved by changing the plasma power and the Ar/Cl₂ ratio, since a higher power and a shift of the Ar/Cl₂ ratio from reactive Cl toward inert Ar lead to a stronger contribution of physical etching via ion bombardment, in other words sputtering. The micro-LEDs are then compared based on monochromatic cathodoluminescence (monoCL) images, which reveal differences in CL intensity profiles, and time-resolved cathodoluminescence (TR-CL) experiments, which provide spatially resolved information about the charge carrier decay of the LEDs so that the local CL lifetime near the sidewall can be investigated.

The focus of this study is on blue InGaN/GaN LEDs grown by metalorganic vapor-phase epitaxy on a c-oriented sapphire substrate inside a 3 × 2 in. close-coupled showerhead reactor (Thomas Swan). The LED stack consists of an approx. 5.7 μm thick unintentionally doped GaN (u-GaN) buffer and n-GaN layer, a 25-fold InGaN/GaN superlattice, and an InGaN/GaN double quantum well (QW), which is composed of 2.7 nm thick quantum wells and a 12.0 nm thick quantum barrier. Due to local variations in the growth conditions, the emission wavelength at low excitation is 460 ± 15 nm depending on the wafer position, which results in 450 nm emission under electrical injection. The stack is terminated by a 100 nm thick u-GaN capping layer on top of the quantum wells instead of an electron blocking layer and a p-GaN layer, so that the structure consists of a fully functional LED stack except for the missing p-doping. The p-doping was omitted in order to keep the structure and, thus, the interpretation as concise as possible, since the separation of the generated charge carriers by the built-in field of the pn-junction is prevented, which would be an additional factor to consider when interpreting the data.

The LEDs were structured according to the schematic in Fig. 1 in order to obtain wedge-shaped triangular mesas with a large sidewall-area-to-volume ratio. First, a SiO_x layer was deposited on the LED wafer by plasma-enhanced chemical vapor deposition (PECVD) (Oxford Instruments, PlasmaPro 80). A photoresist was then spun on and lithographically patterned so that the SiO_x layer could subsequently be structured by careful over-etching with buffered 12.5% hydrofluoric (HF) acid to form a hard mask. Due to the isotropic etching behavior, the size of the SiO_x mask is reduced more or less uniformly from all sides, making the tips sharper but not affecting the angles of the triangular structure, resulting in small tips often less than 200 nm in width, which would be impossible to achieve via standard i-line contact photolithography. Afterwards, the wafer was cut in half. Each half was etched under different conditions in an Ar/Cl₂ ICP (Oxford Instruments, PlasmaPro 100 Cobra), leading to the investigated triangular mesas. The etching parameters were varied by changing the table RF power and the ratio of Ar to Cl₂ to steer the nature of the etching process, as a higher plasma power and an increase in the Ar/Cl₂ ratio should favor physical etching at the expense of chemical etching. Hence, the micro-LEDs labeled “stronger chemically etched” were etched with a table RF power of 200 W and a ratio of 5 SCCM Ar/50 SCCM Cl₂, whereas the micro-LEDs labeled “stronger physically etched” were etched with a table RF power of 500 W and a ratio of 5 SCCM Ar/5 SCCM Cl₂. In both cases, the pressure was 30 mTorr and the ICP power was 1000 W. The resulting etching depth was approximately 0.65 μm for the stronger physically etched LEDs and approximately 1.1 μm for the stronger chemically etched LEDs, so that in both cases the etching terminated at the n-GaN layer and far away from the quantum wells. The etched wafer halves were then diced into chips and the SiO_x hard mask was removed via HF etching with a buffered solution for 10 min. As part of the investigation, some chips were either annealed or hydrogenated. Annealing was carried out for 5 min at temperatures between 300 and 500 °C in a rapid thermal annealing system (AST, SHS 100) under nitrogen atmosphere, while hydrogenation was performed in the above-mentioned PECVD system for 5 min at 40 °C with a table RF power of 200 W and an ICP power of 1000 W, an H₂ gas flow of 50 SCCM, and a pressure of 30 mTorr.

One part of the measurements was carried out using a field emission scanning electron microscope (SEM) (TESCAN, Mira 3 GMH) equipped with a CL system (Gatan, MonoCL4). The CL system collects the CL signal via a parabolic mirror and passes it through a Czerny-Turner monochromator so that it can either be spectrally resolved with a CCD camera (Andor Technology, iDus DU420A-BU2) or spectrally selected to form a monoCL image. In the latter case, the intensity for each scanning point is counted by a photo-multiplier tube. An acceleration voltage of 10 kV and a beam current of either 146 or 259 pA were used during the measurements. To ensure comparable results during a series of monoCL images, parameters that influence the CL intensity, such as the distance between the parabolic mirror and the sample, the beam current, or the gain voltage of the photo-multiplier tube, were kept as constant as possible. Additionally, another field emission SEM (Thermo Fisher, Quattro S) with a different CL system (Delmic, SPARC) was used for TR-CL measurements. This CL system employs a streak camera (Hamamatsu, Streakscope C10627) to record the decay traces and is synchronized with a picosecond beam blanking unit that controls the generation of the electron beam pulses. The electro-optical lenses of the SEM are optimized to achieve conjugate blanking, which is necessary for the electron beam to shift by less than 50 nm during pulse generation. The selected lens parameters together with the acceleration voltage of 10 kV result in a beam current of approximately 20 pA, measured via a Faraday cup for pulsed operation at a repetition rate of 5 MHz. The resulting streak maps are automatically analyzed by spectral integration over the wavelength range where the peak intensity falls to 10% of its maximum, resulting in the decay trace.

Figure 2 shows exemplary secondary electron (SE) and monoCL images from a bird's eye view of the tip area of one of the wedges, which was either (a) stronger chemically etched or (b) stronger physically etched. While the micro-LEDs hardly differ in shape, the stronger chemically etched LED has a comparably homogeneous CL intensity across the entire surface and is slightly brighter near the sidewall, whereas the CL intensity of the stronger physically etched LED is significantly reduced near the sidewall of the LED. In other words, the stronger chemically etched LED behaves similarly to an unetched planar LED layer in terms of CL behavior, unlike the other LED, which is strongly influenced by the sidewall. Similar behavior can also be observed for circular LEDs, so that the observation is not limited to the selected geometry. The commonly accepted explanation for the reduction in CL intensity near the sidewall is that etching causes plasma damage to the sidewall, increasing the number of non-radiative recombination centers and thus decreasing the CL intensity due to non-radiative surface recombination. While this explanation applies to the stronger physically etched LED depicted in Fig. 2(b), it does not adequately explain why the CL intensity near the sidewall is not markedly reduced for the stronger chemically etched LED, as a reduction in quantum efficiency would nevertheless be expected.

A possible explanation for this behavior is that the reduction in radiative recombination events is compensated or even overcompensated by enhanced light extraction near the sidewall,¹⁵ effectively keeping the collected CL intensity at a constant or even higher level. The reason for this is that the light is more likely to be reflected or scattered directly at the sidewall, resulting in a higher proportion of the light leaving the LED without being trapped by total internal reflection. Therefore, a simple analysis of intensity is not suitable for drawing final conclusions. To investigate this further, TR-CL experiments were performed, since the charge carrier lifetime is affected by changes in the local quantum efficiency, but not by changes in the light extraction efficiency. When exciting the stronger physically etched or the stronger chemically etched LEDs at their center with a distance of 5 μm from each sidewall, the decay traces are quite similar [see Figs. 3(a) and 3(b)]. This is because the decay behavior can be considered the bulk material response, which should be comparable since both LEDs are composed of the same wafer. Differences arise when the LEDs are excited with a distance of 150 nm from each sidewall, as can be seen in Figs. 3(c) and 3(d). For the stronger physically etched LED, the corresponding CL decay curves for different total irradiation times in Fig. 3(c) coincide with the excitation pulse. This can be attributed to the fact that the CL signal decays too fast to resolve the CL lifetime, which is reciprocal to the decay rate and is mainly determined by the faster contribution of the radiative and non-radiative recombination lifetime. Since polar InGaN/GaN LEDs with a large QW thickness typically exhibit a radiative lifetime of significantly more than one nanosecond at room temperature,^17,18 it is likely that the CL lifetime is strongly dominated by the much faster non-radiative recombination processes. In contrast, the CL signal decays much slower for the stronger chemically etched LED under the same excitation conditions, as illustrated in Fig. 3(d), suggesting that non-radiative recombination contributes less to the total recombination. More precisely, when the LED is excited close to the sidewall, the CL lifetime is only slightly reduced compared to that at a much larger distance to the sidewall, as can be inferred from the comparison of the CL decay [cf. Figs. 3(b) and 3(d)]. It is reasonable to assume that this small reduction in CL lifetime at a distance of 150 nm compared to 5 μm is primarily due to faster non-radiative recombination in the near-surface region since quantum well conditions do not change. Simultaneously, the measured CL intensity is increased by approximately 15% at 150 nm distance to the sidewall. It is generally known that planar LEDs exhibit spatial variations in the CL intensity, which are caused among other things by threading dislocations and fluctuations in QW composition and thickness. However, the tendency is too large to be explained by these local variations alone, suggesting that the high CL emission at the tip of the wedge, as shown in Fig. 2(a), is likely caused by enhanced light extraction in the presence of reduced quantum efficiency. It should be noted at this point that the intensity and CL lifetime close to the sidewall of the stronger chemically etched LEDs vary greatly from LED to LED, as both values are very susceptible to local surface damage. However, when the lifetime close or far away from the surface is nearly comparable, the intensity tends to be higher due to outcoupling effects. Overall, the observations are in good agreement with the results of Finot et al.,¹⁹ who performed line scans from the sidewall to the center of a rectangular LED mesa and observed different trends for CL intensity and CL lifetime close to the sidewall due to enhanced light extraction. Based on simulations, they also argued that strain relaxation occurs close to the sidewall, leading to a reduction of the quantum-confined Stark effect and, thus, to higher recombination rates. This effect, however, has only a small spatial extent and could therefore not explain their observed change in CL lifetime.¹⁹ Since the structural geometry is different in the present case, this effect could be an additional contribution to the observed reduction in CL lifetime.

All of the above considerations were based on measurements for a low total irradiation time with the electron beam. Considering this additional parameter, another intriguing difference between the LEDs with different etching parameters can be observed. In the case of the stronger physically etched LED, the maximum CL intensity and the general decay behavior remain qualitatively unchanged with increasing total irradiation time. The same is true for the stronger chemically etched LED when the excitation area is placed far away from the sidewall. Conversely, both properties change strongly with the total irradiation time when the excitation occurs close to the sidewall. More specifically, the maximum CL intensity and the CL lifetime decrease with the irradiation time until the decay curve matches the excitation pulse at an even longer total irradiation time than shown in Fig. 3(d), similar to the stronger physically etched LED. It is important to note that this low-energy electron beam irradiation (LEEBI) effect has a very limited spatial extent, as shown by the insets of Fig. 3(d), which serve as a comparison of the monoCL image of the same LED before (top image) and after (bottom image) 2 min of irradiation. The CL intensity decreases where the electron beam was placed during irradiation, symbolized by a green square, and also slightly further in the surroundings, indicating that the number of non-radiative recombination centers is increased by the electron beam in the irradiated area and that some of the charge carriers injected in the near vicinity are also affected by diffusion toward the irradiated region. Since the LEEBI effect only occurs when the electron beam is placed close to the sidewall and not far away from it, there is strong evidence that it is closely related to the sidewall condition.

The most straightforward explanation for this effect is that the electron beam directly or indirectly generates non-radiative recombination centers, leading to a reduction in CL intensity and lifetime. For example, Endo et al. have shown that a high-energy electron beam can generate recombination centers that are related to nitrogen displacement.²⁰ However, such direct generation of point defects or Frenkel pairs by knock-on damage seems unrealistic, as the generation rate should only weakly depend on whether the sidewall or the bulk is excited. In addition, Suihkonen et al. stated that the Frenkel pair generation threshold primary electron energy is 150 and 500 keV for the N- and Ga-sublattice,²¹ which is much higher than the 10 keV used in these experiments. An indirect method to generate carbon-related point defects could be the binding of carbon atoms to nitrogen or gallium atoms at the surface as a side effect of carbon deposition, whereas it is known that such defects are closely related to the typical defect luminescence of GaN.^22,23 It is known that focusing the electron beam on a region of interest on the surface leads to the cracking of residual hydrocarbons contained in the vacuum near the focal point or already adsorbed at the sample surface during sample preparation and storage.²⁴ These cracked hydrocarbons diffuse in the gas phase before they either form a chemical bond with the atoms at the surface or polymerize with the hydrocarbons already bound to the surface, resulting in the growth of a carbon-rich layer that represents the commonly observed surface contamination. Since the reduction in CL intensity and lifetime is only observed during excitation near the sidewall, only the incorporation of carbon on the etched sidewall and not on the pristine top surface could be considered for the decrease in non-radiative lifetime.

To investigate the influence of hydrocarbons on the CL intensity and, more specifically, to check how the distance between the origin of the cracked hydrocarbons and the sidewall affects CL intensity, the sample was tilted by 30° and the electron beam was scanned over a nominally 500 × 500 nm² reduced area, varying the distance to the sidewall. The result is depicted in Fig. 4 in the form of a color-coded monoCL image, which shows the relative change of CL intensity before and after the experiment. Consequently, pixels with a blueish or reddish color indicate that the CL intensity is reduced or enhanced by the electron beam irradiation, while a violet hue indicates that the CL intensity is not changed by the experiment. Overall, the CL intensity remains mostly unchanged by the experiment, as demonstrated by the fact that the LED is predominantly violet in color, but a small decrease in CL intensity can be observed at the sidewall. For better illustration, only the pixels with a strong blue component are shown in the inset of Fig. 4, i.e., those for which the CL intensity was reduced. Based on the first two measurement points i and ii, where the electron beam was focused on the etched n-GaN, one could argue that incorporation of carbon via cracked hydrocarbons diffusing to the sidewall plays a role since the CL intensity is reduced at the sidewall, although there is obviously no direct interaction between the electron beam and the LED. However, no changes in CL intensity at the sidewall are observed for iii, vi, and vii, even though the distance to the sidewall is similar or even closer than for the previous two measurement locations. Moreover, a significant reduction is observed only when the electron beam is placed directly on the sidewall and close to the QW (cf. iv and v), indicating that a direct interaction of the electron beam with the LED sidewall, rather than the incorporation of carbon at the sidewall, is the main reason for triggering the LEEBI effect. In addition, carbon-related defects usually leave a fingerprint in the spectrum, which should therefore become stronger with increasing irradiation time. However, apart from the intensity of the QW emission, the spectrum remains almost unchanged. In this context, it is worth noting that the CL reduction that occurred at the sidewall for i and ii is probably caused by electrons backscattered from the n-GaN surface, which should preferentially exit toward the sidewall due to the tilt of the sample during the experiment.

An alternative approach to explain the LEEBI effect is that the electron beam does not generate non-radiative recombination centers directly, but activates initially passivated non-radiative recombination centers, which would effectively lead to the same result. Hydrogen seems to be a reasonable candidate as it is known that hydrogen can passivate dopants or point defects in GaN.^25–32 Furthermore, hydrogen is present as a gas impurity in the chamber during the ICP process and can also be supplied by the etched LED as it is incorporated into the nitride crystal in the cooldown phase after growth.^25,33 Lastly, it is also known that hydrogen bonds react sensitively to electron beam irradiation, which actually led to the first successful p-GaN layer via LEEBI presented by Amano et al.³⁴ They observed that a Mg-doped layer exhibited p-type conductivity after electron beam irradiation, which was later attributed to the breaking of Mg–H complexes by the electron beam. However, this results in relatively inhomogeneous doping profiles, as the reactivation of Mg can only take place within the interaction volume of the electron beam, which led to the search for alternatives and the discovery by Nakamura et al. that annealing the sample in a hydrogen-free atmosphere with temperatures above 400 °C also leads to the activation of the p-doping.²⁵ 

Applying these findings to the present case, it should be possible to reduce the CL intensity near the sidewall by annealing the LED at temperatures on the order of several 100 °C, thereby removing hydrogen that passivates point defects near the sidewall and therefore increasing the surface recombination velocity. Accordingly, Fig. 5 shows the SE and monoCL images of different stronger chemically etched LEDs before and after annealing at 400 °C [(a)–(c)] and at 500 °C [(d)–(f)] for 5 min under a nitrogen atmosphere. While the CL intensity near the sidewall is not affected by annealing at 300 °C (not shown here), changes can be observed when the sample is annealed at 400 or 500 °C, respectively. The CL intensity is slightly reduced at the edges after annealing at 400 °C, whereas the CL intensity reduction at 500 °C extends even further into the LED. This reduction in CL intensity is once again accompanied by a reduction of CL lifetime and cannot be attributed to a change in light extraction, as the topography is not altered by annealing, so it points toward higher non-radiative activity. At this point, it should be noted that post-growth annealing can have a strong impact on the defect landscape. It has recently been argued that annealing can promote diffusion of point defects;³⁵ however, there is no evidence that this diffusion is surface-selective. In addition, as mentioned above, annealing in the same temperature range leads to a reactivation of Mg. However, since in this case the investigated LED structure does not contain Mg for p-doping, a possible contribution of Mg reactivation to the observed change in CL intensity can be excluded. Thus, this experiment, which demonstrates that the CL intensity reduction can be achieved by either LEEBI or annealing, is a good indicator that non-radiative point defects at the sidewall are initially passivated by hydrogen. This is not entirely surprising, as several studies revealed that hydrogen is a good candidate for binding to point defects, such as gallium or nitrogen vacancies^26–31 or interstitials.³² 

Since the ICP etching conditions of the stronger chemically etched and the stronger physically etched LEDs are still quite similar, one could assume that hydrogenation during etching should, in principle, be possible in both cases, but it was only clearly observed for the former case. In this context, the work of Chen et al. is worth mentioning, who have shown that even bombardment with hydrogen ions can break the hydrogen passivation of point defects when GaN-based LEDs are hydrogenated with a hydrogen plasma in a PECVD.²⁷ This balance between passivation and depassivation during ICP etching should shift toward depassivation for the stronger physically etched LED, as it relies more heavily on etching the sidewall by sputtering. Consequently, hydrogenation of the stronger physically etched LEDs should lead to improved CL intensity near the sidewall if the point defects at the sidewall can be passivated by hydrogen bonds. Based on Figs. 6(a)–6(c), showing the monoCL intensity before and after hydrogenation of the triangular wedge-shaped LED, it is clear that the process had a positive effect, as the CL intensity near the sidewall was greatly improved. The topography was not altered during the process except from the removal of some surface contamination around the LED. At the same time, the CL intensity at the center of the LED is quite comparable, so the improvement in CL intensity is mainly due to a change in quantum efficiency and not a change in light extraction efficiency. This was confirmed by an increase in CL lifetime after hydrogenation, which is not shown here for the sake of brevity. After hydrogenation of the stronger physically etched LED, the CL intensity near the sidewall is similarly sensitive to focused electron beam irradiation as the stronger chemically etched LED. This leads to the conclusion that the achieved improvement in CL intensity, which is likely caused by the involvement of hydrogen based on the design of the experiment, can be reversed by LEEBI. Similarly, the CL intensity remains unaffected when focusing the electron beam far away from the sidewall, which is to be expected since hydrogen diffusion in n-GaN or unintentionally doped GaN is very weak, if not negligible, unless the surface is damaged.^36,37 Consequently, hydrogen in-diffusion should only occur from the sidewall and not too far into the LED.

Micro-LEDs etched under stronger chemically and stronger physically etching conditions were compared in terms of their CL behavior near the sidewall. In both cases, the CL intensity near the sidewall decreases due to an increase in non-radiative activity, but this decrease is partially compensated or even overcompensated by an increase in light extraction, which is especially prominent for the micro-LED with stronger chemical etching. When exciting the stronger chemically etched LED near the sidewall, LEEBI leads to a reduction in CL lifetime and maximum CL intensity with increasing irradiation time, which was attributed to the breaking of hydrogen bonds formed with point defects at the sidewall during the ICP etching. The same reduction in CL intensity near the sidewall is observed when the micro-LED is annealed for several minutes at temperatures of about 400–500 °C, leading to a thermally activated breaking of the hydrogen bonds, which further supports the idea that the sidewall is initially passivated by hydrogen. We suggest that the hydrogenation process is prevented for the stronger physically etched LED due to a strong contribution of sputtering in the etching process, but a post-etching hydrogenation step can recover the CL intensity near the sidewall. Generally, the annealed, stronger chemically etched LED behaves similarly to the stronger physically etched LED, while the same applies to the hydrogenated, stronger physically etched LED, and the stronger chemically etched LED. This illustrates the significance of the etching parameters for the sidewall characteristics, because although the etched sidewalls appeared similar in terms of macroscopic topography, they behaved very differently. Overall, our results point out strategies to improve the efficiency of micro-LEDs by passivation of point defects at the sidewall through optimization of the etching process. Moreover, they indicate that under suitable etching conditions a controlled incorporation of hydrogen at the sidewall is possible, so that an additional hydrogenation process is not required. This is advantageous because hydrogenation does not seem practical as a post-etching treatment for a conventional LED, as the hydrogen would diffuse into the p-GaN layer and compensate for Mg, limiting the area for electrical injection.

The support of the Braunschweig International Graduate School of Metrology B-IGSM is gratefully acknowledged. Furthermore, this project was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy—EXC-2123 QuantumFrontiers—390837967. The Oxford PECVD and the TESCAN SEM were funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—Project Nos. 455444208 and 224612727, respectively. The help of Matthias Hoormann in annealing the samples is greatly appreciated.

The authors have no conflicts to disclose.

Stefan Wolter: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Software (lead); Validation (lead); Visualization (equal); Writing – original draft (lead); Writing – review & editing (equal). Vladislav Agluschewitsch: Conceptualization (equal); Resources (equal); Software (supporting); Visualization (equal); Writing – review & editing (equal). Silke Wolter: Resources (equal); Visualization (equal); Writing – review & editing (equal). Frederik Lüßmann: Resources (equal); Visualization (equal); Writing – review & editing (equal). Christoph Margenfeld: Resources (equal); Visualization (equal); Writing – review & editing (equal). Georg Schöttler: Conceptualization (supporting); Formal analysis (supporting); Visualization (equal); Writing – review & editing (equal). Jana Hartmann: Funding acquisition (supporting); Project administration (equal); Supervision (supporting); Visualization (equal); Writing – review & editing (equal). Andreas Waag: Funding acquisition (lead); Project administration (equal); Supervision (lead); Visualization (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.