Hydrogen iodide (HI) neutral beam etching characteristics of InGaN and GaN for micro-LED fabrication

We investigated the etching characteristics of hydrogen iodide (HI) neutral beam etching (NBE) of GaN and InGaN and compared with Cl2 NBE. We showed the advantages of HI NBE versus Cl2 NBE, namely: higher InGaN etch rate, better surface smoothness, and significantly reduced etching residues. Moreover, HI NBE was suppressed of yellow luminescence compared with Cl2 plasma. InCl x is a product of Cl2 NBE. It does not evaporate and remains on the surface as a residue, resulting in a low InGaN etching rate. We found that HI NBE has a higher reactivity with In resulting in InGaN etch rates up to 6.3 nm min−1, and low activation energy for InGaN of approximately 0.015 eV, and a thinner reaction layer than Cl2 NBE due to high volatility of In-I compounds. HI NBE resulted in smoother etching surface with a root mean square average (rms) of 2.9 nm of HI NBE than Cl2 NBE (rms: 4.3 nm) with controlled etching residue. Moreover, the defect generation was suppressed in HI NBE compared to Cl2 plasma, as indicated by lower yellow luminescence intensity increase after etching. Therefore, HI NBE is potentially useful for high throughput fabrication of μLEDs.


Introduction
Extended reality technology connects the real world and the virtual world through the use of virtual reality and augmented reality (AR) devices [1]. These devices, especially AR displays, Nanotechnology Nanotechnology 34 (2023) 365302 (8pp) https://doi.org/10.1088/1361-6528/acd856 * Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. require a high resolution exceeding 4000 pixels per inch (ppi), high contrast of more than 100 000:1, low power consumption, and high luminance [2,3]. Micro-LEDs (μLEDs) are promising candidates for meeting these requirements. Here, μLEDs for an AR display must have 100 000-nit luminance and be small (chip size for each color: less than 3 μm, resolution: over 4000 ppi), and they must be able to display a full range of colors from red to blue [4][5][6]. An InGaN/GaN stacked multiple quantum well (MQW) structure is suitable as a matrix substrate for fabricating μLEDs [7,8]. The emission color can be varied from blue to red changing the indium content in the QWs from 15% to 35% [9]. Moreover, the emission intensity is able to increase with increasing a number of stack layers of MQW. Thus, the InGaN/GaN MQWs structure has a very flexible color adjustment and high emission intensity.
μLEDs are fabricated by plasma etching of semiconductor LED epitaxial wafers. However, sidewall surface defects occur as a result of the strong UV irradiation and ion bombardment from the plasma discharge. These defects reduce the internal quantum efficiency (IQE) of μLEDs as they trap holes and electrons and degrade the probability of recombination [10][11][12][13]. As a result, the external quantum efficiency (EQE) is decreased and its degradation becomes marked as the chip size decreases, which enlarges the sidewall area to volume ratio. On the other hand, neutral beam etching (NBE) is a defect-free etching that leads to no reduction in EQE for chip sizes larger than 6 μm [14]. In particular, μLED chips have been fabricated using Cl 2 NBE. The peak EQE current densities were very low even when the chip size was as small as 6 μm. This means that NBE is suitable due to being defect-free for μLEDs fabrication.
Etching 15%-indium-content InGaN/GaN MQW blueemitting (≈440 nm) μLEDs with chlorine gas has a low etching rate due to the InGaN layer (4.5 nm min −1 for GaN and 2 nm min −1 for In 0.15 GaN) [14] and forms indium residue [15]. This etching residue comes from low-volatility indium chloride remaining on the surface, and it causes recombination centers and disturbs the regrowth process [15]. Moreover, if red-light-emitting μLEDs are to be etched, the indium content ratio increases from 15% to 35%. It is predicted that InGaN etching would even be slower in this case because a 30%indium-content etching rate was 1 nm min −1 [16,17]. Here, the etching rate should be increased by enhancing the volatility of indium halide. One way of doing so is to produce compounds of high volatility. At 0.1 Pa, which is the process pressure, the boiling point of indium tri-chloride (InCl 3 ) compounds is 302°C [18]. On the other hand, if we use iodine as an etching gas, the etching compound is indium tri-iodide (InI 3 ), and its boiling point is 127°C [19]. In other words, the iodide is a higher vapor pressure indium compound compared with the chloride and thus should have higher volatility. Using HI gas plasma etching has already been carried out on the Ga-V and In-containing compounds by previous reports [20]. However, the effect of the InGaN and GaN etching by HI plasma has not been enough discussion yet. In this study, we investigated the etching characteristics of InGaN and GaN single layers in hydrogen iodide (HI) NBE as compared with Cl 2 NBE.

Experiment
For the NBE equipment, the inductively coupled plasma (ICP) is discharged and a high aspect ratio carbon aperture electrode is inserted under the plasma [21]. When accelerated ions pass through the carbon electrode, they are efficiently neutralized by charge exchange and irradiate the sample as a directional neutral beam while retaining their kinetic energy. By changing the bottom electrode's bias power, the neutral beam irradiation energy, i.e. the kinetic energy, can be controlled in the range from 5 eV to 1 keV [22]. The neutralization ratio is more than 95% [21]. Additionally, the substrate surface is in the shadow of the high-aspect-ratio carbon aperture [23,24]. As a result, the NB can suppress to generate of the etching defects because the high-aspect-ratio carbon aperture electrode prevents UV irradiation [25][26][27][28][29][30]. Thus, the NBE is a both defect-free and anisotropic dry etching, it has the potential to apply good performance for micro-/nano-scale pattern etching [31][32][33], and the reduction of IQE is suppressed [17,27,29,[34][35][36][37].
In our experiment, the HI etching gas was flowed at 50 sccm. The plasma and process chamber pressures were 1 and 0.1 Pa, respectively, and the source power was 1000 W at a 13.56 MHz of ICP with pulsed time modulation at 10 kHz and a 50% duty ratio. The gas flow rate, chamber pressure, and source power of the Cl 2 NBE were 40 sccm, 0.1 Pa, and 800 W, respectively. The HI and Cl 2 NBE were conducted in different chambers in order to prevent gas contamination and match the etching environment of the μLED to those of the previous reports [14]. The bias input power density, which is the applied area ratio of bias power, and the kinetic energy was at 43 mW cm −2 and 11 eV. The stage temperatures were set to 30°C, 80°C, or 130°C.
Unintentionally-doped GaN single-layer (thickness: ∼4 μm) samples were grown on c-plane sapphire substrates by metal-organic chemical vapor deposition and 100 nm thick InGaN layers were grown on the above GaN template substrates having a nominal In concentration of 10%. A 100 nm thick SiO 2 layer was then deposited on the sample surface by using plasma enhanced CVD and was processed into 10 μm squared patterns by photolithography to serve as a hard mask for the NBE.
The etch rate was measured by a surface profiler and the surface was observed with a scanning electron microscope (SEM) (S-5200 by Hitachi, Ltd.) observation. Before measuring the etching depth and the cross-sectional view, the hard mask was removed by 1 wt% HF wet etching. Surface composition and chemical states after NBE were evaluated by x-ray photoelectron spectroscopy (XPS). Ga3d and In3d peaks were fitted by a Gaussian function to distinguish the surface components, which turned out to be Ga-Cl or Ga-I, Ga-O, Ga-N, and Ga metal for Ga3d, and In-Cl or In-I, In-O, and In-N for In3d. The defects caused by the etching were evaluated by investigating the reduction in PL emission intensity. The surface roughness was measured by an atomic force microscope (AFM). AFM was carried out 3 μm × 3 μm for the lateral area and the z-axis was set at 100 nm to compare the surface roughness differences between each gasses NB. The etching depth for the PL measurement was 75 nm for the GaN single wafer when HI NBE was used. A 20 nm thick Al 2 O 3 layer was deposited on the surface by atomic layer deposition to prevent the influence of water and oxygen adsorption on PL intensity. To assess the effect of near-surface defects, the excitation laser wavelength was 266 nm, of which the penetration depth for GaN is approximately 50 nm. The sample stage was kept at room temperature.

Results and discussion
The etching rates of GaN and In 0.1 Ga 0.9 N were measured for each temperature. Figure 1 shows the temperature dependence of the etch rates for GaN and InGaN single layers. In the case of Cl 2 , the etch rate was slower for InGaN than for GaN; it was half that of GaN at 30°C and 80°C. On the other hand, the etching rate of the HI NBE was faster for the InGaN layer at all temperatures and it increased with increasing temperature.
On the basis of the etching rate as a function of substrate temperature, the activation energies of reactions between InGaN and GaN in case of Cl 2 and HI NBE were estimated using the Arrhenius plots shown in figure 2. As shown in table 1, the activation energy in the case of the HI NBE was one-sixth that of the Cl 2 NBE for InGaN. In addition, the activation energies for GaN and InGaN were almost the same in the case of the HI NBE, whereas the activation energy for InGaN was much higher than that of GaN in the case of the Cl 2 NBE. These results show that the HI NBE leads to higher reactivity and volatility compared with Cl 2 NBE. Namely, HI NBE is much more reactive to In (indium) than Cl 2 in InGaN etching.
To understand the surface reaction on InGaN, XPS was used to evaluate the surface component ratio in the case of HI and Cl 2 NBE. Figure 3 shows the Ga3d ((a) Cl 2 NBE and (b) HI NBE) and In3d5/2 ((c) Cl 2 NBE and (d) HI NBE) spectra with fitting results of the InGaN sample etched at 130°C. The surface reactions include both the reaction with the etching gas which forms Ga-Cl, Ga-I, In-Cl, and In-I bonds and the reaction with oxygen after etching. Since the sample was exposed to the atmosphere in the time between the etching and the XPS measurement, part of the NB reaction layer was replaced with oxygen; in particular, chlorine would have had an enhanced oxidation effect. Figure 4 shows the results for (a) Ga3d (Cl 2 NBE), (b) Ga3d (HI NBE), (c) I3d (Cl 2 NBE), and (d) I3d (HI NBE). For the Ga3d of Cl 2 and HI NBE, the component ratios of the surface reaction layer (Ga-Cl, Ga-I and Ga-O ratios) remained almost the same at all temperatures. According to the vapor pressure for ClGa 3 at −16°C and GaI 3 at 69°C, chloride and iodide desorption were promoted. These results show that the Ga had high reactivity with chlorine and iodine.
For the In3d of the Cl 2 NBE, however, the remaining Cl and O increased with increasing process temperature. InCl 3 molecules have difficulty desorbing from the surface due to the low vapor pressure. As a result, a thicker reaction layer formed. In contrast, the In3d of the HI NBE was decreased and the component ratio of the HI NBE decreased closer to the vapor pressure of InI 3 molecules (127°C). In this case, In-I compounds may have volatilized from the surface through the NB irradiation and heating. These results suggest that the HI NBE achieved a highly volatile reaction for In. Figure 5 shows (a) a schematic illustration of the SEM observation area, high-resolution cross-sectional SEM of InGaN layer etched by (b) Cl 2 and (c) HI NBE at 130°C. In the case of the Cl 2 NBE, the etching profile was rough on the   NBE showed tapered and progressed with sidewall etching, in contrast, that of HI NBE progressed without any sidewall etching and taper. Therefore, the smooth sidewall etching was achieved with HI NBE. Figure 6 shows the AFM surface profile of (a) initial InGaN, (b) Cl 2 and (c) HI NBE. AFM results in figure 6(b) of Cl 2 NBE reveal a large amount of etching residue. On the other hand, the surface of HI NBE has no observable etching residue. The root mean square (rms) of the InGaN surface roughness by AFM measurement was 4.3 nm of Cl 2 NBE, 2.9 nm of HI NBE, and 2.5 nm of the initial state (before etching). HI NBE can provide smooth InGaN surface etching with almost no change from the initial. Moreover, the height distribution was analyzed to investigate the differences in the rms as shown in figures 7(a) and (b). The peak around 0 nm of HI NBE was higher than Cl 2 NBE. It indicates that the surface of HI NBE is flatter than that of Cl 2 NBE. According to figure 7(b), Cl 2 NBE had stronger peaks of height distribution, especially around 60 nm. As a result, the height distribution of around 60 nm was the etching residue due of almost the same as the etching depth, on the other hand, HI NBE has almost no etching residue.
The etching mechanism is illustrated in figure 8. In the case of InGaN, indium compounds formed by Cl 2 NBE generate etching residue and the low volatility and low activation energy results in a low etching rate. On the other hand, the HI NBE produces a smooth etching interface and the etching rate is high because of the high volatility and low activation energy.
Surface defects were further investigated by PL. In figure 9(a), the PL spectra are normalized with the peak value of the band-edge (BE) emission for each sample and are displayed at the same magnification. The peak values for yellow luminescence (YL) slightly increased from 1.33 (initial) to 2.15 (Cl 2 NBE) and 2.46 (HI NBE). The R-value describes the difference in ratio of the PL intensity of YL to that of the BE emission between the etched and as-grown samples     and NBE [37] which was investigated for defect generation by dry etching for GaN HEMT device as shown in figure 9(b). After etching, the R-value increased to 66% (Cl 2 NBE) and 116% (HI NBE) due to the defect concentration increment from the initial. On the other hand, R-value of NBE was significantly less than that of Cl 2 plasma etching (210%) as compared with the previous report [37]. Previous studies on using Cl 2 NBE for fabricating μLED reported that the surfaces recovered after GaN regrowth [15] and the EQE was stable at low current density (1 A cm −2 ) [14]. Thus, our results prospected that the etching interface of HI NBE can recover as well or better than that of Cl 2 NBE because it does not have any etching residue.

Conclusion
We investigated the etching characteristics of HI NBE and compared with Cl 2 NBE for InGaN and GaN. The etch rate of the InGaN layer by the HI NBE was faster than that of the GaN at all temperatures and increased with increasing temperature. For the Arrhenius plot, the activation energy in the case of the HI NBE was one-sixth that of the Cl 2 NBE for InGaN. HI NBE was more reactive to In (indium) than Cl 2 in InGaN etching. It was shown by XPS measurement also that a thinner reaction layer than Cl 2 NBE thanks to the high volatility of In-I. HI NBE provided a smoother etching surface than Cl 2 NBE because the high volatility to the Incompounds did not generate the etching residue. Moreover, the HI NBE suppressed the defect generation during the etching of the GaN surface as compared with the Cl 2 plasma. Therefore, the HI NBE has the potential to fabricate high throughput and high-quality μLED devices because of achieved a faster etching rate with a smooth etching surface than Cl 2 NBE and suppressed the etching defect than conventional plasma.

Acknowledgments
Authors would like to thank Innolux, Hsinchu, Taiwan for assistances and supports.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).

Funding
This research was supported by Adaptable and Seamless Technology transfer Program through Target-driven R&D (A-STEP) from Japan Science and Technology Agency (JST) Grant Number JPMJTR20R9. This work was supported in part by the National Science and Technology Council, Taiwan, under Grant MOST 111-2221-EA49-181 and MOST 111-2221-EA49-013.