I. INTRODUCTION

Gallium nitride (GaN)-based high electron mobility transistors (HEMTs) are regarded as promising candidates for next-generation power and high-frequency devices due to their wide bandgap, high breakdown voltage, high current density, and excellent thermal stability ^{[1- 4]}. In AlGaN/GaN heterostructures, a two-dimensional electron gas (2DEG) is formed at the interface through strong spontaneous and piezoelectric polarization effects ^{[5- 7]}, which enables outstanding device performance in power conversion, RF communication, and optoelectronics ^{[8- 10]}.

A crucial step in fabricating AlGaN/GaN HEMTs is the selective etching of GaN over AlGaN. Because the AlGaN barrier is typically only 15-30 nm thick, even slight overetching can reduce the 2DEG density and degrade channel conductivity, while insufficient etching may cause leakage paths and device instability ^{[11- 13]}. This requirement is especially critical in normally-off p-GaN gate HEMTs, where the p-GaN layer in the access region must be removed without damaging the thin AlGaN barrier. Any unintentional etching of the underlying AlGaN directly decreases the 2DEG density and results in serious device performance degradation. For this reason, achieving high GaN-over-AlGaN selectivity and minimizing etch-induced damage on the exposed AlGaN surface are essential for reliable device operation. Achieving highly selective GaN etching with minimal plasma-induced damage is therefore essential for reliable device fabrication ^{[14, 15]}.

Various inductively coupled plasma reactive ion etching (ICP-RIE) chemistries have been investigated for selective GaN/AlGaN etching. Among them, BCl₃/SF₆ mixtures have attracted attention since fluorine species form stable Al-F passivation layers that suppress AlGaN etching, with reported selectivities up to ~40:1 ^{[16- 18]}. Oxygen-containing plasmas such as O₂/BCl₃ have further improved selectivity by forming Al-O surface layers, achieving values of ~70:1 ^{[19, 20]}. More recently, atomic layer etching (ALE) has been explored to achieve atomic-scale precision ^{[21- 23]}. However, ALE generally suffers from very low etch rates, such that an etching of ~100 nm thick GaN may require several hours. This severely limits its productivity and practicality for large-scale device fabrication.

While previous works have mainly optimized plasma parameters such as power, pressure, and gas ratio, substrate temperature has received less attention despite its strong influence on etch selectivity and post-etch surface states. Furthermore, few studies have systematically correlated temperature effects with both surface chemistry and electrical characteristics.

In this work, we address these issues by employing a customized ICP-RIE system designed for precise substrate temperature control using a circulating oil-bath system. The system also utilizes a 27.12 MHz ICP source and a helical resonance structure. Compared to the conventional 13.56 MHz, the higher driving frequency of 27.12 MHz increases electron oscillation and electron density, while the helical design enhances plasma coupling and energy transfer stability. Together, these features significantly elevate electron density, increase gas dissociation efficiency, and enable a high-density radical environment while minimizing physical damage. Using this platform, we investigated the effects of gas composition and substrate temperature on GaN/AlGaN etching behavior. Surface and electrical properties were comprehensively analyzed through X-ray photoelectron spectroscopy (XPS), atomic force microscopy (AFM), and Schottky-diode leakage measurements. A record-high GaN-over-AlGaN selectivity of 78:1 was achieved, while simultaneously demonstrating improved surface smoothness and lower leakage current at an optimized temperature of ~30 °C.

Fig. 1. Schematic illustration of the custom ICP-RIE system used in this study, equipped with substrate temperature control.

II. EXPERIMENTS

Epitaxial heterostructures with a conventional AlGaN/GaN HEMT configuration were employed in this study. Two types of samples were prepared: (i) AlN/AlGaN buffer / GaN channel (~1.2 μm) / Al_0.25Ga_0.75N barrier (20 nm) / GaN cap (240 nm) and (ii) AlN/AlGaN buffer / GaN channel (~1.2 μm) / Al_0.25Ga_0.75N barrier (20 nm) without the GaN cap. The use of these two structures enabled a direct comparison of GaN and AlGaN etching characteristics under identical plasma conditions.

Selective etching was performed in a custom ICP-RIE system (CN1, CS-AJOU-2220). The ICP source was operated at 27.12 MHz and the RF bias at 12.56 MHz. Chamber pressure and ICP/bias powers were fixed at 40 mTorr and 1000/25 W, respectively, unless otherwise specified. Substrate temperature was precisely regulated by a constant-temperature bath (Labcommerce, COB10) that circulated a perfluorinated coolant (Galden, HT-230). The coolant maintained the temperature of the heat exchanger beneath the substrate stage, and helium backside gas was used to ensure efficient thermal transfer, thereby stabilizing the wafer surface temperature within ±3 °C throughout processing. This configuration enabled systematic exploration of the substrate temperature effect beyond conventional chiller-based control.

The plasma chemistry consisted of a BCl₃/SF₆ mixture, with the SF₆ fraction varied from 10% to 35% at a constant total flow of 90 sccm. After determining the gas ratio yielding the highest GaN/AlGaN selectivity, the substrate temperature was varied between 17.5 °C and 80 °C to evaluate temperature-dependent etch characteristics. The etch depth was measured using a KLA-Tencor Alpha-Step D-500 profilometer. The etching times were fixed at 60 s for GaN and 250 s for AlGaN, which resulted in 90-210 nm of GaN being removed and 5-20 nm of AlGaN being etched, depending on the condition.

Post-etch surface characterization was performed by AFM (Park Systems NX10) to evaluate root-mean-square (RMS) roughness, and by XPS (Thermo Fisher Scientific K-Alpha Plus) to analyze chemical residues such as Al-F and Al-O. To assess the electrical impact of etching, Ni/Au Schottky electrodes were deposited by e-beam evaporation, and current-voltage (I-V) characteristics were measured using a Keithley 4200A parameter analyzer. Etch selectivity was defined as the ratio of the GaN etch rate to the AlGaN etch rate, with each condition confirmed through repeated experiments. To ensure reliability and reproducibility, measurements were performed on at least three devices under identical conditions to verify the consistency of the results. The data are presented as the arithmetic mean of the measured values, with error bars representing the standard deviation to indicate the error range.

III. RESULTS AND DISCUSSION

1. SF₆ Concentration

The effect of SF₆ concentration in the BCl₃/SF₆ plasma was investigated at a fixed substrate temperature of 17.5 °C. As shown in Fig. 2(a), the GaN etch rate increased with the addition of SF₆, reaching ~3.2 nm/s at 20%. This enhancement is attributed not to a simple increase in chlorine radicals but to modifications in plasma chemistry. At low SF₆ fractions, F atoms react with B species to form volatile BF₃, thereby reducing B-rich residues that can inhibit etching. In addition, the presence of SF₆ promotes the formation of mixed BCl_xF_y radicals and enhances BCl₃ dissociation efficiency, leading to higher reactivity of Cl species. These effects facilitate the efficient formation and desorption of volatile GaCl₃, resulting in a higher GaN etch rate. When the SF₆ fraction exceeded 20%, however, involatile GaF_x accumulated on the surface, suppressing GaN etching and decreasing the rate ^{[16, 24- 26]}.

Fig. 2. Etching characteristics at 17.5 °C as a function of SF6/BCl3 ratio: (a) GaN etch rate, (b) AlGaN etch rate, and (c) GaN-over-AlGaN selectivity.

In contrast, the AlGaN etch rate decreased sharply with increasing SF₆ up to 20% and continued to decrease more gradually at higher fractions, as shown in Fig. 2(b). This suppression is caused by the formation of stable, non-volatile AlF₃ passivation layers, which act as an effective etch stop. As a result of these contrasting trends, the GaN-over-AlGaN selectivity shown in Fig. 2(c) reached a maximum of ~78:1 at 30% SF₆. Notably, this high selectivity is enabled by the 27.12 MHz high-frequency source and the helical resonance structure, which generate a high-density radical environment. This environment promotes the formation of volatile Ga-Cl byproducts on GaN and strong Al-F passivation on AlGaN. Under this condition, GaN etching remained efficient, while AlGaN removal was strongly inhibited by fluorine-based passivation. These findings confirm that SF₆ plays a dual role: it enhances GaN etching at low concentrations by removing B-related inhibiting species and activating chlorine chemistry, while simultaneously suppressing AlGaN etching through Al-F passivation.

2. Substrate Temperature

The effect of substrate temperature was investigated under the optimized gas ratio of 30% SF₆ in BCl₃/SF₆ plasma. As shown in Fig. 3(a), the GaN etch rate exhibited little variation with temperature, reflecting its volatile GaCl₃ pathway that is relatively insensitive to thermal conditions ^[27]. The slight increase at lower temperatures is attributed to the longer radical residence time, which enhances surface adsorption and Ga-Cl formation, with facilitating efficient ion-assisted removal. By contrast, the AlGaN etch rate increased significantly with temperature (Fig. 3(b)), owing to the thermal instability of the AlF₃ passivation layer. This contrast is summarized in Fig. 3(c): the GaN-over-AlGaN selectivity was maximized at low temperatures, but it decreased sharply once the temperature exceeded 40 °C. These results highlight that strict temperature control is essential for maintaining high selectivity, and that the optimal range lies around 30 °C.

Fig. 3. Etching characteristics as a function of substrate temperature at an optimized SF6/BCl3 ratio of 30%: (a) GaN etch rate, (b) AlGaN etch rate, and (c) GaN-over-AlGaN selectivity.

XPS spectra provide further evidence of these temperature-dependent mechanisms (Fig. 4). The Al³⁺ peak associated with Al₂O₃ appeared at ~74 eV, while that of AlF₃ was located at ~76 eV. Strong Al-F signals were observed at 17.5 °C and 30 °C, but at 50 °C and 80 °C the Al-F peak intensity decreased and shifted toward Al-O-F bonding, indicating breakdown of fluorine passivation. The trend of Al-F intensity was consistent with the selectivity behavior in Fig. 3(c), demonstrating that Al-F stability is the key factor for achieving high selectivity. At lower temperature, the increase in Al-O bonding reflected greater accumulation of AlCl₃ by-products, which oxidized into AlO_x upon air exposure. This indicates unintended surface oxidation at very low temperature. Fig. 4(e) further quantifies the chemical state evolution with temperature. The area intensities of Al-F and Al-O components both increased at lower temperatures, reflecting greater retention of reaction by-products. Among them, the Al-O component includes contributions from AlCl₃ residues that oxidized into Al₂O₃ upon air exposure. This explains why, although low temperatures enhance selectivity, excessive by-product accumulation at 17.5 °C degrades surface quality.

Fig. 4. XPS spectra of Al 2p core level for etched AlGaN surfaces at (a) 17.5 °C, (b) 30 °C, (c) 50 °C, and (d) 80 °C, showing contributions from Al–F (~76 eV) and Al–O (~74 eV) bonding. (e) Comparison of peak area intensities of Al–F and Al–O as a function of substrate temperature.

AFM images (3 μm × 3 μm scan area) confirmed the temperature dependence of surface morphology (Figs. 5(a)-(d)). The RMS roughness increased from ~0.9 nm at 17.5 °C to ~3.8 nm at 80 °C. Both 17.5 °C and 30 °C produced relatively smooth surfaces, but the 30 °C sample exhibited the most uniform morphology. The slightly higher roughness at 17.5 °C is attributed to AlCl₃ residues that oxidized into AlO_x upon exposure to air. At temperatures above 30 °C, roughness increased drastically due to the collapse of Al-F passivation (Fig. 5(e)), exposing Al/Ga sites to direct ion bombardment and chemical attack, which induced severe plasma damage.

Fig. 5. AFM images (3 μm × 3 μm scan area) of etched AlGaN surfaces at substrate temperatures of (a) 17.5 °C, (b) 30 °C, (c) 50 °C, and (d) 80 °C. The color scale bar shown to the right of (a) applies to both (a) and (b), and the scale bar shown to the right of (c) applies to both (c) and (d). (e) RMS roughness values as a function of temperature, indicating a significant increase in surface roughness at elevated temperatures.

The electrical properties measured by Schottky-diode structures provided further insight (Fig. 6). Two Ni/Au contacts (0.1 mm × 0.1 mm) were fabricated to form a back-to-back Schottky configuration, so the measured current corresponds to the reverse-biased junction and is limited by its reverse saturation current. As a result, the I-V curves exhibited a plateau beyond ~5 V, reflecting the surface-limited leakage path rather than avalanche breakdown. The absolute leakage current level varied strongly with etching temperature: ~5.3×10^-4 A/mm at 17.5 °C, minimized to ~4.0×10^-4 A/mm at 30 °C, and increasing markedly to ~1.9×10^-3 A/mm at 80 °C. These results indicate that higher leakage originates from increased trap-assisted conduction along damaged or oxidized surfaces. The lowest leakage at 30 °C confirms that this intermediate temperature balances stable fluorine passivation and limited AlCl₃ oxidation, thereby providing the most stable surface state.

Fig. 6. (a) Schematic of the measurement structure for leakage current characterization using Ni/Au Schottky contacts on etched AlGaN surfaces. (b) Measured I-VA characteristics of the samples etched at different temperatures, showing that leakage current increases with higher substrate temperature, while the lowest leakage is observed at 30 °C.

IV. CONCLUSIONS

In this work, selective etching of GaN over AlGaN was systematically investigated using a customized BCl₃/SF₆ ICP-RIE system equipped with precise oil-bath temperature control and operated at a high ICP frequency of 27.12 MHz using a helical resonance ICP source. The study demonstrated that both gas composition and substrate temperature critically determine etching behavior, surface chemistry, and electrical integrity. By optimizing the BCl₃/SF₆ ratio to 30% SF₆, a maximum GaN-over-AlGaN selectivity of 78:1 was achieved. This record high selectivity was primarily driven by the 27.12 MHz helical resonance ICP source, which generated a high-density radical environment. This environment promoted volatile Ga-Cl formation on GaN and strong Al-F passivation on AlGaN, thereby ensuring a fast GaN etch rate and a suppressed AlGaN etch rate with minimal physical damage. Additionally, precise temperature control preserved the passivation layer's thermal stability. Temperature-dependent experiments revealed that although very low temperatures enhanced selectivity, they also promoted the retention and oxidation of AlCl₃ by-products. Conversely, elevated temperatures (> 40 °C) destabilized Al-F passivation, leading to degraded selectivity, increased surface roughness, and higher leakage current. An intermediate temperature of ~30 °C provided the best balance, simultaneously yielding high selectivity, smooth morphology, and minimized electrical degradation. These results establish that selective, high-throughput, and low-damage GaN etching can be realized through the combined control of substrate temperature and plasma frequency. Compared with previous reports, the present process not only achieved record-high selectivity but also demonstrated superior surface and electrical characteristics, offering both scientific insights and practical advantages for scalable GaN/AlGaN HEMT fabrication.